Technological Developments in Networking,

Education and Automation

Khaled Elleithy · Tarek Sobh · Magued Iskander ·
Vikram Kapila · Mohammad A. Karim ·
Ausif Mahmood
Editors

Technological Developments
in Networking, Education
and Automation

123
Editors
Prof. Khaled Elleithy Dr. Tarek Sobh
University of Bridgeport University of Bridgeport
School of Engineering School of Engineering
University Avenue 221 University Avenue 221
06604 Bridgeport Connecticut 06604 Bridgeport Connecticut
USA USA
[email protected] [email protected]

Magued Iskander Dr. Vikram Kapila

Polytechnic University Polytechnic University
Dept. Civil & Environmental Dept. Civil & Environmental
Engineering Engineering
Jay Street 333 Jay Street 333
11201 Brooklyn New York 11201 Brooklyn New York
USA USA
[email protected] [email protected]

Dr. Mohammad A. Karim Ausif Mahmood

Old Dominion University University of Bridgeport
Monarch Way 4111 School of Engineering
23508 Norfolk Virginia University Avenue 221
USA 06604 Bridgeport Connecticut
[email protected] USA
[email protected]

ISBN 978-90-481-9150-5 e-ISBN 978-90-481-9151-2

DOI 10.1007/978-90-481-9151-2
Springer Dordrecht Heidelberg London New York

Library of Congress Control Number: 2010927951

c Springer Science+Business Media B.V. 2010

No part of this work may be reproduced, stored in a retrieval system, or transmitted in any form or by
any means, electronic, mechanical, photocopying, microfilming, recording or otherwise, without written
permission from the Publisher, with the exception of any material supplied specifically for the purpose
of being entered and executed on a computer system, for exclusive use by the purchaser of the work.

Printed on acid-free paper

Springer is part of Springer Science+Business Media (www.springer.com)

 To our families
Their continuing support makes all our endeavors possible
And
To the memory of Professor Natalia Romalis
Her spirit will always be with us
 Preface

This book includes the proceedings of the 2009 International Conferences on

Telecommunications and Networking (TENE 09), Engineering Education, Instructional
Technology, Assessment, and E-learning (EIAE 09), and Industrial Electronics, Technology &
Automation (IETA 09). TENE 09, IETA 09 and EIAE 09 are part of the International Joint
Conferences on Computer, Information, and Systems Sciences, and Engineering (CISSE 09).
The proceedings are a set of rigorously reviewed world-class manuscripts presenting the state of
international practice in Industrial Electronics, Automation, Telecommunications, Networking,
E-Learning, Instruction Technology, Assessment and Engineering Education.

CISEE 09 was a high-caliber research conference that was conducted online. CISSE 09 received
432 paper submissions and the final program included 220 accepted papers from more than 80
countries, representing the six continents. Each paper received at least two reviews, and authors
were required to address review comments prior to presentation and publication.

Conducting CISSE 2009 online presented a number of unique advantages, as follows:

• All communications between the authors, reviewers, and conference organizing committee
were done on line, which permitted a short six week period from the paper submission
deadline to the beginning of the conference.
• PowerPoint presentations, final paper manuscripts were available to registrants for three
weeks prior to the start of the conference
• The conference platform allowed live presentations by several presenters from different
locations, with the audio, video and PowerPoint presentations transmitted to attendees
throughout the internet, even on dial up connections. Attendees were able to ask both audio
and written questions in a chat room format, and presenters could mark up their slides as
they deem fit
• The live audio presentations were also recorded and distributed to participants along with
the power points presentations and paper manuscripts within the conference DVD.

The conference organizers and we are confident that you will find the papers included in this
book interesting and useful.

Khaled Elleithy
Tarek Sobh
Magued Iskander
Vikram Kapila
Mohammad Karim
Ausif Mahmood

Bridgeport, Connecticut
January 2010

vii
 Acknowledgements

The 2009 International Conferences on Telecommunications and Networking (TENE 09),

Engineering Education, Instructional Technology, Assessment, and E-learning (EIAE 09), and
Industrial Electronics, Technology & Automation (IETA 09) and the resulting proceedings could
not have been organized without the assistance of a large number of individuals. TENE, EIAE,
and IETA are part of the International Joint Conferences on Computer, Information, and Systems
Sciences, and Engineering (CISSE). CISSE was founded by Professors Tarek Sobh and Khaled
Elleithy in 2005, and they set up mechanisms that put it into action. Andrew Rosca wrote the
software that allowed conference management and interaction between the authors and reviewers
online. Mr. Tudor Rosca managed the online conference presentation system and was
instrumental in ensuring that the event met the highest professional standards. We also want to
acknowledge the roles played by Sarosh Patel and Ms. Susan Kristie, our technical and
administrative support team.

The technical co-sponsorship provided by the Institute of Electrical and Electronics Engineers
(IEEE) and the University of Bridgeport is gratefully appreciated. We would like to express our
thanks to Prof. Toshio Fukuda, Chair of the International Advisory Committee and the members
of the TENE, EIAE and IETA Technical Program Committees.

The excellent contributions of the authors made these world-class conference proceedings
possible. Each paper received two to four reviews. The reviewers worked tirelessly under a tight
schedule and their important work is gratefully appreciated. A complete list of reviewers is
provided in the book.

Khaled Elleithy
Tarek Sobh
Magued Iskander
Vikram Kapila
Mohammad Karim
Ausif Mahmood

Bridgeport, Connecticut
January 2010

ix
 Contents

Reviewers List .............................................................................................................................................. xix

1. Education in the Clouds: How Colleges and Universities are Leveraging Cloud Computing................ 1
David C. Wyld and Rusty L. Juban

2. Weak Orderings of Knowledge States .................................................................................................... 7

Sylvia Encheva and Sharil Tumin

3. Problem Based Learning: Obtaining Enzyme Kinetics Parameters Integrating Linear Algebra,
Computer Programming and Biochemistry Curriculum ....................................................................... 13
A. Regalado-Méndez et al.

4. Adaptive Assessments using Open Standards ...................................................................................... 19

Héctor G. Barbosa León et al.

5. Online Academic Advising Support ..................................................................................................... 25

Fawzi Albalooshi and Safwan Shatnawi

6. Using Computer Game to Share knowledge ............................................................................... 31

Mazeyanti M Ariffin et al.

7. Following the Paper Trail: Measuring the Economic and Environmental Impact of Digital
Content Delivery .................................................................................................................................. 37
Vanessa K. Cox et al.

8. Context Driven Personalized e-Learning Environment Provided as a State Service ............................ 43

Deniss Kumlander

9. Exploring Teachers’ Beliefs about Digital Citizenship and Responsibility.......................................... 49

Harrison Hao Yang and Pinde Chen

10. Faculty Driven Assessment of Critical Thinking: National Dissemination of the CAT Instrument..... 55
Barry Stein et al.

11. Enhancing the Instruction of Introductory Electric Circuit Courses using MATLAB ......................... 59
Thomas Yang et al.

12. Istopolis: Design Issues and Evaluation’ Results of an Integrated Web-Based

Educational Application ....................................................................................................................... 65
C. T. Panagiotakopoulos et al.

xi
xii CONTENTS

13. TEAC2H-RI: Educational Robotic Platform for Improving Teaching-Learning Processes

of Technology in Developing Countries............................................................................................... 71
Juan J. González et al.

14. Selecting Patterns for the Evaluation of Systems ................................................................................. 77

Dmitry Zub and Erki Eessaar

15. Effectiveness of Collaborative Learning in Teaching Information Systems ........................................ 83

Ioana Tuugalei Chan Mow

16. Implementation of a Curriculum Management System at a Faculty of Medicine – Lessons

Learned from a Pilot Project ................................................................................................................. 89
Aarti Paul et al.

17. Use of Ontology for Reusing Web Repositories for eLearning ............................................................ 97
Punam Bedi et al.

18. Actual and Future Trends of the E-education and Its Costs Specification in the Czech Republic ..... 103
M. Hedvicakova and P. Maresova

19. Modern Improvements in the Digital Logic Laboratory..................................................................... 109

K. M. Noga and M. Radwanski

20. e-Cheating and Calculator-Technology .............................................................................................. 115

Zeenath Reza Khan

21. Perspectives on Critical Thinking through Online Discussion Forums in Engineering

Mathematics ....................................................................................................................................... 121
S. M. Jacob and H. K. Sam

22. Digital System Description Knowledge Assessment .......................................................................... 127

K. Jelemenská et al.

23. Wikibooks as Tools for Promoting Constructivist Learning in Higher Education: Findings
from a Case Study .............................................................................................................................. 133
Ilias Karasavvidis

24. The Impact of the Context of an Information System on Strategic Information

Quality - The Institute of Technology Sector in Ireland ..................................................................... 139
Laurence P. Elwood and Owen V. Foley

25. Can Mobile Computing Help Reverse the Decline in Informatics Studies? ....................................... 145
D.Vitkute-Adzgauskiene and V. V. Fomin

26. Towards an Academic Social Network for Bologna Process ............................................................. 151
João C. Ferreira and Porfírio P. Filipe

27. Supporting Knowledge Discovery in an eLearning Environment Having Social Components ......... 157
Paola Monachesi et al.
 CONTENTS xiii

28. Computer Numerical Control, Mass Customization and Architectural Education

in an Emergent Economy ................................................................................................................... 163
Neander F. Silva et al.

29. Application of Project Based Study in the Learning Process: A Case Study ..................................... 169
Ashfaque A Chowdhury et al.

30. Interdisciplinary Automation and Control in a Programmable Logic Controller (PLC)

Laboratory .......................................................................................................................................... 175
Jack Toporovsky et al.

31. Stereo Spectral Imaging System for Plant Health Characterization ................................................... 181
S. C. Yoon and C. N. Thai

32. Transition Network for Automation Modeling: A Case Study ........................................................... 187
Yuval Cohen et al.

33. Computational Modeling and Simulation of a New Energy-Saving Sealing Unit

in Industrial Applications ................................................................................................................... 193
Zheng (Jeremy) Li

34. Design of An Intelligent Solar Tracking System Using PIC18F452 Micro Controller ...................... 197
Mostefa Ghassoul

35. Adaptive Two Layers Neural Network Frequency Controller for Isolated Thermal
Power System ..................................................................................................................................... 203
Ognjen Kuljaca et al.

36. Optimal Controller Comparison Using Pareto Fronts ........................................................................ 209

David Moore

37. Interaction with Objects in Virtual Environments .............................................................................. 215

Jesús A. Verduzco-Ramírez et al.

38. One Approach for Training of Recurrent Neural Network Model of IIR Digital Filter ..................... 219
Stefanova S. A.

39. Classification of the Power Transformers using Dissolved Gas Analysis .......................................... 225
T. C. B. N. Assunção et al.

40. Modelling Control of Pore Number and Radii Distribution in Single-Cell Electroporation .............. 231
Sadhana Talele and Paul Gaynor

41. Single Allocation P-Hub Median Problem to Monitor Land Borders by Using
Unmanned Aircraft ............................................................................................................................ 237
Yasemin Akkuş and İnci Sarıçiçek
xiv CONTENTS

42. LPT to RS232 communication converter ........................................................................................... 243

Grofu Florin and Cercel Constantin

43. Simulation model of Coremaking Plant ............................................................................................. 247

Stefan Belev and Miroslav Penchev

44. Depth Calculation using Computer Vision and Sift .......................................................................... 253
Edson CAVALCANTI Neto et al.

45. Dynamic of New Technologies in Switching Power Devices IGBTs ............................................... 257
Christian Ayala Esquivel et al.

46. Anisotropic Percolating Pathways in the Thin Films of Polymeric PEDT/PSS Complex
and their Relation to the Electrical Conductivity as Revealed by the Mesoscale Simulation............. 263
Toomas Kaevand et al.

47. Predicting Trading Signals of Sri Lankan Stock Market Using Genetic Algorithms
and Neural Networks .......................................................................................................................... 269
M. M. K. Dassanayake and Chandima Tilakarathne

48. An Automatic Measurement Algorithm for the Diameters of Carbon Nanotubes by

Using Image Processing ................................................................................................................... 275
Yoichiro Iwasaki et al.

49. Object Detection in Foveated Images ................................................................................................. 281

Abulfazl Yavari and H.R. Pourreza

50. Digital Semi-Controlled Rectifier using a Microcontroller ................................................................ 287

Emmanuel E. Rached et al.

51. Generalized Minimum Variance Controller with Dynamic Pole Assignment to Improve
Performance in Industrial Applications .............................................................................................. 291
M.A. Paz Ramos et al.

52. Automated Detection of Sunspots and Sunspot Groups in Full-Disk Solar Images ........................... 297
Sarab F. Majed et al.

53. Stochastic Model Based Approach for Biometric Identification ....................................................... 303
Mofakharul Islam et al.

54. A New Stochastic Model Based Approach for Object Identification and Segmentation in
Textured Color Image......................................................................................................................... 309
Mofakharul Islam and Paul A. Watters

55. An Unsupervised Stochastic Model for Detection and Identification of Objects in Textured
Color Images Using Segmentation Technique ................................................................................... 315
Mofakharul Islam and Paul A. Watters
 CONTENTS xv

56. Automated Classification of Sunspot Groups with Support Vector Machines ................................... 321
Mehmmood A. Abd et al.

57. A Novel Ballistics Imaging System for Firearm Identification .......................................................... 327
Dongguang Li

58. Color Seal Segmentation and Identification ....................................................................................... 333

Dongguang Li

59. Efficient SOPC-based Multicore System Design using NOC............................................................ 339

Vignesh Veerapandian and Xingguo Xiong

60. Architecture and Initial Performance Indicators of New OPCUA Automation Protocol
Stack Java Implementation ................................................................................................................. 345
Mikko Salmenperä and Mikko Salonen

61. Automatic Segmentation of Cardiac Cavity Images Using Collinear and Triangle Equation ............ 351
Riyanto Sigit et al.

62. Both Side More-Output Transmission with Harmonic Gear.............................................................. 357

Jozef Haľko

63. Intelligent Plasma Monitoring Systems Based on Optical Methods and Chromatic Modulation....... 361
Dimitrios Tomtsis

64. A Neoteric Chaotic & Adaptive Watermarking Scheme for Colored Images ................................... 367
Navita Srivastava and Achintya Singhal

65. Rotary Transducer .............................................................................................................................. 373

Ilie Borcosi et al.

66. Regression Analysis about Humidity Elimination from Diesel Fuel Via Bioorganic
Compounds to Increase Antifouling Action ....................................................................................... 377
C. G. Tsanaktsidis et al.

67. Optimal Threshold for Locating Targets Within a Surveillance Region Using a Binary
Sensor Network .................................................................................................................................. 381
Zhenxing Luo and Thomas C. Jannett

68. A Proposal of a Nonlinear Observer Applied to the Heat Exchange Phenomena in a

Double Pipe Heat Exchanger .............................................................................................................. 387
J. F. Briones et al.

69. Vibration Signal Processing by Deconvolution Method .................................................................... 393

Popescu Ion Marian and Runceanu Adrian

70. An IT Change Management Tool to Optimize Scheduling and Planning .......................................... 399
Denílson Cursino Oliveira and Raimir Holanda Filho
xvi CONTENTS

71. A Low-Power Content-Addressable Memory (CAM) using Pipelined Search Scheme .................... 405
Yibo Song et al.

72. Design and Optimization of Piezoelectric Dual-Mode Micro-Mirror ................................................ 411

Jichao Zhong et al.

73. Device for Measurement of Clamped Joints Friction Torque ..................................................................417

Eva Batešková et al.

74. Self-Optimization for Dynamic Scheduling in Manufacturing Systems ............................................ 421

Ana Madureira and Ivo Pereira

75. A Decomposition Method in Modeling Queuing Systems ................................................................. 427

Eimutis Valakevicius et al.

76. Modular Design and Structure for a Mobile Sensory Platform ......................................................... 433
Ayssam Elkady et al.

77. wradvs: A Didactic Server for IPv6 Stateless Autoconfiguration ..................................................... 443
Gabriel Abadi et al.

78. Securing P2P Wireless Communications by Deploying Honeytokens in a Cooperative

Maritime Network .............................................................................................................................. 449
Jonathan White et al.

79. Integrating Information Security into Quality Management Systems ................................................ 455
Margareth Stoll

80. Evaluation of Asynchronous Event Mechanisms for Browser-based Real-time

Communication Integration ................................................................................................................ 461
Keith Griffin and Colin Flanagan

81. An Ontology-based Intrusion Detection for RFID Systems ............................................................... 467

M. Esposito and G. Della Vecchia

82. A New Algorithm and its Implementations for RC Filters ................................................................. 473
Guoliang Zeng

83. A Computationally Dynamic Resource Management for Multicarrier Communications

in Downlink for OFDMA Systems ..................................................................................................... 479
Kabaou Mohamed Ouwais et al.

84. A General Method for Synthesis of Families of Orthogonal Complementary Codes ........................ 487
B. Y. Bedzhev and M. P. Iliev

85. E-Speed Start- A Window Based Additive Increase, Multiplicative Decrease Network
Congestion Control Technique ........................................................................................................... 493
I. K. Oyeyinka et al.
 CONTENTS xvii

86. Common-Wavelengths Design Scheme for Efficient Management of Spare Resources in

WDM Networks ................................................................................................................................. 499
Ardian Greca

87. ICMP Covert Channel Resiliency ...................................................................................................... 503

Kristian Stokes et al.

88. Chip-to-chip Free-space Optical Interconnection Using Liquid-crystal-over-silicon Spatial

light modulator ................................................................................................................................... 507
Sadia Sultana et al.

89. GA based Neural Network Retraining using Adaptive Genetic Operations ....................................... 511
C. I. Bauer et al.

90. Decision Support for IP Uplink and Routing Choice ......................................................................... 517
Peng Su and Zhengping Wu

91. Differential Cryptanalysis of Blow-CAST-Fish ................................................................................. 523

Xiaoling Sun and Meiqin Wang

92. A Keyless Polyalphabetic Cipher ....................................................................................................... 529

William J. Spezeski

93. Vertical Handover between WiFi Network and WiMAX Network According to IEEE 802.21
Standard .............................................................................................................................................. 533
Mariem Thaalbi and Nabil Tabbane

94. Study on the Impact of Node Density and Sink Location in WSN..................................................... 539
Plamen Z. Zahariev et al.

95. An Algorithm for Synthesis of Families of Generalized Orthogonal Complementary Pairs .............. 543
B. Y. Bedzhev and M. P. Iliev

96. Low Message Overhead Clustering Algorithm for Wireless Sensor Networks.................................. 549
Batista et al.

97. On the Design of SOHO Networks .................................................................................................... 555

Lawan Ahmed Mohammed

98. Study of UMTS Handover Scenarios ................................................................................................. 561

B. Issac

99. Reduction of Energy Consumption in WSN using an Approach Geocast .......................................... 567
Harilton da S. Araújo et al.

100. Conflicts in Online Social Network Privacy Policy ........................................................................... 573

Zhengping Wu and Yuanyao Liu
xviii CONTENTS

101. Developing a Feedback Model for Managing Congestion in Packet Switched Wide Area
Networks ............................................................................................................................................ 579
E. F. Ogbimi et al.

102. Voice Gateway to IP-PSTN Integration and NGN Architecture ........................................................ 585
Geneflides Laureno da Silva and Raimir Holanda Filho

103. An Algorithm for Route Selection on Multi-sink Wireless Sensor Network Using Fuzzy Logic ...... 591
Líliam Barroso Leal et al.

104. Mining Mobile Sequential Patterns in Wireless Cellular Networks ................................................... 597
Joshua G. Bradley and Sherif S. Rashad

105. Agile Software Development Methods and its Advantages ............................................................... 603
Soly Mathew Biju

106. Toward a High Performance Grid Site RO-14-ITIM and the Contribution to WLCG ....................... 609
F. Farcas and G. Popeneciu

107. A MFI4OR-based Approach of Semantic Annotation for Web Service............................................. 615

Cheng Zeng et al.

108. Ethernet Node for Serial Communications ......................................................................................... 621

A. S. Popa et al.

109. EvalTool for Evaluate the Partitioning Scheme of Distributed Databases ......................................... 627
Adrian Runceanu and Marian Popescu

110. Automatic Creation of Daily News Digests from Internet News Sites............................................... 633
David B. Bracewell

Index ............................................................................................................................................................ 639

 Reviewers List

Abbas Riazi Hiten Choudhury Peng Su

Ahmad Sofian Shminan Ilias Karasavvidis Pinde Chen
Alejandro Regalado Ilie Borcosi Popescu Marian
Alen Hajnal Inci Saricicek Razib Hayat Khan
Andrew Dimmond Indrajit Mandal Reza Vahidnia
Anu Gupta Ioana Chan Mow Riyanto Sigit
Ardian Greca Jan Genci Sadhana Talele
Ausif Mahmood Joao Ferreira Sadia Sultana
Barry Stein Jozef Halko Sat Sharma
Biju Issac Jozef Mašcenik Seibu Mary Jacob
Bingyang Zhou Kabaou Mohamed Ouwais Shafqat Hameed
Bo Yuan Katarina Jelemenska Shen Chao
Bojan Medjo Keith Griffin Smita Shukla
Chi Thai Krystyna Maria Noga Sridhar Chandran
Chris Panagiotakopoulos Liciniu-Alexandru Kovacs Stefan Belev
Christian Ayala Esquivel Lucia A. Fernandes Sylvia Encheva
David Moore Luy Nguyen Tan Teodor Iliev
David Wyld Maduranga Dassanayake Teresa Cristina Bessa Assunção
Denilson Oliveira Magued Iskander Thomas Yang
Deniss Kumlander Margareth Stoll Vanessa Cox
Dimitrios Tomtsis Martina Hedvicakova Vini Kanvar
Dmitry Zub Massimo Esposito Xingguo Xiong
Edson Cavalcanti Neto Mazeyanti M Ariffin Yenumula Reddy
Enrique Araiza Montes de Oca Michel Owayjan Yoichiro Iwasaki
Eric Gamess Mikhail Stetsenko Yuanyao Liu
Ernest Hibbs Mofakharul Islam Yuval Cohen
Fawzi Albalooshi Mohammad Karim Zheng (Jeremy) Li
Grofu Florin Mohammad Muqri Zhenxing Luo
Harrison Yang Monika Verma
Hector Barbosa Leon Ognjen Kuljaca

xix
 Education in the Clouds: How Colleges and
Universities are Leveraging Cloud Computing
David C. Wyld and Rusty L. Juban
Southeastern Louisiana University
Department of Management – Box 10350
Hammond, LA 70402-0350 USA

Abstract- Cloud computing has been touted as a revolutionary and other smart, Web-connected devices in the past 10 years
concept in computing. In this article, we introduce the cloud (see Figure I).
computing concept, and then discus the implications for the use
of the cloud model in higher education. Due to budget
constraints and the power of computing “on-demand,” many
believe that colleges and universities will be at the forefront of
the shift to using cloud-based resources. We examine the impact
of cloud computing on both research and university computing
operations. We see that globally, this shift is already making a
difference in how universities procure and use IT resources. In
conclusion, we advance a six-stage Cloud Migration Strategy for
introducing and implementing cloud computing in institutions.
We also discuss the necessary changes in policies and
procurement, as well as security and reliability concerns for
institutions implementing cloud computing.

I. INTRODUCTION

A. A Different Way of Computing Fig. I. From the Internet to Cloud Computing.

I need a computer. Actually, I need the processing power
of hundreds of computing hours. Heretofore, if I was a Now, many think that cloud computing will be “the next
researcher running data or testing a model, that meant using big thing.” Indeed, Gartner [2] believes that in the end, the
solely the computing power available on my campus impact of the cloud model will be “no less influential than
computing system. For a major operation, that might mean e-business” (n.p.). If industry analysts are correct, we thus
waiting behind other faculty and student projects, and then stand at an inflection point – a true paradigm change – in
having to run my data over days at a time. Today, that the evolution of computing.
computing power can be had at my fingertips in a matter of The basic idea behind cloud computing is that anything
minutes - or even seconds. that could be done in computing – whether on an individual
Likewise, my email, my files, my programs were all PC or in a corporate data center – from storing data to
formerly on my computer – or on my campus’ mainframe. communicating via email to collaborating on documents or
Today, those operations - and my data - may reside on crunching numbers on large data sets - can be shifted to the
servers in Washington State – or in Bangalore. And this may cloud. As can be seen in Table I, cloud computing
not just be for my computing needs. Today, it may be for my encompasses a wide variety of offerings, including: SaaS
entire campus and all of the institution’s students and faculty. (Software as a Service), PaaS (Platform as a Service), and
Welcome to the world of cloud computing! IaaS (Infrastructure as a Service) [3].
Cloud computing has now become “shorthand” for the
B. The Cloud Computing Concept larger trend of computing services delivered over the Internet
The Economist [1] reminds us that: “Computing has [4]. From the perspective of the market analyst, IDC [5],
constantly changed shape and location—mainly as a result of cloud computing represents “an emerging IT development,
new technology, but often also because of shifts in demand” deployment and delivery model, enabling real-time delivery
(n.p.). We have seen revolutionary computing technologies - of products, services and solutions over the Internet.” As
truly “game changing” concepts - come about roughly once Lohr [6] characterized it: “Cloud computing — in which vast
each decade in the “modern era” of computing since around stores of information and processing resources can be tapped
1945 when computing came to mean computations from afar, over the Internet, using a personal computer, cell
performed by a machine, not by man. From the mainframe phone or other device — holds great promise…to cut the
era of the 1960s to the advent of minicomputers in the 1970s, costs, complexity and headaches of technology for
the personal computer in the 1980s, the growth of the Internet companies and government agencies.”
and the Web in the 1990s, and the explosion of cell phones

1
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_1, © Springer Science+Business Media B.V. 2010
2 WYLD AND JUBAN

Table I II. Cloud Computing in Universities Today

VARIANTS OF CLOUD COMPUTING

Level Label Description For universities, migrating to cloud-based services affords

User Level SaaS Companies host applications them the ability to provide improved collaboration and
“Software as a in the cloud that many users research capabilities, while at the same time, providing an
Service” access through Internet opportunity to cut IT costs while providing the same – or
connections. The service being
sold or offered is a complete better – levels of computing services. Magnified by the need
end-user application. to pare overhead costs at a time when public and private
Developer PaaS Developers can design, build, institutions are grappling with significant budget shortfalls,
Level “Platform as a and test applications that run cloud computing allows universities to not just use the
Service” on the cloud provider’s
infrastructure and then deliver
resources of commercial cloud providers – many of which
those applications to end-users are available to them either for free or at reduced costs. With
from the provider’s servers. the cloud model, students and faculty can take advantage of
IT Level IaaS System administrators obtain the ability to work and communicate from anywhere and on
“Infrastructure general processing, storage,
as a Service” database management and
any device using cloud-based applications.
other resources and The benefits for higher education center upon the
applications through the scalability and the economics of cloud computing. These will
network and pay only for gets be discussed in subsequent sections.
used.
Source: Adapted from Rayport and Heyward [3]. A. Scalability of Resources
One of the most important impacts of cloud computing will
Certainly, one of the hallmarks of cloud computing is that be the notion of computing power on-demand, in that: “When
it enables users to interact with systems, data, and each other you radically democratize computing so that anyone has
in a manner “that minimizes the necessary interaction with access at any moment to supercomputer-type capacity and all
the underlying layers of the technology stack” [7]. According the data storage they need” [11]. This “democratization” of
to the Cloud Computing Manifesto: “The key characteristics computing processing and storage power could have
of the cloud are the ability to scale and provision computing profound implications in everything from scientific inquiry
power dynamically in a cost efficient way and the ability of (by making no problem too big to compute) to new enterprise
the consumer (end user, organization or IT staff) to make the formation (by drastically reducing the need for upfront
most of that power without having to manage the underlying investment in IT resources – and the people to support and
complexity of the technology” [8]. maintain them) to public agencies (by making IT more
The Economist [9] captured the meaning of this trend in affordable and available to governments at all levels and in
stating: “The plethora of devices wirelessly connected to the all locales). Thus, we may be seeing a truly new era, where
Internet will speed up a shift that is already under way: from through democratizing computing technology, this will help
a ‘device-centric’ to an ‘information-centric’ world….(and) to bring “the benefits of high-powered computers and
as wireless technology gets better and cheaper, more and communications to all” [12].
more different kinds of objects will connect directly to the Cloud computing is a revolutionary concept in IT, due to
cloud.” Technology guru Clay Shirky perhaps put it best an unprecedented elasticity of resources made possible by the
when he said: “What is driving this shift is a change in cloud model. In everyday use, elasticity is commonly thought
perspective from seeing the computer as a box to seeing the of not just as the ability of an object to stretch out when
computer as a door” [3]. The emerging cloud computing needed, but to also contract as necessary (think of a rubber
paradigm is thus based on a “user-centric interface that band or a bungee cord). In computing terms, elasticity can be
makes the cloud infrastructure supporting the applications defined as: “The ability of a system to dynamically acquire or
transparent to the user” [10]. release compute resources on-demand” [7]. Under the cloud
model, organizations that need more computing power have
C. Overview the ability to “scale-up” resources on-demand, without
How does this new, on-demand, information-centric model having to pay a premium for that ability. Say, for instance,
of computing fit in the world of higher education – and what that a researcher or a department has large, batch-oriented
does it entail for research, for collaboration and for processing tasks. The individual or group can run the
communication in colleges and universities? This article operations far faster than previously and at no additional
examines the early evidence from the field and discusses the costs, “since using 1000 servers for one hour costs no more
practical and institutional implications. It concludes with a than using one server for 1000 hours” [13]. This unique
Cloud Migration Strategy for college and university IT attribute of cloud computing is a commonly referred to as
executives to follow as they seek to best integrate cloud “cost associativity,” and it allows for computational needs to
computing into their overall IT strategies. be addressed far faster and far cheaper than in the past. In
short, cloud computing gives organizations – even individual
users – with unprecedented scalability. Graphically, this is
 EDUCATION IN THE CLOUDS 3

depicted in Figure II (The Cost Associativity Principle in been devoted to the concept of “freeconomics,” most notably
Cloud Computing). the recent book by Wired magazine editor Chris Anderson
[19] entitled, Free: The future of a radical price. Most
consumer-level cloud offerings would be labeled a
“freemium,” which is a free version that is supported by a
paid, premium version. Such freemiums are becoming an
emergent business model, as they are particularly popular
among online service and software companies [19]. And,
when faced with competing against “free” alternatives, older,
more established companies have seen users migrate to the
gratis alternative. Indeed, some see an entire “Culture of
free” emerging, where from music to entertainment to news
to software, people are coming to expect that free is the price
Fig. II. The Cost Associativity Principle in Cloud Computing they should pay [20].
Source: Based on Armbrust, et. al. [13]; Reynolds and Best [14].
In the corporate computing market, as software, hardware
and processing power, and storage capacity become more and
Additionally, where in the past only the largest universities
more commoditized, cloud computing becomes a free – or
have had supercomputing capabilities cloud computing, with
lower cost – alternative to the way things have been done for
number-crunching capabilities available on an on-demand
decades. As DiMaio [21] remarked: “Why should I bother
basis, affords researchers anywhere to scale their computing
looking for an email client to replace Outlook and coexist
power to match the scale of their research question - bringing
with my newly installed OpenOffice, if I can get email and
supercomputing to the mainstream of research. As Delic and
office suite as a service with somebody like Google at a
Walker [15] characterized it, cloud computing might just
fraction of the cost and - most importantly - giving up the IT
“enable new insights into challenging engineering, medical
management burden too? Why are we talking about moving
and social problems,” as researchers will now have newfound
servers from Windows to Linux when the real question is
capabilities “to tackle peta-scale type(s) of problems” and to
why do we need to have our own servers in the first place?”
“carry out mega-scale simulations.” Craig A. Stewart,
Already, there have been many campuses that have
Associate Dean for Research Technologies at Indiana
switched to Google or Microsoft-hosted email. Google and
University, recently remarked that with cloud computing,
Microsoft host email for over four thousand colleges and
“You reduce the barrier to use advanced computing
universities, not just in the U.S., but in over 80 countries
facilities” [16].
worldwide. In fact, almost half of all campuses are now
We have seen the reduction of barriers already paying
making use of hosted email services [13]. The switch to
dividends in research. At pharmaceutical giant Eli Lilly,
hosted services is paying significant dividends for the early
researchers needed to queue their projects to run in Lilly’s
adopting institutions. By switching to Gmail, Notre Dame
internal data center. This process to provision enough server
reports that it saved $1.5 million in storage and other tech
capacity for their respective projects often meant a delay of
costs, while at the same time, finding that their students’
up to two months waiting on their data run. Today however,
satisfaction with the campus’ email rose by over a third!
with cloud computing, research scientists can today provision
Likewise, institutions (such as Arizona State and Washington
the necessary processing capacity for their projects in five
State) are consistently reporting at least six figure annual
minutes. This allows researchers at Lilly and other research
savings from switching to Google or Microsoft hosted
organizations to crunch data and test theories in ways that
systems [22]. Even more importantly, by switching to hosted
may have gone unexplored in the prior era where they would
email and productivity software, the job and focus of college
have been dependent solely on in-house computing resources
IT staff can be changed. As Pepperdine University’s CIO
[17]! Similar experiences are being reported at universities,
Timothy Chester observed, his smaller IT staff can be more
both in the U.S. and abroad. For instance, at the International
productive. He recently stated: “We want our staff working
Institute of Information Technology in Hyderabad, India,
more with students and faculty and less on the nuts and bolts
Associate Professor Vasudeva Varma reports that the ability
of delivering technology” [22].
to run more data from more experiments more quickly has
Certainly, as challenging budgetary times have been
resulted in more publications for faculty in the Institute’s
forecast to persist across higher education for the next few
Search and Information Extraction Lab, which he heads [18].
years – at least, there will likely be even greater pressures on
B. Economics of Computing colleges and universities to replace “paid” software and
There is much discussion about the whole concept of computing resources with “free” or low-cost cloud
“free” pricing for products and services today – and many of alternatives. From the cloud provider standpoint, Google has
the email, storage, hosting, and applications that are at the stated that its incentive in providing such free services to
forefront of cloud computing today are indeed free. The most universities is to create “relationships for life” with students
notable of these are the product offerings of Google (Gmail, and faculty [23].
GoogleApps, GoogleDocs, and others). Much attention has
4 WYLD AND JUBAN

III. ANALYSIS faculty could have access to the scale and type of computing
power enjoyed only by elite institutions [26].
Many in higher education are coming to concur with the
opinion of Keahey [24], who plainly stated: “I believe that IV. CONCLUSION
this (cloud computing) will be the model of the future. This
will absolutely drive cost savings for the universities.” A. A Cloud Migration Strategy for Higher Education
Across higher education, the cloud computing landscape It is important to bear in mind the advice of Higginbotham
should be quite active over the next few years, as we will see [27] that “cloud computing is a tool, not a strategy.” IT
both coordinated efforts and “rogue” operations that will test leaders in higher education will thus be well-advised to take a
how and where cloud computing can be effectively applied. programmed, assessment of how cloud computing can fit into
As we have seen, colleges and universities will in many their overall IT strategy, in support of the mission and overall
instances lead the way. These entities will continue to do so, strategy of their institution. This should take a 6-step process,
based on their need for computing power on demand and for as shown in Figure III.
providing the types of ready – and in many cases free – IT
resources – to their faculty and students. With pressure to
reduce the fixed costs of higher education – and IT being a
very rich target – the shift to cloud may be more forced in
some cases than may be dictated by the on-the-ground
circumstances. Indeed, some of the most exciting uses and
best practices for cloud computing could well come from the
world of higher education.
We have seen predictions that due to the cost and
operational benefits of cloud computing, more and more
companies will find themselves outsourcing most – if not all
– of their IT to cloud providers, creating what has been
termed as “serverless organizations” [25]. Indeed, it has been Fig. III. From the Internet to Cloud Computing.
predicted that organizations of all sizes will find it beneficial
to concentrate on and optimize their business processes by The Cloud Migration Strategy begins with learning about
outsourcing the IT function. So, why not “serverless the basics of cloud computing – through attending seminars,
universities”? By outsourcing almost all of IT and all data networking, talking with vendors, and reading articles such
storage/handling – this may be a viable proposition for as this one. Given that cloud computing represents a new
colleges and universities, particularly as cloud offerings paradigm in computing technology, it will be important for
expand and are made more secure and reliable. technology transfer to occur – the “techies” in and outside of
As we have seen in this article, there are certainly the institution will need to go the extra mile to educate and
discussions and embryonic efforts underway – both in the inform the “non-techie” amongst their ranks and
U.S. and abroad - as public and private universities examine constituencies as to the merits and value of cloud computing.
how to best made the cloud-concept work for they and their It will be especially important to devote sufficient funding for
students and faculty. Universities are beginning to work research to establish how cloud computing is working – or
collaboratively in the cloud to pool their IT resources. not working – in various areas in the university and across
Already, this has occurred in Virginia and North Carolina. In institutions, so as to ground policies and develop best
the Commonwealth, a dozen colleges and universities have practices in regards to the use of cloud computing.
come together to form the Virginia Virtual Computing Lab Then, IT executives should conduct an honest assessment
[16]. Such efforts allow institutions to cut their IT costs by of their institution’s present IT needs, structure, and capacity
reducing their need for software licensing, for upgrade utilization. In a cloud computing environment, where
capabilities, and for perhaps maintaining their own data resources can be added – or subtracted – based on needs and
centers, all while improving the IT resources for their faculty demand, it will be critical for IT managers to honestly assess
and students. Already, by shifting to cloud offerings, North their institution’s IT baseline for faculty, students and
Carolina State University has been able to dramatically lower operations. In looking at data center utilization, it will be
expenditures on software licenses and simultaneously, reduce vital to look at what resources are used all the time and are
the campus’ IT staff from 15 to 3 full-time employees [18]. necessary for day-to-day operations to establish a baseline for
Additionally, there have been calls for the federal internally-hosted operations. Only then can one look at
government to take the lead to create a cloud computing whether to continue to host “excess” capacity in the data
environment, to be available for use by all colleges and center or to contract for cloud services as needed to scale-up
universities nationwide. In doing so, proponents argue for the to meet demands for greater amounts of computing resources.
economic and educational benefits that such a resource University IT leaders should then pick one area – even one
would provide, as it would democratize computing specific project - to “cloud pilot” and assess their ability to
technology and “level the playing field” so all students and manage and bring such a project to fruition. As with any new
technology, we are seeing a great deal of pure
 EDUCATION IN THE CLOUDS 5

experimentation with cloud computing – “science project” examine how cloud computing will secure and procure their
like work for the most part up till now. All of us who use the computing environment.
Internet are experimenting with cloud applications in our Indeed, one of the principal concerns about cloud
daily lives – from Twittering to Gmail to using photo-sharing computing whether it is secure and reliable. Unfortunately,
sites. In the same way, we are seeing organizations worries over cloud reliability and availability – or
conducting cloud computing trials - what Woody [17] termed specifically, the lack thereof when such instances arise - are
as “science experiments” in the use of the technology. Such not just theoretical, as there have been well-publicized
efforts that are far away from their core IT operations and outages of many of the most popular public cloud services
many times on (or trying to connect) the periphery of the [30, 31]. And, as Schwartz [32] astutely pointed-out, when
organization. . Many times – even in the public sector and cloud service outages or inaccessibility occur, “most of the
especially on campuses, these experiments may be “rogue” risk and blame if something goes wrong will fall directly on
operations – taken on by individuals and units to test the the shoulders of IT -- and not on the cloud computing service
utility of the technology. These are important efforts, and providers.”
they should be supported – and reported within and outside Security concerns may indeed impede the shift to cloud-
the institution – so that others in the IT and the wider based models. As with prior shifts in information technology
community can learn of the successes – and the downsides – with the advent of the Internet and the Web, the introduction
of operating in the clouds. Thus, it will be vitally important to of e-mail, and the explosion of social media, their growth and
share both “best practices” and “lessons learned” in cloud adoption rates have been slowed by initial fears – some
computing. Indeed, many predict that such “science projects” justified and some very unjustified – over security concerns
in large and small organizations will drive the eventual and the loss of control over data and operations [33].
acceptance and adoption of cloud computing [28]. Certainly, privacy and security questions will need to be
After the internal assessment and external outreach addressed as institutional data and applications move into a
stemming from the pilot effort, they should then conduct an cloud environment. Indeed, analogies have been drawn
overall IT cloud-readiness assessment to determine if they between the advent of cloud computing today with the
have data and applications that could readily move to a cloud introduction of wireless technologies a decade ago [34].
environment and if a public/private/hybrid cloud would be Finally, security is undoubtedly a hard metric to quantify.
suitable or useable for these purposes and rank-order And, all too often, from the perspective of Golden [35] and
potential projects. Finally, it is time to begin a cloud rollout other observers, the IT community has a somewhat damaging
strategy – gaining buy-in from both institutional leadership tendency to treating all risks – whatever the real nature of
and IT staffers and communicating with both internal and them – as the very worst case scenario and not judging the
external stakeholders as to the goals, progress, and true impact – and likelihood – of their occurrence.
costs/benefits of each cloud project. This is where the cloud Finally, universities’ often outdated and byzantine
goes from being a test effort to become more mainstream in procurement rules and regulations, some of which may even
the way the university manages its data, its operations and its preclude the use of cloud computing in select instances, will
people. It becomes part of “normal” operations, just as other need to be changed to be more cloud-friendly and encourage
prior tech innovations (from telephony to fax to the Internet the savings and efficiencies that can come from this new
to email and to social media) have become IT tools, used in model of IT [36]. There will also need to be changes made in
support of the institution’s IT strategy and more importantly, not just the language, but in the mindset of contracting for
its overall strategy. computing services. For while IT administrators look at
At this point, the process enters the final stage – call it capacity and systems, end users look to performance. As
“continuous cloud improvement” – to where the institution Jackson [37] recently put it, the key metric will now become:
continues to move appropriate data and applications to the “When I sit down at that computer, do I see the functionality
cloud – and perhaps even back from the cloud to internally- I need?”
hosted operations, if necessary, based on a thorough and In time, we may look back on the latter portion of this first
continuous assessment of the appropriate use of cloud decade of the new millennium as a true turning point in the
technologies for their particular university. history of computing. The transition however will take years,
perhaps even decades, and “we're not going to wake up
B. Implications for Higher Education tomorrow and get all our computing requirements through a
The shift to more cloud-based applications will indeed socket in the wall” [38]. However, all signs point to a true,
bring newfound capabilities to communicate, collaborate and campus-led revolution in computing.
conduct research to university faculty, staff and students.
However, it will also necessitate a flurry of policy decisions REFERENCES
that will need to be made and operational rules that will need
to be implemented. For instance, there will have to be IT [1] The Economist Editorial Staff, “Let it rise,” The Economist, October
23, 2008. [Online]. Available:
policy decisions made as to who can access what files and http://www.economist.com/specialreports/displayStory.cfm?story_id=1
what type of access they will have (i.e. read-only, editing 2411882 [Accessed: March 3, 2009].
access) [29]. The shift will also necessitate institutions to [2] Gartner (2008), Press Release: Gartner says cloud computing will be
as influential as e-business, June 26, 2008. [Online]. Available:
6 WYLD AND JUBAN

http://www.gartner.com/it/page.jsp?id=707508 [Accessed: February http://blogs.gartner.com/andrea_dimaio/2009/04/08/open-source-in-

24, 2009]. government-is-irrelevant/ [Accessed: July 18, 2009].
[3] J.F. Rayport and A. Heyward, Envisioning the cloud: The [22] J. Caplan, “Google and Microsoft: The battle over college e-mail,”
next computing paradigm, March 20, 2009. [Online]. Available: Time, August 14, 2009. [Online]. Available:
http://www.marketspaceadvisory.com/cloud/Envisioning-the- http://www.time.com/time/printout/0,8816,1915112,00.html
Cloud.pdf [Accessed: March 24, 2009]. [Accessed: August 27, 2009].
[4] J. Hempel, “Salesforce hits its stride,” Fortune, February 16, 2009. [23] S. Coughlan, “Google's e-mail for universities,” BBC News, June 11,
[Online]. Available: 2007. [Online]. Available:
http://money.cnn.com/2009/02/16/technology/hempel_salesforce.fortu http://news.bbc.co.uk/2/hi/uk_news/education/6741797.stm
ne/index.htm?postversion=2009021710 [Accessed: March 8, 2009]. [Accessed: September 1, 2009].
[5] F. Gens, “Defining ‘cloud services’ and ‘cloud computing,’” IDC [24] J. Staten, “Cloud maturity is accelerating,” Forrester.com, July 10,
Research, September 23, 2008. [Online]. Available: 2008. [Online]. Available:
http://blogs.idc.com/ie/?p=190 [Accessed: May 2, 2009]. http://blogs.forrester.com/it_infrastructure/2008/07/cloud-
[6] S. Lohr, “I.B.M. to help clients fight cost and complexity,” New York maturity.html [Accessed: February 28, 2009].
Times, June 15, 2009. [Online]. Available: [25] Appirio, Inc., Press Release: Top 10 cloud computing predictions for
http://www.nytimes.com/2009/06/15/technology/business- 2009. December 18, 2008. [Online]. Available:
computing/15blue.html?_r=1&emc=eta1 [Accessed: June 17, 2009]. http://www.appirio.com/about/pr_predictions-09_121808.php
[7] K. Langley, “Cloud computing: Get your head in the clouds,” [Accessed: January 12, 2009].
Production Scale, April 24, 2008. [Online]. Available: [26] W. Hurley, “Higher education needs a national computing cloud,”
http://www.productionscale.com/home/2008/4/24/cloud-computing- InfoWorld, January 25, 2009. [Online]. Available:
get-your-head-in-the-clouds.html [Accessed: June 3, 2009]. http://www.infoworld.com/d/cloud-computing/higher-education-needs-
[8] The open cloud manifesto: A call to action for the worldwide cloud national-computing-cloud-404?source=fssr [Accessed: June 1, 2009].
community (Draft 1.0.9). OpenCloudManifesto.org, March 30, 2009. [27] S. Higginbotham, “Cloud computing is a tool, not a strategy,”
[Online]. Available: GigaOm, February 19, 2009. [Online]. Available:
http://opencloudmanifesto.org/opencloudmanifesto1.htm [Accessed: http://gigaom.com/2009/02/19/cloud-computing-is-a-tool-not-a-
April 1, 2009]. strategy/ [Accessed: March 11, 2009].
[9] The Economist Editorial Staff, “On the periphery,” The Economist, [28] J. King, “5 key questions about cloud storage,” Computerworld, July
October 23, 2008. [Online]. Available: 13, 2009. [Online]. Available:
http://www.economist.com/specialreports/displaystory.cfm?story_id=1 http://www.computerworld.com/s/article/340471/Cloud_Storage_Illum
2411896 [Accessed: March 3, 2009]. inated [Accessed: July 23, 2009].
[10] IBM, White Paper - Seeding the Clouds: Key Infrastructure Elements [29] D. Beizer, “Cloud computing comes into focus,” Government
for Cloud Computing, February 2009. [Online]. Available: Computer News, June 11, 2008. [Online]. Available:
ftp://ftp.software.ibm.com/common/ssi/sa/wh/n/oiw03022usen/OIW03 http://gcn.com/articles/2008/06/11/cloud-computing-comes-into-
022USEN.PDF [Accessed: February 22, 2009]. focus.aspx [Accessed: February 7, 2009].
[11] N.G. Carr, “The many ways cloud computing will disrupt IT,” [30] S. Gaudin and D. Nystedt, “Gmail outages could turn off enterprises,”
InfoWorld, March 25, 2009. [Online]. Available: Computerworld, September 10, 2009. [Online]. Available:
http://www.tmcnet.com/usubmit/2009/03/25/4084363.htm [Accessed: http://www.computerworld.com/s/article/342708/Gmail_Outages_Coul
April 1, 2009]. d_Turn_Off_Enterprises?taxonomyId=16 [Accessed: September 12,
[12] J. Foley, “Report: Cloud computing could be bigger than the Web,” 2009].
Information Week, March 20, 2009. [Online]. Available: [31] C. Waxer, “Is the cloud reliable enough for your business?”
http://www.informationweek.com/cloud- Computerworld, June 1, 2009. [Online]. Available:
computing/blog/archives/2009/03/google_report_c.html?catid=cloud- http://www.computerworld.com/action/article.do?command=printArtic
computing [Accessed: April 24, 2009]. leBasic&taxonomyName=Servers+and+Data+Center&articleId=33915
[13] M. Armbrust, A. Fox, R. Griffith, A.D. Joseph, R. Katz, A. Konwinski, 2&taxonomyId=154 [Accessed: June 15, 2009].
G. Lee, D. Patterson, A. Rabkin, I. Stoica and M. Zaharia, Above the [32] E. Schwartz, “The dangers of cloud computing,” InfoWorld, July 7,
clouds, University of California, UC Berkeley Reliable Adaptive 2008. [Online]. Available:
Distributed Systems Laboratory, February 2009. [Online]. Available: http://www.infoworld.com/article/08/07/07/28NF-cloud-computing-
http://d1smfj0g31qzek.cloudfront.net/abovetheclouds.pdf [Accessed: security_1.html. [Accessed: August 1, 2009].
March 10, 2009]. [33] Kelton Research, 2009 Cloud Computing Survey, February 2009.
[14] E. Reynolds and C. Bess, “Clearing up the cloud,” Cutter IT Journal, [Online]. Available:
June/July 2009, vol. 22, no. 6/7, pp. 14-20. http://www.avanade.com/people/thought_detail.aspx?id=70 [Accessed:
[15] K.A. Delic and M.A. Walker, “Emergence of the academic computing May 8, 2009].
clouds,” Ubiquity, August 5, 2008, vol. 9, no. 31, pp. 2-5. [34] D. Beizer, “NIST creates cloud-computing team,” Federal Computer
[16] J.R. Young, “3 ways web-based computing will change colleges,” The Week, February 25, 2009. [Online]. Available:
Chronicle of Higher Education, October 24, 2008. [Online]. http://fcw.com/articles/2009/02/25/nist-cloud-computing.aspx
Available: http://chronicle.com/cgi2- [Accessed: July 2, 2009].
bin/printable.cgi?article=http://chronicle.com/free/2008/10/5611n.htm [35] B. Golden, “The case against cloud computing, part two,” CIO,
[Accessed: June 5, 2009]. January 29, 2009. [Online]. Available:
[17] T. Woody, “The future of cloud computing,” Portfolio, March 9, 2009. http://www.cio.com/article/478419/The_Case_Against_Cloud_Comput
[Online]. Available: http://www.portfolio.com/views/columns/dual- ing_Part_Two [Accessed: June 4, 2009].
perspectives/2009/03/09/Todays-Weather-Report [Accessed: May 8, [36] M. Weigelt, “Kundra aids search for procurement leader,” Federal
2009]. Computer Week, July 31, 2009. [Online]. Available:
[18] K. Ng, “Campus on a Cloud,” FutureGov, August 2009, vol. 6, no. 4, http://fcw.com/Articles/2009/08/03/WEEK-Kundra-aids-OFPP-
pp. 34-38. search.aspx [Accessed: August 7, 2009].
[19] C. Anderson, Free: The future of a radical price. New York: Hyperion. [37] J. Jackson, “New metrics needed for cloud computing,” Government
[20] Knowledge@Wharton, “How about free?” University of Pennsylvania, Computer News, July 30, 2009. [Online]. Available:
March 4, 2009. [Online]. Available: http://gcn.com/Articles/2009/07/30/New-metrics-for-cloud-
http://knowledge.wharton.upenn.edu/article.cfm?articleid=2169 computing.aspx?p=1 [Accessed: August 3, 2009].
[Accessed: June 10, 2009]. [38] K. Melymuka, “The end of corporate IT,” Computerworld, May 9,
[21] A. DiMaio, “Do vendors have a vision for the future of government?” 2005. [Online]. Available:
Gartner, April 8, 2009. [Online]. Available: http://www.computerworld.com/managementtopics/management/story/
0,10801,101573,00.html [Accessed: January 30, 2009].
 Weak Orderings of Knowledge States
Sylvia Encheva and Sharil Tumin*
Stord/Haugesund University College, Faculty of Technology, Business and Maritime Sciences, Bjørnsonsg. 45,
5528 Haugesund, Norway
University of Bergen, It Dept, P.O. Box 7800, 5020 Bergen, Norway*
[email protected], [email protected]

permitting them to make the most of the facilities the system

offers, [9]. In that framework which supports personalized
Abstract - This article discusses assessment of knowledge states of learning a new topic is recommended to a student after
students enrolled in a new subject. The above-mentioned assessing his/her knowledge. The system developed as a result
knowledge states belong to a single study unit. Applying methods of that framework responds to students needs according to
from the theory of weak orderings assures also firm quantitative
their learning orientations.
support for subsequent tuning of current course materials, hints
and diagnostic tests.
Artificial intelligence techniques are widely used for
facilitating a learning process tailored to individual needs.
Among the most successful computer-based instructional
I. INTRODUCTION tools, seriously relying on such techniques, are intelligent
tutoring systems.

E
An intelligent tutoring system is defined in [25] as a
stablishing the knowledge state of a student in a "computer system" based on "artificial intelligence" designed
subject is crucial for providing him/her with individual help. to deliver content and to provide feedback to its users. In [26]
Subsets of relevant examination questions and certain skills an intelligent tutoring system is considered to be a software
from a branch of knowledge are listed as examples of agent that provides customized and adaptable instructions and
knowledge states in [12]. It is quite important to underline that feedback to learners. The topic received a lot of attention from
not all possible subsets of such items turn out to be knowledge the research community since the 1980's. However, it seems
states. that the process of providing adequate help to users based on
Outcomes of diagnostic tests usually give an indication their individual responses is still open.
about the general status of students’ knowledge. At the same A cognitive model represents the knowledge that an ideal
time, they do not address the problem of classification of student would possess about a particular subject. Developing
knowledge states. The latter is strongly related to a cognitive tutor involves creating a cognitive model of
identification of existing misconceptions and lack of adequate students’ problem solving by writing production rules that
knowledge. This work presents an approach that can be used characterize the variety of strategies and misconceptions
in the process of solving these problems. students may acquire, [16, 17, and 18]. Cognitive tutors have
Based on students' responses a content provider can follow been successful in raising students' math test scores in high
the process of response clusterings and make appropriate school and middle-school classrooms, but their development
adjustments in contents of a problem or the related materials has traditionally required considerable time and expertise, [8].
or both. Response clusterings can be used to track down the An investigation of whether a cognitive tutor can be made
effect of assistance provided to each student and thus more effective can be found in [15].
contribute to a better individualized assistance. The process of delivering hints in an intelligent tutoring
The rest of the paper is organized as follows. Section 2 system has been discussed in [24]. Taxonomy for automated
contains information about related work. Section 3 is devoted hinting is developed in [20]. The role of hints in a Web based
to ordered sets. Section 4 relates various personal responses learning systems is considered in [7] and [8]. Usability issues
and Section 5 is devoted to a system description. Section 6 of a Web-based assessment system are discussed in [13] and
contains the conclusion of this work. [21].
A model for detecting students’ misuse of help in intelligent
tutoring systems is presented in [3]. A proliferation of hint
abuse (e.g., using hints to find answers rather than trying to
II. BACKGROUND
understand) was found in [1] and [15]. However, evidence
that when used appropriately, on-demand help can have a
Navigational structures in an automated tutoring system positive impact on learning was found in [19] and [23].
should prevent students from becoming overwhelmed with
information and losing track of where they are going, while

7
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_2, © Springer Science+Business Media B.V. 2010
8 ENCHEVA AND TUMIN

III. ORDERED SETS Theorem 1 along with several axioms, [5] provides an
existing social welfare function.
Two very interesting problems about orderings are
considered in [5], namely the problem of determining a Theorem 1 There is a unique social welfare function
consensus from a group of orderings and the problem of satisfying Axioms 0-4. It is given by: i is preferred to j if
making statistically significant statements about ordering. and only if the sum of the row of A corresponding to j is
An ordered set (or partially ordered set or poset) is an
larger than the sum of the row of A corresponding to i , and
ordered pair ( P, ≤) of a set P and a binary relation ‘ ≤ ’
otherwise the social ordering is indifferent between i and j .
contained in P × P called the order (or the partial order) on
Two elements a and b where a ≠ b and a, b ∈ P are
P such that ‘ ≤ ’ is reflexive, antisymmetric and transitive.
comparable if a ≤ b or b ≤ a , and incomparable otherwise.
A relation I is an indifference relation when given AIB
neither A > B nor A < B has place in the componentwise If ∀a, b where a, b ∈ P are comparable, then P is a chain.
ordering. A partial ordering whose indifference relation is If ∀a, b where a, b ∈ P are incomparable, then P is an
transitive is called a weak ordering. antichain.
Let ω1, ω2, ω3 be weak orderings. The ω2 is between ω1
and ω3 if each decision made by ω2 is made either by ω1 or
B. Knowledge spaces
ω3 and any decision made by both ω1 and ω3 is made by ω2,
i.e. ω1 ∩ ω3 ⊆ ω2 ⊆ ω1 ∩ ω3.
The distance is defined as Within the theory of knowledge spaces, [11], [12] a
d(ω1, ω2) + d(ω2, ω3) = d(ω1, ω3 ). knowledge domain is described by a set of problems Q. Then
The distance is a metric in the usual sense, it is invariant under the knowledge state of a student is a set Q’ ⊆ Q that the
permutation of alternatives, and the minimum positive student can solve. A knowledge structure of a subject is a set
distance is 1. If two weak orderings agree except for a set of of all knowledge states.
alternatives that is an open interval in both, then the distance A knowledge space is a closure system, [11] where closure
may be computed as if the alternatives in this set were the system on a finite set M is a set F of subsets of M such that
only objects being ranked. M ∈ F and C , C 1 ∈ F Ÿ C ∩ C 1 ∈ F
If given two alternatives, a person is finally choosing only one.
The natural extension to more than two elements is known as the
'majority rule' or the 'Condorcet Principle'. A relation
IV. GEOMETRIC INTERPRETATION OF KNOWLEDGE STATES
R ( L1 , L2 ,..., Lk ) is constructed by saying that the pair
(a, b) ∈ R if (a, b) belongs to the majority of relations Li . Initially we are trying to establish the status of students’
knowledge in a topic with respect to three knowledge states
The following three linear orderings a b c , b c a , and
called a, b and c. This is realized by considering students’
c a b , leading to R = {( a, b), (b, c), (c, a)} (three-way answers to three questions in a multiple choice test that is a
tie), illustrate the 'paradox of voting', [4]. part of a Web-based system. Students' responses are saved in
The probability of the voting paradox for weak orderings is a database. Depending on whether a provided answer is
calculated analytically for the three-voter-three-alternative correct, partially correct or incorrect, the student is suggested
case. It appears that the probability obtained this way is appropriate reading materials, hints or examples. Afterwards
considerably smaller than the corresponding case for linear she can repeat the test. The consecutive test contains new
orderings, [22]. questions but related to the same knowledge states. This
allows the course/test designer to follow each student’s
progress within any knowledge state and to obtain a clear
A. Social Welfare Function
picture about the degrees to which these knowledge states are
reached by a particular student, and by a group.
A 'social welfare function' maps k -tuples of the set of If for example one or more of the knowledge states
linear orderings of any b ⊂ A to single linear orderings repeatedly appear to be difficult to be reached by a single
of B , where A is a set of at least three alternatives, [2]. student or by a group then the corresponding course materials
and the related questions, have to be redesigned.
A social welfare function acts on k -tuples representing
A knowledge state is assumed to be reached by a student if
either weak or linear orderings of k individuals for m the student provides a correct answer to the question
alternatives. The representation of an individual's ordering can presented to her/him. The question is of course related to that
be thought of as a column vector in which an integer in knowledge state. A knowledge state is assumed to be partially
position i represents the preference level the individual reached by a student if the student provides a partially correct
assigns to alternative i . Thus k individuals presented with answer to the presented to her/him question, and knowledge
m alternatives can be illustrated by k -tuples of orderings in state is assumed not to be reached by a student if the provided
a m × k matrix of integers of preference levels. answer is incorrect.
 WEAK ORDERINGS OF KNOWLEDGE STATES 9

Using nested parentheses one can represent the partial With this distribution, one can safely conclude that
ordering structure internally in a given application. Let the knowledge state 'a' has been reached by the majority of
alphabets in this definition be '( ), a, b, c'. The use of students. Another interesting observation is that knowledge
parentheses signifies the position of ordering of a particular state 'c' is reached by 20% of the students. Since the distance
element 'a', 'b', or 'c', where ‘( )' implies correct answer, '(( ))’ from (ab(c)) or (a(bc)) to (c(ab)) is the same, in this particular
implies partially correct answer, '((( )))' implies incorrect case, we can safely conclude that there are two types of
answer. Parentheses are not used for the elements in the center students with different knowledge background in relation to
of the graph. the learning process of this particular study unit. Such
Note that the set {a, b, c} is neither a chain no an antichain, observations are very useful for future tuning of course
i.e. a pair of different elements can either be comparable or materials.
incomparable. When a pattern of clustering does not show any significant
value but is evenly distributed amongst the twelve ordering
patterns (see f. ex. Fig. 5) then one can say that none of the
provided materials was particularly helpful for the students’
study progress. Both the course materials and the questions in
diagnostic tests need to be redesigned.
Fig. 1 Partial ordering (a(b(c)))
Discussions on distances between different orderings are
enclosed in the next subsections.
The ordering ( a (b(c ))) in Fig.1 implies that knowledge state
‘a’ is reached by that student, knowledge state ‘b’ is partially
A. Orderings Where Two of the Knowledge States Either
reached, and knowledge state ‘c’ is not reached by that
Reached or Partially Reached
student.

The observed shortest distances between two orderings of

knowledge states from a shown sequence are considered to be
both either reached or partially reached.

Fig. 2 Partial ordering (a(bc)) 1) From ordering (a(bc)) to ordering ((ac)b)

Fig. 2 illustrates a partial ordering ( a (bc)) where

knowledge state ‘a’ is reached by that student, and both
knowledge states ‘b’ and ‘c’ are partially reached.

Fig. 3 Partial ordering (ab(c))

The ordering (ab(c)) in Fig. 3 implies that knowledge state

‘c’ is partially reached by that student, while knowledge states
‘a’ and ‘b’ are reached.

Fig. 5: Orderings (a(bc)) and (ac(b))

Fig. 4 Partial ordering (a(b))c
In ordering (a(bc)) we observe one knowledge state, ‘a’, to
The ordering ( a (b))c in Fig. 4 implies that knowledge state be reached and two knowledge states, ‘b’ and ‘c’, to be
partially reached, as shown in Fig 5. In ordering ((ac)b) we
‘a’ is reached by that student, knowledge ‘b’ is partially observe two knowledge states, i.e. ‘a’ and ‘c’, to be reached
reached, and the status of knowledge state ‘c’ is unknown and one knowledge state, ‘b’, to be partially reached. Note
since the system has not registered any response from the that knowledge state, ‘a’ is reached in both the orderings
student. (a(bc)) and (ac(b)) while one of knowledge states, ‘c’, is
Suppose a particular study unit exhibits the following partially reached in ordering (a(bc)) and reached in ordering
pattern of clustering (ab(c)) - 50%, (a(bc)) - 10%, and (c(ab)) ((ac)b). This indicates that the course materials, hints,
- 20%. The rest of the 20% are distributed to other orderings questions, etc. related to knowledge states ‘a’ and ‘c’ do not
and are therefore not significant. The significant factor is currently need adjustments. The remaining question is how to
assumed to be 10% or above.
10 ENCHEVA AND TUMIN

improve on the provided assistance in a way that the

knowledge state ‘b’ is also reached. The distance between
ordering (a(bc)) and ordering ((ac)b) is 2.

• From ordering (ab(c)) to ordering (c(ab))

All knowledge states are reached in a different way in

ordering (c(ab)) in comparison to ordering (ab(c)). Thus
knowledge states ‘a’ and ‘b’ being reached in ordering (ab(c))
are partially reached in ordering (c(ab)), and knowledge state
‘c’ is partially reached in ordering (ab(c)) and reached in
ordering (c(ab)). This indicates that the course materials,
hints, questions, etc. related to knowledge state ‘c’ do not
currently need adjustments.
The remaining question is how to improve on the provided Fig. 7: Orderings (a(bc)) and (b(ac))
assistance in a way that knowledge states ‘a’ and ‘b’ remain
reached. In similar occasions, it is recommended to pay This indicates that the course materials, hints, questions, etc.
particular attention to the contents of questions involved in related to knowledge state ‘b’ do not currently need
establishing both orderings. The distance between orderings adjustments. The work that has to be done in this case is
(ab(c)) and (c(ab)) is 4. related to knowledge states ‘a’ and ‘c’. The fact that
knowledge state ‘a’ is only partially reached in ordering
(b(ac)) requires special attention to be paid to the involved
questions and to the related course materials, hints, etc. Since
knowledge state ‘c’ remains partially reached in both
orderings the course developer can concentrate on tuning the
provided explanations, examples, hints, etc. The distance
between orderings (a(bc)) and (b(ac)) is 4.

B. Orderings Where Knowledge States are Reached,

Partially Reached or Not Reached

• From ordering (a(c(b))) to ordering (c(a(b)))

Fig. 6: Orderings (ab(c)) and (c(ab))

• From ordering (a(bc)) to ordering (b(ac))

Both orderings in Fig. 7 have the same structure. Two of the

knowledge states, ‘a’ and ‘b’ are being respectively reached
and partially reached in ordering (a(bc)) and are partially
reached and reached in ordering (b(ac))), where knowledge
state ‘c’ is partially reached in both orderings

Fig. 8: Orderings (a(c(b))) and (c(a(b)))

Two of the knowledge states in Fig. 8, ‘a’ and ‘c’ are being
respectively reached and partially reached in ordering
(a(c(b))) and are partially reached and reached in ordering
(c(a(b)), where knowledge state ‘b’ is not reached in both
orderings. This indicates that the course materials, hints,
questions, etc. related to knowledge state ‘c’ do not currently
 WEAK ORDERINGS OF KNOWLEDGE STATES 11

need adjustments. The work that has to be done in this case is

related to knowledge states ‘a’ and ‘b’. The fact that
knowledge state ‘a’ is only partially reached in the second
ordering requires special attention to be paid to the involved
questions and to the related course materials, hints, etc. Since
knowledge state ‘c’ remains not reached in both orderings the
course developer can work on the provided reading materials,
explanations, examples, hints, etc. The distance between
orderings (a(c(b))) and (c(a(b))) is 2.

• From ordering (a(c(b))) to ordering (c(b(a)))

All knowledge states change their status regarding the

degrees to be reached, Fig 9. This occurrence requires very
serious attention on the involved reading materials,
explanations, examples, hints, questions etc. If it is due to a Figure 10: Orderings (a(c(b))) and (b(c(a)))
particular student response only then individual approach
might be appropriate. The distance between orderings
(a(c(b))) and (c(b(a))) is 4. C. Various Orderings

Occurrences like those in Fig. 11 where one of the orderings

is of type (a(bc)) and the other one is an ordering like
(a(c(b))), (c(a(b))), and (c(b(a))) are treated like the ones in
the previous subsections. In an attempt to save space and keep
to the geometrical interpretation, the couples of orderings to
be considered are indicated with additional arrows instead of
coloring the corresponding paths.

Fig. 9: Orderings (a(c(b))) and (c(b(a)))

• From ordering (a(c(b))) to ordering (b(c(a)))

Knowledge states ‘a’ and ‘b’ in Fig. 10 are being

respectively reached and not reached in ordering (a(c(b))) and
are partially reached and reached in ordering (b(a(c))), where
knowledge state ‘c’ is partially reached in both ordering. This Fig. 11: Various orderings
indicates that the course materials, hints, questions, etc.
related to knowledge states ‘a’ and ‘b’ have to be
reconsidered. Since knowledge state ‘c’ remains partially Remark: With respect to orderings discussed in this section
reached in both orderings the course developer could it is important to establish which group of students have a
concentrate on tuning the provided explanations, examples, clear tendency to move within any of the ordering considered
hints, etc. in subsections A, B, C. These groups are formed based on the
amount of preliminary knowledge, current performance,
learning styles, [14], learning orientations, [10] etc. Such data
can be important while f. ex. updating instructions for future
students and refining contents of questions in diagnostic tests.

V. SYSTEM DESCRIPTION
12 ENCHEVA AND TUMIN

A system prototype is build as a Web-based application Intelligent Tutoring Systems, ITS 2000 Berlin: Springer Verlag (2000)
292—303.
using Apache HTTP server, mod_python module and SQLite
[2] Arrow, K. J. (1951). Social Choice and Individual Values. Wiley, New
database. The mod_python module provides programmable York, (1951).
runtime support to the HTTP server using Python [3] Baker, R.S., Corbett, A.T., and Koedinger, K.R.: Detecting student
programming language. The application components are Web- misuse of intelligent tutoring systems. Lecture Notes in Computer
based users interface, application logic and application Science, Vol. 3220. Springer-Verlag, Berlin Heidelberg New Jork
(2004) 531-540
interfaces written in Python, and a relational database. The [4] Berg, S.: Condorcet's Jury Theorem and the Reliability of Majority
system architecture is presented in Fig. 12. Voting, Group Decision and Negotiation, 5(3), (1996), 229-238
The system contains also an interactive theoretical part [5] Bogart, K. P.: Some social sciences applications of ordered sets. In: I.
(lecture notes), problems to be solved at home, tests assessing Rival, Editor, Ordered Sets, Reidel, Dordrecht (1982), 759–787
[6] Breuker, J., Components of Problem Solving and Types of Problems. In
recall of facts, an ability to map the relationship between two EKAW '94, (1994), 118-136
items into a different context, and conceptual thinking, asking [7] Brunstein, A. and Krems, J.: Helps and Hints for Learning with Web
students to evaluate consequences and draw conclusions, and Based Learning Systems: The Role of Instructions, Intelligent Tutoring
various quizzes. Systems, Lecture Notes in Computer Science 3220, (2004), 794-796
[8] http://www.carnegielearning.com/web_docs/Whitepaper_Edweek.pdf
The users, i.e. expert tutors, teachers, and students interact [9] http://ctat.pact.cs.cmu.edu/
with the system using Web forms. Experts and teachers can [10] http://www.learning-styles-online.com/overview/
submit and update data, while students can only view [11] Doignon, J.-P. and Falmagne, J.-C.: Knowledge Spaces. Springer-
information. Verlag, 1999.
[12] Dowling C.E.: On the irredundant generation of knowledge spaces.
The system provides recommendations on whether or not a Journal of Mathematical Psychology, 37, 1993, pp. 49-62
student needs to take additional classes (courses) based on [13] Jan, N., Shun-Chieh Lin, Nancy P. Lin, Chang, C., A Trend-based
fuzzy dependencies. Prediction System for Web User Behavior, WSEAS Transactions on
Advances in Engineering Education, 5, 2008
[14] Keefe, J. W., Learning style: An overview. In NASSP's Student learning
styles: Diagnosing and proscribing programs, (1979) 1-17. Reston, VA.
National Association of Secondary School Principles.
[15] Koedinger K.R., McLaren B.M. and Roll I.: A help-seeking tutor agent.
Proceedings of the Seventh International Conference on Intelligent
Tutoring Systems, ITS 2004, Berlin, Springer-Verlag, (2004) 227-239
[16] Mayo, M. and Mitrovic, A.: Optimising ITS behaviour with Bayesian
networks and decision theory. International Journal of Artificial
Intelligence in Education 12 (2001) 124-153.
[17] Pecheanu E., Segal C. and Stefanescu D.: Content modeling in
Intelligent Instructional Environment. Lecture Notes in Artificial
Intelligence, Vol. 3190. Springer-Verlag, Berlin Heidelberg New Jork
(2003) 1229-1234.
[18] Renkl, A.: Learning from worked-out examples: Instructional
explanations supplement self- explanations. Learning and Instruction,
12, (2002) 529-556
[19] Stewart, K. L., and Felicetti, L. A., Learning styles of marketing majors.
Educational Research Quarterly, 15(2), (1992) 15-23
[20] Tsovaltzi, D., Fiedler, A., and Horacek, H.: A Multi-dimensional
Taxonomy for Automating Hinting, Intelligent Tutoring Systems,
Lecture Notes in Computer Science 3220, (2004), 772-781
Fig. 12: System architecture [21] Yeum, Y., Lee W., Usability Issues of Web-Based Assessment System
for K-12 and Implementation of PAS System, WSEAS Transactions on
Advances in Engineering Education, 9(4), (2007)
[22] Van Deemen, A., The Probability of the Paradox of Voting for Weak
V. CONCLUSION Preference Orderings. Social Choice and Welfare, 16(2), (1999)
[23] Wood, D.: Scaffolding, contingent tutoring, and computer-supported
learning. International Journal of Artificial Intelligence in Education, 12
The work described in this article provides an approach (2001) 280-292
that helps to locate significant gaps in students’ knowledge. In [24] Zhou, Y., Freedman R., Glass, M., Michael, J. A., Allen A., Rovick, A.
A., and Evens, M. W.: Delivering hints in a dialogue-based intelligent
addition, it gives an opportunity not only to follow the tutoring system, Proceedings of the sixteenth national conference on
learning progress of individuals but also to recognize pattern Artificial intelligence and the eleventh Innovative applications of
clustering’s within a study unit. The course developer can artificial intelligence conference innovative applications of artificial
then easier find out which of the involved reading materials, intelligence, (1999), 128-134
[25] http://design.test.olt.ubc.ca/Intelligent_Tutoring_System
hints and tests need further adjustments and what the future
[26] http://www.cs.ualberta.ca/ zaiane/courses/cmput605/2008-3.html
work should be concerned about.

REFERENCES
[1] Aleven, V., and Koedinger, K. R.: Limitations of Student Control: Do
Student Know when they need help? In G. Gauthier, C. Frasson, and K.
VanLehn (Eds.), Proceedings of the 5th International Conference on
 Problem Based Learning: Obtaining Enzyme
Kinetics Parameters Integrating Linear
Algebra, Computer Programming and
Biochemistry Curriculum
Regalado-Méndez, A. (a), Báez-González, J., G. (b) Peralta-Reyes, E. (a), and Trujillo-Tapia, M., N. (a)
(a) Ingeniería Ambiental, Universidad del Mar-Puerto Ángel, Ciudad Universitaria S/N, Km. 1.5 Carretera
Puerto Ángel–Zipolite, Puerto Ángel, Oaxaca, 70902, México.
(b) Departamento de Alimentos, Facultad de Ciencias Biológicas, Universidad Autónoma de Nuevo León,
Av. Universidad S/N, Ciudad Universitaria San Nicolás de los Garza,
Nuevo León, 66451, México.
E-mail: [email protected]

Abstract-This work is focused on a project that integrates the

curriculum of biochemistry, lineal algebra and computer
programming. The purpose is that students develop a software
tool which calculates enzyme kinetic parameters based on
proposed data. This program calculates such parameters using a
linear regression of one of the linear forms of the Michaelis-
Menten equations; moreover it characterizes the confidence of the
lineal fit with the correlation coefficient. Once the different
proposed steps were accomplished, we concluded that the purpose
was satisfactorily reached with an increment in creative ability.
The important is that the percentage of failure among students
was 57%, 50%, 28% and 18% from 2005 to 2008, respectively.

I. INTRODUCTION
In Mexico, problem based learning (PBL) has not been
developed much, contrary to what is found in European
countries and United States. In the last few years, this type of
learning has become very popular [8], allowing the integration
of different disciplines in to the curriculum. Frequently,
students ask themselves about the usefulness of some courses, Fig. 1. Number of failed and accredited students each year.
as they don’t manage to see any concrete application. To
ensure that students take a primary part in their own academic To improve the Teaching-Learning process in Mexico, it is
formation, the lecturer should assume a facilitator role, and necessary that professors interconnect the knowledge obtained
encourage the students to motivate themselves in searching for by the program curricula not only in a horizontal way but also
possible solutions to a given problem, looking at it from a vertically, with the objective to improve the cognitive
multidisciplinary viewpoint. processes of students in the first semesters, in order to form
As shown in fig.1, students from the Universidad del Mar more critical and less procedural students [9].
have high failure statistics which leads teachers to seek The PBL structure was kept similar to the one adopted by
different pedagogic approaches. A project that allows Offman and Cadet, 2002, with 3 work sessions of 2-3 hours,
integrating courses of biochemistry, linear algebra, and the first with an introductory purpose. In these sessions,
programming is proposed herein. It is intended for use with discussion is encouraged with a minimum involvement of the
undergraduate students of the Universidad del Mar, Mexico. professor (facilitator) [1, 3, 6 and 7]. Each session is spaced by
Until now, it has been applied to students in the Environmental three days; in each a written report that answers specific
Engineering program, but it can be expanded to students of questions is expected.
natural science.

13
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_3, © Springer Science+Business Media B.V. 2010
14 REGALADOǦMÉNDEZ ET AL.

II. PROBLEM DESCRIPTION 4. I did all the indicated activities. YES NO

α- 5. I actively participated in my team and contributed
The PBL problem is: Determine the enzyme
some individual ideas. YES NO
chymoptrypsin kinetic parameters, developing a software
6. I was respectful in team participation (Taken
program. The rates of hydrolysis of N-acetyl-
decisions, agreements, and etcetera). YES NO
phenylalaninamide by α - chymoptrypsin at different
concentrations are as follows [2]: B. Knowledge
TABLE 1 7. I understand all the concepts. YES NO
EXPERIMENTAL DATA 8. Now I could solve the problem the by myself.
[S] (mM) V (mM de Alanine minute-1)
10.5 3.33 × 10-2
YES NO
15.1 4.55 × 10-2 9. From this example, I understand the importance of
20.0 5.56 × 10-2
35.0 7.15 × 10-2 teamwork. YES NO
93.4 10.0 × 10-2
IV. EXPECTED ANSWERS
A. Part I (Session 1)
A. Part I (Session 1)
The facilitator, in this section, should motivate the students
Enzymes are proteins that are highly specialized, due to the
to investigate subjects related with enzyme kinetics. In order to
fact that they catalyze numerous reactions in biological
assist, he can propose the following questions: What are
systems.
enzymes?
In general, an enzyme gives the adequate environment for a
What are they useful for? How do they work? What is their
determined reaction to be, energetically, more favorable. The
reaction mechanism? Do models that allow predicting their
particularity of an enzymatic catalyzed reaction is that it occurs
reaction rate exist? If so, which would be the simplest?
in a specific place of the enzyme, the active site. The molecule
B. Part II (Session 2) that is fixed on the active site and on the enzyme is called
Once the enzymatic kinetics is understood, the facilitator substrate. The enzyme-substrate complex formed is of vital
must introduce the mathematical tool and outline the following importance in order to define the kinetic behavior of the
questions: What is the mathematical form of the models different catalyzed reactions. A simple enzymatic reaction can
encountered? How can we turn the simplest model into a linear be represented by the Michaelis-Menten equation.
form? How can we determine the fit parameters,
B. Michaelis-Menten Equation
mathematically, in a linear equation?
Generally, it is accepted that the catalysis of enzymatic
C. Part III (Session 3) reactions needs the formation of a complex between the
In this section we will apply the programming theory, using enzyme E and the substrate S. This enzyme-substrate complex
flux diagrams to facilitate the program design. To develop ES, can dissociate to reform E and S, or can undergo some
software that will quickly obtain the kinetic parameters for any reorganization and form E and products P. The simplest
case study, using the linear equation of part II, the program mechanism of all can be represented as [2]:
must:
¾ Be able to read the data directly from keyboard ⎯⎯→ ES ←⎯
⎯⎯
k1
→E + P k3
E + S ←⎯⎯ ⎯ (1)
capture or those present in a file with extension .dat. k2 k4

¾ Calculate of kinetic parameters.

¾ Present them on screen and keep save as txt extension When concentration ES is constant, the rate at which it is
file. formed is the same as the rate of its decomposition. Therefore,
¾ Characterize the quality of the adjustment realized. according to the law of mass, the equilibrium state is:
III. STUDENT SELF-ASSESSMENT
k1 [ E ][ S ] + k 4 [ E ][ P ] = k 2 [ ES ] + k3 [ ES ] (2)
In the self-assessment the students are asked to identify their
skills and weaknesses in problems solutions. Students answer
each of the following questions. When he/she finishes, he/she At the beginning of the reaction we have [P] << << [S] and
should grade themselves points between 0 and 10. [S] ≈ cte, with the velocity V of P of formation as V = k3 [ES],
moreover the totality of the enzyme (ETot) can be complexes
A. Attitude with the substrate or not, with [ETot] = [E] + [ES], and the
1. If I didn’t understand how to the begin problem. I maximum rate Vmax is reached when all the enzymes present
tried understanding it. YES NO are involved in the active enzyme-substrate complex, i.e.
2. I was honest with my self-learning (I didn’t copy; I Vmax = k3 [ETot]. Substituting the previous considerations in (2)
did my work without help). YES NO and reorganizing the Michaelis-Menten equation can be
3. I was responsible in my self-learning (I searched for obtained and its description is in the table 2:
information, I always looked more in-depth, I didn’t
stop with the information acquired, and I didn’t V=
[ S ]Vmax
(3)
depend on others to do my work). YES NO [ S ] + km
 PROBLEM BASED LEARNING: OBTAINING ENZYME KINETICS PARAMETERS 15

TABLE 2 Equations (7) can be resolved with the Cramer rule.

VARIABLE DESCRIPTION
Parameters (ao and a1) are given as:
Variable Description
[E] Enzyme Concentration
[S] Substrate Concentration
⎡m ⎤⎡ m 2⎤ ⎡ m ⎤⎡ m ⎤
⎢ ∑ f ( xi ) ⎥ ⎢ ∑ xi ⎥ − ⎢ ∑ xi ⎥ ⎢ ∑ f ( xi ) xi ⎥
[ES] Enzyme-Substrate Complex Concentration (mM)
ki Reaction Kinetic constant
Km Constant for a determined enzyme and a given substrate (mM) ao = ⎣ i =1 ⎦ ⎣ i =1 ⎦ ⎣ i =1 ⎦ ⎣ i =1 ⎦
V Velocity of product formation (mM min-1) 2
m
⎡m ⎤
Vmax Maximum velocity of product formation (mM min-1) m∑ xi − ⎢ ∑ xi ⎥
2

i =1 ⎣ i =1 ⎦
(8)
m
⎡m ⎤⎡ m ⎤
C. Part II (Session 2) m∑ f ( xi ) xi − ⎢ ∑ f ( xi ) ⎥ ⎢ ∑ xi ⎥
As equation (3) is hyperbolic and obtaining the Vmax and km a1 = i =1 ⎣ i =1 ⎦ ⎣ i =1 ⎦
2
parameters directly is complex, a first step is to make the m
⎡m ⎤
m∑ xi − ⎢ ∑ xi ⎥
2

Michaelis-Menten equation linear, as follows: i =1 ⎣ i =1 ⎦

p ( x ) = a0 + a1 x (4) To obtain Vmax and km parameters, from (5) it comes:

Equation (3) can be presented in the next lineal form, which 1 a1

Vmax = and km = (9)
is: ao ao

1 k 1 1 To determine how confident the adjustment of minimum

= m +
V V
{ {{ {
max [ S ] Vmax
(5) squares is, the correlation coefficient can be calculated (r2) as:
( )
f x a1 x ao

⎛ m ⎞ ⎛ m ⎞⎛ m ⎞
m ⎜ ∑ xi yi ⎟ − ⎜ ∑ xi ⎟⎜ ∑ yi ⎟
According to the case, one or the other equation can be ⎝ i =1 ⎠ ⎝ i =1 ⎠⎝ i =1 ⎠
r2 =
adopted. The choice is basically guided by the experimental ⎡ m 2 ⎛ m ⎞2 ⎤ ⎡ m 2 ⎛ m ⎞2 ⎤ (10)
data that, once in the linear form have to be far enough from ⎢ m∑ xi − ⎜ ∑ xi ⎟ ⎥ ⎢ m∑ yi − ⎜ ∑ yi ⎟ ⎥
each other in order to obtain a confident linear curve. The ⎢⎣ i =1 ⎝ i =1 ⎠ ⎥⎦ ⎢⎣ i =1 ⎝ i =1 ⎠ ⎥⎦
following deviations to linearity can be related to impure
substrate or enzyme, to the fact that more than one enzyme Depending of the value of this correlation coefficient, it is
acts on the substrate, or to different types of inhibition of the accepted that:
system [4].
1) Perfect r 2 = 1.0
D. Determination of enzymatic parameters using minimum squares 2) Excellent 0.9 ≤ r 2 < 1.0
Once the Michaelis-Menten linear equation is used, the 3) Good 0.8 ≤ r 2 < 0.9 (11)
minimum squares method is applied, which tries to apply the 4) Medium 0.5 ≤ r 2 < 0.8
best curve upon experimental data. The method involves the 5) Bad r 2 < 0.5
sum of squares of the calculated distances between the ordinate
at an experimental abscissa (x) of the line are polynomial
function P (xi) and the real function f(x) of the experimental E. Part III (Session 3)
data is minimum, i.e. [5]: In this section, kinetic parameters are calculated.

m m
F. Program flux diagram
∑ ⎣⎡p ( x ) − f ( x )⎦⎤ =∑ d
2 2
(6) Before doing the program, a blocks diagram will be
i i i
i =1 i =1 designed, which will help to have a clear idea of the data
needed, of the following steps, and of the desired results. The
Substituting (4) in (6), we observe that there is a family of flux diagram is written in order to be translated in any
linear curves that pass through the points, nevertheless programming language. An example is given in fig. 2.
obtaining the minimum of (6) using the derivate of the Note: this section has greater complexity and the students
function respecting to each of the variables (ao and a1) and develop the cognitive process in accordance to the
setting to zero, a system of linear equations of the following mathematical sequential logic obtained in his/her academic
form is generated: trajectory. In this step, the student constructs new knowledge.

m m
G. Example of developed software
mao + ∑ xi a1 = ∑ f ( xi ) Once the flux diagram is obtained in order to calculate the
i =1 i =1
(7) kinetic parameters of Michaelis-Menten equation, it is
m m m
translated in MATLAB 7.6 language. An example of what can
∑xa
i =1
i o +a1 ∑ x = ∑ f ( xi ) xi
i =1
2
i
i =1
be achieved by students is given, in the format of an
16 REGALADOǦMÉNDEZ ET AL.

executable file, in order to be able to run the program without Square Minimun Method
the use of MATLAB 7.6, as shown in fig. 2. 35
Experimental Date
Linear Adjusment
30
Begin
1/V = 237.3456(1/[S]) + 7.0887

This program calculates enzymatic kinetic 25

parameters by linear regression techniques r2 = 0.9948

1/V(mM/minute)
20
Read the data from File
datosbioq.dat
15

Print data on screen

V and [S]
10

Subprogram minimum
5
0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1
1/[S](mM)
Print Km, Vmax, r^2
In screen

Fig. 4. Linear adjustment of Michaelis-Menten equation.

End
In addition to the parameter values, the correlation
Fig. 2. Example of flux diagram. coefficient is given; in the study case, according to (11) it is
excellent.
If students want to read experimental data, they must check In Fig. 5, velocity degradation profile and experimental data
that the data are correctly written in the required emplacement, with adjustment parameters are shown.
with the right filename (example datosbioq.dat).
To calculate the kinetic parameters we ran the CISSE2009.m 0.11
Michaelis-Menten Kinetic

program from MATLAB 7.6. A window opens that explains Experimental Data
what the program does, as observed in fig. 3. 0.1 Adjusment
V = (0.1414*[S])/(33.4821+[S])
0.09

0.08
V(mM/minute)

0.07

0.06

0.05

0.04

0.03
10 20 30 40 50 60 70 80 90 100
[S](mM)

Fig. 5. Degradation rate profile of Michaelis-Meten equation.

The grade of student’s understanding was greater within the

proposed methodology, given that students participated highly,
showing strong motivation while using programming as a tool
to find the solution of a given problem integrating several
curriculum courses. In Fig. 6 presents a picture of students on
Fig. 3. Enzymatic kinetic parameters obtained.
In fig. 4, the linear adjustment and the experimental data are this activity.
shown, characterizing the adjustment obtained.
 PROBLEM BASED LEARNING: OBTAINING ENZYME KINETICS PARAMETERS 17

Fig. 6. Working students.

Fig. 8. Self-Assessment point.
The most difficult activity for learners was to establish the
flux diagram, as it was in this part where they have to use The methodology proposed considerably reduced the
sequential mathematic logics. student’s failure from 2005 to 2008, as shown in Fig. 9. The
In Fig. 7. we have an analysis of the answers to each percentage of failure in 2005, 2006, 2007 and 2008 was of
question in the student’s self-assessment. Most students had a 57%, 50%, 28% and 18% respectively. Therefore PBL is a
good attitude and understand the concepts. better way to give class.

Fig. 9. Percentage of failed and accredited students each year.

I wish to mention that it is very important to listen to the

Fig. 7. Self-Assessment statistics. students, because they can help to improve the learning
process between professor and students.
In Fig. 8. shown, the self-assessment points of each student
V. STUDENT FEEDBACK
show that only one student self-reported that he failed, because
he likes the traditional learning method more and most were The students stated that, PBL is preferred, because:
comfortable with PBL, as demonstrated by above average i. They can acquire skills at their own pace within a
points. problem solving environment.
The computer programming was a pretext for students’ ii. Learning takes place in a cooperative student-
motivation and they acquired their own responsibility centered, environment.
knowledge to build their own success. iii. It opens their eyes to how university life should be.
18 REGALADOǦMÉNDEZ ET AL.

iv. PBL improved their generic skills. educational role. This supports an updated commitment
v. It makes us feel more comfortable to work in a group to the search for learning alternatives, as well as a
to solve problems. responsibility of institutions to invest in teaching courses
vi. They said that “It really makes us work for ours for the improvement of the educational processes.
studies and we feel intelligent doing it.” ACKNOWLEDGMENT
vii. We feel like engineers when we are doing the group
assignments. The authors wish to thank to colleagues Kathleen Meredith
viii. The given problem excites the students’ curiosity to Brown, Kathleen Dooley, Nina Carol Buck and Aitor Aizpuru
know more. for theirs revisions.
ix. It can be described as a double edged sword, because REFERENCES
we are required to do more than with traditional [1] Dean, R., L., “Laboratory Exercises: Kinetic Studies with Alkaline
learning methods education. Phosphatase in the Presence and Absence of Inhibitors and Divalent
Cations”, Biochemistry and Molecular Biology Education, 2002, Vol. 30
VI. CONCLUSIONS No. 6, pp. 4021-407.
[2] Hemberg T. “Abscising II as an inhibitor of α-quimoptripsinae”, Act
• Once the students concluded each step of the work Chem. Scand., 1967, Vol. 21, pp. 1666.
program, the facilitator of this activity notes a cognitive [3] Magued, I., Innovate Techniques in Instruction Technology, E-learning,
improvement in the global understanding of the involved E-assessment and Education, 2008, pp. 602.
discipline, and the objective is satisfactorily attained. [4] Montgomery R. and Swenson C. A., “Problemas Cuantitativos de las
Ciencias Bioquímicas”, Acribia, España, 1975, pp. 179-186.
• The PBL method is a strong tool to develop the creativity [5] Nieves H. A. and Domínguez S. F., “Métodos Numéricos Aplicados a la
of students, as the prime objective of universities is to Ingeniería”, Continental, México, D.F., 2003, pp. 362-364.
produce critical students. [6] Offmann, B. and Cadet, F., “Problem-based Learning: Redox-active
Disulfides in a Plant Light Switch”, Biochemistry and Molecular Biology
• If the objective was attained for the students which had Education, 2002, Vol. 30 No. 4, pp. 249-254.
the PBL to solve, it is worth noting that the degree of [7] Prausnitz J. M., “Hacia el fomento de la creatividad de los estudiantes”
understanding depends as well of the solidity of the Educación en Química, 1997, Vol. 5 No. 3, pp. 156-159.
knowledge background and the abilities acquired during [8] Savin-Baden, M., “Problem-based Learning: Disciplinary Differences or
Modes of Curriculum Practice? Who promised to deliver what in
the academic curriculum, and of the availability of problem-based learning”, Biochemistry and Molecular Biology
specialists and the leadership of the facilitator involved. Education, 2003, Vol. 31 No. 5, pp. 338-343.
• The use of PBL methodology can make the difference in [9] White H., B., “Problem-based Learning: Commentary Problem-based
the preparation and formation of students, and this is why Learning and Becoming a Physician”, Biochemistry and Molecular
Biology Education, 2003, Vol. 31 No. 5, pp. 337.
it remains important that professors assume their
 Adaptive Assessments using Open Standards
Héctor G. Barbosa León, Francisco J. García Peñalvo, María José Rodriguez-Conde
University of Salamanca, Science Faculty, Plaza de los Caidos s/n, Salamanca, Spain 37008, Technological Institute of Colima, Av. Tecnológico s/n, Villa de
Álvarez, Colima, México 28976
[email protected], [email protected], [email protected]

Abstract - This article presents the conclusions of a research According to the Australian Flexible Learning Framework
workpreparedin the University of Salamanca, Spain, including [3]assessment, especially when is included within a real
the investigation, definition, construction and test of a model to learning task or exercises, could be an essential part of the
define adaptive assessments considering the user´s presentation learning experience, giving to the entire Web site the
needs and also the traditional adaptation process using the level
of complexity for every item. We applied innovatively some
characteristic to adapt itself to the needs of the users.
common open standards and specifications. Conceptualizing the learning process to its basic elements,
we can identify at last the following elements:
• The educational material to be taught by the teacher in a
I. INTRODUCTION classroom.
Students has been adopted the Internet as their first source • The teaching and learning activities that take place in a
of information. The Web and its intrinsic characteristics allow classroom.
the content to evolve from a static to a dynamic organization • The assessment activity to measure the student learning
by including adaptive processes, making the user´s experience and,
to be improved and enhanced [1]. This content, developed in • The report of the score results given by the teachers to the
form of Units of Learning (UoL), include one or several students. This conception is well suitable for the
learning objects and their related multimedia material. traditional educative process.
But we must ensure that those developments have
characteristics like be standardized, compatible and
interchangeable, incrementing the characteristic of actual Web
applications that are cross-platform systems.To achieve those
characteristics, developers defined the Semantic Web, in
which the content is seen in a semantic view instead of been
seen in a syntactical one by using meta-data [2].
Among the huge amount of information in the Internet we
will focus ourselves in instructional and educational
institutions which have been using the Internet to incorporate
the information and communication technologies in the
learning and teaching processes in order to increase the
quality, efficiency and dissemination of education.
We want to emphasize the role of the assessment activity
inside the e-learning process, and how it can help to improve it
to all participants: students, teachers and content designers. FIGURE 1. IMPORTANCE OF ASSESSMENT IN THE E-LEARNING PROCESS
II. IMPORTANCE OF THE ASSESSMENT ACTIVITY IN THE
EDUCATIVE CONTENT In figure I, we characterize the importance of assessment
in the learning flow. As we can note, the tests and evaluations
Traditionally, assessment activity has been seen like task not only are an integral part of the learning process, but also is
aside of the e-learning process and there is a danger in an element that complete and close a circular activity,
focusing research on assessment specifically [3], as this tends contributing as a feedback source for: the users (giving the
to isolate assessment from teaching and learning in general. scores and feedback), for the instructors (by giving support
For us, the assessment activity after the student took the and feedback) and for the instructional designer (to update the
educative lesson, could be seen as an element that closes and contents of the learning system) as well. This circular
complete a circular activity, being an integral part of the conceptualization of the learning process allow us to see the
learning process as well [4]. In addition, the assessment significance of the assessment because it helps to the
activity inside the e-learning process could be used to adapt adaptation of the system by setting a new user knowledge
the system by setting a new user knowledge level, evaluate, level, evaluating and determining new learning profiles, assign
and setting new learning profiles, assign user grades and, in the user grades and, in consequence, performing user content
consequence, performing user content re-adaptation [1].

19
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_4, © Springer Science+Business Media B.V. 2010
20 LEÓN ET AL.

re adaptation. This is how we see the importance of the Those specifications allow the creation of many types of
assessment task for the adaptation process. educational designs following a consistent notation to develop
Conclusively, we would like to mention an assessment a system that could have a homogeneous implementation in
categorization: several courses or learning contexts, giving the systems extra
• Formative assessment: this is an assessment that helps to characteristics like compatibility and exchangeability. In
give a convenient feedback and motivation to the student addition, those specifications define an abstract data model
and do not have scores. Also brings convenient feedback using XML, supporting the deployment of item banks and
to designers of materials. well documented content format.
• Summative assessment: this is a scored assessment and The potential benefit of Computer Based Test (CBT) are
gives a result to measure outcomes and success of the that a test item can attach various multimedia, such as graphic,
learner. audio or video, and reuse the test item easily, mostly if those
• Norm assessment: use the achievements of a group to set are designed and developed with open specifications and
the standards for specific grades and is used in most standards, allowing to different information from test result
universities. from students can be collected and recorded by computers
• Criterion assessment: establish the standard (criterion) effortlessly and immediately in electronic portfolios (e-
and mark students against it. portfolios).
There is another kind of assessment called alternative But most of CBT are linear that makes all students take
assessment. Here, the integration of the assessment activity the same number of test items in a specified order during a
with learning processes and real-life performance as opposed testing section. Nevertheless, this approach seems to imitate
to display of “inert knowledge”. Know as authentic the paper-and-pencil (P&P) test because it merely presents the
assessment it is very much based on the constructivist test in a digital format.
approach that enables students to demonstrate knowledge by IV. CHARACTERISTICS OF ADAPTIVE SYSTEMS
working on authentic task, putting students undercontrol of
their own learning [3]. In contrast to the linear test with fixed test items,
Computer Adaptive Test (CAT), based on Item Response
Theory (IRT) the presentation of each test item and the
III. CHARACTERISTICS FOR INTERNET-BASED SYSTEMS
determination to finish a test are dynamically adopted based
Nowadays, learning systems in the Internet must be on the student´s answers.
flexible and efficient, and one way to accomplish that is to be One aim is to make those systems to work in adaptive
an open and standardized system. The following standards learning systems and, given the fact that the assessment
give to those systems some of the mentioned characteristics: activity is an important and integral part of the e-learning
• The LTS (Learning Technology Standard) process; it is desirable that this evaluation activity could be
specification is a group of agreements about the adaptable as well. If we want the assessment to be
characteristics that a learning element should have. interoperable, compatible, and shareable, we must have to
The use of standards ensure instructional develop a standardized tool [8].
technologies to work with other systems The implementation of the adaptability in the assessment
(interoperability), follow-up information about activity could be performed by giving the instructor:
learners and contents (manageability), generate • Tools to design the assessment with several types of
objects that are usable in other contexts (reusability) resources (text, graphics, video, animations) to improve
and avoid obsolescence (durability) [5]. the student´ understanding in the examination
• IMS Specifications1 are part of the LTS, developed in • Design different types of assessment
1997 as a project of the National Learning • Define content structures and groups of students
Infrastructure Initiative at Educause. Its mission is to • Manage the assessments and the questions linked to each
promote distributed learning environments. For the assessment
IMS many areas require interoperability when • Define the grades for each question and assessment
learning is distributed, thus it details a set of • Define and manage the schedule for the assessment that
specifications that constitute a framework to will be taken.
interchange educational elements [6]. The purpose of our work was to adapt the characteristics
• Inside IMS specification one with special interest for of the questions that are included in one individual’s exam to
our investigation is the IMS QTI (Question and Test their learning style [4],[5]. To do this, the IMS QTI
Interoperability), for sharing test and/or items specification allows the author to link the multimedia
between compatible LMS. It defines a data model for resources that could fit for three different user needs: visual,
the presentation of presentation of questions and test auditive or kinesthetic. That categorization fits with one of the
and the correspondent results. learning styles definition2.

1
http://www.imsglobal.org
2
http:///www.ldpride.net/learningstyles.MI.html
 ADAPTIVE ASSESSMENTS USING OPEN STANDARDS 21

Our aim was to define the primary resource (that is, the author in the definition stage, accordingly to his/her personal
material presented to the user by default) and the equivalent criterion.
resource (presented to the user taking into account his/her One of the main characteristics of the IMS LD (Learning
learning styles, physical disabilities, or preferences). For users Design) specification is its potential to define adaptive
with disabilities the user interface must be capable to make a characteristics covering the student preferences, prior
transformation of the display using flexible controls. This knowledge and/or learning needs. To do this it is necessary to
adaptation is performed by matching the characteristics use this specification in the level B to define individual
defined in the assessment item meta-data with those defined in interactions because in this level we can use some elements
the user profile in the Learning Management System. like <properties> and <conditions>.
Adaptation is another way to adjust to the user needs. This
process is performed when the student is answering the
questions, following the next sequence [9]:
• At the authoring stage, all items are categorized to
estimate their level of difficulty.
• When the user starts the exam, he/she is presented with a
pretest question to estimate his/her ability.
• The “best” next item is administered and the student
responds.
• A new ability level is computed based on the responses of
the student.
• Steps 3 and 4 are repeated until a stopping criterion is
met.
Questions would be given randomly to eliminate cheating.
The new ability level is computed based on equations
described in the figure 1, used in the MANIC [10] system
during the quiz sessions to compute the new estimated ability
level.
The equation (1) is for correctly answered questions and
the equation (2) is for wrongly answered questions.

NewValue = OldValue+OldValue*(level of Question / 10) (1)

NewValue = OldValue+OldValue*(5-level of Question / 10) (2)

V. ADAPTATION PROCESSES

I. Adaptation in the final presentation to the user

The learning object could adapt their final presentation
taking into account the needs or preferences of the users. For FIGURE 2 . ADAPTATION PROCESSES
this, the learning object containing an adaptable exam includes
content for each learning style through the use of a specific The learning style values could be set from values stored
environment. The LMS access the appropriate environment in the user model in the LMS and stored in property elements
that fit to the user learning style or preference, showing the (<locpers-property>, <globpers-property) to perform the
multimedia material [9], [11], [12],[13]. (Fig. 2). adaptation. Finally, the <on-completion> element could be
II. Adaptation in the complexity level used to set the actions that will be done once certain action is
performed. To perform adaptability to the learning style the
Another process of adaptation is the level of complexity LMS access the environment according to the need or
of the questions that are presented to the student (Fig. 2). The selection of the user. From this, the questions set are presented
questions are selected by their level of complexity, taking into to the student applying the adaptation algorithm proposed by
account the response of the student to the last question [14].
answered. If he/she answer correctly then the next question is Adaptation rule RUL1 is to perform adaptability
of the same or higher complexity, if not, then the next question according to the ability level and RUL2 is to adapt the
is of a lower complexity. This is the traditional adaptation presentation to the user learning style.
process used by some developments in this area3. Please recall
that the level of complexity of each item is given by the item’ RUL1= IF <student>::(response,true)
THEN
newLevel = oldLevel+
3
http://www.imsglobal.org/cp
oldLevel(oldLevel/10)
22 LEÓN ET AL.

ELSE style (or, in consequence, for the generic learning style),

newLevel = oldLevel+ selecting the single items from the database that could
oldLevel(5-oldLevel/10). evaluate a specific unit of learning. In this level we include
RUL2= IF <student>::(LS_visual)
THEN
IMS LD labels in the imsld:methods section to define
show item (newLevel,visual) sequencing rules to show the items according to their level of
ELSE complexity.
show item (newLevel,verbal).
C. Third level of integration
VI. PACKAGE CONFIGURATIONS
In this level the testConstructor could construct a package
The IMS CP (Content Packaging) specification is used containing a complete exam for each learning style. This
when is necessary to transfer learning objects (lessons, exams package could be seen as a learning object to supply an exam
or other material) between several Learning Management for a specific learning style, including adaptation rules to show
Systems. In this case we can use this specification to package the items according the level of complexity. To define and
and transfer assessment items between LMS. In the case of a construct the items and test, we consider using open
simple element, the package will contain (a) the manifest specifications and standards to ensure the use of the products
(XML file called imsmanifest.xml), (b) the item (QTI XML in a broader number of LMS.
format file) and, (c) any auxiliary files required by the item. The test object is divided in three main testPart(s), one
The IMS QTI and IMS LD (Learning Design) for each kind of supporting material: audio, video or text.
specifications allow their integration to define a learning Inside each testPart there will be four assessment sections
object to evaluate the knowledge acquired by the students [13]:
when interact with a unit of learning (UoL). The IMS LD • assessmentSectionfirstNormalItem: in this section the level
includes activities of course instruction, modules, lessons and of difficulty that will be presented to the student is set. The
exams in a generic context (those considered as formative idea is to present a normal weight item to the student, if
assessment) to support the recent knowledge or to give he/she responds correctly to it, then a heavier item will be
immediate feedback to the student. presented to the user, if he/she responds right again, then a
The main structure defined in a IMS LD object is the heavy level is set.
manifest, containing the organization and resources. Inside the • assessmentSectionlowWeight: include the item set for
organizations section some elements are described, like the questions with light weight.
roles, properties, activities, environments and methods. The
• assessmentSectionnormalWeight: contain the item set for
integration of the specifications could be done defining tags
questions with normal weight.
and instructions in the imsld:properties to control the visibility
and order of the elements and the imsld:conditions to define • assessmentSectionhighWeight: enclose the item set for
decision structures. questions with heavy weight.
The environments section is a container for the • assessmentSectionsetDifficulty: contains preconditions and
environment elements, each could be used to describe branch instructions to guide the flow to the right
assessment items for a particular learning style. These assessmentSection, in concordance to the complexity level
structures (the environment ones) could be executed by the established.
LMS in parallel, allowing multiple students of different
This approach to organize individual assessment items
learning styles to access their own adaptable elements (or an
into complete exams allows us to include adaptive metadata in
adaptable exam).
a form of precondition and branching instructions. The
We propose three levels of integration [11] , going from
precondition directives govern the access to test sections or to
simple single items to complete exams with adaptability
a single item according to a given condition, and the branching
characteristics.
commands allow to the LMS to select the next section, test or
A. First level of integration individual item based on the fulfillment of a condition. We
want to delineate this information to construct the adaptable
In this level, the testConstructor could integrate a single
navigation taking into account the weight or item difficulty.
item in a package containing the item description and the
auxiliary files or the reference to them. The aim for this
package is to be exported to the LMS, to be part of complete VII. ADAPTIVE TEST
exams compatible with the IMS QTI specification or to be
referenced in learning objects using IMS LD labels. An adaptive test is a learning object organized in such
way it can be used for compatible LMS to give each student
with an adapted test using the parameters stored in the user
model. The authoring process allows the definition and
B. Second level of integration construction of different products [11]:
In this level the testConstructor could define a group of single
items to construct a complete exam, for a specific learning
 ADAPTIVE ASSESSMENTS USING OPEN STANDARDS 23

• As separate single items that could be referenced by this group into two: for the first group we applied the test with
compatible external tools using the IMS QTI questions and accompanying multimedia material that matches
specification. their preferences of presentation (audio, text or video); for the
• As complete test that could be used in a session by LMS second group we applied a test that did not match these
as a part of a learning activity. preferences. Also, we applied a test a test to determine the
• As a learning object referenced from a learning unit learning style of each student, prior to the adaptive test.
defined using the IMS LD specification in which some Please, recall that this test also perform an adaptation in
conditions are defined to select the testPart using the the level of complexity of each item, for this we organize the
value stored in the user model. items grouping them according this characteristic and, in a
The elements to perform the adaptation are: higher level, we set a classification according to the
• User preference: stored in the user model with values of multimedia material they are refereeing to.
“visual”, “audio” or “text. The aim of this activity is to evaluate our thesis
• Result of the first normal-level question stored in the hypothesis that, if a student is presented with an adaptive test
unifiedResult variable, with values of “Y” or “N”. This that matches its preferences of presentation of the
variable is special because look to unify the result for the accompanying multimedia material he/she could average
different kind of questions (simple choice, multiple better results in a test.
choice, etc.). The evaluation included the following statistical test:
• The result stored in the following questions in the variable • Final note: To check if the final note of the test is higher
unifiedResult, with values “Y” and “N”. for the students with adapted test.
In the authoring process the author select from the list, • Time to response: To check if the total time to answer the
those items that will be included in the test. The tool shows the test is lower when an adapted test is presented to the
items grouped by the multimedia files referenced by them. student.
Once the selection is made, the tool construct the XML-QTI • Level of satisfaction: To check if the level of satisfaction
file with the testPart, assessmentSection sections and the reported by the student is higher when the test is adapted
reference to the external QTI files for each question and to his/her preferences.
references to the external multimedia files as well. At the end, For the final note test the results were: the average final
a file with the IMS CP instruction is generated and packed as note obtained by the students with an adapted test was 6.6
.zip format; this file is the one that will be used by the LMS. points, whilst for students without adapted test, the final note
Also in the authoring process is possible to define the was of 5.9 points (t test = 0,836).
following kind of test: In the time of response the results were: the average time
• Sequential test: This test does not include branch rules, for an adapted test was 17 minutes whilst for an not adapted
presenting the questions in the sequential order they were test was 20 minutes ( t test=0.964).
selected from the list. The level of satisfaction was obtained from the students
• Random test: This test does not include branch rules, but themselves from a poll after they answer the test. It cover level
the ordering label is set to shuffle. of satisfaction from 1 (lower satisfaction) to 5 (highest
• Adapted test: This test includes the adaptation rules with satisfaction). The results were: for an adapted test 3.9 and for
the branchRule and preCondition labels to control the not adapted test 3.6 (t test=0.821).
items presented to the users. Bring to mind that, the statistical test was applied only to
verify the results for the adaptation process in the presentation
VIII. ADASAT TOOL of the multimedia material, but this test also performs the
adaptation in the complexity level for all participants.
We develop an authoring tool to construct items and test
Also, we think that these results must be taken with some
with adaptation characteristics, following open standards and considerations at the time to make generalizations. In other
specifications like the IMS QTI, CP and LD and the XML words the results obtained in this work may change in other
language. The aim is to allow to the author to define items and
learning environment and stated that this is an internal work.
test using a graphical interface starting from the definition of Upon this, we consider that our conclusions are a science
single questions to the construction of complete test packaged contribution for future works in this area.
using IMS CP and delivered in a .zip package [13], [14].
The application covers the following activities: (1) X. CONCLUSIONS
Question Construction management, (2) Test Construction
management, (3) User management, (4) Test management. Online assessment is an important step inside the e-
learning process, helping to improve the learning and teaching
IX. RESULTS OBTAINED experience.
According to the new developments in the area of e-
To verify that an adapted test is better for the students in learning, we can see that most of these developments look to
terms to ease their answer and understanding, we apply an be compliant with accepted standards like the IMS QTI. This
exam to evaluate the level of knowledge in English language gives the convenience to those works to be interoperable and
to a group of students in the University of Salamanca. We split adaptable to several learning platforms. In concordance,
24 LEÓN ET AL.

referring to the assessment activity, we can think that it must [3] Booth, R., Berwyn, C. The development of quality online assessment in
vocational education and training.Australian Flexible Learning
be interoperable as well, because it is one element of the Framework, Vol.1, 2003, 17.
e-learning process and play an important role inside this task.
Adaptability is another key factor in assessment. Given [4] Barbosa, H., Garcìa, F.,An Authoring Tool to Develop Adaptive
Assessments.In proceedings of 2006 Webist Congress, Setubal,
the fact that assessment is an important element of the e- Protugal, 2006.
learning process and the fact that this process look to be
interoperable, then we can think that the assessment tool could [5] García, F., Berlanga, A., Moreno, M, N., Carabias, J. “HyCo and
authoring tool to Create Semantic Learning Objects for Web-based E-
be used with different educative content administrators with Learning Systems”.In Web Engineering Fourth International
different conceptualizations and ways to design an ways of Conference, ICWE 2004. Published in N.Koch, P. Fraternali, M.
purposes to the students. To face this situation it is necessary Wirsing (Eds.). Series Lecture Notes in Computer Science. Springer
to develop a test with different types of resources, different Verlag. Vol. LNCS 3140, pp. 344-348.
kind of assessments, groups of students, kind of questions, [6] Berlanga A., Garcia, F., “Towards Semantic Metadata for Learning
etc. Elements”. In proceedings of the fifth. International Conference on
Information Technology Based Higher Education and Training ITHET
Under this conceptualization, we created and applied an 2004. Y. Akpinar (Ed.) (2004) pp. 572-577.
adaptive test to a group of students to evaluate the hypothesis
that, if an assessment tool is adapted to the user`s needs or [7] Barbosa, H., Garcia, F. “Setting and Sharing Adaptive Assessment
Assets”. In proceedings of the SIIE congres, Leon Spain, 2006.
preferences we can reduce the time to answer the test while
we increase the level of satisfaction of the user at the same [8] Barbosa, H., García, F. “An Authoring Tool for Adaptative Assessment
Items”. In proceedings of IAIRA Congress. Bucharest, Romania, 2006.
time.
[9] Barbosa, H., Garcia, F. “An Authoring Tool to Construct Adaptive
ACKNOWLEDGMENT Assessments”. In proceedings of the 2006 EIAE Conference.Online,
2006.
We want to thank to the group KEOPS (ref. TSI2005-
00960) of the University of Salamanca for their contributions [10] Mia,k., Stern, Woolf, B. “Curriculum sequencing in a Web-based tutor”.
Lecture Notes in Computer Science, 1452, 1998.
and ideas for the development of this work. In addition, we
want to thank to the Education and Science Ministry of Spain, [11] Barbosa, H., Garcia, F., Rodriguez, M. “Defining Adaptive
Assessments”.In proceedings of the Webist Conference, Barcelona,
National Program in Technologies and Services for the Spain, 2007.
Information Society, Ref.: TSI2005- 00960.
[12] Barbosa, H., Garcia, F. “Constructing Learning Objects for Adaptive
Assessments”. In proceedings of IASTED conference, Chamonix,
HéctorBarbosa thanks the National System of Technological France, 2007.
Institutes (SNIT–Mexico) for its financial support.
[13] Barbosa, H., Garcia, F., Rodriguez, M. “A Tool for Online Assessment in
REFERENCES Adaptive e-Learning Platform”.In proceedings of the eUniverSAL
conference. Virtual, Salamanca, Spain. 2008.
[1] Barbosa, H., Garcia F. Importance of the online assessment in the
elearning process. 6th International Conference on Information [14] Barbosa, H., Garcia F., Rodriguez, M. “Construction of Assessments
Technology-based Higher Education and Training ITHET & IEEE, with Double Adaptation Processes”. In proceedings of EIAE Congress,
Santo Domingo. Virtual, 2008.

[2] Barbosa, H., García, F., Adaptive Assessment Tool for the Hypertext
Composer”, ICTE2005. In proceedings of the 3th.Internatonal Conference
on Multimedia and Information and Communication Technologies in
Education: m-ICTE 2005, Cáceres, Spain, Vol.l 1, pp. 16-20.
 Online Academic Advising Support
Fawzi Albalooshi1 and Safwan Shatnawi2
College of Applied Studies
University of Bahrain
Sakhir Campus, Bahrain Kingdom.
1
[email protected], [email protected]

Abstract- The paper addresses academic advising – an and graduation support system is explained in Section 4 and
important issue that is often not given enough attention by the timetabling support system is explained in Section 5.
students and advisors alike. An online advising system is
presented that can be utilized by students, advisors, and course 2. LITERATURE SURVEY
timetable planners. Students are given informative advice on
which courses to register for in the next semester and are
Although the system described in this paper was primarily
informed of their remaining graduation requirements; advisors motivated by our need to optimize advising, to determine the
are able to see students’ progress towards their graduation courses to be offered and to utilize existing information
requirements; and timetable planners are given statistics on available in the university registration system, there is no
courses and sections requirements for the coming semester. doubt that academics worldwide agree that proper advising is
an important factor for students’ successful progress in
higher education. Many studies have been conducted to
Index Terms- Academic Advising, Automated Advising, Course confirm this matter. Examples of such studies are reported in
Scheduling, Proactive Advising [1] and [3]. As a result, many academic institutes investigated
the use of computer technologies in academic advising to
1. INTRODUCTION overcome the difficulties experienced with traditional
Student advising is one of the important challenges facing methods of advising. A sample list is as follows:
academics, but direct consultation between instructors and
students is another problem. Many students do not take the Reference [1] present a systematic view identifying key
time and effort to see their advisors to plan their timetable subsystems grouped under basic study planning and high
before registration resulting in many registration issues and level advising that can be automated to the benefit of advisers
long queues for advising at registration time. Another and advisees. Under basic study planning, the authors suggest
problem facing academic institutes is planning ahead for the that a number of basic queries be available such as course
courses to be offered for the coming semester and availability, courses in which a student can register, degree
determining the number of sections for each course. This completion requirements, degree credit transfer possibilities,
difficulty is due to the fact that exam results for the courses and time constraints. Reference [2] reports a system that
in which students are registered in the current semester are offers advisees up-to-date online advising and related
not available until exams are over at the end of the semester. information including a recommended list of courses in
Determining students’ graduation status is another time which a student must register in the next semester in order to
consuming effort for academics worldwide. complete his degree requirements. The system has a web-
based main page through which system users such as
In this paper we present a system that accesses students’ students, faculty, and administrative staff are allowed access
transcript records and generates a report for each student to their respective sites. Reference [3] presents the
indicating the courses in which he must register in the motivations to develop FROSH [4] the automated advisor for
coming semester based on rules explained later in the paper. freshman. Reference [5] presents a collaborative course
Such details can only be exact towards the end of the recommender system that recommends elective modules to
semester once grades are input into the registration system. students based on the core modules they have selected. A
During the course of the semester, the proposed system can “More Like This” recommender offers students similar
generate the students’ requirements for courses including the modules to their first choice of electives in cases where no
number of sections required for the coming semester. This places are available or where timetable clashes occur.
will enable the timetable committee to decide on the courses Reference [6] presents InVEStA, an expert system for
and sections to be offered and prepare the most suitable advising students on the courses in which they must register
timetable for the students. Once grades are entered into the next. The recommender system also generates a semester
system at the end of the semester, a more accurate list of schedule based on the courses offered for the semester and
courses and sections can be generated before the start of the the student curricula. Reference [7] presents a web-based
new semester. The proposed system helps students in interactive student advising system for course plan selection
advising, graduation, and timetabling. The students’ advising and registration using Java framework.

25
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_5, © Springer Science+Business Media B.V. 2010
 26 ALBALOOSHI AND SHATNAWI

The works cited in this section are a selection of many TABLE 2

STUDENT TRANSCRIPT TABLE EXTRACTED FROM THE REGISTRATION SYSTEM.
systems developed by academic institutes for their specific
programs that utilize the power of computer technologies to Field Description Data type Domain Example
make student advising easier, accurate, and available to all.
ID Student's ID Number integer 20081234

3. REGISTRATION DATA DESCRIPTION code course code string A{A*}d{d*} CSA111

Our university registration system maintains students’
grade grade obtained string A,B,C,D,F,B+, A+
records, their transcripts, and courses details. Students’ C+,D+,B-,C-
personal details include name, identification number, major ,A-,W,WF,IS,S,
etc. Students’ transcript details include details of courses a
student has completed, the grade obtained in each course, repeated how many times number 0..3 0
semester GPA and completed credit hours, and cumulative this course was
GPA and credit hours. All registration information just taken
described can be accessed as an HTML format that cannot be
handled directly. Web Semantic and text processing is Year the date(year) number dddd 2008
necessary to extract the required information for our next when this course
processing phase. Tables 1 and 2 show student information is registered
extracted from the registration system using our text
semester the semester number 1..3 1
processing approach.
when this course
was registered

TABLE 1
STUDENT INFORMATION TABLE EXTRACTED FROM THE REGISTRATION SYSTEM.
3.1 Text Processing Algorithm
Field Description Data type Domain Example The algorithm used to process the students' transcript file
follows the same steps used in third generation languages
Name Student name string A-Z FAISAL ALI
AHMED AJAJ
compilers. It scans the file character by character, identifies
tokens, and takes actions according to the recognized token.
ID Student's ID Number integer 20081234
Pseudo code
Degree Students string A-Z Office
program (major) management 1- Open the transcript's file for input
S_Cr_hrs Semester Credits number 0-21 9 2- while not end of transcript file do
Attended
3- read the file character by character and identify
C_Cr_hrs Cumulative number 0-180 30 tokens
Credits Attended
4- check if the token is one of the following identifiers:
S_GPA Semester GPA number 0-4 3.2
[course , GPA , name , status, degree, year,
GPA Cumulative GPA number 0-4 3.5
semester, passed]

5- for each identifier in Step 4, take the appropriate

action to extract the required information.
Example of action
Theory:
let  , and β, represent any string, X,C and Y represent
single non terminals (identifier), and β represents a
terminal symbol (information[section, course, name,
cumulative GPA]).
→β
β → X |Y |C
X → Name
Y → ID
C → GPA
 ONLINE ACADEMIC ADVISING SUPPORT 27

Application: The advising process is then invoked using database queries

→ Name: Ahmed Mohammed ID:1001234 GPA:3.2 and data management procedures. Students’ transcripts are
Name → Ahmed Mohammed compared against their programs and the list of courses for
ID → 1001234 which a student must register in the new semester is
GPA → 3.2 generated. For a student, the process determines all
unregistered courses and the prerequisite for each. For each
4. ADVISING AND GRADUATION PROCESSES
unregistered course, if the prerequisite has been successfully
4.1 Data Description
completed the course is listed and can be registered. The
The information derived from the registration system as
process is then repeated for all previously unregistered
described in Section 3 is mainly related to the student study
progress as shown in Tables 1 and 2. The college offers 14 courses and a list of courses in which the student should
programs. In order to generate the list of courses required for register in the following semester is generated. This process
students in different programs, the registration requirements is repeated with every student. A sample report generated by
for all programs must be input to the system. For each this process is shown in Fig 1.
program, the list of courses along with their titles, codes,
credit hours, and prerequisite details must be maintained as The student is considered to be graduated if he/she has
shown in Tables 3 and 4. Each student’s study progress is successfully completed all courses in the program with a
compared with his program as explained in Section 4.2.
specific cumulative GPA and major GPA. The system checks
4.2. Process Description whether the student meets the graduation requirements or not.
The advising process starts by querying the existing Students who satisfy the study program graduation
registration system for student transcripts. For each student, a requirements are considered graduated and a special report
customized integrated web browser component gets the for each student is generated as shown in Fig 2. The report
student’s transcript as an HTML format which is then shows all completed courses, the grade attained in each
converted to text file format. The text file processing process course, and the Cumulative and Major GPAs.
is then invoked to convert textual data to database tables as
shown in Tables 1 and 2.
TABLE 3
TABLE HOLDING DETAILS OF COURSES FOR A PROGRAM

Field Description Data type Domain Example

Code Course code string A{A*}d{d*} CSA111

Title Course title string A-Z Introduction

to PC

Cr_hrs Credit hours number 0-6 3

Type Type of the course number 0-1 1

major course, or
service course

Lab_hrs Lab hours number 0-4 3

TABLE 4
TABLE OF COURSES AND THEIR PRE-REQUISITES

Field Description Data type Domain Example

Code Course code string A{A*}d{d*} CSA113

Pcode Prerequisite string A{A*}d{d*} CSA112

Fig. 1. Screen snap shot for courses to be registered next semester.
code
28 ALBALOOSHI AND SHATNAWI

student is registered. The data table shown in Table 6 holds

the list of courses for which students should register along
with the program in which the student is registered. The
data table shown in Table 7 holds the list of courses for each
program. Both tables are extracted as explained in section
5.2.

Fig. 2. Screen snap shot for graduates report.

Fig. 3. The current schedule web page snaps shot - registration system.
5. COURSE AND SECTIONS STATISTICS GENERATION
TABLE 5
TABLE HOLDING CURRENT SEMESTER SCHEDULE.
5.1. Data Description
Field Description Data type Domain Example
During the current semester, details of courses and sections
to be offered for the next semester are requested by the ID Student's ID Number integer 20081234
registration deanship. In the current semester while students
code course code string A{A*}d{d*} MGTA222
are attending courses, it is difficult to predict accurately
which courses they will pass in order to determine the list of Year the date(year) number dddd 2008
courses and their sections which must be offered for the next when this course is
registered
semester. Therefore, reading students’ transcripts on its own
as explained in Section 4.2 will provide us with incomplete semes the semester when number 1..3 2
information as details of courses in which students are ter this course was
currently registered must be taken into consideration. To registered
read currently registered course details, the system accesses
the students’ current HTML formatted schedules as shown in
Fig 3 using a customized integrated web browser component TABLE 6
TABLE HOLDING COURSES TO BE REGISTERED FOR
and converts it to text document format. The text document Field Description Data type Domain Example
in return is parsed using the algorithm explained in Section
3.1 with a different set of identifiers such as ID, Code, Year, Stno Student ID number dddddddd 20061234
and Semester. This information is then stored in a separate
Code Course code string A{A*}d{d*} CSA112
table as shown in Table 5. In order to generate the statistics,
it is considered that the students will successfully complete PID Program ID Number 1..20 5
their currently registered courses.

Courses completed and currently registered are used as a

base to generate courses to be offered and their section
statistics. In order to specify courses for different programs,
it is necessary to determine the program in which each
 ONLINE ACADEMIC ADVISING SUPPORT 29

5.2. Process Descriptions

The courses to be offered and sections generation process
starts by storing the list of courses to be registered described
in Section 4.2. A code for the program in which each
student is enrolled is also stored in the table shown in
Table 6. The list of sections for a course to be offered is then
generated using the information stored in this table. Courses
are grouped under a program and sections are grouped under
a course. A limit for a section size is set e.g. 25. A report as
shown in Fig 4 is generated showing the number of students
in a program that can be registered in each course.
The main challenge in this process is students’ grades for
the current semester must be available. It is; therefore, better
to execute this process at the end of the semester. To
overcome this limitation and prepare the timetable for the
coming semester before the end of the current semester, an
assumption can be made that all students will pass all courses
in which they are registered. Sections for the coming Fig. 4. Screen snap shoot for time table support system.
semester are then offered, bearing in mind that there will be a
certain failure rate which could result in the possible closing 7. FUTURE WORK
of courses or sections. The process by which courses or The existing system is tailor made for the programs in our
sections can be closed must be simplified. Timetable college. We plan to make improvements to the system so
adjustments can be performed once the current semester that it can be used easily by all other departments and
grades are input to the registration system and processing colleges. We are also strongly motivated to produce a
outlined in this section is repeated using fresh student timetable and load allocation system that interacts with the
information. system described here to generate a timetable that
specifically suits students’ course registration requirements
6. CONCLUSION and at the same time efficiently allocates courses to
The paper presents an online advising system that is run by instructors.
the department. The system is used to advise students ACKNOWLEDGMENT
enrolled in different programs on which courses they should Thanks goes to deanship scientific research at the
register for in the next semester to stay on the correct study University of Bahrain for supporting research published in
path and graduate within the expected time frame. A this paper. We also would to thank Ms Alexandra Fraser for
graduation report can be printed showing the list of courses a proof reading the paper and the anonymous reviewers for
student has completed along with the list of courses to be their comments and feedback. Finally, our thanks also go to
completed if any. The system generates a list of courses and the conference organizers for publishing our work.
the number of sections for each course to be offered in the
coming semester. Such information is based on the courses
completed by the students enrolled in each program and on REFERENCES
the remaining courses to be completed. The system eases [1] P. Bailes , X. Li, B. Lovell, S. Sadiq , X. Zhuo, “The Academic
Advisor’s Apprentice – An Automated Academic Advising System,”
advising and graduation processes, greatly improves CSEE Strategic Development Proposal.
accuracy, and helps in timetabling courses for the coming [2] O. Marques, X. Ding, and S. Hsu, “Design and Development of A
semester. Web-based Academic Advising System,” in 31st ASEE/IEEE Frontiers
in Education Conference, October 10-13, 2001 Reno, NV.
[3] R. M. Siegfried, A. M. Wittenstein, and T. Sharma, "An Automated
TABLE 7
Systems for Course Selection and Scheduling," Journal of Computing
TABLE HOLDING COURSES BELONGING TO A PROGRAM Sciences in Colleges, vol. 18, no. 3, pp. 17-25, February 2003.
Field Description Data type Domain Example [4] R. M. Siegfried, "FROSH: An Expert System for Freshman
Advisement," Journal of Computing in Small Colleges, vol. 9, no. 2,
pp. 88-98, November 1993.
Code Course code String A{A*}d{d*} CSA112
[5] M. P. O’Mahony, and B. Smyth, “A Recommender System for On-line
Course Enrolment: An Initial Study,” in RecSys’07, Minneapolis,
PID Program ID Number 1..20 5
Minnesota, USA, October 19-20, 2007, published by ACM.
[6] D. Pokraja, and M. Rasamny, “Inetractive Virtual Expert System for
Advising (InVEStA),” 36th ASEE/IEEE Frontiers in Education
Conference, San Diego, CA, October 28-31, 2006.
[7] V. R. Naini, R. S. Sadasivam, and M. M. Tanik, “A Web-based
Interactive Student Advising System using Java Frameworks,”
Proceedings of the SoutheasternCon, pp. 172-177, IEEE 2008.
 Using Computer Game to Share knowledge
1
Mazeyanti M Ariffin, 2Nazleeni Samiha Haron @ Baharon and 3Mohd Izad Che Muda
1,2,3
Computer & Information Sciences Dept
Universiti Teknologi PETRONAS
31750 Tronoh, Perak, MALAYSIA

Abstract— sharing knowledge is the key dimension of a

learning organization. Organization that wants to sustain II. LITERATURE REVIEW
their competitiveness must encourage people to learn from A. Taxonomy of knowledge
each other or in other words to share knowledge among
other. Variety mechanisms to share and transfer knowledge Referring to [8] knowledge is actionable information
among knowledge workers are being used nowadays. These that leads to relevant action [9]. Reference [9] further
knowledge sharing mechanisms relatively depends on the added that knowledge is defined as human understanding
nature and the type of knowledge to be transferred. Most of specialized field of interest that has been acquired
organizations are using mentoring, face-to-face meeting, through study and experience. An individual collect
cross-functional and electronics means to share knowledge. knowledge based on learning, studying, thinking and
Scholars claimed that simulation game can encourage experience. People also get knowledge by familiarizing
human learning as players are able to make meaning during with situation, condition or problem in a specific domain
the play. This paper proposes to use simulation game as an or area. According to [9], knowledge includes perception,
alternative to share knowledge. A simulation game called skills, training, common sense and experience.
FireHazard is developed and tested to selected participants. Knowledge helps people to make meaningful conclusion
This paper also discussed on the development and the result on certain problem or condition. It is know-how or
obtained from the study. familiarity of how to do something that enables a person
to perform a specialized task [9]. Referring to [10] further
Keywords — simulation game, human learning, knowledge added knowledge is the ability to turn data and
sharing mechanism information into effective action.
Knowledge can be categorized into two forms; explicit
I. INTRODUCTION and tacit knowledge [11, 12]. Tacit knowledge is embed
As inspired by [1], knowledge has become a new basic in human mind and stems to individual experience [9] ,
economic resource as compared to capital, natural therefore it is hard to articulate or express. In contrast,
resources or labor. Organizations start to realize the explicit knowledge is knowledge that can be codified such
importance of knowledge [2] and its strategic impact as manuals, documents, reports, white papers and
[3,4,5] to sustain its competitiveness in a dynamic spreadsheets which are easier to share among individuals.
knowledge economy. As many organizations aim to ASHEN framework introduced by [13], splits tacit and
increase profit by reducing their production cost, they explicit knowledge into six other types; documents,
made their operations globalize [6,7]. While this can methods, skills, relationships, talent and experience.
reduce the operation cost, it imposes new challenge to B. Knowledge sharing
organization. How to ensure knowledge that resides in the
organization can be shared and transfer effectively despite Knowledge sharing is defined as the transmission of
their geographical location? knowledge from one source to another source and use of
the transmitted knowledge [9]. He further adds that
As such, researchers have proposed techniques and knowledge sharing only happen when there is
mechanism for effective knowledge transfer and sharing. transmission from the sender and absorption by the
One of the methods introduced is by using game and receiver. Referring to [14] support this claim by defining
simulation. It has been proven that this method can aid in knowledge sharing as when the receiver’s (e.g., group,
enhancing learning. Therefore, we have opted for this department, or division) performance is affected by the
method as a mean for transfer and sharing knowledge on experience of the sender. Reference [15] and reference
fire hazard. Fire hazard information is one of the crucial [16] point out that knowledge sharing is a key of a
knowledge that must be possessed by all staff in any learning organization. And learning process only occurs
organization. This paper presents the theories behind when knowledge in one part of an organization is
knowledge sharing and computer game as well as the transferred effectively to other parts and it is used to solve
description of the game that has been developed to problems or simply to create new and creative insight.
validate our study. The success of knowledge sharing can be evaluated
This paper is segmented into several sections : section through changes in performance of receiver [14].
II discuss on literature review to support the study, section
III talks about methodology used, section IV discuss the
result and discussion.

31
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_6, © Springer Science+Business Media B.V. 2010
32 ARIFFIN ET AL.

C. Knowledge sharing mechanisms In some organizations, people share knowledge through

transfer of people, forums (internal
conferences/meetings), boundary spanners, periodicals,
Having known that knowledge can exist in many forms; audits, benchmarking, best practice, guidelines and
organizations need to use different way or mechanism to international teams [17]. Referring to [15] points out that
share knowledge within the organization. It depends on mentoring, chat room, personal intranets and face-to-face
the nature of knowledge. Reference [17] proposed that the conversations also being used by organizations to share
type of knowledge sharing mechanisms is highly depends knowledge. Thus, before organization invests on certain
on the type of knowledge to be shared. As knowledge knowledge sharing mechanisms, they need to identify the
sharing mechanisms are categorized into ‘reach’, goal and the nature of knowledge to be shared. The
‘richness’, ‘tacitness’ and ‘embedness’, mechanisms that ultimate goal of this knowledge sharing mechanism is
has high level of ‘reach’ are more suitable to create facilitate learning process, increase the absorption level of
awareness while mechanisms with high level of ‘richness’ the receivers, and eliminate the geographical barrier.
are often more effective in transferring knowledge among Many scholars like [18] claim that the effective tool to
individuals as shown in Table 1 achieve this goal is to use computer-aided systems.

TABLE 1: Determinants of ‘Rich and ‘Reachness’ [17]

D. Game and simulation
Mechanism’ Desired elements General guidelines Researchers such as [19] and [20] starts to realize the
characteristics potential effects of games and simulation in instructional
Reachness High number of How many receivers is
receivers the mechanism likely to
field years ago. According to [21], game consists of rules
reach at one time? that describe allowable player moves, game constraints,
What is the likely cost privilege and penalties for illegal actions. By referring to
for potential receivers? [22] describe game as rule-governed, goal focused,
Ability to overcome How easy is the microworld-driven activity incorporating principles of
geographical barrier mechanism to reach gaming and computer–assisted instruction while in
receivers in different contrast simulation is the one that consist real settings,
location?
What is the cost
participants’ role, playing rule and scoring system [23].
involved? By referring to [21], games and simulations carries
Ability to overcome Is the knowledge stored distinct criteria and functionalities such as (1)observable
temporal barrier in people or non-people characteristics used in games is more to ‘drawing cards,
medium?(people medium moving pieces etc while simulations provide ‘a scenario
are assumed to be less or set of data’ which to be addressed by the players (2) the
capable in overcoming goal of game is to provide competition and winning state
temporal barrier as they
can be lost more easily
while in simulation participants are given serious
when people leave the responsibilities, and conducting feasible job procedures
organization and possible consequences (3) event sequence in game is
Is the knowledge with linear while simulation provides ‘branching’ or
one or more people? ‘nonlinear’ event (4)rules and regulations in games are
(mechanism in which unnecessarily to be realistic however those in simulations
knowledge is spread to are authentic and closely related to real world.
many individual is more
likely to withstand There are lots of research being done in game and
turnover of people than simulation as a method to enhance learning such as
mechanism which [24,19, 25,26, 27]. Reference [28] proposed that
concentrate with only
one person) simulation can help people to learn without the need to
Ability to overcome What is the likelihood of develop costly training simulations. In other research
functional/departmental the ‘outsider’ coming in conducted by [25], they proposed that computer
barrier contact with mechanism? simulation may be used to help in adults’ learning skills
Richness Ability to transfer a lot How much information
of information at one the mechanism can
particularly in business skills. Simulations and games are
time transfer at one time? well accepted as a method that has impact on teaching and
What is the cost? learning especially in military and aviation settings
Ability to transfer a lot What are the [27,28]. Reference [29,30,25], further support these
of information of communication tools that
different nature at one likely to involve? Can
claims by saying games and simulation are a powerful
time the knowledge receivers alternative to teach and learn as compared to conventional
meet face-to-face with ways since this method is able to facilitate experiential
the knowledge learning. Other study done by [30] suggests that computer
resource/originators?
games help to enhance learning process and retention of
High interactivity Can the receiver and
between receiver and knowledge holder meet knowledge in participants.
sender face-to-face? At what
cost?
III. METHODOLOGY
Focusing on sharing knowledge activity and human
learning, authors developed a simulation game called
FireHazard using evolutionary prototyping phases. The
 USING COMPUTER GAME TO SHARE KNOWLEDGE 33

software used was Game Maker, Game Maker Language scenario, the fire will not fade away and deteriorate. The
(GML) Anime Studio, Adobe Photoshop, Audacity, PHP, FireHazard game is purposely designed to simulate three
MySQL and Apache HTTP Server. different scenarios to maintain consistency. Scenario A
depicts a situation of a Class A type of fire which caused
by ordinary combustible materials such as rubbish and
A. Simulation game architecture paper. Player must decide what to do in this situation and
Figure 1 illustrates the architecture of FireHazard. It which fire extinguisher he must use to put off the fire
consists of four main components which are the game, based on the clue or hint given in the scenario. Scenario B
knowledge repository, scoring system and score database. of the FireHazard game depicts fire in a kitchen caused by
Class A and Class C type of fire. Player should be able to
identify and use fire extinguisher two different fire
extinguishers to put off the fire. Player must think
instantly as the fire is spreading fast and rapidly. Scenario
C illustrates another different scenario which in this case
in a chemical laboratory caused by Class D type of fire. If
player use incorrect type of fire extinguisher, the fire will
grow bigger and uncontrollable.

C. Play the game

Player is required to listen to set of tutorials and videos
on how to use fire extinguisher and how to identify the
type of fire. Later, players are required to play the
FireHazard game to test their ability to apply knowledge
that being transferred to them. Scores will be given if they
manage to successfully apply what they have learnt and
FIGURE 1: Architecture of FireHazard marks will be deducted if they failed in their attempt.
Refer to Figure 2.
Game is the main feature of the application that
contains the client module. The client module provides the
user interface of the application and is linked to
knowledge repository and scoring system. Knowledge
repository stores video and text tutorials regarding how to
use fire extinguisher and how to put out fire properly. The
stored information can be viewed via the Game interface.
Furthermore, scores that are obtained from the game will
be sent to scoring system for calculation using pre-
determined formulas. The scores will then be stored in the
score database. The overall score obtained by each player
can also be retrieved and will be displayed at the Game
interface. The summarized description of each component
is discussed in Table 2.

TABLE 2: Description of FireHazard Main Components

Component Description
Knowledge Consists of tutorials and video clips about how
Repository to put out fire using the correct techniques. The
players have to read the tutorials and watch the
videos before playing the game.
FireHazard game The main portion of the FireHazard application.
The game consists of three levels depicting three
different scenarios. FIGURE 2: Game flow of FireHazard
Scoring System The scoring system receives the scores from the
game and calculates the total score using various
pre-determined formulas.
Score Database Stores the score of all players which consists of
the subtotal score for each level and the total D. Players
score for all three levels.
25 players were chosen for this study. These players
were selected from Health and Safety department,
B. Game Description lecturers and students who have no experience in playing
FireHazard is a fire-fighting computer game with the the FireHazard game but have prior experience in playing
aim to educate players on how to react in the case of fire. other computer game.
Players should be able to identify the type of fire
extinguisher to be used to put off different type of fire. If
players failed to use the correct fire extinguisher for the
34 ARIFFIN ET AL.

IV. RESULT & DISCUSSION B. Results

A. User Interface TABLE 3: result of playing FireHazard simulation game

FIGURE 3: FireHazard interface

Table 3 shows the overall scores obtained by all players

for each scenario. The results exhibit that all 25 players
obtain consistent scores throughout their games.
Additionally, most of the players got more than 80 marks
as the total score. Based on the results, we can infer that
the players have successfully practice and applied the
newly acquired knowledge obtained.

V. CONCLUSION

FIGURE 4: FireHazard interface

FireHazard is suitable to simulate ‘real’ situation in a
controlled environment. As conducting the real situation is
risky and costly, FireHazard simulation game able to
provide ‘artificial’ experience to the players. By playing
Figure 3 and Figure 4 show sample interface of FireHazard it helps players to improve their learning
FireHazard simulation game. In this simulation game, process by allowing players to learn by taking
players are being exposed to several types of fire; range responsibilities in performing specific job procedures and
from Class A to Class D. Three different scenarios are faced possible consequences in a controlled environment.
being developed to encourage players to apply their Game and simulation provides experiential learning by
knowledge based on the previous encounter. FireHazard giving opportunity to learn thorough experience and
simulation game also provides hint and guidance to the discovery.
players when they provide wrong answers.
Knowledge repository is chosen as the transfer method.
This method of transfer is suitable for large audience
because experts do not have to be available to perform the
transfer since the experts in a large organization cannot
afford to cater to all members in the organization.

REFERENCES
[1] Peter F. Drucker, Post Capitalist Society, HarperBusiness, p 8,
1993
[2] Yozgat, U. Yönetim Bilişim Sistemleri. İstanbul: Beta Basım
Yayım, 1998.
[3] Conner, K.R., Prahalad, C.K.. “A resource-based theory of the
firm: knowledge versus opportunism”. Organization Science. 7/5:
477-501. 1996
[4] Nahapiet, J., Groshal, S. “Social capital, intellectual capital, and
the organizational advantage”. Academy of Management Review.
40/2:242-266, 1998
 USING COMPUTER GAME TO SHARE KNOWLEDGE 35

[5] Flaherty, M. Global Operations Management, McGraw-Hill, [18] P.S. Goodman, E.D. Darr, "Computer-aided systems and
Singapore 1996 communities: Mechanisms for organizational learning in
[6] Yip, G. Total Global Strategy, Prentice Hall, New Jersey. 1992 distributed environments", Management Information Systems
Quarterly, Vol. 22 pp.417 - 440. 1998
[7] Tiwana, A. The Knowledge Management Toolkit, Prentice-Hall: [19] Ruben, B.D, Simulations, Game and Experience Learning,: the
Saddle River, 2000. quest for a new paradigm for teaching and learning, Simulation
[8] Elias M. Awad, Hassan M. Ghaziri, Knowledge Management and Game 30(4) 498-505,1999
Upper Saddler River, New Jersey : Pearson Education, Inc., 2004 [20] Abt, C.A Serious games. New York: Viking, 1970.
[9] Applehans, Globe, Laguero. Managing Knowledge: A Practical [21] Gredler, M.E, Educational games and simulations: A technology
Web-Based Approach. 1999 in search of a (research) paradigm, In D Jonassen (Ed.)
[10] Polanyi, M. The Tacit Dimension, Routledge & Kegan Paul, Handbook of Research for Educational Communications and
London. 1966. Technology (pp521-540). New York: Macmillan 1996
[11] Nonaka, I. and Takeuchi, H. The Knowledge-Creating Company, [22] Driskell, J.E & Dwyer, D.J Microcomputer videogame based
Oxford,University Press, New York. 1995 training. Educational Technology, pp11-16. 1984
[12] Snowden D “ A Framework for creating a sustainable [23] Christopher, E. M. Simulation and games as subversive activities,
programme” CBI Guide to Knowledge Management, London Simulation and Gaming, pp 441-455.1999.
Caspian Publishing/Confederation of British Industry 1998 [24] Schank, R. & Neaman, A Motivation and failure in educational
(Republish in Knowledge Management Year Book 1998 simulation design. In K. D. Forbus & P.J Feltovich (Eds), Smart
Butterworth Heinemann, April 1999 machines in education: The coming revolution in educational
[13] Argote, L., Ingram, P., Levine, J. & Moreland, R. “Knowledge technology. Cambridge, MA, pp. 37-69, US: the MIT Press.2001
transfer in organizations: learning from the experience of [25] Washburn, J., Gosen, J. An exploration of game-derived in total
others,” Organizational Behavior and Human Decision Processes, enterprise simulations. Simulation & Gaming, pp 281-296, 2001.
82(1), 1-8. 2000
[26] Baker, E.L, & O’ Neil, H. F. Jr. Evaluation and research for
[14] Goh, S.C “Toward a learning organization: the strategic building technology: Not just playing around Evaluation and Program
blocks”, SAM –advanced Management Journal, Vol. 63 No. 2, pp. Planning, pp 169-176. 2003
15-22. 1998
[27] O’ Neil, H. F. Jr. , & Andrews, D. (Eds). Aircrew training and
[15] Garvin, D., “Building learning organizations”, Harvard Business assessment. Mahwah, NJ: Lawrence Eribaum Associates. 2000
Review, July-August, pp. 78-91.1993
[28] Mayer, R. E., Mautone, P.m & Prothero, W. Pictorial aids for
[16] Chai, K H, Y S Shi and M J Gregory, “Bridging islands of learning by doing in a multimedia geology simulation game.
knowledge: A framework of knowledge sharing in international Journal of Educational Psychology, pp 171-185,2002.
manufacturing networks In Managing Operations Networks”,
Proceedings of the EurOMA 1999 Annual Conference, 7-8 June [29] Rosenorn, T., & Kofoed, L. B. Reflection in Learning Processes
1999, Venice, Italy, edited by E Bartezzaghi et al., pp. 853-860, through simulation/Gaming. Simulation & Gaming, pp432-
1999 440,1998
[17] Goh, S.C, “Managing Effective Knowledge transfer:an integrative [30] Westbrook, J.I & Braithwaite, J. The health care game: an
framework and some practice implications”, Journal Knowledge evaluation of heuristic, web based simulation, Journal of
Management, Vol 6 No 1. pp23-30. 2003 Interactive Learning Research , pp89-104,2001
 Following the Paper Trail:
Measuring the Economic and Environmental Impact
of Digital Content Delivery
Vanessa K. Cox
Ruth C. May
Stanley L. Kroder
Geralyn McClure Franklin
University of Dallas
College of Business
1845 E. Northgate Drive
Irving, TX 75062

Abstract- We provide an overview of a framework for

measuring the economic and environmental impact of digital
content delivery currently being pilot tested in the College of
Business at the University of Dallas. This framework is based
on Elkington’s Triple Bottom Line (TBL) model of
sustainability which includes Economic (Profit),
Environmental (Planet) and Social (People) dimensions [1].
We first describe the development of a rubric for quantifying
online digital content followed by an overview of the process
for assessing the environmental and economic effects of digital
versus traditional paper delivery. Results are based on
archival data generated with the rubric and survey data
collected from 1,039 online students in the spring and summer
semesters of 2009. We conclude with suggestions for
establishing measurement models at other institutions where
fostering partnerships between university administrators,
faculty and students holds promise for increasing the Fig. 1. Triple Bottom Line Dimensions:
sustainable impact of eLearning.
University of Dallas Paper and Printing Pilot Study
I. INTRODUCTION
II. BACKGROUND
While universities are increasingly focused on integrating As concerns for the environment have increased among
sustainability concepts into their curricula, few have organizational stakeholders and competition for brand
considered the potential impact that their eLearning loyalty from customers has intensified, more firms are
programs may already be having on the environment and looking for ways to make sustainability a key pillar of their
their bottom lines. Online classes result in some portion of competitive strategy [3]. The definition of sustainability has
course content not printed on paper, thereby reducing the evolved over the past 20 years, but is most often linked to
carbon footprints of institutions and their students. the Brundtland Report issued in 1987 which proposes that
Although academic institutions aspire to be recognized as sustainable development “meets the needs of the present
champions of sustainability [2], recent progress made by without compromising the ability of future generations to
corporations in measuring and reporting their sustainability meet their own needs.” [4]. While institutions of higher
efforts have not been matched by educational institutions, learning have generally lagged the private sector in using
particularly in the domain of online course delivery. sustainability for competitive advantage, recent progress by
Consequently, the purpose of our paper is to present a colleges and universities to close this gap between industry
framework for measuring the economic and environmental and education demonstrates a growing awareness among
impact of digital content delivery currently being pilot tested higher education administrators of the potential benefits to
at the University of Dallas (UD). Using archival and student be gained by making sustainable policies and practices a
survey data from the spring and summer semesters of 2009, centerpiece of their strategy. Even AACSB International –
we demonstrate the economic and environmental value of The Association to Advance Collegiate Schools of Business,
digital content delivery versus traditional paper and printing. the premier accrediting body for business schools globally,
The sustainable benefits of eLearning, at least in part, lie at has been promoting sustainability and business school
the intersection of the Profit and Planet dimensions of leadership in this effort with speakers and sessions at
Elkington’s TBL as depicted in Fig. 1. various conferences and seminars as well as its own
Sustainability Conference [5]. Yet what cannot be measured

37
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_7, © Springer Science+Business Media B.V. 2010
38 COX ET AL.

cannot be managed effectively, much less used for strategic digital content for the COB consisted of 92 online course
advantage. sections for spring and summer 2009 combined.

The Graduate School of Management at UD, now part of 1) Digital Content: Calculating the total number of virtual
the College of Business, has been involved in distance pages required establishing a definition of what constitutes a
learning since 1969 beginning with a regional microwave virtual page as well as specific guidelines to ensure a
video network. As the Internet emerged in the 1990s, the consistent, accurate page count. We defined a virtual page
school leveraged the new technology by launching three as either a) actual pages in a paginated document or b) the
online credit-bearing courses. This initial offering has equivalent number of pages for scrolling digital content in a
evolved over the past decade to include the full Masters of Microsoft Word document. The following items were
Business Administration (MBA) core curriculum and a considered when counting the virtual pages in each online
dozen specialized MBA concentrations, with no residency course.
requirement. The expansion of the college’s online delivery
has provided an opportunity to reduce costs associated with • Definitions: The total number of virtual pages includes
traditional paper and printing content production and to obvious printable downloads, nonobvious printable
reduce the carbon footprint of the institution. In early 2009, digital content and nonprintable digital content.
a pilot study was initiated to document these savings which Examples of obvious printable downloads include
are a natural byproduct of online education. The following syllabi and XanEdu (vendor) course packs.
section describes the process used to account for digital Nonobvious printable digital content includes items
content not printed and the resulting savings in dollars and such as chat sessions as well as directives to students in
carbon emissions. digital format. Nonprintable digital content includes
grade book and exam items protected by plug-ins that
III. METHODOLOGY prohibit the printing of designated content.

A. Sample • Guidelines: Specific counting protocols of course

The sample consists of online MBA students in the spring content items were established and applied by a single
and summer 2009 semesters. While traditional brick and counter in both semesters to ensure consistency. This
mortar MBA classes in the COB have companion websites approach reduced the risk of double counting items
for each class section, neither the digital content nor the such as graphics, pictures or links. eBooks or external
students in these on-ground classes were included in this content located on other web sites were excluded from
study. Rather, only fully online MBA students were invited the page count.
to participate in an automated survey as part of a broader
project focused on the impact of eLearning on all • Multiple User Access: Once the total number of virtual
dimensions of the TBL model, much of which extends pages per course was determined, we multiplied this
beyond the scope of this paper. figure by the number of students enrolled per course
(i.e. total course seats) to determine total digital content
A week prior to launching the survey, students were provided in each online course. This approach is
invited to participate in the study via a notification message consistent with considering the sum of all copies of a
in the online course system. The online survey was inserted traditional course textbook sold to every student
into the opening frame of one online course per student purchaser versus assuming that a single copy of a
regardless of the number of online courses they were taking textbook is shared by all students in a class. This
to avoid duplicate responses. Participation in the survey perspective represents more accurately the total amount
was optional. Out of 644 online students invited to respond of digital content provided by the institution in a single
in the spring 2009 semester, 539 provided complete data semester.
resulting in a response rate of 84%. Of the 621 online
students enrolled in the summer 2009 term, 500 provided 2) Printing Behaviors: In order to determine the number of
useable responses for a response rate of 81%. The combined virtual pages offered by the COB that ultimately remained
sample from both semesters totaled 1,039 for an overall virtual (i.e. never got printed), students were asked about
response rate of 82%. Across both semesters the student their printing behaviors in their online courses as part of the
sample ranged in age from 20 to 63 with an average age of student survey. Specifically, students were asked to
37. Fifty-one percent of the respondents considered the approximate the percentage of obvious downloadable
Dallas /Fort Worth (DFW) metroplex to be their home base documents and nonobvious printable content that they
and 49% percent of the respondents resided outside the would typically print in an online class. In both document
DFW metropolitan area. categories they were asked to select a percentage that they
typically printed across all online courses from a drop down
B. Data Collection list ranging from zero to 100 % in 10% intervals. We
In addition to the student survey data, archival data was elected to ask students about their printing behaviors in
generated using an original rubric designed to count the general rather than ask for a specific page count or rate for a
number of virtual pages available to students in all COB specific class because we believed that the probability of
online courses. In spring 2009, 48 online sections were getting reliable responses about printing habits at such a
offered, and in summer term 44 online sections were level of specificity was relatively low and would be more
conducted. Thus, the archival base used to compute total likely to reduce our response rate to the survey.
 MEASURING THE ECONOMIC AND ENVIRONMENTAL IMPACT 39

Students were also asked “What type of printer do you B. Economic and Environmental Consequences
typically use for printing course related materials?”. They Providing digital content versus traditional paper and
could select one of the following: small desktop printer, printing delivery results in trees not destroyed to produce
shared workgroup within their company or work group, or a paper, energy not used to print content on paper, and carbon
business center printer such as Fed-Ex Kinkos or a hotel. dioxide not emitted in the printing process. Collectively,
these outcomes result in economic savings and
IV. RESULTS environmental benefits for the planet. To quantify these
benefits, we relied on the Hewlett-Packard (HP) carbon
A. Digital Content footprint calculator.
Table I summarizes the virtual page count in all categories:
obvious printable, nonobvious printable, nonprintable and HP has a strong focus on global citizenship and
total virtual pages. For both terms combined, the college sustainability with corporate goals that pertain specifically
provided over 1.5 million pages of digital content. to environmental outcomes [6] [7]. Its printer division is a
However, the net number of virtual pages that remained in leader in the imaging market and has a “green” focus in its
digital format (i.e. were not printed by students) could only product and delivery. The HP Carbon Footprint Calculator
be derived by applying the data captured on student printing is designed to assess printer energy consumption and
behaviors. Specifically, students in the spring 2009 term associated carbon output as well as quantity of paper used
printed an average of 53% of obvious downloadable pages and related monetary costs. The calculator generates
and 30% of nonobvious printable content. Students in the estimates of energy consumption during use of a printer and
summer 2009 term printed an average of 49% and 32% in emissions of carbon dioxide from production of estimated
the same categories, respectively. volumes of paper consumed during printing [8]. The
savings derived by the calculator provided explicit figures
The net results for the institution are illustrated in Fig. 2 for the study with the exception of the CO2 emissions which
where 543,275 pages remained virtual in the spring semester had to be converted from pounds of CO2 to a Metric Tonne
and 320,028 remained virtual in the summer term. Thus, a Carbon Equivalent (MTCE), the standard unit of CO2
total of 863,303 pages of digital content provided by the emissions that we adopted for this study [9].
institution remained in digital format, with their
corresponding economic and environmental consequences, Table II provides the gross institutional savings in
even after students’ printing behaviors were taken into kilowatts, tonnes of CO2 emissions, pounds of paper and
account for both semesters. dollars generated across all online courses in spring and
summer 2009. While the savings depicted in Table II are
TABLE I real savings from an institutional viewpoint, the printing
RUBRIC SUMMARY
FOR ALL ONLINE COURSES
habits of students must be taken into account in order to
determine the net real environmental benefits of the digital
content delivery. After accounting for both printing habits
University of Obvious Nonobvious Nonprintable Total
Dallas Printable Printable Pages Virtual
and types of printers used by students, the net results for the
Term Pages Pages Pages planet remain positive as illustrated in Table III.
Spring 2009 544,151 399,593 7,808 951,552
Summer 2009 421,757 147,722 5,381 574,860
Year to Date 965,908 547,315 13,189 1,526,412
TABLE II
CARBON FOOTPRINT CALCULATOR RESULTS
FOR TOTAL DIGITAL CONTENT

Gross Institutional Savings

Term Year

Savings In: to
Spring Summer
2009 2009 Date

Energy( kWh) 154,991 143,654 298,645

Carbon Emissions- 103.16 93.31 196.47

CO2 (tonnes)a
Paper (lb) 10,462 6,327 16,789

Fig. 2. Printing Behaviors Energy and Paper

(USD) $ 20,243 $17,098 $ 37,341
40 COX ET AL.

TABLE III • Approximately 10 Round Trip flights from New York

CARBON FOOTPRINT CALCULATOR RESULTS
AFTER ADJUSTING FOR STUDENT PRINTING
to Los Angeles not taken [12].
• Nine average passenger vehicles not driven for one year
[13].
• 2.35 football fields of trees in a section of the Amazon
NET ENVIRONMENTAL BENEFITS
Rain Forrest not cut and burned [14].
Term Year
While the measurement process described here makes an
Savings In: to incremental contribution to the challenge of measuring and
Spring Summer
2009 2009 Date reporting the sustainable impact of online delivery, it is
subject to the following limitations:
Energy (kWh) 33,538 31,949 65,487
• The measurement model does not take into
Carbon consideration all aspects of digital content delivery at
Emissions-CO2 24.85 22.20 47.05
(tonnes)a
UD – COB (i.e. Hybrid or blended learning as well as
web enhanced instructional companion elements) so the
Paper (lb) 5,978 3,531 9,509 results are understated.
• The pilot study is located at a private university in a
a
A tonne (t) or Metric Tonne Carbon Equivalent (MTCE) is a large metropolitan area, and thus the results are
measurement of mass equal to 1,000 kg or 2,204.622262 lbs, institution specific.
or approximately the mass of one cubic metre of water [10].
• The digital count of virtual pages included only fully
online courses in a Masters of Business Administration
In summary, for the first eight months of 2009, after program delivered in an asynchronous delivery model,
accounting for student printing behaviors, the digital and therefore, may not translate well to undergraduate
delivery of all online content in the COB resulted in 65,487 programs with more synchronous content and
kilowatts of energy not used, 47.05 tonnes of CO2 not interaction.
emitted, and 9,509 pounds of paper not printed for a total • While the survey response rate was excellent from
institutional savings of $37,341. graduate level students at UD, the same level of interest
or participation may not be replicated with an
V. CONCLUSIONS undergraduate population.

This paper reported on the results to date of a pilot study Despite these inherent limitations, the pilot program
currently underway in the College of Business at the demonstrates the value of considering economic and
University of Dallas to develop a framework for measuring environmental consequences of digital content delivery for
the economic and environmental impact of digital content the institution and the planet. Moreover, while the pilot
delivery. While many universities have stepped up efforts framework described here is only one approach to
to integrate sustainability concepts into their curriculum and measuring the sustainable impact of online learning, our
their on-ground campus operations, few [11], if any, have intention is to provide an initial blueprint for other
attempted to quantify the positive environmental effects that educational institutions to follow in:
their eLearning operations are generating while producing
cost savings as well. • Determining the volume of digital content they provide
and the corresponding economic benefit.
Using survey data from 1,039 online MBA students in the • Determining the offsetting effects of student printing
spring and summer semesters of 2009 and archival data behaviors on the institution’s digital content delivery
created via virtual page counts, we found positive economic strategy.
and environmental benefits from digital delivery for the first • Determining the combined effect of institutional and
eight months of the year. Cumulatively, digital delivery of student choices on the broader environment.
content resulted in 298,645 kilowatts not used, 196.47
tonnes of CO2 not emitted and 16,789 pounds of paper not University administrators, faculty and students have the
printed for a total of $37,341 saved by the University of opportunity to positively influence the economic and
Dallas. environmental dimensions of the TBL with regard to digital
content. Specifically, administrators can be more proactive
After taking into account the printing behaviors of online in adopting institutional strategies to increase online
students, which is a more conservative measurement offerings in various academic units and can mandate the
approach, the online operations of the COB still produced continuous tracking of benefits and costs associated with
environmental benefits totaling 65,487 kilowatts saved, digital delivery. Administrators can also consider
47.05 tonnes of CO2 not emitted and 9,509 pounds of paper establishing green “stretch goals” for their online operations
not printed. The savings in carbon emissions alone equates on both a gross and per capita basis.
to the following:
Students have a significant role in determining how much
of an institution’s digital content remains virtual as a result
 MEASURING THE ECONOMIC AND ENVIRONMENTAL IMPACT 41

of their printing behaviors. Therefore, it is important to The authors also thank Jeff Borden, Senior Director of
inform students of the power that they have to reduce carbon Teaching and Learning for Pearson-eCollege. Mr. Borden
emissions simply by becoming more aware of how much has led his company’s global efforts to deliver high-quality
they print and how their printing decisions have direct online courses for hundreds of higher educational
environmental consequences. A leading edge example of institutions since 2002 and he provided valuable advice to
institutional influence on student printing behaviors is a our team during the design of the survey and data collection
pilot program involving six colleges and Amazon.com. The process.
partnership provides designated students at the pilot
universities with a Kindle DX and access to required course Finally, the authors acknowledge with grateful appreciation
readings. The e-book reader is designed to meet the needs the contributions of Nestor Via, Graduate Assistant in the
of students by offering a larger screen and searchable online learning business unit of the University of Dallas,
content embedded within required reading materials while College of Business. Mr. Via conducted the count of all
motivating them to reduce their printing of reading materials virtual pages in both spring and summer terms to ensure the
[15]. In general, today’s students are likely to be more integrity and consistency of the data.
environmentally conscious than their predecessors [16], but
they cannot contribute to a solution if they are not made REFERENCES
fully aware of the problem and their potential to influence
its outcome. [1] J. Elkington, “Towards the sustainable corporation: Win-win-win
business strategies for sustainable development,” California Management
Review, 36, no.2, pp.90-100, 1994.
Likewise, instructors should be encouraged to focus on [2] D. Ferrer-Balas, J. Adachi, S. Banas, C.I. Davidson, A. Hoshikoshi,
their roles in the sustainable efforts of their institutions by A.Mishara, Y. Motodoa, M. Onga, M. Ostwald, “An international
rethinking their design of course content in keeping with the comparative analysis of sustainability transformation across seven
universities,” International Journal of Sustainability in Higher Education,
“less is more” proposition. For instance, instructors should vol. 9, no.3, pp.312, 2009.
consider the following: How effective is pure text content [3] M. Porter, M. Kramer, “Strategy & Society: The link between
and how is it likely to be perceived by students, particularly competitive advantage and corporate social responsibility,” Harvard
those in the Gen Y demographic? Recent studies show that Business Review, pp 1-2, December, 2006.
[4] W. Adams, “The future of sustainability: rethinking environment and
Gen Y individuals prefer learning through multi-media and development in the twenty-first century,” The world conservation union,,
multi-tasking designs rather than traditional text delivery online, last updated May 2006, http://www.iucn.org.
[17]. With this reality in mind, instructors should consider [5] B-school facilities resource center, online,
what might be more effective pedagogical options relative to http://www.aacsb.edu/resource_centers/fcilities/buildinggreen.asp .
[6]HP Global Citizenship Report: Environmental sustainability, online,
text for increasing student engagement and reducing the September 24, 2009,
temptation to print. These options might include relevant http://www.hp.com/hpinfo/globalcitizenship/gcreport/envsustainability.
videos and student-directed content “scavenger hunts” based html .
around a central theme or problem which prompts students [7]Letter from Mark Hurd, HP Chairman, CEO and President,
http://www.hp.com/hpinfo/globalcitizenship/gcreport/intro/ceoletter.html
to focus on collecting and analyzing virtual content as [8] HP Carbon footprint calculator for Printing,
opposed to printing and memorizing basic text. This http://www.hp.com/go/carbonfootprint .
approach to content design is consistent with John Dewey’s [9] MtCO2- Metric Tonne Carbon Equivalent, online, 10/06/2009,
philosophy which is, “Give the pupils something to do, not http://www.mtco2e.com .
[10] Tonne, online, http://en.wikipedia.org/wiki/tonne .
something to learn; and if the doing is such a nature as to [11] R. Roy, S. Potter, “Designing low carbon higher education systems:
demand thinking; learning will naturally result [18].” Environmental impacts of campus and distance systems,” International
Journal of Sustainability in Higher education, vol.9, No. 2, pp. 117-118,
In conclusion, our results demonstrate the value of not just 2008.
[12] C.Dunn, “Reduce your carbon footprint by half in three steps: Plant
following the paper trail, but in documenting and measuring Green”, online, July, 2009, http://planetgreen.discovery.com .
it in order to better guide organizations in achieving [13] Emission Facts: Greenhouse Gas Emissions from a Typical Passenger
sustainable strategies. We hope that these initial efforts will Vehicle- US EPA, online, July, 2009,
prompt greater consideration and discussion of the potential http://www.epa.gov/OMS/climate/420f05004.htm .
[14] How your pollution is calculated – Fight Global Warming-
benefits of digital content delivery and the corresponding Environmental Defense Fund, online, July, 2009,
challenges involved in measuring them. http://fightglobalwarming.com .
[15] S. Peters, “Universities turn to Kindle- Sometimes to save paper,”
ACKNOWLEDGMENTS New York Times, online, July 30, 2009,
http://greeninc.blogs.nytime.com/2009/07/30/universityies-turn-to-kindle-
sometimes-to-save-paper/ .
The authors of the paper thank John Frey, Americas [16] J. Pryor, S. Hurtado, J. Sharkness, W. Korn, L. DeAngelo, L.
Sustainability Executive with Hewlett-Packard Company. Romero, S.Tran, “Political engagement among college freshmen hits 40-
Under Frey’s leadership, Hewlett-Packard has been named year high,” Higher Education Research Institute (HERI): The American
Freshman: National Norms Fall 2008, online,
one of the 100 most sustainable corporations in the world by http://www.gsei.ucla.edu/heri/pr-display.php?prQry=28.
the Global 100 each year since the inception of the list. In [17] C. Dede, “Planning for neomillennial learning styles: Shifts in
addition, HP continues to be listed on key sustainability students’ learning style will prompt a shift to active construction of
indices including the Dow Jones World Sustainability Index knowledge through mediated immersion,” Educause Quarterly, pp. 8-9.
2005.
and the FTSE4Good. Mr. Frey provided important input [18]S.Khurana,About.com,online,
and clarification related to the methodology of the Hewlett- http://quotations.about.com/od/stillmorefamouspeople/a/JohnDewey1.htm .
Packard carbon calculator and the paper and printing aspects
of the study.
Context Driven Personalized e-Learning Environment
Provided as a State Service
Deniss Kumlander
Department of Informatics, TTU
Raja 15, 12618
Tallinn, Estonia
e-mail: [email protected]

Abstract—The main topic of this paper is the life-long education. delivering it to students. Therefore the university approach
The rapidly evolving world permanently challenge both doesn’t work very well in many cases of the modern, rapidly
individuals and governments as the productivity and educational changing world. Constant technology, science, best practices,
level of the state work-force is one of the most important factors standards and so forth changes cause various types of
of the long term success in both financial and social sectors. The unemployment, low productivity of work-force since skills and
paper describes an e-education system that should be funded by knowledge become outdated very quickly. This is a serious
the state and exposed to all citizens in order to constantly challenge for all involved parties: society, business and
upgrade their skills. The broad potential auditorium and targets individuals and therefore it should be solved together in a
of such system define certain unique requirements that should be
synergy.
met by the context driven personalized approach of delivering
information as proposed in the paper. It makes the process of The concept of the life-long education is the first step on
acquiring knowledge simpler for all categories of users and this road. Unfortunately in many cases it is left to individuals to
stabilizes the virtual learning environment from the data follow it, while others do minor investments into that mostly
protection point of view. motivating people to continue the education on constant basis.
In the result many people spent their time trying to find and
Keywords-e-education;artificial intelligence; life-long learning; acquire knowledge in parallel instead of following mentor
state services; context.
advices consuming it in a prepared form. Fortunately
everything started to change (so far slowly) and the importance
I. INTRODUCTION of an electronic learning as an alternate solution to the
There are two different approaches to the educational traditional education has been recognized. The quick
system in the modern society. One of those is university centric development of electronic channels and world globalization [6,
suggesting to concentrate only on educating young people. It 7] help us to organize now the re-education much better than it
proposes that correctly organized educational process gives sued to be in the past. Unfortunately those attempts mostly
enough skills to upgrade knowledge later on without mentors. local and lack the strategic, global power that could be provide
Here the state builds up centers of massive education for young by a state financing and driving the development of such virtual
people and fund those also as research facilities in order to education environment. At the same time, upgrading to the new
concentrate science in one place and produce this way highly level of virtual education means also facing completely new
skilled work-force, which is so much required by the modern challenges in process of building such systems. There are
world [1]. This approach has both positive and negative sides already millions of students attending e-courses in developed
[2, 3]. For example it was criticized as been mostly looking countries and this number is growing permanently. They come
into the past accumulating the knowledge in order to spread it from very different education levels, age and society groups.
without major updates of courses during sufficient periods of Therefore the online education based on average student
time [3]. Universities do argue that this is much more stable doesn’t work as well as it was hoped and requires
method of educating as proper understanding of fundamental personalization of the delivered content. Fortunately computers
science (basics, techniques) will give to students a good base and artificial intelligence give us just enough tools to deliver
for acquiring whatever additional knowledge in the future [5]. such personalized content as described in this paper.
Having agreed with it, we still like to highlight the sufficient
gap between the knowledge obtained in universities and the II. PERSONALIZED E-EDUCATION PROVIDED AS A STATE
knowledge students will need dealing with specific tasks at SERVICE
their future workplaces. Moreover the ability to acquire new
knowledge in the future doesn’t necessary mean that the person A. Auditorium
will do that. It could be a problematic or extremely time The kind of the system aimed in the paper has certain
consuming for him without having a good understanding of difference from other e-Education systems described a lot in
what is going on in the modern technology and having to parse the literature and the most important one is the targeted
a huge amount of information sources. Notice that the last auditorium. If the standard literature case is moving the
mentioned tasks are tasks of the teacher in normal educational process from offline to online aiming still the
circumstances, who should concentrate the knowledge before

43
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_8, © Springer Science+Business Media B.V. 2010
44 KUMLANDER

university courses and students, then the state service is mainly with the main participants’ activities at work or home.
directed to broader auditorium of citizens of different ages. So, Therefore we should foresee different level of education and
here e-learning participants can be those who completed the previous experience of users. Besides the ability of individuals
last study long time ago, people who spent a lot of time on to acquire knowledge is known to be sufficiently varying
other primary activities, individuals who wish to obtain modern dependently on their age and it makes students also different.
knowledge or acquire a completely new profession. Moreover Finally it should be possible to define any course on packets
we should still count with potential participants from suitable to students’ availability as highlighted earlier and be
universities or even schools. Therefore it is important to available online on demand.
personalize the delivered courses as much as possible it order
to increase a probability of successful acquisition of knowledge C. Responsibilities
by each participants. It is incorrect to assume that the organizer of the virtual
As a concluding note for the auditorium description we learning environment is responsible only for the content of
would like to mention that all this defines a challenging task for courses, i.e. quality of the learning process. Unfortunately there
the system to acquire the context of students in order to are a lot of other responsibilities, which should be taken into
personalize delivered information correctly. account building such a service.
First of all the service operates with information about the
B. Targets user and in many cases it is more just a username and a
The high level task of the proposed virtual educational password. Notice that here we don’t mean only the protection
system is to serve citizens of a state looking for possibilities to of the data in the communication channel between the e-
upgrade their knowledge. Besides, the system should be learning server and client computers. The personalization of the
intelligent enough to find a knowledge delivery workflow system means that it should collect a lot of context about the
suiting for a particular individual basing on her previous consumer and all this information is stored on the server in
experience, background and availability. forms of different logs, patterns and so forth. Moreover even
the information on which courses were taken, when and what
All this must be explained in more details in order to marks she got should be classified. Only the person (i.e. owner)
identify correctly the scope and complexity of the goal. First of could expose this information, for example temporary to the
all such a virtual environment should be a proactive way of a employer, and nobody else.
state to deal with a potential unemployment and prevent the
drop of workforce skills. The reactive approach is usually too Secondly the course authors are usually also highly
expensive due a need to pay social benefits and have a low interested in protecting the content of lessons as much as
level of technological development loosing competition to possible. So, there is a clear demand to build virtual
other states. All this expenses are carried by the entire society environments that will be both easy to consume by students
in form of different taxes and the government task is to avoid and hard to misuse or replicate without the author permission.
them if possible. That is why the paper proposes the system to It is quite a challenge task and the easiest way to protect the
be developed as a service provided by the state. It generally content is to set a copyright on this. Unfortunately it is not
means that it should be enough in many cases, so authors will be thankful if the system
will contain other protection mechanisms.
1. Supported by the state (financing);
The protection of such private data is the most important
2. Advertised by the state; challenge for the system developers. They need to ensure that
3. Integrated into the general state infrastructure and data neither can be stolen nor made available outside due any
made available basing on dedicated infrastructure kind of errors or backdoors of the system.
elements implemented by the stated.
If the first two elements are easy to understand then the last III. GENERAL MODEL
one can easily be missed. Unfortunately it is not possible to This chapter continues the description of the virtual
build such a system from scratch. It should be an evolutional learning environment introduced on our last paper [10] and
approach when the educational system follows other state proposes additional artificial intelligence elements supporting
services and relies on the same delivery infrastructure in order the personalization based on the acquired context. Therefore in
to make it available (in both meanings: ability to access and a the beginning an overview of the system is given to describe
constant availability). The educational content (data flow) is the topic defined so far: prerequisites for building such a
much bigger than other information service like registers and system and interfaces of then model.
so demands sometimes completely new infrastructure system to
be built up. Fortunately state usual can afford financing and A. Prerequisites
supporting that considering the social cost it will carry vice
The prerequisites will be described basing on a particular
versa.
example of one European country – Estonia. The country has
Moreover the system should not unify or demand too much an informational system available to citizens to request their
from consumers in terms of spent time and required knowledge information hold by different state registers. The system is
level in order to participate. An ideal life-long education orients developed centrally by a special agency and mostly
on gradual upgrade of work-force skills organized in parallel concentrates information available internally from different
 CONTEXT DRIVEN PERSONALIZED E-LEARNING ENVIRONMENT 45

departments like Real-estate agency, Police, Automobiles should stimulate producing personalized courses trying
registry and so forth. Moreover the system does allow citizens to keep simple courses set on minimum level and make
to send requests and documents to the state in order to obtain a this route available only for distinguished lectors. The
new passport for example. All departments of the state should interface should contain also templates making the
process such electronic documents signed by so called “ID- course building process both simpler and standardized.
card” and there should be no difference between paper and
electronic channel of communication. The information system Notice that for the simplification we assume that all
has a single entry point. courses do not require a mentor to help students to
acquire the knowledge, although this element can also
This is an example of the system that normally should be in be included if required.
place before building the state-wide fully functional
educational system. Here we would like to list some reasons Student Author Business company
why we believe such system is required:
1. The process of delivering information to citizens is
verified and in use, so citizens know how to find and access it; Calls brokers
2. Citizens do trust to such system much more than to any
other especially if the state is taking seriously the protection of
private data; HTML UI Engine Security server
3. The security (log in) is well established and is normally
supported by some kind sophisticated mechanism of passwords Communication
distribution (in Estonian case: ID-cards issued by the state, server
which are equivalent to passports, i.e. well protected including Courses
warehouse Templates,
the law); standards and AI engine
4. The developers’ team is acquired and educated; referencies
relationship to external contractors (including the internet
providers) is well established. Context
Results: engine
B. Overview of model interfaces Statistic engine
Such a broad system like described here has several
complex interfaces designed to different type of actors.
Figure 1. General model
1. Common part.
First of all each interface does contain the security 3. Consumers (students) interface
elements that can be inherited from the pre-system like The nature of this interface is obvious: students do
the one described earlier. Secondly, there are a set of participate in courses by passing different topics of
databases on the back end that contain course step by step and (optionally) pass an exam in the end.
materials, results and other personal information on This is the interface we are going to personalized first
participants and so forth. Thirdly there are of all by using AI and context engines shown on the
infrastructure elements that do support the figure 1.
communication like a load-balancing server, which
should spread the load between a set of servers, since 4. Third-party interface
the state wide educational system is likely to be heavily The system should not be restricted to the course
used. Finally there should be a communication server (teacher) – student communication. There are other
that should allow communication between all involved groups of people including business community that
parties outside the learning process in order to provide may be interested in this process retrieving information
a) general feedback on courses, b) define interests from if it is exposed or providing requests. This is a way job
business community to design courses or stimulating seekers could expose their professional level to
participations – for example a job advertisement companies. It is also an interface for companies to
defining passing certain courses as a minimum stimulate creation of a work-force they do require by
requirement and so forth. motivating authors to create certain courses and
2. Courses’ authors interface. students to participate and obtain high marks.

This interface allows building and uploading course

materials into the system following certain system
standards. If the author is not interested in
personalization then those standards are quite simple
and define types of files, structuring and examination
rules. At the same time the supervisor of this system
46 KUMLANDER

IV. CONTEXT DRIVEN PERSONALIZATION The next important parameters the system could react on is
the available time slots. The minimum requirement would be to
A. Personalisation using context preserve the position of a lecture where the student had to stop
The context-aware system is a very popular topic nowadays last time as the delivered content is now unique and so could be
in different artificial intelligence systems starting from web hard to reproduce exactly. Ideally the delivered portions should
pages up-to automobile user interfaces for drives. be dependent on availability of a person. For example she
could define in advance time size of available time slots (for
Generally, context-aware system development means that example 15 minutes each) and the system will deliver lessons
the computer reaction can be organized using more than just of that size dividing those on topics and subtopics making the
the direct data input. Instead computer system should consider acquired knowledge tractable on the lowest grain (up to a
circumstances under which the input is provided sensing the lesson page).
environment in which the request is formulated and form the
response basing on those parameters [10]. Some authors say The progress of learning could also be measured during the
that it is nothing else that a system ability to adapt to specific education and delivered in different level of explanation
circumstances under which the system operates or reacts on the provided. The explanation could be short to keep the speed of
request and ability to collect and recognize the context moving over topics or could be very details if the person have
information [11, 12]. Therefore the context-aware approach is no ideas at all about the explained concepts (or previous
mainly used to increase relevancy of the response (service) experience with those), so will require slowing down in this
provided by the system and is targeted to bridge the gap particular concept or a slow delivery of the knowledge in total.
between formal and human to human type interaction [13]. The next important stage of the education process that can
Unfortunately majority of scientists do restrict themselves be a subject on context driven personalization is the
with the most obvious context they can acquire – the location. examination. Sometimes results of the education process are a
Fortunately for us it is not the only one and the virtual valuable achievement exposed to the business community.
education system is the perfect example of a system that could Those can be used to sort individuals basing on their skills and
obtain and process other types of context in order to make the therefore the system should prevent the misuse of the education
process of education very personal. process by persons attempting to get higher marks than they
actually are worth of. Therefore the study and examination
The previous experience and educational level can be process should be closely watched in order to identify the
identified as one context that can be used to build up personal person taking the exam. There are hard and soft security system
courses. It can be acquired in a form of tests and questionnaires based on biometrical patterns that can be recorded
where a student defines both her current knowledge and topics communicated to a person. The basic idea here is to learn
she wishes to learn. The standard environment would leave it computer system to identify humans as they do in their
up to users to select courses or lessons she likes to attend, but everyday life distinguishing one person from another. Hard
sometimes courses and lessons interconnection is too complex methods, like face recognition, require certain hardware to be
to understand or even hidden from the student. Therefore an used and therefore are not possible to be applied in many cases.
artificial system based on clearly defined rules will be a perfect Fortunately there are soft methods as well. One of the simplest
mentor existing inside the educational system. If courses are “soft” biometrical identification methods is keystroke patterns,
divided into isolated lessons, then it will be possible to skip where the system monitors the keystroke typing dynamic, used
certain, that the student doesn’t need, or replace them with symbols and words. It is well-known that writing/typing
others bridging the gap in certain areas of student knowledge. dynamics are assumed to be unique to a large degree among
This will ensure that all prerequisites for a successful different people [6]. The same is known for the selection of
understanding of next lessons are provided to the student. words individuals use or keys on the keyboard (consider
Moreover, if the previous case will do some minor changes “delete” or “backslash” keys, which are nearly identical in
in a course, then more sophisticated engine could combine functionality). Moreover the keystroke dynamic is not one
different lessons building a course on demand from available factor to be measured but is a set of characteristics [7, 8, 9]:
parts. Such artificial intelligence could easily produce non 1. Duration of the keystroke, hold time;
envisioned courses basing on personal characteristic either 2. Latency between consecutive keystrokes;
without authors supervision or minimizing it sufficiently. 3. Overall typing speed;
Obviously this is possibly only if the number of available
4. Habits by using additional keys on the keyboard like
lessons form a big enough database and those are well isolated.
typing numbers via number pad part;
The production of the delivered context basing on previous 5. Frequency errors and backspace vs. delete key using;
knowledge is not the only way to process a context available and some others
for the system. Another important property the system could
monitor and measure is how well the student acquires the
knowledge. For example the course could be a combination of The pattern captured during tests and learning process can
explanation and practical tasks. The number of practical tasks be reused during the examination phase and identify the student
could vary depending on how well previous were made in doing the exam. If there is any mismatching then the system
order not to slow down quick brains providing enough tasks to should either have some kind person-to-person interaction
slow catchers. possibility making possible to uniquely identify knowledge of
the person or should have a concept of reliability of exam
 CONTEXT DRIVEN PERSONALIZED E-LEARNING ENVIRONMENT 47

marks, which is delivered together with those, so interested The more sophisticated protection mechanism can be build
parties could reevaluate those if needed. basing on the acquired context during the process of education.
The first problem that occurs quite often is to a) detect the node
Recent researches have reported that up to 65% of people the course leaked out and b) prove who the real author of the
over the middle age have certain disabilities and therefore course is. A so called “watermarking” could greatly help us
could like to modify the user interface via which the content is resolving those two issues. It is possible to make the delivered
delivered. This is another very important personal context dynamic even if quite static files are delivered to the
characteristic, which the system should react on. Notice that in student by embedding into those invisible digital marks stating
many cases this is ignored in systems directed to students as the owner of the content and the student it is delivered to. The
quite a minor portion of young people suffer from any kind of process is basically similar to adding watermarks into
disabilities. In the state wide system such ignorance is not banknotes except the fact that in the digital case we would like
possible any longer. The two elementary possibilities are to keep them as hidden as possible in order to prevent
required as a minimum. A possibility to increase the font size removing of those. Notice that those marks are added right
of the delivered content is one of those. Notice that although prior delivering the content to the student inside the virtual
many browsers nowadays offer such functionality, not all web environment web server.
pages react on this correctly. The second one is possibility to
deliver voice explanations. It will decrease the overall pressure Besides, the earlier described personalization of the
on eyes for the education process participants and allow delivered context means that all lessons are divided into very
consuming the content by persons with disabilities. Moreover it small chunks which are delivered on demand basing on
could detach the education process from the computer as there concrete person characteristics. In those small parts certain
are a lot of devices able to reproduce voice recording but not lessons can be omitted or replaced with others from completely
able to show web pages due the restricted screen size (or no other topics. All this could make the process of restoring the
screen at all) – consider as an example cell phones and mp3- original content and reusing it to be very complex. It will
players. require a lot of time and a certain level of knowledge making it
even more complex that creating a course from scratch.
The age or disabilities is not the only reasons for slowing Summarizing, the personalization divides the content into
down the educational process. The overall environment blocks making he course hard to acquire as a whole.
students are located in could also promote or suppress the
ability to understand information including the type of device
the content is delivered too. Obviously even the sophisticated V. CONCLUSION
PDA device need restructuring delivered information as it The virtual educational system described in the paper is an
doesn’t have the same screen as a usual computer. Besides important component ensuring the social and financial stability
writing a text using the same keyboard at home or in the train is of any state. The work-force productivity is one the most
sufficiently different. Therefore the monitored keystroke patter important factors defining the success of the state industry and
used so far and property of the remote device could and should innovation. The level of unemployment and professionalism
be used to adjust materials correspondingly. (and consequently of salaries) defines the health of society
including such as the level of crime or development of all state
B. Protecting the content institutions. Therefore it is important to fund and developed
Nowadays, when knowledge becomes more and more such systems centrally by the state. The broad goal of the
important once again, many teachers are concerned about system makes it hard to build as it should serve educational
protect of their intellectual product (the course) from needs of completely different people. Therefore the
unauthorized copying. This becomes extremely important in personalization of the delivered services becomes extremely
the virtual environment, which is easy to access and even easier important developing such systems. Fortunately the state either
to copy and reproduce materials. contains enough information about their citizens that can be
consumed or the state system is trusted enough to enable
There are several ways to avoid this issue. First of all a context-driven collection of such data. Later the context can be
copyright can be defined, which means a legal protection. converted into the concrete delivery of the education content to
Unfortunately it is not always good enough in virtual the concrete person during the concrete time-interval making
environments since the course can be used in other countries the lesson unique and completely suitable to the available
producing certain barriers for courts and laws. Secondly the moment of life-time of the student. All this is designed to
author rights can be acquired by the state and deliberately increase the knowledge acquisition level from the online
unmonitored. The primary goal is to educate citizens and if it is system and close the gap between online education and
achieved, then reuse of the course by anybody else can be teacher-student offline collaboration, which is usually much
ignored, as long as citizens do understand what courses they more intensive and so productive.
could take free of charge via the state system. The third and the
last elementary protection mechanism is the language barrier,
REFERENCES
i.e. the national language used to build up the course prevents
the same course to be transferred to countries with other [1] D. Mowery, R. Nelson, B. Sampat, and A. Ziedonis, “Ivory Tower and
Industrial Innovation,” Stanford Business Books, California: Stanford,
languages solving a problem with countries where the law 2004.
cannot be enforced to restore the course ownership.
48 KUMLANDER

[2] K. Fischer, “The university as economic savior: as higher education the 25th conference on Proceedings of the 25th IASTED International
replaces industry in some cities, colleges face both support and Multi-Conference: artificial intelligence and applications, 2007, pp.
unrealistic expectations,” Chronicle of Higher Education, 2007. 317-322.
[3] J. Youtie, and P. Shapira, “Building an innovation hub: A case study of [8] A. K. Jain, A. Ross, and S. Prabhakar, "An Introduction to Biometric
the transformation of university roles in regional technological and Recognition," IEEE Trans. on Circuits and Systems for Video
economic development,” Research policy, vol. 37 (8), pp. 1188-1204, Technology, vol. 14 (1), 2004, pp 4-19.
2008. [9] A. K. Jain, L. Hong, and S. Pankanti, “Biometric Identification,”
[4] D. S. Siegel, D. A. Waldman, L. E. Atwater, and A. N. Link, “Toward a Communications of ACM, vol. 43 (2), 2000, pp. 91-98.
model of the effective transfer of scientific knowledge from [10] D. Kumlander, “Soft Biometrical Students Identification Method for e-
academicians to practitioners: qualitative evidence from the Learning,” in Advances in Computer and Information Sciences and
commercialization of university technologies,” J. of Engineering and Engineering, Springer, Netherlands, 2008, p. 114 – 118.
Technology Management, vol. 21(1-2), pp. 115-142, 2004. [11] P. Chang, A. Gul, "Towards Context-Aware Web Applications," 7th
[5] B. Readings, Education / Teaching, Harvard University Press, 1997. IFIP International Conference on Distributed Applications and
[6] E. González-Agulla, E. Argones-Rúa, C. García-Mateo, and Ó W. M. Interoperable Systems (DAIS), Springer, LNCS 4531, 2007, pp. 239–
Flórez, “Development and Implementation of a Biometric Verification 252.
System for E-learning Platforms,” EDUTECH, Computer-Aided Design [12] W.Y. Lum, F.C.M. Lau, “A context-aware decision engine for content
Meets Computer-Aided Learning, IFIP 18th World Computer Congress, adaptation,” IEEE Pervasive Computing, vol. 1, no.3, 2002, pp. 41–49.
2004, pp. 155-164. [13] J. Pascoe, “Adding Generic Contextual Capabilities to Wearable
Computers,” Proc. 2nd Int’l Symp. Wearable Computers, IEEE CS
[7] O. Guven, S. Akyokus, M. Uysal, and A. Guven, “Enhanced password
Press, Los Alamitos, Calif., 1998, pp. 92–99.
authentication through keystroke typing characteristics,” Proceedings of
Exploring Teachers’ Beliefs about Digital Citizenship
and Responsibility
Harrison Hao Yang
State University of New York at Oswego
Oswego, NY 13126, USA
[email protected]

Pinde Chen
South China Normal University
Guangzhou, 510631, China
[email protected]

Abstract – Promoting and modeling digital citizenship and This paper reviews related studies connected with the digital
responsibility can be a challenge to today’s teachers. This paper citizenship movement and teachers’ beliefs gleaned from the
presents a review of related studies connected with the digital literature. It then provides a study which explores teachers’
citizenship movement and teachers’ beliefs gleaned from the beliefs about digital citizenship and responsibility.
literature. It then reports a descriptive study which explores
teachers’ beliefs about and opinions on digital citizenship and II. LITERATURE REVIEW
responsibility.
A. Digital Citizenship
The movement to address and characterize digital citizenship
I. INTRODUCTION started in the UK, where educators have been working toward
Today’s world has moved from the industrial age into the establishing protocols for good digital citizenship since the
information age. The rapid technological change and mid-1990s [5]. According to Digizen.org, owned and operated
proliferation of information resources are lineaments of our by London-based nonprofit Childnet International,
contemporary society. Multimedia on the Internet, Digital citizenship isn't just about recognizing and
telecommunications, wireless applications, electronic portable dealing with online hazards. It's about building safe
devices, social network software, Web 2.0, etc., are all spaces and communities, understanding how to manage
radically redefining the way people obtain information and the personal information, and about being internet savvy -
way to teach and learn [1]. Findings of a 2007 national-wide using your online presence to grow and shape your
survey study Teens and Social Media conducted by the Pew world in a safe, creative way, and inspiring others to do
Internet and American Life Project show that some 93% of the same [6].
teens use the internet, and more of them than ever are treating In the US, Ribble, Bailey, and Ross defined digital
it as a venue for social interaction – a place where they can citizenship as “the norms of behavior with regard to technology
share creations, tell stories, and interact with others. In use” [4]. They identified nine elements of digital citizenship:
addition, content creation by teenagers continues to grow, with digital access, digital commerce, digital communication, digital
64% of online teenagers ages 12 to 17 engaging in at least one literacy, digital etiquette, digital law, digital rights and
type of content creation, up from 57% of online teens in 2004 responsibilities, digital health and wellness, and digital
[2]. security. Exploring these elements has led an array of studies
While the emerging technologies provide more opportunities focusing on the implementation of good digital citizenship in
and possibilities to enhance existing teaching and learning schools, such as assessments, plans, strategies, activities, etc.
activities, there is considerable evidence that technology [3] [4] [5] [7] [8] [9] [10] [11] [12]. As Ribble, Bailey, and
misuse and abuse is widespread and can be found inside and Ross indicated “digital citizenship speaks to several levels of
outside the school today: Web sites to intimidate or threaten responsibility for technology. Some issues may be of more
students, downloading music illegally from the Internet, concern to technology leaders while others may be of more
plagiarizing information using the Internet, using cellular concern to teachers” [4]. The recent National Educational
phones during class time, and playing games on laptops or Technology Standards for Teachers 2008 (NETS·T 2008)
handhelds during class, etc [3] [4]. As a result, digital prepared by the International Society for Technology in
citizenship has been received the increased attention of many Education (ISTE) has attested the digital citizenship movement
educators and has been considered as one of top priorities in and specialized its Standard IV for teachers to “Promote and
numerous school districts. “Everyone – administrators, board Model Digital Citizenship and Responsibility” [13]. Standard
members, teachers, parents and students – need to be involved IV requests teachers “understand local and global societal
in the dialogue about the appropriate use of technology” [3]. issues and responsibilities in an evolving digital culture and

49
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_9, © Springer Science+Business Media B.V. 2010
50 YANG AND CHEN

exhibit legal and ethical behavior in their professional to investigate teachers’ beliefs about digital citizenship and
practices.” In order to guide teachers meeting this standard, responsibility in particular.
NETS·T 2008 Standard IV has four indicators covering
III. THE STUDY
different aspects of digital citizenship and responsibility:
• Indicator A: advocate, model, and teach safe, legal, The purpose of this study was to examine teachers’ beliefs
and ethical use of digital information and technology, about and opinions on NETS·T 2008 Standard IV - Promote
including respect for copyright, intellectual property, and Model Digital Citizenship and Responsibility. The
and the appropriate documentation of sources. following questions guided this study:
• Indicator B: address the diverse needs of all learners 1. What were teachers’ beliefs about and opinions on
by using learner-centered strategies providing each indicator of NETS·T 2008 Standard IV?
equitable access to appropriate digital tools and 2. How could teachers adopt and implement the four
resources. indicators of NETS·T 2008 Standard IV into their
• Indicator C: promote and model digital etiquette and teaching practices?
responsible social interactions related to the use of A. Participants and the Course
technology and information. The participants of this study came from four sections of
• Indicator D: develop and model cultural students (n = 87) who were enrolled in the asynchronous online
understanding and global awareness by engaging with course entitled Multimedia and Internet for Educators, offered
colleagues and students of other cultures using digital- at a university in the northeastern region of the United States
age communication and collaboration tools [13]. during the winter, spring, summer, and fall terms in 2009. Most
B. Teachers’ Beliefs of the participants were part-time in-service teachers located in
In describing a working conception of beliefs, Bryan the same geographical area, who were pursuing graduate level
summarized “beliefs are psychological constructions that (a) education programs in content areas of literacy, biology,
include understandings, assumptions, images, or propositions chemistry, earth science, mathematics, social studies,
that are felt to be true; (b) drive a person’s actions and support technology, etc.
decisions and judgments; …” [14]. It is important to realize The course Multimedia and Internet for Educators provides
that pre- and in-service teachers’ opinions and briefs shape an introduction and guide to pre- and in-service teachers to
their actions. As Clark and Peterson pointed out: current and emerging technology. It intends to help PK-12
…the research shows that thinking plays an important educators not only to use multimedia and Internet resources in
part in teaching, and that the image of a teacher as a their own educational endeavors, but also integrate technology
reflective professional…is not far-fetched. Teachers do into their work environments as teachers. Since spring 2001,
plan in a rich variety of ways, and these plans have real this course has been joined in the State University of New
consequences in the classroom. Teachers make York Learning Network (SLN) as one of its asynchronous
decisions frequently (on every two minutes) during learning network courses offered to both on and off-campus
interactive teaching. Teachers do have theories and students and is an elective graduate course. In each module of
belief systems that influence their perceptions, plan, and this course, students were required not only to complete and
actions [15]. submit the assignment, but also to collaborate on specific
Similarly, Cuban stated, “The knowledge, beliefs, and learning activities required during a certain time frame: (a)
attitudes that teachers have shape what they choose to do in reading each “Mini-Lecture” and supplemental materials, (b)
their classrooms and explain the core of instructional practices responding to the questions in each Mini-Lecture at the
that have endured over time” [16]. Pajares found there was a “Online Discussion” section, (c) responding publicly to some
“strong relationship between teachers’ educational beliefs and or all of the questions submitted by classmates, and (d)
their planning, instructional decisions, and classroom replying to classmates who responded to the question.
practices” and that “research on the entering beliefs of pre- Asynchronous messages posted by the participants were
service teachers would provide teacher educators with threaded and could be sorted by author or by date for the
important information to help determine curricula and program participants’ viewing and reviewing preference.
direction and interpretation of knowledge and subsequent B. Data Collection
teaching behavior” [17]. To investigate teachers’ beliefs about and then to promote
In sum, the literature suggests that it is critical to investigate their understanding of digital citizenship and responsibility, in
the pre- and in-service teachers’ beliefs about technology the discussion area of Module 1 - Introduction of Multimedia
innovation and integration. Specifically, the pre- and in-service and Internet the participants were asked to provide the written
teachers’ beliefs and opinions indicate their willingness to responses to the following statement:
adopt a new teaching behavior such as promoting and People are excited by the power and popularity of the
modeling digital citizenship and responsibility in their Internet, in your opinion, what are the
professional practices. Although previous studies have ethical/social/human issues we should consider when we
provided a variety of practical ways on the implementation of implement digital technologies, especially the Internet-
good digital citizenship in schools, such as assessments, plans, based technologies into today's education?
strategies, activities, etc. in general, few studies have attempted
 EXPLORING TEACHERS’ BELIEFS ABOUT DIGITAL CITIZENSHIP AND RESPONSIBILITY 51

As indicated in Table 1, a total of 299 responses from • [Student A-Fall responded to Student B-Fall’s
participants were collected and analyzed by the instructor of response]
the course, who is the first author of this study as well. I think that if a teacher chooses to use the internet
during any type of project it is essential that the
TABLE 1 students learn it is not appropriate to copy someone
NUMBERS OF PARTICIPANTS AND RESPONSES
Participants Responses
else’s work. On top of that I think it is the teacher’s
Winter 09 26 92 job to double check each student’s work and make
Spring 09 20 69 sure they haven’t done so. I remember from student
Summer 09 21 91 teaching placement when I gave an essay assignment
Fall 09 20 47 and had a student plagiarize I took them aside and
Total 87 299
asked why they did so. They told me they had never
gotten caught before so they thought they could get
C. Findings
away with it. They understood they did something
Among 299 responses, while most of them addressed issues
wrong but had never gotten caught before. The
ranged over two or more indicators of NETS·T 2008 Standard
reinforcement of being caught would help prevent it
IV, some of them focused on one indicator of NETS·T 2008
in the future.
Standard IV. Table 2 listed numbers and percentages of
responses that pertained to each indicator. • [Student C-Fall responded to Student B-Fall’s
response]
TABLE 2 Whenever students are assigned to write an essay in
DISTRIBUTION OF PARTICIPANTS’ RESPONSES which they need to do research I feel the teacher
Frequency Percent should always give a mini lecture on the importance
Indicator A 210 70.2
Indicator B 44 14.7 of citing work to avoid plagiarism as well as making
Indicator C 125 41.8 students aware of the consequences they will face if
Indicator D 107 35.8 they are caught in the act of plagiarism. Students
should also be reminded of how to determine if a
First of all, the vast majority of responses (70.2%) were source is reliable and creditable as well as review on
falling into the Indicator A – “advocate, model, and teach safe, how to site works (or a handout) if necessary. My
legal, and ethical use of digital information and technology, content area is business but I always discussed
including respect for copyright, intellectual property, and the citations, references, and plagiarism with my students
appropriate documentation of sources.” The following online as it is important in every subject.
discussions in fall 2009, which involved three participants Secondly, although there were fewer responses (14.7%) on
exchanging their ideas on the plagiarism, were typical the Indicator B - “address the diverse needs of all learners by
examples: using learner-centered strategies providing equitable access to
• [The initial discussion on the plagiarism posted by appropriate digital tools and resources,” participants
Student A-Fall] thoroughly discussed this important area regarding the
Researching for papers the internet makes the implementation for their teaching practices. As were evident in
availability to plagiarism a lot easier for students. the following participants’ discussion in spring 2009:
Students even have the option to purchase papers of • [The initial discussion on the digital access posted by
the internet. The authenticity of individual work may Student A-Spring]
be more difficult to determine with the access of the The major social issue that I have been faced with is
internet. the fact that many of our students do not have access
• [Student B-Fall responded to Student A-Fall’s initial at home and it is important to be fair when giving
post] assignments which require this. In fact, I have gone
I agree that it is much more convenient and easy for so far as to meet students at the local library to help
students to plagiarize while using the World Wide them with a project that I assigned, so that they would
Web (sometimes without even knowing it!). Due to have the time outside of class, but my support, since
their early start to using the internet and their lack of they may not be as familiar with the technology as
knowledge about plagiarism this could definitely be a their peers.
problem that might occur when using the internet in • [Student B-Spring responded to Student A-Spring’s
the classroom. I think as educators we need to address initial post]
this problem, educate our students about plagiarism I applaud you for going to the local libraries to help
and teach them how to use the information correctly. your students. Why don’t they use the computers in
If a teacher chooses to use the internet for a research school? There are a number of students who don’t
project I think it is essential to address this issue in the have a computer at home do you have alternative
very beginning and spend time discussing how to version of the assignment for them?
complete a works cited to use the information legally. I do not think of lack of computers at home as a social
issues. Good thinking. I hope this is the reason why
52 YANG AND CHEN

some schools are developing laptop programs for their their new form of bathroom graffiti and harassment.
students. Teachers can't witness this cyber-bullying and parents
• [Student A-Spring responded to Student B-Spring’ don't even know it's going on most of the time. I
response] think this gives the bullies an even bigger sense of
Usually it's just one or two students who don't have power. They are unseen, and sometimes even
the access and it's for those kids who I will go above nameless, except for their screen name, which maybe
and beyond - although I could provide an alternate is known or unknown. When students are given
assessment, it is important to me that they have the computer usage time, it needs to be carefully
authentic experience like their peers. monitored by the network administrator and the
• [Student C-Spring responded to Student B-Spring’s supervising teacher. A students' history can be
response] tracked, but it is difficult for a teacher to monitor 20
My school had the laptop program (it was a students' activity when all it takes is a click of the
predictable disaster, and no longer is offered), and mouse to close a window you're not supposed to be in.
only the students who were leasing one of the laptops What are some effective ways to monitor computer
($$) got to take them home. The students who could usage by a classroom of students? How can educators
not afford to lease them only had access to them try to eliminate cyber-bulling in their schools?
during the day. They had to return them before going • [Student B-Winter responded to Student A-Winter’s
home. So, the disadvantaged students who probably initial post]
did not have a computer at home to begin with still As I think about your question about how educators
couldn't have access at home because they couldn't can try to eliminate cyber-bulling in our schools it just
afford to lease the laptops and take them home. From brings up such a difficult task. It makes me think of
the financial standpoint, it makes sense. The district just regular bullying and how much of it occurs
can't afford to have kids "lose" $800 laptops for which outside of school. How do we combat that? I've
there will be no financial recovery. talked with parents before whose children are on the
• [Student D-Spring responded to Student A-Spring’ social networks and they said from their standpoint,
initial post] it's just lots of monitoring. If anything, it seems like
As a teacher I am also facing with the same social cyber-bullying is easier to monitor because they leave
issue in my own classroom. It is great that students a trail of everything they are doing.
can access their homework from the internet or • [Student C-Winter responded to Student A-Winter’
interact with each other in a math game, but only for initial post]
the students that have a computer at home. Many of Great post and great questions! I found this article
my students do have a computer, but a few don't. The Tools Help Families Combat Cyber Bullying with
ones that do have a computer don't always have Alerts, Tips about a tool called "CyberBully Alert"
internet access at home. I choose not to really assign and thought the information was interesting to share -
any homework that requires using the internet for this “One such tool, recently unveiled by Vanden, allows
reason. We use it within our classroom. children to instantly notify selected adults when they
• [Student E-Spring responded to Student D-Spring’ are bullied or harassed online. CyberBully Alert
response] (http://www.cyberbullyalert.com/) also documents the
I agree, and only require use of computers when I am threatening message by saving a screen shot of the
supplying them in my classroom via laptop carts. It child's computer when the child triggers an alert. The
surprises me that so many people fail to consider Web-based solution is compatible with most
students lacking access to a computer when creating computers and Web browsers. It requires a single
assignments. download and registration. Parents then select which
Thirdly, approximately 42% of the responses addressed e-mail addresses and phone numbers will receive the
various aspects on the Indicator C – “promote and model alerts. Alerts are sent through e-mail and text message
digital etiquette and responsible social interactions related to when a child clicks on the CyberBully Alert icon…
the use of technology and information.” The following online The creators of CyberBully Alert also offer a School
discussions in winter 2009, which involved three participants Incentive Program” (cited from Information Week,
exchanging their ideas on the cyber-bulling, were typical October 1, 2008).
examples regarding one of new-fangled and concrete issues Finally, there were approximately 36% of the responses
related to Indicator C: falling into the Indicator D – “develop and model cultural
• [The initial discussion on the cyber-bully posted by understanding and global awareness by engaging with
Student A-Winter] colleagues and students of other cultures using digital-age
Bullying has always been an issue in schools among communication and collaboration tools.” Participants reported
students. Now they are able to do it behind computer their deep awareness of this indicator. As were evident in the
monitors and cell phones. Some of this is being done following participants’ discussion in summer 2009:
right from computers and cell phones at school. It's • [The initial discussion posted by Student A-Summer]
 EXPLORING TEACHERS’ BELIEFS ABOUT DIGITAL CITIZENSHIP AND RESPONSIBILITY 53

My concern is that when students can be a part of the • [Student C-Summer responded to Student A-
“social clique” created with technology, how will that Summer’s response]
affect their ability to be social in the real world? This is an interesting thought because there are many
When someone is either at the keyboard or with communication resources that students can use. There
headsets on, will their real world collapse into email, is a fine line between replacing technology-based
YouTube and iPod? Will families still have the same communication with face to face communication.
connections, when each as an individual has the Personally, I believe the technology-based
ability to create their Second Life avatar, design their communication that we have today is helping our
own room and hang out with people who are only like society stay in touch. Although it is not face to face
them? Are we building communities for our children communication, it does not mean that people are
that are physical and in real time? I realize these are replacing technology with going out and meeting
questions, but when our students are being prepared people. This technological communication surge is
for a high technology workplace, are their social and simply helping stay in touch with people on a 24-7
moral obligations to each other as community citizens basis.
being addressed? I also want to note that as our youth
IV. CONCLUSION
are exposed the ethical and social differences of other
countries, other religions and the range of human The findings of this study may produce useful information
behavior throughout the world, will our culture be that will enable administrators, program planners, and teachers
affected? That is where educators have the moral to better understand teachers’ beliefs about digital citizenship
obligation of understanding how we are exposed to and related responsibilities. As Ribble and Bailey pointed,
differences and how they are discussed can have “The endless list of misuse and abuse includes hacking into
impact our personal lives and future choices. school servers, using e-mail to intimidate or threaten students,
• [Student B-Summer responded to Student A- illegally downloading music, plagiarizing information from the
Summer’s initial post] Internet, using cellular phones during class time, accessing
The issues you have raised should be a concern for pornographic Web sites, and playing video games during class.
everyone. More and more people are not Therefore, if you are using technology in your district, you
communicating face to face. I think the implications must begin to deal with digital citizenship in a significant way”
that this can have on society can be detrimental to [10]. To capture the improvements in digital citizenship and
how we all communicate with each other. I sometimes responsibility, teachers must be prepared – through sharing and
find myself texting to a friend or family member when expressing their beliefs and opinions – to learn what digital
I could easily call them and verbally say what I citizenship is and what it can do. Indeed, there have been some
needed to. Progressively, since the continuing line of of valuable resources and activities related to digital citizenship
new advancements in technology, the idea of existing for teachers and educators working on curriculum. For
community has declined in society. It’s scary to think example, for teachers to teach Internet safety, some of lessons
that more and more people have replaced the ideas of and activities including electronic simulations are available
community with thoughts concerning only online [18]:
themselves. As educators I think it is important that • Cybersmart (http://www.cybersmart.gov.au/) is
we continually stress the importance of community developed by the Australian Communications and
and help students to understand the value of working Media Authority, and “provides activities, resources
together. By exposing students to the benefits of and practical advice to help young kids, kids, teens
community it will help them understand that in order and parents safely enjoy the online world” [19].
to thrive, community cooperation is essential. Also, I • i-SAFE (http://www.isafe.org/) is endorsed by the
think that our culture will undoubtedly be affected. U.S. Congress, and “incorporates classroom
Cultural diffusion does not only take place through curriculum with dynamic community outreach to
means on face to face interaction. With the internet empower students, teachers, parents, law enforcement,
students can view videos of virtually any society that and concerned adults to make the Internet a safer
they wish. Students can find information on remote place” [20].
villages in the Amazon rainforest, something few • NetSmartz Workshop (http://www.netsmartz.org) “is
students of past generations could do. I do believe that an interactive, educational safety resource from the
this exposure to other cultures will be beneficial to National Center for Missing & Exploited Children®
students. Educators can utilize this resource to help (NCMEC) and Boys & Girls Clubs of America
teach students to understand other cultures and (BGCA) for children aged 5 to 17, parents, guardians,
societies, working to break stereotypes and prejudices. educators, and law enforcement that uses age-
I hope that affects on our culture will be beneficial; appropriate, 3-D activities to teach children how to
however educators need to understand their stay safer on the Internet” [21].
importance to this cultural diffusion. • ID the Creep (http://www.idthecreep.com) is an
online game sponsored by the National Center for
54 YANG AND CHEN

Missing & Exploited Children and The Ad Council [8] M. S. Ribble and G. D. Bailey, “Digital citizenship: Focus questions for
implementation,” Learning & Leading with Technology, vol. 32 issue 2,
for young students to practice identifying pedophiles pp. 13-15, 2004.
online. [9] M. S. Ribble and G. D. Bailey, “Developing ethical direction,” Learning
If the strong beliefs can be built, then better teaching & Leading with Technology, vol. 32 issue 7, pp. 36-37, 2005.
behavior and learning activities can be developed and [10] M. S. Ribble and G. D. Bailey, “Monitoring technology misuse & abuse:
A five-step plan for creating a digital citizenship program in your
implemented. For this reason, beliefs about digital citizenship school,” T. H. E. Journal, August 2004.
and responsibility among teachers may be an important < http://www.digitalcitizenship.net/uploads/T.H.E.JournalArticle.pdf>
consideration in the process of incorporating NETS·T into real- [11] M. S. Ribble and G. D. Bailey, “Is digital citizenship a problem in your
world teaching, with options for more concrete actions for school?” Technology Turning Points, December 2005.
<http://www.digitalcitizenship.net/uploads/TechnologyTurningPointsTea
teachers. cherTalk.pdf>
[12] M. S. Ribble and G. D. Bailey, Digital citizenship in schools, ISTE,
REFERENCES 2007.
[1] S. Wang and H. Yang, “Editorial,” International Journal of Technology [13] International Society for Technology in Education, National education
in Teaching and Learning, vol. 3 issue 2, 2007. technology standards for teachers, 2008.
[2] A. Lenhart, M. Madden, A. R. Macgill, and A. Smith, “Teens and social <http://www.iste.org/Content/NavigationMenu/NETS/ForTeachers/2008
media: The use of social media gains a greater foothold in teen life as Standards/NETS_for_Teachers_2008.htm>
they embrace the conversational nature of interactive online media,” Pew [14] L. A. Bryan, “Nestedness of beliefs: Examining a prospective elementary
Internet & American Life Project, 2007. teacher’s belief system about science teaching and learning,” Journal of
<http://www.pewinternet.org/~/media//Files/Reports/2007/PIP_Teens_So Research in Science Teaching, vol. 40 issue 9, pp. 835-868, 2003.
cial_Media_Final.pdf.pdf > [15] C. M. Clark and P. L. Peterson, “Teacher’s thought process,” in M. C.
Wittrock (ed.). Handbook of research on teaching, New York: McMillan
[3] M. S. Ribble, “Point of view on technology drivers’ licenses,” District
Publication, 1986.
Administrator Magazine, 2004.
[16] L. Cuban, “Computers meet classroom: Classroom wins,” Teachers
< http://www.digitalcitizenship.net/uploads/TechDL.pdf>
College Record, vol. 95 issue 2, pp. 185–210, 1993.
[4] M. S. Ribble, G. D. Bailey, and T. W. Ross, “Digital citizenship:
[17] F. M. Pajares, F. M., “Teachers’ beliefs and educational research:
Addressing appropriate technology behavior,” Learning & Leading with
Cleaning up a messy construct,” Review of Educational Research, vol. 62
Technology, vol. 32 issue 1, pp. 7-11, 2004.
issue 3, pp. 307-332, 1992.
[5] M. Villano, “Text unto others…as you would have them text unto you,”
[18] A. McAnear, “Digital citizenship,” Learning & Leading with
T. H. E. Journal, September 2008.
Technology, 2007.
<http://thejournal.com/articles/2008/09/01/text-unto-others-as-you-
<http://www.iste.org/Content/NavigationMenu/Publications/LL/LLIssues
would-have-them-text-unto-you.aspx>
/Volume_34_2006_2007_/March_6_/34604m.pdf>
[6] What makes you such a good digizen? < http://www.digizen.org/>
[19] < http://www.cybersmart.gov.au/>
[7] J. Sturgeon, “Raising good digital citizens: Teachers are developing new
[20] < http://www.isafe.org/>
ways for students to follow good internet and media use practices,” Cable
[21] < http://www.netsmartz.org/overview/aboutus.htm>
in Classroom, November 2008.
 Faculty Driven Assessment of Critical Thinking:
National Dissemination of the CAT Instrument
Barry Stein, Ada Haynes, Michael Redding, Kevin Harris, Mark Tylka, and Elizabeth Lisic
Tennessee Technological University

Abstract- This paper reports the latest findings of a project to Tennessee Technological University (TTU) has been en-
nationally disseminate the CAT© instrument, a unique interdis- gaged in an extended effort during the last 9 years to develop
ciplinary assessment tool for evaluating students’ critical think- and refine an instrument to assess critical thinking that over-
ing skills. Tennessee Technological University originally part- comes many of the weaknesses of other existing tools. Pree-
nered with six other institutions across the U.S. to evaluate and
minent theoreticians and educators in the area of learning
refine the CAT instrument beginning in 2004. Through these
efforts a test with high face validity, high construct validity, high sciences and assessment participated in the project. Unlike
reliability, and that is culturally fair was developed In the cur- many other available assessment tools, the Critical thinking
rent national dissemination project, we are collaborating with Assessment Test (CAT) instrument uses short answer essay
over 40 institutions across the country to disseminate this unique responses to assess critical thinking. Although essay ques-
instrument and support the creation of assessment based efforts tions can be harder to score than multiple choice questions,
to improve student learning. These dissemination efforts involved they provide a better understanding of students’ thought
training representatives from other institutions to lead scoring processes and ability to think critically and creatively when
workshops on their own campuses as part of a broader effort to confronted with real world problems [6]. In addition, the CAT
improve student learning. A variety of findings indicate these
instrument is unique in that it utilizes a campus’s own faculty
dissemination efforts are successful. National user norms that are
being collected from this project indicate that largest gains in to evaluate student responses. Ewell [7] has noted that one
undergraduate critical thinking appear to occur in the junior and problem with many existing assessments is that “we let as-
senior years. 1 sessment get excessively distanced from the day-to-day busi-
ness of teaching and learning.” The CAT instrument allows
INTRODUCTION faculty to directly observe students’ critical thinking and un-
derstand their students’ weaknesses. This activity helps moti-
In our increasingly technological and information driven vate faculty to consider changes in pedagogy that might im-
society, the ability to think critically is a cornerstone to both prove students’ critical thinking skills [8]. This becomes in-
workplace development and effective educational programs. creasingly important as accrediting agencies such as ABET
The importance of critical thinking skills in higher education increase their focus on efforts to improve students’ critical
has become increasingly apparent. Accrediting agencies such thinking [1].
as Accreditation Board of Engineering and Technology TTU’s approach to developing the CAT instrument has
(ABET) recognize the need for higher order thinking skills been to involve faculty in the identification of a core set of
and real world problem solving in their accreditation standards skills they believe to be an important part of critical thinking
[1]. Faculty also recognize the importance of critical thinking across many disciplines (see Table 1). TTU worked with a
skills. According to Derek Bok [2], over ninety percent of the diverse group of institutions nationwide (University of Texas,
faculty in the U.S. feel that critical thinking is the most impor- University of Colorado, University of Washington, University of
tant goal of an undergraduate education. Despite the recog- TABLE 1
nized importance of critical thinking skills, it is difficult to SKILL AREAS ASSESSED BY THE CAT INSTRUMENT
find effective tools for assessing these skills. Many of the
assessment tools are plagued by problems related to validity, Evaluating Information
reliability, and cultural fairness [3]. According to Bransford, Separate factual information from inferences.
Interpret numerical relationships in graphs.
Brown, and Cocking [4] “a challenge for the learning sciences Understand the limitations of correlational data.
is to provide a theoretical framework that links assessment Evaluate evidence and identify inappropriate conclusions.
practices to learning theory.” Pellegrino, Chudowsky and Creative Thinking
Glaser also note that assessment instruments need to be devel- Identify alternative interpretations for data or observations.
oped based upon principles of learning and cognition [5]. Identify new information that might support or contradict a hypothsis.
Explain how new information can change a problem.
Learning and Problem Solving
1
Separate relevant from irrelevant information.
Partial support for this work was provided by the National Science Founda- Integrate information to solve problems.
tion’s CCLI Program under grants 0404911 and 0717654. Any opinions, Learn and apply new information.
findings, and conclusions or recommendations expressed in this material are
Use mathematical skills to solve real-world problems.
those of the authors and do not necessarily reflect the views of the National
Communication
Science Foundation.
Communicate ideas effectively.

55
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_10, © Springer Science+Business Media B.V. 2010
56 STEIN ET AL.

Hawaii, University of Southern Maine, and Howard Universi- nister the CAT instrument to their students and to conduct
ty) to further develop, test, and refine this instrument. The scoring workshops with their own faculty to score their stu-
results of that project revealed the CAT instrument had high dents’ responses on the CAT instrument. Since institutional
face validity when evaluated by a broad spectrum of faculty interest in the CAT instrument greatly exceeded that level of
across the U.S. in STEM and non-STEM disciplines, had good participation supported by NSF, we created an alternative me-
criterion validity when compared to other instruments that thod of institutional participation that was self-supporting.
measure critical thinking and intellectual performance, had We are currently collaborating with over 40 institutions across
good scoring reliability, good test/re-test reliability, and good the country. Grant funded participants were selected to insure
construct validity using expert evaluation in the area of learn- diversity in mission, institutional characteristics, ethnici-
ing sciences [8]. ty/racial demographics, and geographic location. The collabo-
Student response to the CAT instrument has also been very rating institutions range from community colleges to large
positive. Many students indicate appreciation for the real research institutions and include private, as well as, publicly
world problem solving tasks that make up the test [9]. Using funded institutions.
questions that are interesting and engaging helps motivate
students to perform their best. Materials
Performance on the CAT instrument is significantly corre- A variety of training materials were developed to support
lated with other measures of student performance including the training of new scoring workshop leaders and to help those
cumulative college grade-point average (r = 0.295), entering leaders train their own faculty to effectively score the CAT
SAT or ACT scores (r = 0.528, and r = 0.560, respectively), instrument on their respective campuses. These materials in-
and performance on several other tests that are thought to cluded a narrated overview of the development process for the
measure certain skills associated with critical thinking such as CAT instrument that streams from our website
the California Critical Thinking Skills Test (r = 0.645) and the (www.CriticalThinkingTest.org), a detailed training manual
CAAP Critical Thinking Module (r = 0.691). These correla- that covers issues from sampling and obtaining institutional
tions provide support for the criterion validity of the CAT review board approval to administering the test, recruiting
instrument. However, the magnitude of the correlations also faculty scorers, conducting the scoring workshop, and inter-
indicate that the CAT instrument measures something differ- preting institutional results. A detailed scoring guide and
ent than the other assessment tools [9]. Performance on the multi-media computer based training modules were also de-
CAT instrument has also been found to significantly correlate veloped to explain question scoring.
with several key questions on the National Survey of Student
Engagement (NSSE). One NSSE question that is significantly Train-the-Trainer Workshops
negatively correlated with students’ performance on the CAT Participating institutions sent two to three representatives to
instrument is the extent to which students think their courses a two-day training workshop where they received extensive
emphasized rote retention [10]. In contrast, CAT performance training in scoring the CAT instrument and preparation to lead
is significantly positively correlated with NSSE questions that a scoring workshop on their own campus. The workshop also
reflect the extent to which students’ think their college expe- covered other topics related to understanding and using differ-
rience emphasized critical thinking and real-world problem ent assessment designs, sampling, and using the scoring ses-
solving. sion as an opportunity for faculty development. One of the
Extensive work has been done to refine the scoring process advantages of the CAT instrument is that it gives faculty a
for the CAT instrument to improve scoring accuracy and re- good understanding of student weaknesses. Coupling this
liability. For example, the scoring process uses multiple scor- experience with a discussion of ways to improve student
ers for each question and uses a detailed scoring guide that has learning creates an effective faculty development experience
been refined with extensive faculty input to ensure scoring that can lead to real improvements in instruction and student
consistency. Although scoring accuracy is typically a problem learning.
for essay type tests, the scoring process for the CAT instru-
ment overcomes many of the problems typically associated RESULTS
with scoring essay tests.
The major focus of the new work reported in this paper in- The effectiveness of the train-the-trainer workshops were
volves efforts to disseminate the instrument to a broad range evaluated in a number of ways including; participant surveys,
of institutions across the U.S. These efforts involved training direct observations by our external evaluator, follow-up onsite
representatives from other institutions to lead scoring work- observations of campus scoring sessions, and scoring accuracy
shops on their own campuses as part of a broader effort to checks of tests scored by each institution.
improve student learning.
Training Workshop Surveys
METHOD At the conclusion of each of the training workshops partici-
pants were asked to complete an anonymous survey to eva-
Funding from the National Science Foundation supported luate various components of the workshop training. The sur-
collaboration with 20 institutions across the country to admi- vey used a 6 point Likert type scale: 6 – Strongly Agree, 5 –
 FACULTY DRIVEN ASSESSMENT OF CRITICAL THINKING 57

Fig. 1 Fig. 2
Average Training Workshop Evaluation Responses (with 95% CI) Results of Scoring Accuracy Checks Across Institutions as Percent Error

Agree, 4 – Slightly Agree, 3 – Slightly Disagree, 2 – Disagree, about the success of this national dissemination model comes
1 – Strongly Disagree. Eighty-five participants were included from checking the accuracy of scoring performed by each par-
in this analysis. ticipating institution. Experienced scorers at our institution
The mean response on each survey item is shown in Figure rescore randomly selected tests from each scoring session at
1 (error bars indicate 95% confidence interval). The results other institutions. The sample includes about 15% to 20% of
indicate that the workshop evaluations have been quite posi- the tests scored by each institution. Error rates below 5% are
tive across the topics surveyed. considered acceptable. Thus far, overall accuracy has been
very good. Figure 2 shows the overall test scoring error found
Qualitative Evaluation of Training Workshops across 21 institutional scoring sessions that have undergone
Observations by our external evaluators and project staff accuracy checks. About 1 in 7 institutions has been found to
are used formatively to continually improve the training work- deviate more than 5% from our experienced scorers. All insti-
shops. For example, we noted early in the project that partici- tutions receive a scoring accuracy report together with a ques-
pants in the training workshops may find it difficult to process tion by question analysis of accuracy, and suggestions for im-
information from too many different sources (the workshop proving scoring accuracy in subsequent sessions. Recommen-
leader, the multimedia training modules, and the scoring dations are given to institutions that deviate more than 5% on
guide). This led to changes in how the multi-media training how to appropriately adjust their scores for comparison to
module was used in the workshop. It is now used during the other institutions.
second day as a resource for explaining scoring in situations
where scorers may need further explanation. Many other National Norms
changes were also made in response to qualitative evaluations Another goal of the current project is to begin assembling
(e.g., extending the discussion of sampling methods and ex- national norms for various populations of students including
ploring methods for obtaining representative samples at dif- community college students and students at four-year institu-
ferent institutions, and encouraging participants to develop tions. Although many institutions will track their progress by
discipline specific analog learning activities that correspond to
questions on the CAT instrument). 20.0

19.0
Onsite Evaluation of Scoring Workshops
Observations by our external evaluators and project staff 18.0
who have attended scoring workshops at other institutions
Mean

have been particularly useful. These observations have con- 17.0

firmed that procedures outlined in our training manual must be
carefully adhered to in order to efficiently and accurately 16.0

score tests. Institutions that have deviated in important ways

15.0
from those guidelines experienced numerous problems. These
observations have led to increased emphasis and training sup- 14.0
port to explain various procedures associated with test scoring.
Freshman Sophomore Junior Senior

Scoring Accuracy Checks Fig. 3

One of the most important pieces of evaluative information Mean Scores on the CAT Instrument for Students at 4-year Institutions
58 STEIN ET AL.

comparison to their own test scores over a period of time, oth- and can begin to explore modifications in teaching methods
ers have requested information that would allow comparison that might address these weaknesses. Many of the participants
to larger populations. Figure 3 shows the mean score on the in our regional training workshops feel that the CAT instru-
CAT instrument for approximately 3000 students of different ment is particularly useful because it helps faculty understand
class standings at four-year institutions across the country. how they can develop discipline specific assessments that en-
The largest gains in critical thinking appear to occur in the courage students to think critically instead of just encouraging
junior and senior years of undergraduate education. The aver- rote retention. This becomes increasingly important as accre-
age senior score was about 51% of the maximum possible diting agencies such as the Accreditation Board of Engineer-
score on the CAT instrument (maximum score = 38 points). ing and Technology (ABET) increase their focus on efforts to
These scores indicate there is considerable room for improve- improve student learning.
ment. The fact that no student or has obtained a perfect score
or a score of zero at any institution suggests the absence of a REFERENCES
ceiling or floor effect.
[1] Accreditation Board of Engineering and Technology, Criteria for Accre-
CONCLUSION diting Engineering Programs: Effective for Evaluations During the 2006-
2007 Accreditation Cycle. Baltimore, MD: Accreditation Board of Engineer-
ing and Technology, 2005
There have been several significant outcomes of the current [2] D. Bok, Our Underachieving Colleges: A candid look at how much stu-
national dissemination project. We have found that our dis- dents learn and why they should be learning more. Princeton: Princeton
semination model for training representatives at other institu- University Press, 2006.
[3] U.S. Department of Education, National Center for Education Statistics.
tions to use the CAT instrument at regional train-the-trainer The NPEC Sourcebook on Assessment, Volume 1: Definitions and Assess-
workshops is successful. This finding is supported by infor- ment Methods for Critical Thinking, Problem Solving, and Writing, NCES
mation from participant surveys, onsite observations, and 2000—172, prepared by T. Dary Erwin for the Council of the National Post-
scoring accuracy checks. We do, however, continue to modify secondary Education Cooperative Student Outcomes Pilot Working Group:
Cognitive and Intellectual Development. Washington, DC: U.S. Government
our training methods and supporting materials to improve ef- Printing Office, 2000.
fective dissemination. [4] J. Bransford, A. Brown, and R. Cocking, (Eds.) How People Learn:
We have found that many institutions are interested in find- Brain, Mind, Experience, and School. Washington, D.C.: National Academy
ing ways to engage faculty in quality improvement efforts. Press. 2000.
[5] J. Pellegrino, N. Chudowski, and R. Glaser (Eds.) Knowing What Students
Faculty involvement in the scoring of student essay exams Know. Washington, D.C., 2001.
greatly facilitates this process. We are also expanding our ef- [6] B. Stein, A. Haynes, and M. Redding. “Project CAT: Assessing Critical
forts to provide links to institutional and grant funded projects Thinking Skills,” in Proceedings of the National STEM Assessment Confe-
that have positively impacted student performance on the CAT rence, D. Deeds and B. Callen , Eds. NSF, 2007.
[7] P. Ewell. “The examined life: Assessment and the ends of general educa-
instrument. This type of information is just beginning to tion,” Association of American Colleges & Universities General Education
emerge from a broad range of institutional initiatives. We are and Assessment Conference, 2004.
also collaborating with several other NSF funded projects that [8] B. Stein, A. Haynes, M. Redding, T. Ennis, and M. Cecil. “Assessing
are using the CAT instrument to help identify potential student Critical Thinking in STEM and Beyond,” in Innovations in E-learning, In-
struction Technology, Assessment, and Engineering Education, M. Iskander,
gains in critical thinking and real-world problem solving that Ed. Springer, 2007, pp.79-82.
may result from innovative educational pedagogies. [9] B. Stein, A. Haynes, and K. Harris. “Assessing and Improving Critical
The CAT instrument is one of the few interdisciplinary as- Thinking Using the CAT Instrument,” ABET Best Assessment Processes
sessment instruments available that also provides an opportu- Symposium XI. Indianapolis, IN: Accreditation Board of Engineering and
Technology, 2009.
nity for faculty development. By participating in the scoring [10] M. Cecil, “The Relationship Between Student Responses on the Nation-
process, faculty become aware of their students’ weaknesses al Survey of Student Engagement (NSSE) and Performance of the Critical-
Thinking Assessment Test (CAT),” Dissertation, Tennessee Technological
University, 2006.
 Enhancing the Instruction of Introductory Electric
Circuit Courses using MATLAB

Thomas Yang, Member, IEEE, Ilteris Demirkiran, Member, IEEE, Jianhua Liu, Member, IEEE,
Albert Helfrick, Senior Member, IEEE, David Pedersen, and Christopher Grant

MATLAB ideal for assisting classroom instruction. It is also

Abstract—This paper discusses the implementation and capable of producing interactive simulation. This can enhance
assessment of the integration of MATLAB usage into the the quality of presentation with graphical simulations [4].
instruction of an introductory circuit analysis course at Embry- Although there have been several studies on using
Riddle Aeronautical University (ERAU). This teaching practice
seeks to enhance student learning by delivering relevant course
MATLAB in electrical engineering curriculum [5-10], we
materials using MATLAB’s graphical representation and believe there has not been enough emphasis within
problem solving capabilities. Also, students learn to plot voltage/ educational institutions on the inclusion of this software in the
current waveforms and other important graphs via simple undergraduate curriculum, especially in introductory circuit
MATLAB programs. The effectiveness of the teaching practice is analysis courses for non-electrical engineering majors. For
assessed through assignments and a survey. Assessment data example, at ERAU, engineering students study the basics of
indicates that integration of MATLAB into course instruction
promotes learning and enhances students’ computer skills. In
MATLAB briefly as freshmen in the course “Introduction to
addition, students highly appreciate the effectiveness of this Computing for Engineers”. However, the use of MATLAB is
teaching practice. not introduced in sophomore and junior years when students
learn discipline-specific subjects. Although many senior
projects involve heavy use of MATLAB, it is often difficult to
refresh students’ MATLAB skills at the senior level because
Index Terms— MATLAB, circuit analysis, assessment, they have not been consistently using the software.
engineering education In this paper, we describe our efforts in addressing this
inadequacy in a junior level Electrical Engineering (EE)
course by integrating MATLAB in the illustration of
I. INTRODUCTION
important concepts in basic circuit analysis. Our objectives are

W idely considered as the “language of technical

computing” [1], MATLAB is used extensively in the
scientific and engineering communities for research and
to enhance students’ computer skills and improve their
understanding of important circuit analysis concepts through
graphic representations and visualization in MATLAB.
development. With its universal approach to use the matrix as The effectiveness of this teaching practice is assessed
a fundamental data structure, its large library of pre-made through homework assignments and an end-of-course student
functions, and readily available toolboxes, Matlab finds survey. Collected assessment data confirmed the benefits of
application in all fields of engineering [2]. Hence, the ability MATLAB usage in classroom instruction and its positive
to develop solutions to technical problems using MATLAB is impacts on student motivation.
becoming increasingly important for graduates of engineering Our efforts were funded by ERAU Center for Teaching and
schools. MATLAB allows students to focus on course work Learning Excellence (CTLE) in Spring 2008 under
and applications rather than on programming details. Writing “Advancing Teaching Practice Grant”. It also serves as the
MATLAB codes to perform numerical calculations only initial study for ERAU College of Engineering’s additional
requires a fraction of the time it would take to write a program efforts to incorporate MATLAB component in undergraduate
in a lower level language. MATLAB helps engineers better curriculum from freshman to senior years.
understand and apply concepts in applications [3]. Powerful
graphical capabilities and simple grammar structures make
II. IMPLEMENTATION
This work was supported by Center for Teaching and Learning Excellence Our teaching grant project was implemented in three
at Embry-Riddle Aeronautical University, Daytona Beach, Florida. sections of “Electrical Engineering I” in Spring 2008. This is
T. Yang, I. Demirkiran, J. Liu, A. Helfrick, C. Grant are with College of an introductory course in circuit analysis and also the first
Engineering, Embry-Riddle Aeronautical University, Daytona Beach, Florida
32114, USA (Corresponding author: Thomas Yang, phone: 386-226-7098; e- course in Electrical Engineering (EE) taken by non-EE majors.
mail: [email protected]). The total number of students in these sections is 102. Majority
D. Pedersen is with Center for Teaching and Learning Excellence at are junior and senior students majoring in Aerospace
Embry-Riddle Aeronautical University, Daytona Beach, Florida 32114.

59
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_11, © Springer Science+Business Media B.V. 2010
60 YANG ET AL.

engineering (AE), Mechanical Engineering (ME), Civil Sample instruction materials are included in Section III of
Engineering (CivE) and Engineering Physics (EP). this paper.
We used MATLAB to illustrate many important concepts in After students were taught how to use MATLAB to plot
circuit analysis, including: desired graphs, they were assigned simple problems for
practice purposes. Sample MATLAB codes were posted on
A. Solving linear systems of equations resulted from Nodal
Blackboard as examples. In MATLAB assignments, we
Analysis and Mesh Analysis
emphasize that the results obtained from computer software
As fundamental circuit analysis methods, Nodal Analysis should confirm, validate and match the results from analysis
and Mesh Analysis utilize Kirchhoff’s current and voltages by hand. For example, the solution of a linear system of
laws to solve linear circuit problems. Often, it is tedious to equations from Nodal Analysis obtained in MATLAB should
solve the resulting linear systems of equations by hand. In match the solution obtained by hand. Also, in determining the
MATLAB, the solution can be obtained via simple commands. types of passive filters, the answer can be found by comparing
In our project, we introduce two straightforward ways to the magnitudes of the transfer function at DC (Direct Current)
perform this task, i.e., solution based on matrix inverse and and at very high frequency. This answer should be confirmed
solution based on Gaussian elimination. by the magnitude response plot generated in MATLAB.
B. Voltage and current across capacitors and inductors In final review of course materials, MATLAB was again
used to assess students’ understanding of key concepts. Here
The current and voltage waveforms are plotted and
are some examples:
compared using MATLAB. The relationship between the
waveforms, which involves differentiation and integration, is
1. To evaluate students’ understanding of the relationship
clearly shown when they are plotted in the same window.
between the voltage and current across capacitors/inductors, a
Therefore, the related formulas become easier to understand
multiple choice question is constructed with MATLAB. For
and memorize.
example, a voltage waveform across a capacitor is plotted
C. AC Circuit Analysis together with three possible current waveforms. Students are
The AC current and voltage waveforms are plotted in the asked to identify which of the three is the current
same MATLAB window. The phase difference of the two corresponding to the given voltage.
waveforms, which is key to understand and analyze AC
circuits, is easily seen. 2. To evaluate if students are able to distinguish three types
of second-order transient responses, i.e., overdamped,
D. First and Second order transient analysis for circuit underdamped and critically damped responses, multiple plots
behavior after switches change position
are generated through MATLAB. Students are asked to
The model to describe transient process is differential determine the type of the transient response for each plot.
equations. Merely solving the differential equations does not
provide much insight into the dynamic evolution of the 3. To assess students’ ability to associate frequency
physical process. Using MATLAB, the transient responses are selectivity of bandpass filters with quality factor and
accurately plotted which give clear illustration of the physical bandwidth, magnitude responses of multiple bandpass filters
meaning of the solution. Moreover, the forced and natural with same center frequency are plotted using MATLAB.
responses are also clearly shown together with the complete Students are asked to order the quality factors and bandwidths
response. of those bandpass filters.
E. Frequency characteristics of passive filters and the
phenomenon of “Resonance” Sample review questions are included in Section 3 of this
paper.
Frequency response of passive filters is critically important
to understand the principles behind modern electronic
equipments. In our project, the magnitude response of the
filter transfer function is plotted using MATLAB. Important
concepts, such as resonance frequency, cutoff frequency,
bandwidth, and quality factor, are all clearly illustrated in the
plots. This makes the understanding of these concepts almost
effortless.
 ENHANCING THE INSTRUCTION OF INTRODUCTORY ELECTRIC CIRCUIT COURSES 61

III. SAMPLE INSTRUCTIONAL MATERIALS C. Response of a second-order underdamped transient

circuit
A. The relationship between an inductor voltage and
current Complete response of underdamped second-order circuit
2
complete response
0.1 exponential decay factor
1.5 sinusoidal factor
i(t), (mA)

forced response
0.05
1

Current, A
0
0 5 10 15 0.5
time, (ms)
0.4
0
0.2
vL(t), (V)

0 -0.5

-0.2
-1
-0.4 0 0.005 0.01 0.015 0.02 0.025 0.03
0 5 10 15
time, (ms)
Time, s

B. Complete, natural and forced responses of first order D. Using magnitude responses of second-order bandpass
RC transient circuit filters to illustrate the concepts of resonance, quality factor
Complete, natural and forced response of RC circuit
and bandwidth
15

Magnitude response of second-order band-pass filter

1
10 vc(t) Q=0.25
0.9 Q=0.5
vcN(t)
Q=0.75
vcF(t) 0.8
5 Q=1
0.7 Q=2
Volts

Q=4
Amplitude ratio

0.6 Q=8
0
0.5

0.4
-5
0.3

0.2
-10
0 0.5 1 1.5 2 0.1
time, (s)
0
-2 -1 0 1 2
10 10 10 10 10
Normalized frequency, w/wn
62 YANG ET AL.

E. Sample Review Question: given the capacitor voltage, challenging subject under our new teaching practice.
choose the corresponding capacitor current Assessing their subjective opinions is very helpful also
original capacitor voltage choice A because student motivation is of paramount importance in
6 0.8
STEM (Science, Technology, Engineering, and Math)

i(t), (amperes)
disciplines. Therefore, we designed an end-of-course survey to
v(t), (volts)

0.6
4
0.4 assess our teaching practice from the students’ perspective.
2
0.2 Below is the list of survey questions we used:
0 0
0 2 4
time, (seconds)
6
-6
0 2 4
time, (seconds)
6
-6
1. I can use MATLAB to perform matrix and vector
choice B
x 10
choice C
x 10
operations.
0.8 6
i(t), (amperes)

0.6 i(t), (amperes) 2. I can use MATLAB to solve linear systems of equations
4
0.4 resulted from Nodal Analysis and Mesh Analysis.
2
0.2
3. I can use MATLAB to plot data and mathematical
0 0
0 2 4 6 0 2 4 6 functions such as voltage and current across resistor, capacitor
time, (seconds) time, (seconds)
or inductor.
-6 -6
x 10 x 10

4. I can use MATLAB to plot responses of first and second

F. Sample Review Question: for each of the following order transient circuits.
figures, determine the types of second-order response (over-
damped, under-damped or critically damped) 5. I can use MATLAB to plot frequency responses of
passive filters based on transfer functions and determine the
Response of circuit A
2 types of the filters.
Current, A

1
6. I have a better understanding of relationships between
0
0 1 2 3 4 5 voltages and currents across capacitors/inductors through
Time, s visualization by MATLAB.
Response of circuit B
1
Current, A

7. MATLAB plots of first-order transient responses help me

0.5
understand the dynamic evolution of voltages/currents during
0
0 1 2 3 4 5 transient process.
Time, s

2
Response of circuit C
8. MATLAB plots of second-order transient responses help
Current, A

me understand the differences among underdamped,

0
overdamped and critically damped transient processes.
-2
0 5 10 15 20 25
Time, s 9. MATLAB plots of passive filters’ frequency responses
help me understand the importance of transfer function.

10. MATLAB plots of passive filters’ frequency responses

IV. ASSESSMENT AND OUTCOME help me understand the connection between filters’ circuit
Based on students’ performance in MATLAB assignments, structure and practical systems’ user requirements.
it was found that most students were able to perform basic
mathematical operations in MATLAB to assist the solution of 11. MATLAB plots of various bandpass filters’ frequency
fundamental circuit problems. In addition, many students responses help me understand the concepts of quality factors,
gained an understanding of the advantages and limitations of bandwidth, frequency selectivity and resonance.
solution by hand analysis and solution via MATLAB. In
particular, they realized that computer software is a powerful 12. Overall, I appreciate the incorporation of MATLAB into
aid to problem solving, but by no means a substitution of the instruction of Electrical Engineering I.
theoretical analysis.
One of the major reasons we choose the course “Electrical For each of the above questions, students’ will provide a
Engineering I” to incorporate MATLAB illustration and score ranging from 1 to 5, with 1 being “strongly disagree”
visualization is that, unlike mechanical systems, electricity and 5 being “strongly agree”. Among these questions, 1-5 are
cannot be directly seen, heard or felt, especially for students concerned with students’ ability to use MATLAB to assist
who are new to the EE field. Thus, it is important for us to circuit problem solving, and 6-11 evaluates the improvements
assess whether students feels more comfortable about this in students’ understanding of course materials. Question 12 is
 ENHANCING THE INSTRUCTION OF INTRODUCTORY ELECTRIC CIRCUIT COURSES 63

the overall evaluation of the teaching practice. consistently throughout the course of their college study. If
Survey results indicate that the majority of students believe more engineering courses, especially at the sophomore and
the graphic illustration in MATLAB substantially improves junior level, incorporate MATLAB in their instruction, only a
their understanding of relevant course materials. The average modest amount of effort is need to refresh students’ MATLAB
score for question 12 was 4.5/5. Specifically, the instruction of skills in each course.
second-order transient analysis (question 8) and passive Second, a balance needs to be achieved between the amount
filters’ frequency responses (questions 9-11) benefit most of course materials covered and the amount of time the
from visualization in MATLAB. Also, the plotting of instructor and students devote to MATLAB instruction and
voltage/current across capacitors and inductors (question 6) assignments. In promoting our teaching practice, we need to
helps students quickly grasp the relationship between the be cautioned against letting the MATLAB take too much of
voltage and current waveforms. instruction time and student efforts in assignments. The course
Also, we found students did appreciate the usefulness of materials need to be the top priority, and the MATLAB
MATLAB as a software tool to facilitate efficient solution of element should not increase students’ burden significantly.
electrical engineering problems. However, concerning After all, student motivation should never be compromised,
students’ ability to perform analysis using MATLAB especially for STEM disciplines.
(questions 1-5), there is much variation among students’
responses, with average scores varying between 3.5 to 4.2. REFERENCES
There are many students who are not proficient in using basic [1] MathWorks, “Getting Started with MATLAB”, Natick, MA, 1996.
MATLAB commands. This further confirms the necessity of [2] A. Haubold, “Matlab for First-year College Engineers”, The 37th
ASEE/IEEE Frontiers in Education Conference, pp. F1H-7 – F1H-12,
enhancing MATLAB instruction in engineering curriculum 2007.
consistently throughout undergraduate years. [3] MathWorks, “Learning Matlab”, Version 6, Natick, MA, 2001.
[4] A. Azemi, C. Stoock, “Matlab for First-year College Engineers”, The
26th ASEE/IEEE Frontiers in Education Conference, Vol. 2, pp. 599 –
602, 1996.
V. CONCLUSION/DISCUSSION [5] M. Durán, S. Gallardo, S. Toral, R. Martínez-Torres, F. Barrero, “A
In this paper, we described our recent efforts in integrating learning methodology using Matlab/Simulink for undergraduate
electrical engineering courses attending to learner satisfaction
MATLAB into the instruction of the course “Electrical outcomes”, International Journal of Technology and Design Education,
Engineering I”. It is concluded that the visualization of circuit Vol. 17, No. 1, pp. 55-73, 2007.
analysis concepts in MATLAB facilitates instruction and [6] A. Azemi, E. Yaz, “Utilizing Simulink and MATLAB in a graduate
nonlinear systems analysis course”, The 26th Frontier in Education
promotes student learning. Meanwhile, the new teaching Conference, Vol. 2, pp. 595 – 599, 1996.
practice demonstrated to the students the advantages and [7] E. Yaz and Azemi, “Utilizing MATLAB in Two Graduate Electrical
limitations of computer software in engineering problem Engineering Courses”, The 25th Frontiers in Education Conference,
Vol. 1, pp. 2c6.1-2c6.4, 1995.
solving, and the relationship between the results obtained by [8] S. Ayasun, C. Nwakpa, “Transformer Tests Using MATLAB/Simulink
software tools and by theoretical analysis. and Their Integration into Undergraduate Electric Machinery Courses”,
There are some additional observations based on the 2006 Wiley Periodicals Inc., pp. 142-150, 2006.
[9] V. Pires, J. Silva, “Teaching nonlinear modeling, simulation, and control
findings of this project. Here we discuss two of them: of electronic power converters using Matlab/Simulink”, IEEE
First, the basic MATLAB knowledge students learned in the Transactions on Education, Vol. 45, pp. 253-261, 2002.
freshman year still need to be substantially reviewed in our [10] R. Patel, T. Bhatti, D. Kothari, “MATLAB/Simulink-based transient
stability analysis of a multi-machine power system”, International
project, because students have not been using MATLAB Journal Electrical Engineering Education, Vol. 39, pp. 320-336, 2003.
 Istopolis: Design Issues and Evaluation’ Results of
an Integrated Web-Based Educational Application
C. T. Panagiotakopoulos T. S. Papatheodorou G. D. Styliaras
Departmnet of Primary Education, Departmnet of Computer Departmnet of Cultural Heritage
University of Patras, Greece. Engineering and Informatics, Management and New Technologies,
[email protected] University of Patras, Greece. University of Ioannina, Greece.
[email protected] [email protected]

Abstract – In this study the design issues of a web-based These systems offer exceptional and continuously enhanced
application, named Istopolis, are described. Istopolis is an capabilities to educators, who eventually have the potential to
integrated system aiming to support History and Culture communicate with their students and assign them lessons by
courses of primary and secondary education in Greece. This using educational scenarios as well as exercises and projects.
hypermedia application system can be installed on a server
allowing access both to students and teachers. Authorized
Exercises and projects can be reviewed online, providing
teachers are able to modify or add modules with text, virtual students with feedback in terms of educational comments and
material and video, and also modify or add scenarios towards an grades [2] [10]. Thus, the actual virtual class can be realized.
optimum exploitation of the initially specified educational In the present work, the structuring (design and
modules. Furthermore, results of the summative evaluation development) of the Istopolis application is presented. This
obtained by a sample of active teachers are presented, showing application constitutes an integrated web-based hypermedia
that problems such as “lost in cyberspace” have been evaded educational system aiming to support History and Culture
and that, in general, educators are rather positive against the courses of the Greek primary and secondary education. The
developed application in terms of usability, quality of media and structuring and interconnection of the various sub-units of
aesthetics.
Istopolis are also described, which aim at serving all the users
(e.g. students, educators and administrator) interacting with
I. INTRODUCTION
the system. Finally, results of a predictive evaluation are
presented, in terms of usability, aesthetics, quality of included
Online systems aiming to support education are firstly met
media, and hierarchy of learning effect of differentiated
in the decade of ’60 with the appearance of the first online
information transmission media. The data used to perform
bibliographic retrieval system [1]. Systems developed from
this evaluation, were acquired from a sample of 36 educators,
1963 to 1995 could be classified as the first generation of
who used the developed system in real conditions.
educational online systems.
After the wide popularization of Internet (April 1995),
II. ISTOPOLIS APPLICATION: IDENTIFICATION
online educational systems met an increasing implementation,
and can be classified in two categories [2]: (a) oriented to
The name of the presented application comes from parts of
teacher-to-student communication [3] and (b) oriented to
the Greek words Isto-ria (history) and Poli-tismos (culture). It
student-to-student communication [4]. The generic attributes
is an integrated hypermedia web-based educational system
of systems from both categories present many similarities
constructed during 2007-2008 in the context of the Unit
with those of the first generation online educational systems.
“Chrysallides” (Development of integrated educational
Along with the emergence of the second generation of
activity products), of the Action “Pleiades” (Development of
systems an effort to support the learning process took place.
Educational Software and Integrated Educational Activity
This effort was mainly based on elements of the traditional
packages for Greek Primary and Secondary Schools &
class [2].
Distribution of Educational Software Products to Schools),
With the wide diffusion of Internet and WWW, online
with the Research Academic Computer Technology Institute
systems started to be employed as sources of information
(RACTI http://www.cti.gr) as the organization responsible for
retrieval [5] and means of design and development of learning
implementation monitoring. It was funded by the Greek
environments [6] [7].
Ministry of Education & Religious Affairs and the
The third generation of web-based learning systems or
Operational Program for the Information Society, Measure
online educational systems, as they are met in the literature, is
1.2, 3rd Community Support Framework.
based on novelties regarding the means of control and the
The system was integrated through a continuous formative
direction of information flow [8]. Contemporary educational
evaluation from the respective organization responsible for
online systems are nowadays exclusively based on Internet.
monitoring and controlling (i.e. RACTI) technical and
Along with these systems, the term “virtual teaching”
educational issues. It was finally positively evaluated by the
emerged [9].
Greek Pedagogical Institute from a technical as well as an

65
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_12, © Springer Science+Business Media B.V. 2010
66 PANAGIOTAKOPOULOS ET AL.

educational perspective. their original source,

The developed application can be installed on a networked - approximately 400 min of video, distributed in video clips
computer, which educators and students are both able to lasting from 3 to 6 minutes,
access. It is worth noting that Istopolis is a content - links to webpages with extra visual and text educational
independent application. material relevant with the content of each module.
The application’s integration was realized through the (b) 40 two-houred educational scenarios for the software
cooperation of several specialists from various fields, such as exploitation in a school class level, which also determine the
educational design, programming, video creation, sound, level of use, that is the educational grade that this software
video and image processing, text editing, design editing, will be used (primary education – secondary education).
educational scenarios’ construction, content writing and text
narration. B. The educational scenarios
The generic internal structure of the system is illustrated in Each authorized student can easily explore and study the
Fig. 1. content of the application. The effective exploitation of the
content though, is realized with the assistance of the
originally created educational scenarios that come along with
the software or by subsequent scenarios manually created
from authorized educators. Creation of new scenarios is
always necessary in order to assist the learning process of
specific students or specific classes with alternative
methodologies, addressing their personal skills, preferences
and needs.
The scenarios essentially comprise of activities that should
be performed by students of a class and include exercises,
written projects (assignments) as well as instructions for their
effective fulfillment. This way, the student becomes familiar
with the educational content of Istopolis.
The scenario’s instructions encourage the students to go
back to specific software units or external links, study text
documents or original sources, in order to be able to work out
Fig. 1. The internal structure of the system. the exercises or compose the projects they were assigned.
The exercises included in the scenarios can be classified
The administrator can assign use rights to educators and into three types: open, multiple choice and one-to-one
register their properties such as the school and the class they correspondence. The application has the potential to
are teaching to. Each educator in his turn can assign use rights automatically evaluate the multiple choice exercises as well
to students of the classes he is teaching to. Through the as the one-to-one correspondence ones, depending on the
Istopolis environment, educators are also able to assign to the settings designated by the scenarios’ composer in advance. In
students of a class, one or more scenarios for processing this direction, the teacher may change the weight of each
purposes, in order to actively engage them in the application’s question or exercise, before they have been assigned to a
content. class, determining an alternative weighting list.
The software is accompanied by a Content Management Assignments are composed by the student through a proper
System (CMS), which addresses the need for: environment facilitating a small hypertext editor that
a) Current content improvement incorporates all the common capabilities for text processing.
b) Inserting new integrated modules The small editor also supports interconnection of text with
c) Inserting new scenarios towards a more effective content other kind of information (hyperlinks), located in the system’s
exploitation. content or in the World Wide Web. Finally, it provides
The administrator is capable of assigning specific use rights capabilities of image insertion and integration with the under-
to appropriate educators who will have the responsibility to process-text.
operate the CMS. As the above show, the application-user interaction is
boosted in a considerable level.
A. Content
The student may have access to the content that is III. IMPLEMENTATION DETAILS
comprised of:
(a) 78 modules containing data regarding the Ancient, the The system was built by using HTML, PHP, Adobe Flash,
Byzantine and the contemporary History and Culture. For the Java, Javascript and MySQL. Apart from Flash, all software
creation of these modules the following were used: employed is based on open and acceptable standards [11] and
- approximately 400 pages 4 size, of original text written by runs in a school laboratory environment. A MySQL database
archaeologists, historians and philologists, stores the system’s content along with user permissions and
- approximately 4000 pictures with legends and references in
 ISTOPOLIS: DESIGN ISSUES AND EVALUATION’ RESULTS 67

parameters about assigned projects. The system’s content is within a specific period of time, determined by the user.
generated on the fly by using PHP scripts that draw the Finally, through the “ap” choice modules and
appropriate content from the database. Exercises are complementary texts are classified according to their
implemented with Javascript and Java. The timeline is based association with a geographical area. If the module or the text
on Javascript and Dynamic HTML, whereas Flash is can be associated with a certain location, it then appears
employed for implementing the map interface. Flash positioned as a point on the map. Otherwise, it appears under
communicates directly with the database via Javascript in the map of the region it was associated with.
order to display the appropriate map and the points on the On the other hand, the user can access a specific content in
map. Flash is used also for displaying the virtual interface and a combinative way through the “Timeline” and the “Map”. In
the menu on the upper right of every screen. Finally, the other words the user can access a specific content through the
content management interface is based on PHP scripts. combined use of time and space.
As many researchers in the domain of hypermedia
IV. STUDENT’S INTERFACE applications have already reported, due to several design
issues, the phenomenon of “lost in cyberspace” has been
The application’s information flow, from the user’s observed. This phenomenon disorganizes and misleads the
perspective, depends on the kind of the user (administrator, user [12]. To avoid such problems, we did not connect with
teacher, and student). hyperlinks any words or phrases within the main and the
The general pattern of the information flow concerning the complementary text of each module. Instead, we used a term
student’s access to the software’s content is shown in Fig. 2. dictionary in every module that the user can access by
pressing a button in each module’s environment. All the small
dictionaries of the modules act as subsets of a bigger
dictionary that the user can find in the root menu and
approach it alphabetically. Moreover, all hyperlinks of the
module’s menu operate and expand inside the specific
module, apart from the “Resources”, the “References” and the
“Links”.

Fig. 2. The data flow of student’s interface.

A. Content’s access
The student can access the software’s content in three
Fig. 3. An actual screenshot of the student’s interface.
different ways – choices. These choices are the “Thematic
Catalogue”, the “Timeline” and the “Map” (Fig. 2, Fig. 3).
B. Student’s Book
The shortest way to access the content is through the
An important “field” of the application from the student’s
“Thematic Catalogue” choice. Within the “Thematic
perspective, is the “Student’s Book”. In that “field” the
Catalogue” modules and complementary texts are classified
student can find:
through their positioning in a thematic hierarchy.
a) The scenarios that his teacher sent him for processing.
With the “imeline” choice, modules and complementary
b) The tasks that the student has fulfilled (exercises and
texts are classified through their positioning in chronological
assignments) and wants to be published in order to be
periods. This classification enables both unit and text search
68 PANAGIOTAKOPOULOS ET AL.

evaluated by his teacher. collection of the student’s projects, evaluation, and provision
Note, that the field “Student’s Book” is accessible by all of grades and constructive feedback to the students.
authorized users. This fact emerged under the consideration b) The construction of new scenarios.
that in this way, a better competition between the students Information flow of the teacher’s interface is depicted in
will be created and maintained since all of their projects and Fig. 5:
worked out exercises will be in public view.
Moreover, in that field every student can view the
evaluation of the projects submitted. Evaluation includes both
the project grade and the comments justifying the received
grade.

C. The form of the modules

All units’ content appears in a specific structure, which
allows the student to have easy access (always with less than
three mouse clicks [13]) in every sub section, which either is
in text (written text, scanned prototype text) or visual form
(pictures - videos).
Every module consists of a main title and text, a video
presentation, a wide group of high definition images and
maps, original sources (if such exist), bibliography,
complementary texts, interconnections with other module’s
content or with World Wide Web’s URLs and a small Fig. 5. The structure of data flow in teacher -mode.
dictionary hosting the module’s terms.
Graphically, each module’s content structure appears in The interface through which the teacher is able to perform
Fig. 4. the above tasks is illustrated in Fig.6:

Fig. 4. The data flow structure of every module.

Fig. 6. An actual screenshot of the teacher’s interface.

V. TEACHER’S INTERFACE
VI. INTERACTION
The teacher can be informed by the system for the
According to several researchers [14], the interaction
scenarios already assigned to a class in order to be effectuated between educational software and user is an essential element
along with those that remain unassigned. He is also able to
positively contributing to the learning process. Istopolis offers
extract a scenario in XML format, change or enhance it and
several capabilities concerning the interaction with the
import the modified scenario back to the server. The online
student, which include:
tools the teacher is offered are associated with:
1. Navigation in 3D space in order to discover the
a) The class administration, such as: assignment of a
software’s research and configuration tools, after the
scenario or a set of scenarios to a class, establishment of an
student has logged in. This way the students are
appropriate hierarchy for the scenarios to be accomplished
welcomed by an environment which follows a gaming
when more than one scenarios are assigned to a class,
fashion that is more close to the student.
 ISTOPOLIS: DESIGN ISSUES AND EVALUATION’ RESULTS 69

2. Content search using simple research methods, such as standard deviation and the mode (the value that occurs with
the modules’ catalogue. the highest frequency) with the respective rates are shown in
3. Content search using complex search tools, coupling Table 1.
intense semantic charge with simple graphic interface
(such as the Thematic Catalogue, the Timeline or the TABLE 1
QUESTION, MEAN, STANDARD DEVIATION AND MODE
Map) and exploit the intrinsic hierarchical information Question (usability) Mean St.D. Mode
structuring. U1. At the very beginning working with Istopolis 5
3.97 0.97
4. Content search through complex combinations of search was: (36.1%)
criteria regarding "subject - location - time". U2. In general, you found the software’s 4
4.13 0.76
handling: (41.7%)
5. Access to same content through several paths. U3. Did you understood the way this software 4
6. Project writing through a small hypertext processor with 3.92 0.81
operates: (52.8%)
complete editing capabilities and project publishing. U4.Do you think that the students will be able to 3
3.58 0.69
7. Content collection, temporary storage and reuse in the use it: (52.8%)
projects. U5. Can you create a scenario with exercises 4
3.56 0.98
and projects through the software: (33.3%)
8. Fill out and submission of self-evaluation exercises. 4
The above are only the most important from the whole of U6. Can you assign a scenario to the students: 3.92 0,84
(47.2%)
the interactive functionalities offered to the students. U7. The software favors interaction between
4.17 0.74
4
Regarding the student-educator interaction, it is noted that student and content: (52.8%)
U8. Navigation in the software’s units was 5
the software is designed to be used inside the class, while the performed:
4.44 0.59
(50.0%)
educator is present supervising the students. U9. The use of the search tools (Thematic 5
4.21 0.72
Catalogue, Timeline, Map) is: (69.4%)
VI. APPLICATION’S EVALUATION U10. Functionalities of the “Student’s Book” for
4
the teacher and the student is in terms of 4.03 0.61
(63.9%)
usability:
The application, during its implementation, was structured U11. Did you feel “lost” at any time, while 1
through a continuous formative evaluation by the supervising 1.50 0.56
navigating in the software: (52.8%)
organization as well as the development team. The summative U12. Did you know at all times your exact
4.19 0.78
5
evaluation was performed with multiple groups of users and position in the software: (41.7%)
multiple evaluation subjects.
In the following, the results of a predictive summative Usability effects were analysed using a chi-square test.
evaluation received from a sample of 36 active teachers of Teachers found the software easy to handle (U3: x2=17.11;
primary and secondary education are presented. df=3; p< .001). This result is replicated when teachers
The evaluation procedure was performed with the sample comment whether their students will be as well able to handle
divided (for practical reasons) into two groups. The software it (U4: x2=9.5; df=2; p< .01). As a result, the high level of
was initially presented to each group by one researcher for usability makes educators feel confident to assign scenarios to
two teaching hours, while each participant was working at a their students (U6: x2=12.67; df=3; p< .01). From the
personal computer. After the presentation the participants teachers’ point of view, the navigation through modules poses
were asked to work with the developed software for three no usability constraints (U8: x2= 17.11; df=3; p< .001).
teaching hours. Finally, each participant was asked to provide In addition, the sample was asked to answer two more
feedback on specific questions filling an appropriate questions:
anonymous questionnaire. - In general, were there any points, where due to difficulties
The variables of the scale used for the evaluation [14] [15], in using the software you finally did not work with it?
aimed in collecting data regarding the applications’ usability, - In general, did you find any problems using the software
the aesthetics, the quality of the media included, and the or not?
establishment of an hierarchy for the learning effect of the Answers regarding the 1st question were 97,2% negative.
information transmission media from the teacher’s This means that only one subject claimed that there was a
perspective. point where working with the software was not possible due
Analysis of the self-evaluation of the participants’ to difficulties in its use. However, when this subject was
knowledge in basic issues of computer use (Microsoft asked with a follow-up question to spot this point, no answer
Windows/Office, Internet) showed an average of 3.9 in the was given. Answers regarding the 2nd question were 100%
five grade Likert scale (knowledge: none = 1, 2, 3, 4, 5 = negative.
excellent) with st.d. 0.7. The findings described above, show that the participants in
the evaluation believe that the software is easy and flexible in
A. Usability its use. Furthermore, the software’s design met with success
We asked the sample to provide answers to specific regarding the aim to avoid the “lost in cyberspace”
questions with expected answers within the five-grade Likert phenomenon.
scale (not at all = 1, 2, 3, 4, 5 = absolutely or difficult = 1, 2,
3, 4, 5 = easy). The questions, the answer’s average, the B. Aesthetics and quality of media
The sample was asked to evaluate the software’s aesthetics
70 PANAGIOTAKOPOULOS ET AL.

and the quality of information transmission media used by it. application, in its current form, comprises an integrated
The outcome is presented below (Table 2): platform aiming to support History and Culture courses of
primary and secondary education in Greece. It is a web-based
TABLE 2 application offering the potential to modify or enhance its
QUESTION, MEAN, STANDARD DEVIATION AND MODE
Question (aesthetics – quality of media) Mean St.D. Mode content through a CMS. Using the presented application, the
1Have you found image and video quality fair 5 educators are able to assign exercises and projects to their
4.08 0.97
enough? (41.7%) students through educational scenarios, evaluate them online
2. In which level did you find the software’s 5 and provide them with grades and constructive feedback.
4.02 0.91
images good regarding their quality? (38.9%)
3. Was the display’s color aesthetics 4
The results of the final predictive summative evaluation
3.92 0.73 with a sample of 36 teachers showed that the developed
appealing? (47.2%)
4. How much distinctive – readable was the 4 application benefits the educational process while being
3.97 0.81
text? (36.1%) handy, very appealing in terms of aesthetics and of high
5. Are the design and the presentation of the 4 quality with regards to the quality of information transmission
3.94 0.83
application’s screen appealing? (41.7%)
6. Is there cohesion regarding the presentation 4 media.
3.89 0.75
of the screen in every module? (52.8%)
A7. In general, how do you consider the
3.94 0.86
4 REFERENCES
software from an aesthetics perspective ? (36.9%)
[1] C. P. Bourne, and B. H. Trudi, “A History of Online Information
In general, application’s aesthetics were assessed as being Services, 1963–1976”, Cambridge, Mass.: MIT Press, 2003.
[2] R. Sheizaf, B. Miri, D-G. Yuval, and T. Eran, “QSIA – a Web-based
of high quality (A7: x2=9.78; df=3; p< .05).
environment for learning, assessing and knowledge sharing in
Establishment of a hierarchy regarding the quality of the communities”, Computers & Education 43, (3) 273–289, 2003.
information transmission media’ factors (text, pictures, [3] Y. J. Dori, M. Barak, & N. Adir, “A Web-based chemistry course as a
graphics-maps, videos) was analyzed with the assistance of means to foster freshmen learning”, Journal of Chemical Education,
80(9), 1084–1092, 2003.
the Friedman’s test and showed an important advantage of
[4] M. Guzdial, and J. Turns, “Effective discussion through a computer-
video (x2=25.43; df=3; p< .001) followed by pictures (Table mediated anchored forum”, The Journal of the Learning Sciences, 9(4),
3). 437–469, 2000.
[5] C. A. Merlic, B. C. Fam, M. M. Miller, “WebSpectra: Online NMR and
IR Spectra for Students”, Journal of Chemical Education, 78, 118–120,
C. Hierarchy of information transmission media’ learning
2001.
effect [6] Dori, Y. J.; Barak, M., “Virtual and physical molecular modeling:
The sample was asked to give their opinion for which Fostering model perception and spatial understanding”, Educational
information transmission media is considered as helpful for Technology & Society, 4, 61–74, 2001.
[7] Donovan, W. J.; Nakhleh, M. B., “Students’ Use of Web-Based
the student in order to better understand the concepts in the
Tutorial Materials and Their Understanding of Chemistry Concepts”,
developed application. Data analysis through the Friedman’s Journal of Chemical Education, 78, 975–980, 2001.
test showed again an important advantage of video (x2=19.23; [8] Nachmias, R., Mioduser, D., Oren, A., & Ram, J., “Web-supported
df=3; p< .001) followed by pictures (Table 3). emergent-collaboration in higher education courses”, Educational
Technology & Society, 3(3), 94–104, 2000.
[9] Sheremetov, L., & Arenas, A. G., “EVA: An interactive Web-based
TABLE 3
RANK OF MEDIA’ QUALITY AND EFFECT collaborative learning environment. Computers & Education, 39, 161–
Media Quality - Mean Rank Effect - Mean Rank 182, 2002.
Videos 1.92 1.72 [10] Panagiotakopoulos, C., Papaterpos, C., & Papatheodorou, Th.,
Pictures 2.06 2.53 “Presentation of a hypermedia system supporting History: Basic
characteristics and content’s evaluation”, Proceeding of 1st Educational
Graphics – maps 2.78 2.83
Conference for the Introduction and Use of ICT in Education (P.
Text 3.25 2.92
Politis, ed.), 759-765, 2009.
[11] Rafaeli, S., Barak, M., Dan-Gur, Y. & Toch E., “QSIA – a Web-based
The previous results show for one more time that an image environment for learning, assessing and knowledge sharing in
on its own is able to paint a thousand words. For this reason, communities”, Computers & Education, 43, 273–289, 2004.
[12] Otter, M. & Johnson, H., “Lost in hyperspace: metrics and mental
during the design of an educational material this should be models”, Interacting with Computers, 13(1), 1-40, 2000.
seriously taken under consideration [16]. [13] Panagiotakopoulos, C., Pierrakeas, C. & Pintelas, P., “Educational
Finally, all participated teachers, who answered a relative Software Design”, Patra: Hellenic Open University Publications, 2005.
question, stated that they would like to use the developed [14] Picciano, A.G., “Beyond student perceptions: Issues of interaction,
presence, and performance in an online course” Journal of
software in their classroom, as part of their everyday teaching Asynchronous learning networks, 6(1), 21-40, 2002.
practice. [15] Panagiotakopoulos, C., Pierrakeas, C. & Pintelas, P., “Educational
Software and its Evaluation”, Athens: Metaihmio Publications, 2003.
VII. CONCLUSIONS [16] Spiliotopoulou-Papantoniou, V., Karatrantou, A., Panagiotakopoulos,
C. & Koustourakis, G., “Visual representations of the internet in greek
school textbooks and students’ experiences”, Education and
In this study, issues regarding the design and development Information Technology, 14, 205-227, 2009.
of the application named Istopolis were analyzed. This
 TEAC2H-RI: Educational Robotic Platform for
Improving Teaching-Learning Processes of
Technology in Developing Countries
Juan J. González1, PhD Jovani A. Jiménez2, PhD Demetrio A. Ovalle3
National University of Colombia
Carrera 80 numero 65-223 (M8A-309) Medellín-Colombia
1
[email protected], [email protected]; [email protected]

Abstract—The technological development achieved in each of the Pedagogy, which also claim, as the constructivism theory, that
epochs of the human history has changed the means of the knowledge is not transferred but created in the learner’s
production, communication, transportation and education. There mind, but additionally, that the method to achieve that is
are many results because of the incorporation of technology on through the construction of something tangible by the learner
education, including remote laboratories, online courses, and [3] [4].
educational robotics. In recent years, the latter has become very
Papert and his research group at the Massachusetts Institute
popular because of its significant contributions to the learning
process, strengthening of cognitive skills, including creativity, of Technology (MIT), based on the Construccionism
design and problem solving skills. However, this phenomenon has Pedagogy, developed a programming language for kids, called
been focused in the developed countries, and many of the Logo. Logo had a Graphic User Interface which allowed
applications of the developing countries rely on a technological programming a turtle robot. Afterward, LEGO and
transference from the developed countries. This model is not the Fischertecnnik with a similar concept developed kits to build
most suitable due to the fact that the tool should be adapted to the robots using many bricks, including tires, gears, lights,
environment and its user, and not the opposite. This article sensors. The Media Lab of the MIT and other institutions used
presents some of the main contributions in this issue and outlines these kits to explore the results of the interaction with them.
the TEAC2H-RI kit. This kit was developed based on the
Most of the researchers concluded that theses kits not only are
Colombian educational environment characteristics and the
experience in robotics of the research group GIDIA. It will be useful for the learning process but also for the strengthening of
used in Senior High School Students. the creativity, design, troubleshooting, team work and social
awareness, among other cognitive skills.
Index Terms—Society and Computing, Educational Robotics, These kits produced by Develop Countries have been
Robotic Swarm, Constructivism, Construccionism implemented in these regions, and in some cases in
Developing Countries. Although, the last one have social-
I. INTRODUCTION economic conditions very different from those of the formers,
As well as the societies evolve so do its means of then the performance of these kits without the correct reading
production, communication, transportation and education. In of the context could generate no satisfactory results. It is
the last one, one of the most important moments was in the because by this methodology the user and his environment has
XIX century when many researchers proposed the switch from to be adapted to the educational tool, but it is the tool that
the passive paradigm to an active paradigm where the should be adapted to the user and his environment. This
participation of the learner is more active. It comes as a result phenomenon is very common in the technological
of the works of pedagogues, educators and psychologist, implementation in the Latin American Countries and it is very
including, Johann Heinrich Pestalozzi, Friedrich Froebel and criticized by many different authors. [5]
Maria Montessori [1]. These previously mentioned In order to provide an autochthonous solution to this
approaches, along with the subsequent, are the root of the problematic the Artificial Intelligence in Education Research
development of the constructivism theory of the Swiss Group of the National University of Colombia addresses these
Psychology Jean Piaget. In this theory the knowledge is not difficulties in the research project: Educational Robotics:
assume like a fluid which can be transferred from one person Intelligent Machine in Education.
to another, but like a process where the knowledge is created This paper is organized as follows: Section II briefly
in the learner’s mind [2]. Based on it, the South African reviews some approaches in Educational Robots. Section III
mathematician Seymour Papert proposed the Construccionism presents some cases of the implementation of educational
robots in developing countries. Section IV describes the

71
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_13, © Springer Science+Business Media B.V. 2010
72 GONZÁLEZ ET AL.

proposed methodology. Section V presents the kit TEAC2H- developed and Latin American countries. There is very clear
RI (Technology, Engineering And sCience eduCation through that the research crisis covered almost the whole region, and is
Human-Robot Interaction). Finally, section five presents the less severe in Chile, Brazil, Argentina and México. The
conclusions and future work. research capacity of any country is one of the most important
resources to affront the social, ecological, technological, and
II. EDUCATIONAL ROBOTICS economical challenges. Therefore, it can be understood why
Since the creation of the Logo Turtle in the 60’s by the the countries with a low number of researchers per million
Media Lab MIT until our days the advancement in Educational inhabitants, have a lower progress than those with a greater
Robotics has covered not only the primary school students but number of researchers.
also the high school, undergraduate and graduate students, and TABLE 1
even the industry workers. Some of the most important RESEARCHERS PER 1 000 000 INHABITANTS (TAKEN FROM [12])
educational kits are: LEGO Minstorms, Logo, Fischertechnik Researchers per 1 000 000 inhabitants (FTE)
[6], Pioneer, K-Team, IroRobot y Handy Board [7], [8], [9]. Argentina 980
The three formers are used in education and toy industry. On
Bolivia 120
the other hand, the last five could be implemented in the
Brazil 629
school, university and in the industry. These kits are very
useful to explain and verify algorithms, to learn some concepts Canada 4157
of programming languages and to encourage the strengthening Chile 833
of different cognitive skills [10]. Colombia 151
From the different options available LEGO Minstorms and Ecuador 69
Fischertechnik are the most widely used in the world because Guatemala 25
its relative low cost, in comparison with the others, and the Japan 5573
educational resources that provide. The philosophy of both is
Mexico 460
very similar because is based on interlocking bricks. One of
Panama 62
these kits is RoboExplorer. It is manufactured by
Fischertecnik, and it is consists of ultrasonic range sensors, a UK 2881
color sensor, a track sensor, a NTC resistor and a USA 4663
photoresistor. The locomotion is based on two servomotors
and a crawler drive. Also, with this kit is provided the ROBO Aware of this fact, some research centres and universities
interface, and the programming software ROBOPro [11] have implemented different educational robotics courses, to
As can be seen the characteristics of these kits made of find a solution to these issues. Some of them are the Carnegie
them a very useful tool in the teaching of important subjects as Mellon University (CMU) and the MIT. Next is presented
mathematics, physics, electronics, and algorithms. These kits some of their results
at the university, in careers related with Computer Science, In Qatar and Ghana the Carnegie Mellon University
contribute with a deeper understanding of the algorithms associated with some higher educational institutions of these
concepts. On the other hand, in industry there are many regions, developed a course on autonomous robots with the
applications, including the process simulation, training in the aim of strengthening the knowledge on algorithms and the
control machine. However, its implementation must be creativity in the undergraduate Computer Science students.
adjusted to the particular context of the region. The following The educational robotic tool was the MIT Handy Board, the
section presents some approaches in this matter. software was the Interactive C and the hardware the bricks of
LEGO. At the end of the courses was concluded that robotics
III. EDUCATIONAL ROBOTICS IN DEVELOPING COUNTRIES as an educational tool in Computer Science meet the proposed
objectives and also increase the interest of the learners by
The great deal of problems in the majority of the developing technological topics and research. Another important
countries, especially in the Latin American Countries, requires conclusion was that the interaction, for the goal
that the inhabitants face them not only with a solid knowledge accomplishment, among the team members strongly depends
in their speciality area but also with high research skills. on the cultural environment. This fact has to be observed at the
Furthermore, it is necessary to increase the attraction of school design stage of a course in this thematic [13].
students for careers related with technologic production, such In Costa Rica the Massachussets Institute of Technology
as engineering. Statistically (see Table 1), it can be seen that (MIT) worked with a school located in one of the rural zone of
there is a research crisis in Colombia and in almost all the this country. The educational robotic tool was the LEGO
Latin American Countries, e.g. in Colombia in 2006 there was Mindstorms and the software was the Yellow Brick Logo.
only 151 researchers per million inhabitants, whilst for the From the results some of the conclusions were that educational
same year in USA and Japan there is 4663 y 5573 researchers robotics is a very useful tool for the student learning process
per million inhabitants, respectively [12]. In the table I, can be support and also for the strengthening of the creativity and the
seen the number of researchers per million inhabitants in troubleshooting skills. [14].
 TEAC2H-RI: EDUCATIONAL ROBOTIC PLATFORM 73

As was illustrated before there are many robotic educational demonstrated that the active role of the student in his/her
tools, however, because they were developed in countries with learning process can be done through the interaction with the
socio-economical conditions very different from the Latin kit TEAC2H-RI.
American countries, its effectiveness is lower in those In the second workshop the objective is to increase the
countries than a autochthonous propose made from a fair motivation of the learners for robotics, especially for the kit
reading of the context. Moreover, they are very expensive in TEAC2H-RI. The success of the course depends on this fact
comparison with the economical resources of the majority of then it should be handled with caution. First of all, we will
the educational institutions of these countries, restricting the present a movie or documental about robotics. Then, they will
widely spread of this tool. Due to this fact, it is necessary to be induced to observe and analyze the behavior of the robotic
make a deeper reading of the Latin American reality and bring swarm. After, teams of four students will be formed and each
an autochthonous solution according to the region necessities of them will work with a tutor and a robot. It is done in order
[5]. From the educational view, one important constraint of the to make of robotics meaningful material for them [16].
majority of these kits is the specific number of pieces that they In the third workshop the objective is that the learners
have, which influence the mind of the learner to develop some understand the theories that rule the behaviour of the robots
specific prototypes. Technically, one of the most important through the modification of the parameters of them. The tutor
constraints is that each brick has a group of equations that will stimulate the interest of the learners to interact with the
rules its behavior which can not be modified by the user [15], kit, and also he/she will provoke situations where the learners
additionally the learner can not integrate new bricks of his/her could conclude some important theories that rule robotics. For
own creation. These constraints restrict the learning process, example, the tutor will incite the learners to move the batteries
and the strengthening of creativity and design. position and observe how the change in the centre of mass
affects the robot’s stability. There are many other important
IV. PROPOSED METHODOLOGY concepts that could be understood through the interaction with
the kit, including overlapping of light beams, angular
The Artificial Intelligence in Education Research Group of momentum, friction, resistance [17]. Some of the skills that
the National University of Colombia, addressed the will be strengthened are analyses, problem solving, and the
difficulties mentioned above through the research project, team work.
Educational Robotics: Intelligent Machine in Education. This In the next workshop the objective is that the teams design,
project is ongoing. In this project, based on a objective view of and develop an addition, improvement, or a modification on
the Colombian reality and the experience in robotics of the the robotic platform. Also, the learners could choose to design
Research and Development Group in Artificial Intelligence of and develop a small automation system which must be
the same institution, was developed the robotic swarm associated with the topics of the course. For both activities, the
TEAC2H-RI (Technology, Engineering And sCience tutor will be a supporter to the learner during his design and
eduCation through Human-Robot Interaction) which will be construction stage. The skills to be strengthened in this
implemented in the Senior High School students of the workshop are the design, cooperation, creativity and research.
Metropolitan Area of Medellin. In the fifth workshop the objective is to strengthen the
In this zone will be chose ten schools and to each of them writing skills. In order to do that, each student will be asked to
will be delivered a kit TEAC2H-RI, composed of 4 robots. To write an article, which must contain abstract, introduction,
be eligible, each school must fulfil with some minimum theoretical support and their experience in the previous
requirements, i.e. a computer room. If the school not fulfils workshops. Other issues that have to be mentioned in the
this requirement, our university will provide a special room for article are: “what are the characteristics of the device that you
this purpose. Each course will have duration of 20 hours developed in the previous workshop?”, “Name an industrial
distributed in five weeks, with seven workshops. sector where it can be implemented”, “Why did you choose
The aim of the first workshop is to get the commitment of this design over others?” among others. Finally, the article
the teachers with this process, by means of the increasing of should have the conclusions and references section. This
their comfort with the topics of the course and dismissing the strategy is used because people like to write about their
negatives or technophobic paradigms. This will ensure that, interests [18].
even when the course has finished, the tool remains its The last objective is to promote the research discussion
pertinence to be used in the institution. spaces among the students. For this purpose, each group will
For this purpose, it will be presented in the workshop, the present, in front of their partners, the results obtained in the
topics to be studied and the expected results. Additionally, it fourth workshop. The other students in the audience will be
will be illustrated some pedagogies and theories of active encouraged to criticize constructively the projects, to let the
education, such as the Piaget’s Constructivism theory, the team to identify the improvements or limitations of the
Ausubel’s learning theory and the Contructionism of Papert. prototype. It is also with the aim to strengthen the team ability
Afterward, it will be exposed and argued the reasons to use to appreciate the diverse points of view, and as a consequence
active pedagogies of education as their teaching methods expand his community sense [17], [13].
supporting this in the social, economical and productive Finally, the Artificial Intelligence in Education Research
necessities of the Colombian context. Finally, it will be Group, based on the previous experience in educational topics,
74 GONZÁLEZ ET AL.

will develop the test to evaluate the effect of the course in the groups, making with this that the group and intergroup
learners, testing the performance of the students before and interaction skills be strengthened. This is the fundamental
after of their participation in the course. model of relations in enterprise and research projects. One
Despite de fact that the course finishes in this stage, the deeper explanation is giving next: every member of the group
support of our group will continue, because through the web interact with his/her partner to get the desired result in the
page, http://xue.unalmed.edu.co/~dbetanc0/index-3.html, we robotic agent, but also the interactions of each team with the
will present updated information about the kits, such as others must be coherent to get the desired behaviour of the
designs, algorithms or manuals that could improve the robotic swarm.
performance of the robot. Furthermore, the students could Following, it is explained each component of the robotic
interact with the Developer research group of the kit through swarm.
the web page or the e-mail, facilitating in this way the
feedback of the concepts and validity of the kit through time. A. Sensors
Next, it is mentioned some software and hardware
characteristics of the kit TEAC2H-RI. Traditionally, the educational robotic systems for the school
do not accept devices of other manufacturer. This fact makes
V. EDUCATIONAL ROBOTIC KIT: TEAC2H-RI difficult the access to spare parts and the design of new
2
prototypes. On the contrary, TEAC2H-RI is an open and
TEAC H-RI is a robotic swarm composed of four robotic modular architecture, i.e. the accessories that come with the kit
agents, i.e. three sons and one mother. It is a hybrid system can be improved or replaced. Even more, new accessories can
because it has distributed and centralized characteristics. It is be added, being therefore more adaptable to other
distributed because the son-robots, with similar behaviour as technologies. These concepts are very important for the
the mother-robot, could interact among them. It is centralized strengthening of the design, creativity and development skills
because the mother-robot could perform the role of group in the user. The kit has five different sensors for each robotic
leader. See Fig. 1 where a mother-robot is interacting with a agent. In these sensors the learner can experiment and modify
son-robot. the parameters of the equations that rule them. Also, additional
The robots not only interact among them, but also with the pieces are provided to the learner, in order that he/she may
user by means of sound signals (buzzer) and light (LEDs). It develop new sensors. In Fig. 2 it can be seen the infrared
facilitates to the user to conclude if the sensor connections or sensor. In 2a the emitter of the infrared sensor does not have
the robot behaviour is appropriated, and hence take the any resistance. In 2b a resistance of 10ȍ is put on the emitter.
corrective actions [19]. In 2c the Velcro, used to hold the sensor to the robot, can be
Each robot, when has been properly configured by the user, observed. In 2d the sensor is in the front side of the robot. Due
it could carry out navigation activities and obstacle avoidance. to the fact that the resistance of the infrared sensor emitter can
Even so, the mother-robot is the only one that could also send be changed, the maximum distance sensed can be modified. In
and receive the information from and to a remote PC, either by this manner and based on the textbook, the concept of
way of a radio link or a data cable. The modes of navigation electrical resistance, current and voltage can be understood.
are: track follower, by means of the QRB1114, obstacles
avoidance through infrared sensors or touch sensors.

a b

Fig. 1 Mother and Son Robot Interacting

c d
This robotic swarm will be used in the workshops with the
schools. In each workshop the maximum number of Fig. 2 Infrared Sensor a) without resistance b) with a 10ȍ resistance c)
participants is 16, divided uniformly in four groups and a with the Velcro at the bottom side d) in the front side of the robot
robot assigned for each team. This work topology with the kit
TEAC2H-RI facilitates not only the cooperation among the
members of each team, but also the cooperation among the
 TEAC2H-RI: EDUCATIONAL ROBOTIC PLATFORM 75

In the kit each robot has infrared sensors to avoid obstacles phobia for the technological topics [6]. The figure 3 was taken
or to follow tracks, photoresistance to recognize day and night, from the textbook.
touch sensor to avoid obstacles, sound sensor to detect claps,
and also it can be added other sensors including, temperature VI. CONCLUSION AND FUTURE WORK
sensor using NTC resistance and ultrasonic range sensor. The production cost of each robot is around $65US, which
is very attractive in comparison with other options in the
B. Mechanical Design market. This fact facilitates the expansion of the educational
robotics to a wider number of regions.
The chassis of both robots is very similar; its size is 23cm The implementation of educational robotics in developing
long, 14cm wide and 10cm high. The chassis has a first level countries must be done with a previous reading of the
where the motors card is and a second level where the Main economical and social context. It is important to select the best
Board is. In the outline of both levels, it is glued male Velcro. tool that fit the reality. In our project, this reading was done
It allows changing the position of each sensor and explores the and a new educational tool was developed: TEAC2H-RI.
operation in function of the position. The design of the chassis, The open architecture of TEAC2H-RI allows implementing
which could be assembled or disassembled, is ideal for new devices in the robot. This fact is very important because
experimentation and understanding of physical and the learner would like to improve the performance of his/her
mathematical concepts. For example, when it is modified the robot through the design and development of new sensors or
distance between the motors, the position of the sensors or the actuators. This characteristic, strengthen the creativity and
distribution of the mass in the platform , it could be design skills.
understood some concepts, including centre of mass, torque The proposed methodology is related to the steps to follow
and beam interference. An additional advantage of the in a research project. In this way, implementing this
mechanical design of TEAC2H-RI is that its components are methodology could be strengthened the research skills of the
very cheap and easy to acquire in Colombian market. Fig. 3, learner. It is extremely important in the developing countries,
shows the mechanical construction of the chassis of one like Colombia.
TEAC2H-RI robot. In this first approach, the social group chose to work on is
the Senior High School students from the Metropolitan Area
of Valle de Aburra – Antioquia – Colombia. However, future
works will cover not only the Metropolitan Area but also other
Colombian and Latin American places.
In upcoming projects will be developed new programming
languages with an easier interaction of the learners with the
educational tool TEAC2H-RI. Additionally, it will be brought
an introduction in the programming world.
Currently, there is a human-robot interaction but it is very
weak and it is necessary to consider the addition of new
hardware and software devices which must adapt the kit to the
user necessities [19].
Fig. 3. First construction stage of the chassis of one TEAC2H-RI robot. The kit TEAC2H-RI should not be taken as the panacea for
the educational problems in Colombia, neither in the
C. Programming Language developing countries, because it is necessary the collaborative
work between Government, School, University and Private
It will be used the programming language CodeWarrior of Sector to evolve to the suitable educational model [5, 3], in
the Metrowerks Company, which is supported in C. However, which the teacher will give a greater commitment and effort
we are considering developing an easier programming but also will receive better fruits.
language to the user, which not required previous knowledge
in algorithms and give an introduction to this subject. [20] [21] REFERENCES

D. Textbook
[1] Zuckerman O., “Historical Overview and Classification of Traditional
and Digital Learning Objects”. MIT Media Laboratory, 2006
The textbook contains the topics of physics, mathematics, [2] Piaget, J. Inhelder, B. 1966. La psychologie de L’enfant. Paris. P.U.F.
electronics, algorithms, and artificial intelligence that will be [3] Papert S. 1990. A Critique of Technocentrism in Thinking about the
addressed in the course. This is one of the most important School of the Future.1990 M.I.T. Media Lab Epistemology and Learning
Memo No. 2.
issues, which has to be handled carefully, because require a
[4] Papert S. 1980. The Gears of my Childhood. Mindstorms: Children,
language, concepts and adequate examples, to catch the Computers, and Powerful Ideas. Basic Books, New York.
attention of the learners. If it is not regarded with great [5] UN Millennium Project 2005. Innovation: Applying Knowledge in
caution, the reverse effect can be caused, i.e. rejection or development. Task Force on Science, Technology, and Innovation.
76 GONZÁLEZ ET AL.

[6] Chiou A. 2004. Teaching Technology Using Educational Robotics. En: [14] Urrea C. 2001. Designing Robotic Artifacts: Rural School-Community
Scholarly Inquiry in Flexible Science Teaching and Learning Collaboration for Learning. 2001 Media Lab MIT.
Symposium. (Sydney, Australia, Octubre 1, de 2004) [15] Zuckerman O., Resnick M. 2003. System Blocks: A Physical Interface
[7] Anderson M., Thaete L, Wiegand N. Player/Stage: A Unifying Paradigm for System Dynamics Learning. MIT Media Laboratory. 2003.
to Improve Robotics Education Delivery. RSS Workshop on Research on [16] Ausubel, D.P. 1963. The Psychology of Meaningful Verbal Learning: An
Robotics in Education 2007 (Georgia, USA, June 30 , 2007) Introduction to School Learning, Grune & Stratton Inc., New York, 1963
[8] Barrera P., Martin F., Robles G. Cañas J., Matellan V. 2005. Modelos y Bureau of Educational Research, University of Illinois. Grune &
herramientas de software libre en el campo de la robótica. En las Stratton, Inc.
memorias del congreso Tecnologías del Software Libre (La Coruña, [17] McKerrow P. Robotics research and education with LabVIEW. In
España, Julio 7-8, 2005) Proceedings of the Apple University Consortium Conference (New
[9] Martin F. 2001. Robotic Explorations. A hands-On Introduction to South Wales, Australia, Abril 25-28, 2000)
Engineering. New Jersey Prentice Hall. 2001. [18] Acuña A. 2006. Projects for Educational Robotics: Engines for the
[10] Blank D. 2006 Robots Make Computer Science Personal. In Journal of Innovation. In the proceedings of the Fourth International Conference on
Communications of the ACM. Vol. 49. Ed. ACM Press, New York, NY, Multimedia and Information and Communication Technologies in
25-27. Education. Current Developments in Technology-Assisted Education
[11] Fischertechnik. http://www.fischertechnik.de/en/index.aspx. Fecha de (Sevilla, España, Noviembre 22-25, 2006)
acceso: 09 de October de 2009 [19] T. Fong, I. Nourbakhsh, and K. Dautenhahn, “A survey of socially
[12] UNESCO Institute for Statistics, 2006 interactive robots,” Robotics and Autonomous Systems vol. 42, no. 3–4,
[13] Dias M., Browning B., Mills-Tettey A., Amanquah N. (2007) Robotics pp. 143–166, 2003
Education in Emerging Technology Regions. En AAAI Spring [20] Piperidis, S. Doitsidis, L. Anastasopoulos, C. Tsourveloudis,
Symposium: Robots and Robot Venues: Resources for AI Education. N.C. 2007. A Low Cost Modular Robot Vehicle Design for Research
(Menlo Park, USA, 26-28 Marzo 2007) and Education In proceeding of the 15th
[21] Wedeward K., Bruder S., Sarkodie-Gyan T. 2001. Exploration in
Robotics: Outreach to Secondary Education. In Proceedings of the AAAI
2001 Spring Symposium Series. (California, USA, Marzo 26–28, 2001)
 Selecting Patterns for the Evaluation of Systems
Dmitry Zub, Erki Eessaar
Tallinn University of Technology, Estonia
[email protected], [email protected]

Abstract - Earlier, we proposed a general pattern-based eval-

uation method of systems. The method is based on the use of II. BASIC CONCEPTS
Analytical Hierarchy Process and the use of patterns as criteria. A. Usability
This paper describes a process of selecting patterns that would be
used during the evaluation. This process is an important step in Usability requirements are non-functional requirements.
the method. Firstly, we describe the process, based on an exam- According to [3] a system with good usability should (a) be
ple. More precisely, we select a set of usability patterns that one easy to learn, (b) make it possible to complete common tasks
could use to evaluate e-learning systems. After that, we propose quickly, (c) be well memorable, and (d) make it possible to
a generalized process for selecting patterns. recover from errors quickly.
While these factors explain what usability is all about, it is
I. INTRODUCTION
rather difficult to use them in practice. Many methods exist
There exist several types of requirements for software sys- that help us to evaluate usability. Folmer and Bosch [4] pro-
tems. For example, functional requirements specify the func- pose a possible classification of these methods. According to
tionality of the system while quality requirements show how them there are three types of methods. The purpose of usabili-
well a system should perform its intended functions. ty inquiry methods is to get information about likes, dislikes,
Performance is a very significant quality requirement. It and needs of users by talking to them and observing them
indicates how well a system should complete its tasks (speed, working. The examples of such methods are focus groups [5],
accuracy and the resources it takes). Another important quality interviews [3], and questionnaires [3]. In usability testing
requirement is usability. It shows how easy is to learn to use a potential users complete some typical tasks by using the sys-
system, how efficient it is in daily use, and some other aspects. tem. For example, in “Thinking aloud” method a user verba-
In some types of software systems, like educational soft- lizes his/her thoughts or opinions while interacting with the
ware, the usability is especially important. If students, instead system [5]. In usability inspection methods usability experts
of studying, have to wonder how computer software works, (and sometimes other software developers, users, and experts
then it disturbs the learning process. We conducted a study at in other areas) examine usability related aspects of the system.
the Tallinn University of Technology (TUT) [1]. During the One of the best-known inspection methods is “Heuristic eval-
study, we compared a TUT in-house developed e-learning uation” [6, 7]. In this method usability experts examine the
system with some internationally known e-learning systems. system that is being tested and decide whether the system cor-
The goal of the project was to evaluate the TUT system responds to some evaluation criteria (heuristics). Schmettow
against other systems and to give some practical recommenda- and Niebuhr [8] propose a pattern-based usability inspection
tions about improving its usability. One of the results of this method, according to which patterns should be used as heuris-
project was a general evaluation method of systems that was tics in usability inspections. We also use patterns in our eval-
developed to repeat such evaluations in the future. uation method. Therefore, in the next section, we shortly in-
In this method, we used patterns as criteria for the evalua- troduce the concept of patterns.
tion of systems. In usability evaluation of e-learning systems
[1, 2], the selection of patterns was based on the experience of B. Patterns
an expert. The first goal of this paper is to explicitly define a Christopher Alexander, one of the well-known proponents
procedure of selection of patterns. The second goal is to illu- of the pattern concept, describes patterns [9]: “Pattern is a
strate the selection process based on usability patterns and three-part rule, which expresses a relation between a certain
propose a set of usability patterns that one can use for the usa- context, a problem, and a solution.” Initially this concept was
bility evaluation of e-learning systems. used in architecture, but later it became popular in other areas
The rest of the paper is organized as follows. Firstly, we in- like software design [10], database design [11], and pedagogy
troduce the basic concepts that are related to our research and [12].
shortly describe our pattern-based evaluation method. Next, Several formats exist for writing down a pattern. However,
we explain the selection process of patterns based on a con- there are some parts that are used in all the formats. Every
crete example. After that, we give an overview of related pattern has the name and usually it provides some kind of hint
work. In the next section, we discuss our selection process to the solution of the problem, the description of which is in
and present a generalized procedure. Next, we describe possi- the pattern. The body of each pattern consists of at least three
ble directions for future work. Finally, we draw conclusions. parts. First part describes context (that means background or

77
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_14, © Springer Science+Business Media B.V. 2010
78 ZUB AND EESSAAR

situation) in which a problem can occur. Secondly, the prob- So if there are more such possible alternatives (systems),
lem itself is described. And finally, the solution to the prob- then it is necessary to further limit the set of systems.
lem is described together with discussion of how to implement One should select the systems that have high popularity or
this solution. Now, we describe our method and how it was quality (in terms of the goal of the AHP model). For the
used for the evaluation of usability of e-learning systems. selection of alternatives one could use the results of sur-
veys, opinions of experts, and personal experiences. If
C. A Pattern-based Evaluation Method of Systems. one selects systems that have low quality according to ex-
We use Analytic Hierarchy Process (AHP) [13] for the perts, then the good result of evaluation means only that s
evaluation of systems. The general idea of our proposed me- has a good quality compared to the other selected sys-
thod is to compare a system s, which we want to evaluate in tems. Our approach is not well usable if one wants to
evaluate a groundbreakingly new system that has few or
terms of an overall goal g, with similar systems by using a set
no similar systems.
of patterns that correspond to g.
3. Selection of patterns (criteria) for the evaluation. We
We have to create a hierarchical model to solve a complex
discuss the selection of criteria in the next section.
evaluation problem, according to AHP. On the highest level of 4. Analysis of the patterns, which are selected for the evalu-
this model is the goal, which is the overall objective that we ation, by comparing them pairwise in relation to the goal
want to achieve. For instance, the goal may be to find relative g of the evaluation. The result of the step is an AHP
usability of different e-learning systems. model that shows us the relative importance of each pat-
At the next level of the model are objectives. They are the tern in terms of g.
main aspects that we should take into account while working 5. Writing of tests that are needed to evaluate the systems in
towards achieving our goal. These objectives may have some terms of the selected patterns.
sub-objectives that are located at the next level of the model. 6. Performing the tests based on the selected patterns.
We propose to use patterns as criteria to evaluate how well 7. Analysis of the results of the tests. Pairwise comparison
systems satisfy objectives. of the systems in terms of each pattern based on the re-
The selected patterns must correspond to the overall goal of sults of tests.
the model. In case of evaluating an e-learning system, we 8. Calculation of final weights for each system based on the
might use web-usability and e-learning patterns as criteria. relative importance of patterns. The resulting model
And finally, at the lowest level of the model, there are al- shows us how well each selected system (including s) cor-
ternatives. They are objects, between which the choice will be responds to the goal of evaluation.
made. We propose to use systems as alternatives. One of the 9. Identification and solving of the main problems that were
systems must be a system that we want to evaluate. discovered during the testing. It means partial redesign
So in general, we are evaluating which of the alternatives and reimplementation of system s.
corresponds better to our objectives. AHP method uses pair- 10. Retesting of s to find out whether it has improved.
wise comparisons where elements of lower level are compared 11. Analysis of the patterns used in the evaluation.
with each other in relation to some element of the model that III. SELECTION OF PATTERNS
is located at the next higher level. Therefore, objectives are
In our previous study [1], the selection of patterns was
compared with each other in terms of the goal. Possible sub-
based on the experience of an evaluator. In this paper, we
objectives are compared pairwise in terms of the objective.
explicitly explain the method for selecting patterns. Firstly,
Finally, alternatives are compared with each other against ob-
we introduce the method based on an example.
jectives (or sub-objectives). For example, in [1], we used an
Initially, we have to select collection of patterns that we
in-house developed e-learning system and in addition three
will use for the analysis. We select Web Usability (WU) pat-
other e-learning systems as alternatives.
tern language [15] because all the systems under discussion in
The steps of our method are the following.
the study [1] are web-based systems and we used patterns
1. Selection of the overall goal g of the evaluation. Based
from this language in [1]. Therefore, we can compare the se-
on that we have to select criteria (patterns) and alterna-
lection based on the experience of an evaluator and the selec-
tives (systems).
2. Selection of systems (alternatives) for the evaluation. tion that is proposed in this paper.
One of the alternatives must be the system s that we want One should take into account the following aspects during
to evaluate. Other alternatives must be systems that are the selection of patterns.
similar to s. The first criterion for selecting a system, • The number of patterns that will be used for the evalua-
with which we want to compare s, is its availability. The tion. If we use AHP method, then probably nine is the
evaluators must have access to the selected systems so maximum number of patterns one could use (see sec-
that they can perform tests based on these systems (see tion II) as criteria. We can use more patterns if the crite-
steps 5–6). Forman and Selly [14] note that psychologi- ria have sub-criteria but it complicates the evaluation. On
cal as well as mathematical evidences suggest “when de- the other hand, WU language contains 79 patterns [15].
riving priorities from paired comparisons, n, the number • Functionality of the system. If a pattern is about a func-
of factors being considered must be less or equal to nine”. tion that is not necessary or important for the system,
 SELECTING PATTERNS FOR THE EVALUATION OF SYSTEMS 79

which we want to evaluate, then we do not have to con- opinion some patterns could belong to more than one group.
sider the pattern. The next step in the process of selecting patterns is to ex-
• A pattern p could be unsuitable for the evaluation because clude patterns that one cannot use for the evaluation due to
it is not important in terms of the overall goal of the anal- their unsuitability for evaluating already existing systems.
ysis or the evaluators do not have enough information
about systems to evaluate them in terms of p. C. Excluding Patterns from the Evaluation
• Some patterns could be related to each other. For in- As we mentioned in the previous section, WU pattern lan-
stance, they could be similar to each other. In this case, guage has some patterns that deal with the initial planning of
we should decide what patterns to choose from the set of systems (Group “Getting started”). Most of the patterns in
related patterns. this group are not suitable for our evaluation. In other groups,
The first step of pattern selection is the classification of pat- there are also some patterns that deal with rather technical
terns. Firstly, we should examine what information the se- issues that are hard (or impossible) to check in the evaluation
lected pattern language already contains. Hence, we now de- process. For example, pattern “Cache transactions”.
scribe the format and grouping of patterns in WU language. The only pattern, which could be useful for us from the set
“Getting started”, is “Two years old browser”. The reason is
A. Groups of Patterns that it describes not only how to test but also what to test.
The patterns of WU language are described in a classical (Check whether the system works in computers with old
format close to the one that is used in [9] by Alexander. hardware and with different web-browsers.) Hence, we pro-
Along with traditional pattern information (like name, context pose to use this pattern in the evaluation.
etc) we could use information about related patterns and star
ranking of patterns. D. Finding the Best Candidates from the Groups
Patterns of WU language are also grouped in categories ac- The goal of this step is to determine for each remaining
cording to a possible order of their use. The groups range group one or more patterns that are the most important in the
from “Getting started with the site” to “Dealing with context of the evaluation.
workflow and security”. Patterns About the Effectiveness of use of the System
Unfortunately, the existing grouping of patterns of WU lan- In this group there are several patterns about the speed of
guage is not very helpful. The reason is that patterns that could using the system like “Download times” and “Content before
be of interest to us are located in different groups. Therefore, graphics”. We do not include these patterns in the evaluation.
in the next section we divide the patterns into other groups We assume that nowadays there are not many problems with
based on the problem, the solution of which they specify. internet speed and there is no need to download large files in
e-learning systems. As you can see, selection of patterns re-
B. A Classification of Web Usability Patterns quires assumptions about systems that we evaluate.
After examining the patterns and problems that they de- Search for information is very important task for all sys-
scribe, we decided to group them based on the types of prob- tems, not only educational ones. Hence, the majority of the
lems that the patterns help us to solve. This way it would be patterns that describe how to organize navigation are impor-
easier for us to select patterns for the evaluation. The groups tant. Therefore, we discuss them next in more detail.
(and the problem areas) are the following. (See table one for Navigational Patterns
number of patterns in each group) Firstly, we shortly describe patterns (from [15]) that in our
• Getting started – development process of systems (from opinion cover navigational aspects.
setting of business objectives to testing). “Site map” – Provide a single page that gives visual and
• Effectiveness of use – problems with navigation, search, textual representation or overview of the entire site.
speed of using the system. “Search box” – Allow visitors to search from a site.
• Functional patterns – patterns that give recommendations “Sense of location” – In case of each page, provide visitors
about the functionality of systems and organization of clear clues about where they are on the page.
functions within a system. “Canonical location” – Make sure that the page name is
• Help patterns – the application of these patterns makes it displayed prominently
easier for users to learn to use the system. “Breadcrumbs” – Do not rely solely on breadcrumbs for
• Content patterns describe what information should be navigation unless you are short of space.
published at the site and how the text should be written. “Site logo at top left” – Clicking on the logo is a way to go
• Design patterns in our case describe how to implement to the homepage.
functions of the system, and how to organize the build- “Navigation bar” – Provide a bar on the home page and
ing-blocks of systems. possibly on other pages that allow users to go to the main sec-
• Technical patterns concern such things like the use of tions of the site.
HTML, security, and others. “Go back to safe place” – Allow users to mark the place
These groups were quite roughly defined. Therefore, in our they want to return if they are lost.
80 ZUB AND EESSAAR

“Back button” – Do not disable back button of browsers. enables a user to get necessary information about the task
Design transactions the way that they can be undone. without browsing help documentation.
If we look at the patterns described above, then we see that Design Patterns
there are some patterns that could be grouped together. For There are a lot of design patterns according to our classifi-
example, “Search box” can be an addition to “Site-map”. cation. Large part of them could be useful in our opinion be-
“Sense of location” is more general variant of “Canonical lo- cause they describe how to improve different aspects of usa-
cation”. “Canonical location” provides a recommendation to bility. Let us describe those patterns in more detail.
use cross-links across site hierarchy. One can use the solu- “Follow standards” and “Keep it simple” are rather abstract.
tions from “Breadcrumbs”, “Navigation bar”, and “Site logo” However, if these two patterns are followed, then it would be
patterns to create these links. “Back button” (or “Site logo”) easier to learn and to use a site.
can be used as a variant of “Go back to safe place”. There- “What you see is what you can use” – If a user sees on the
fore, we think that from these patterns the most important are site only things that he/she can use in the learning process and
“Site-map”, “Sense of location”, and “Go back to safe place”. he/she is not distracted by something else, then it is also good
Content (and Text) Patterns for learning.
This group of patterns explains what information should be “Mandatory fields” – Avoid the use of mandatory fields,
on a web-site and also how it should be represented. In our and while using them provide information about how to fill
opinion, most of content patterns (like “Links to many sites” them in forms.
and “Human touch”) are not very important for e-learning “Oblique Landmarks” – Indicate somehow whether some
systems. From this group we add to the evaluation “Natural features of a system are usable indirectly or when a user is in a
metaphor”. If this pattern is used, then probably it is easier to certain state.
learn to use the system and to understand it. In addition, “Button gravity” – Users usually expect “Submit” buttons
“Home page” pattern could be used in a sense that it would be at the bottom of the page. Therefore, you should locate them
nice if links to all main features of the system would be avail- there (or make them more visible).
able at home page. “Pipeline interaction” – Give users chance to revisit the ear-
There is a similar situation with the text patterns. Most of lier steps of the workflow and correct provided data if needed.
them provide recommendations about how text should be Selecting the Final Subset of Patterns
represented (“Design page for scanning”, “Short text”, and Based on the previous discussion, we have the following
“Separate print pages”). That is surely important for learning patterns left as candidates.
materials, but not for the system itself. From that category of
patterns, we include “Acceptable wording” for the evaluation. 1. Two year old browser 11. Sense of progress
If the system uses well-known terms, then it is easier for users 2. Site-map 12. Context sensitive help
to understand the new system and subsequently learn to use it. 3. Sense of location 13. Keep it simple
Technical Patterns (HTML, Security) 4. Go back to safe place 14. Follow standards
We found no suitable patterns in this group as all of them 5. Natural metaphors 15. What you see is what you
rather technical. 6. Home page can use
Functional Patterns 7. Acceptable wording 16. Mandatory fields
The most important functional pattern in the context of usa- 8. No unpleasant sur- 17. Oblique landmarks
bility evaluation is probably “No unpleasant surprises” – sys- prises 18. Button gravity
tem behavior should be predictable and reliable. Also “Ex- 9. Exploit Closure 19. Pipeline interaction
10. Feedback
ploit closure” can be suitable – this pattern means that before
completing some main tasks it should be guaranteed that a
We found too many patterns (19) that can be used in the
user has completed all the necessary preliminary tasks.
evaluation. How can we reduce the amount of patterns to nine
Help Patterns
[14]? Firstly, the list contains some patterns that specify a
Help patterns are quite important from the point of view of
possible solution (or part of solution) to some other more gen-
usability and they improve the ease of learning and the effec-
eral (abstract) pattern. For example, if we use pattern “Button
tiveness of use of a system. We have selected three patterns
gravity”, then it means that we also use pattern “Follow stan-
from that area. If a system gives a user a “feedback", then
dards”. If users expect that “Submit” buttons are located at
he/she understands better what is going on while working with
the bottom of the page, then it means they are used to that
the system. “Sense of progress” helps user to understand bet-
because of previous experience and it became a standard of
ter the task he/she is completing. “Context sensitive help”
TABLE I
PATTERN GROUPS AND NUMBER OF PATTERNS
Groups of patterns GS Eff Funct Help Cont Des Tech In all
Number of patterns 11 21 8 5 10 20 13 88
Of them selected 1 3 2 3 3 7 0 19
 SELECTING PATTERNS FOR THE EVALUATION OF SYSTEMS 81

some kind. So, we should remove the more specific pattern Welie [16] and Tidwell [17].
from the list of selected patterns and only use the more general There exist some examples of the use of patterns for the
pattern. According to this principle, we propose to eliminate evaluation of systems. Kramberg [18] examines how much
pattern 11 (as it is a sub-pattern of 3), patterns 5 and 7 (and IBM WebSphere software supports the use of control-flow
keep pattern 13), keep pattern 17 and eliminate pattern 15. In patterns from [19]. Kramberg [18] does not select a subset of
addition, we keep pattern 14 and eliminate pattern 18. Pattern patterns for the evaluation. There exists a proposal [8] to use
19 is a sub-pattern of 11, which in its turn sub-pattern of 3. patterns in the usability inspection. Unfortunately the method
This way, we have eliminated six patterns (5, 7, 11, 15, 18, of selecting patterns is not clearly defined. Moynihan et al.
19) and decided to keep four patterns (3, 13, 14, 17). [20] describe a prototype of an expert system for the selection
The second way to decrease the number of patterns is to of software design patterns. But the goal of this system is
look through all of them once again and to decide if some of “leading a designer to a suitable design pattern to the problem
the patterns are really important for the systems, which we at hand” [20] and so it cannot be used to evaluate a system.
want to evaluate. For example, while we think that “Oblique During our selection process, we tried to classify the pat-
Landmarks” pattern increases the usability of systems, it is terns. The idea of classifying patterns is not new. One of the
possible that in e-learning system there are no functions that reasons for classifying patterns is the big amount of pattern
are available only to some special types of users. “Home collections in the literature. Henninger and Correa [21] state
page” pattern states that we should draw the attention of po- for instance that there exist 14 collections of user interface
tential users to a web-site with a home page. That is not im- patterns containing 425 patterns in total. Hence, some papers
portant for learning. The requirement that all the important discuss how patterns and pattern languages can be organized.
functions should be visible at home page can in some way Van Welie and van der Veer [22] discuss how pattern lan-
satisfied with “Site-map” pattern. “Exploit closure” pattern guages in interaction design can be structured. There are at-
concentrates more on how system should be designed so that tempts, like [23], to create repositories of patterns. One of the
all the tasks would be completed in the right order. “Sense of problems of storing many patterns in one repository is that
location” should provide users necessary information so that it different authors may use somewhat different formats for de-
would be clear in what order tasks should be completed. Fi- scribing patterns. In order to solve this problem in Human
nally, “Context sensitive help” is quite important because if a Computer Interaction, a unified pattern format was developed
user needs to understand how a function works, then he/she named Pattern Language Markup Language (PLML) [24].
can get the necessary information without searching that in V. A GENERALIZED PROCEDURE
help files. A recommendation to give information about fill-
ing “Mandatory fields” in forms is similar to the “Context We have developed a method for evaluating systems by us-
sensitive help”, so we do not include “Mandatory fields” in ing patterns as the criteria. In [1, 2], we demonstrated how to
the final subset of patterns. use the method to evaluate usability of e-learning systems
At the end, we have selected patterns 1–4, 8, 10, 12–14 for based on usability patterns. In [1], we selected patterns for the
the evaluation. In the next section, we compare this selection evaluation based on the personal experience and knowledge of
with the one that was made in our earlier research [1]. an evaluator without trying to explicitly explain the process of
selection. In this study, we repeated the selection process and
E. Comparison of the two Selections. divided it into steps so that it would be possible to reason
In our previous research [1] two types of patterns were used about it. The process itself consists of four steps.
for the evaluation: web-usability patterns and e-learning pat- Step 1: Selection of one or more pattern languages (sets of
terns. Therefore, five patterns of each type were selected. For patterns) or individual patterns. The selected patterns form the
the web-usability, the following patterns were selected. “Two initial set S1 of patterns, from which we select the final set of
year old browser”, “Site-map”, “Sense of location”, “Feed- patterns that will be used for the evaluation.
back”, and “Follow standards”. All these patterns were se- Step 2: One has to look through all the patterns in S1 and
lected in the current research as well. It means that the choice divide them into groups. All the patterns in one group must
of patterns in [1] was good and repeatable. But it was based help us to solve similar problems.
solely on the experience of one expert and the process of mak- Step 3: Some patterns might be unsuitable for the evalua-
ing this choice was not explicitly explained. tion process (at least for the evaluation of existing systems).
One has to eliminate these patterns from S1.
IV. RELATED WORKS
Step 4: From each group (step 2) one has to select a subset
As it was already mentioned in Section II, a pattern de- of patterns. The selected patterns are the main criteria and
scribes “a solution to a problem” so that the solution could be form set S2. The cardinality of S2 must not be bigger than
used over and over again in different contexts. There exist nine. Selection must be based on the suitability of the patterns
collections of patterns (pattern languages) that describe prob- to the systems in the analysis. For example, while “Interna-
lems and solutions of a specific domain. Examples from the tionalization” is important for the sites of organizations that
field of usability are Wu Language [15] and collections of van operate in several countries, it is not so important for an
82 ZUB AND EESSAAR

e-learning system where learning is going on in one language. AHP, should not be bigger than nine. For instance, WU lan-
In our opinion a pattern p, which defines a general rule, is guage of usability patterns contains 79 patterns. Therefore, it
very suitable for the evaluation if there are one or more must be possible to select a set of patterns for the evaluation
“sub-patterns” that can describe specific (variants) of solution from a bigger set of patterns.
for a problem defined in p. We can use this principle also for In this paper, we proposed a systematic and structured
selecting a subset of patterns from each group (step 4). If a process how to select a set of patterns that one could use to
selected pattern p has a set of specific sub-patterns, then it is perform evaluations of systems. We illustrated the process by
also possible to use these patterns as sub-criteria of the main selecting a set of usability patterns that one can use to evaluate
criterion (p) in the AHP analysis. The amount of sub-criteria e-learning systems.
of a criterion should also be no more than nine, because we
REFERENCES
have to compare the sub-criteria pairwise.
The proposed process still needs decisions of experts who [1] D. Zub and E. Eessaar, “Pattern-Based Usability Evaluation of E-
Learning Systems,” International Conference on Engineering Education,
perform evaluation. However, the procedure should facilitate Instructional Technology, Assessment, and E-learning (EIAE 07), pp.
the work of evaluators. If decisions, which are made during 117–122, Springer, 2008.
the selection, are explicitly documented, then it also allows [2] D. Zub and E. Eessaar, “Pattern-Based Usability Analysis: Comparison
with a Case-Study,” International Conference on Engineering Education,
interested parties to better understand the evaluation results. Instructional Technology, Assessment, and E-learning (EIAE 08),
To get the final set of patterns for the evaluation, we can re- Springer, 2009, in press.
peat step 4 once more to compare patterns from different [3] J. Nielsen, “The Usability Engineering Life Cycle,” Computer, vol. 25,
pp. 12–22, 1992.
groups. The pattern-based evaluation method is not well usa- [4] E. Folmer and J. Bosch, “Architecting for Usability: A Survey,” The
ble if one wants to evaluate an aspect a of systems, and a has Journal of Systems and Software, vol. 70, pp. 61–78, 2004.
few corresponding patterns that specify the best practices. [5] J. Kontio, L. Lehtola, and J. Bragge, “Using the Focus Group Method in
Software Engineering: Obtaining Practitioner and User Experiences,”
To facilitate the process of pattern selection, the following International Symposium on Empirical Software Engineering,
additional information about each pattern should be recorded pp. 271–280, 2004.
in the repository of patterns: (a) a classification of patterns [6] J. Nielsen, Usability Engineering. Academic Press, 1993.
[7] J. Nielsen, “Heuristic Evaluation,” In: Nielsen J., Mack R. L. (eds.)
based on the problems the patterns deal with; (b) is this pattern Usability Inspection Methods. New York: John Wiley & Sons, 1994, pp.
suitable for the evaluation process?; (c) the history of using 25–62.
the pattern for the evaluation of systems; (d) does this pattern [8] M. Schmettow and S. Niebuhr, “A Pattern-based Usability Inspection
Method: First Empirical Performance Measures and Future Issues,” HCI
provide a general advice or some practical solution?; (e) what 2007, vol. 2 of People and Computers, pp. 99–102, 2007.
is the rating of the pattern?; (f) how is this pattern connected [9] C. Alexander, The Timeless Way of Building. New York: Oxford Uni-
to other patterns in the collection? versity Press, 1979.
[10] E. Gamma, R. Helm, R. Johnson, and J.O. Vlissides, Design Patterns –
VI. FUTURE WORK Elements of Reusable Object-Oriented Software, Addison-Wesley, 1994.
[11] D.C. Hay, Data Model Patterns: Conventions of Thought. New York:
The future work in this area includes the creation of a sys- Dorset House Publishing Company, 1996.
[12] J. Bergin “Fourteen Pedagogical Patterns,” Fifth European Conference
tem, which supports the evaluation process. Among other on Pattern Languages of Programs, 2000.
things it should support the management of patterns and all the [13] T.L. Saaty, “How to Make a Decision: The Analytic Hierarchy Process,”
necessary information for the selection (see Section V) of pat- European Journal of Operational Research, vol. 24, pp. 19–43, 1990.
[14] E. H. Forman and M.A. Selly, Decision by Objectives. World Scientific
terns. Another direction of research is to study the following Publishing Company, 2002.
aspects of pattern-based testing process. [15] Web Usability Patterns. Retrieved September 25, 2009, from
• What kind of tests to use in the evaluation? http://www.trireme.com/WU/home.htm
[16] M. van Welie, Interaction Design Patterns Library. Retrieved September
• How many patterns to evaluate in one test? Should a test 25, 2009, from http://www.welie.com/patterns/
cover one pattern or a set of them? This information [17] J. Tidwell, Designing Interfaces. Retrieved September 25, 2009, from
could be later stored in a repository so that the future use http://designinginterfaces.com/
of this test would be easier. [18] V. Kramberg, “Pattern-Based Evaluation of IBM WebSphere,” BPEL
Master’s Thesis, Universität Stuttgart, 2006.
• Who will be the testers of the system and how many [19] V.M.P. van der Aalst, A.H.M ter Hofstede, B. Kiepuszewski and A.P.
people should participate? Barros, “Workflow patterns,” Distributed and Parallel Databases,
• How to determine the relative importance of the problems vol. 14, pp 5–51, July 2003.
[20] G.P. Moynihan, A. Suki, D.J. Fonseca, “An Expert System for the Se-
discovered during the testing, and how to classify them? lection of Software Design Patterns,” Expert Systems, vol. 23,
pp. 39–52, 2006.
VII. CONCLUSIONS [21] S. Henninger and V. Correa, “Software Pattern Communities: Current
We have proposed an evaluation method of systems that is Practices and Challenges,” 14th Conference on Pattern Languages of
Programs, 2007.
based on Analytical Hierarchy Process (AHP) and patterns. [22] M. van Welie and G. C. van der Veer, “Pattern Languages in Interaction
Patterns are criteria, based on which is possible to evaluate Design: Structure and Organization,” Interact’03, pp. 527–534, 2003
how well systems correspond to different objectives. Psycho- [23] A. Gaffar, D. Sinnig, H. Javahery, and A. Seffah, “MOUDIL: A Com-
prehensive Framework for Disseminating and Sharing HCI Patterns,”
logical and mathematical evidences suggest that the amount of CHI 2003 workshop on HCI Patterns: Concepts and Tools, 2003.
criteria (that correspond to the objectives), that one can use in [24] S. Fincher, “Perspectives on HCI Patterns: concepts and tools (introduc-
ing PLML),” Interfaces, vol. 56, pp. 26–28, 2003.
 Effectiveness of Collaborative Learning in Teaching
Information Systems
Ioana Tuugalei Chan Mow
National University of Samoa

Abstract: This paper is based on the second phase of an collaborative learning could be proven effective in teaching
investigation which explored student attitudes and experiences in computing subjects, this would have a direct beneficial
collaborative and individualised learning in computer based bearing on resource implications.
instruction. It attempted to answer the questions a) whether The current research was conducted in two phases. The
Samoan students prefer learning in a cooperative learning
environment as opposed to individualised learning and secondly
research was structured such that the first phase initiated a
b) whether collaborative learning is more effective than general comparative investigation of the attitudes and
individualised learning? Research results indicated that students performance of students in collaborative learning as compared
enjoy and show positive attitudes towards collaborative learning to individualised learning. This initial phase spanned three
and that they perform well in a collaborative environment weeks. The second phase provided a semester long and more
provided the essential conditions for effective collaborative in-depth study of the essential conditions necessary for
learning are satisfied or present. optimal collaborative learning.

II METHODOLOGY
I INTRODUCTION
The research design used in the study was quasi-experimental,
This paper describes the second phase of an investigation
where the independent variable (learning environment) was
which explored student attitudes and experiences in
manipulated and subsequent changes in the dependent
collaborative and individualised learning in computer based
variables (student attitudes to learning environment and
instruction. It attempted to answer the questions
student performance) were investigated. The quasi-
• Do Samoan students prefer learning in a cooperative
experimental design was chosen for both phases in the study,
learning environment as opposed to individualised
as its strict control of variables enabled a distinct focus on
learning?
causal relationships between learning environment and student
• Is collaborative learning more effective than attitudes to learning.
individualised learning?
The Computing Department at the National University of The population for the second phase was a ‘captive’ one
Samoa offers courses in Computer Science and Information consisting of students in a Computer Information systems
systems. One of the challenges in teaching Information class. The students in this phase had already completed an
Systems is the provision of students with skills in developing Introduction to Information Systems class in Information
real life systems in the classroom. In the workplace, these systems analysis and design. Phase 2 involved a much smaller
systems are developed using the team approach.Therefore, the group of students (9 students) and group selection was on
use of collaborative learning in the learning environment for ability so that each group of three was of mixed ability (high,
Information systems coordinates well with the team approach low and average). The composition of the group is an
to Information systems design in real world applications. important factor in CL and research in CL recommend
Another motive for undertaking this study was the interest in heterogeneous groups as some difference in view points is
the cultural effects of collaborative learning in the field of required to initiate interactions between group members [2].
computer education in Samoa. Samoan culture is communal Ability was based on previous grades in CSB 283, the
and based on communal practices[1]. Family members prerequisite class to this Information Systems class.
cooperate and share in all manner of responsibilities ranging
from domestic tasks to maintaining family lands and providing III PROCEDURES
for the welfare of the family.
A major drawback in Educational computing in Samoa has Phase II of the research was conducted on students in an
been the high cost of computer technology, manifesting itself Information systems and database design class at the local
in the shortage of computers and the high student to computer university. The class was divided into heterogeneous groups of
ratio. Enrolment numbers for computer courses have always threes according to ability (high, average, and low ability).
been restricted by the availability of computers. If Selection and composition of the groups were decided by the
lecturer, who was also the principal researcher.

83
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_15, © Springer Science+Business Media B.V. 2010
84 MOW

To provide the context rich environment of the real world, the Evaluation of group projects involved one group presentation
groups were assigned two real life projects. The first project for the Library system and written documentation such as data
involved the development of a computerised system to flow diagrams, entity relationship diagrams, data dictionary,
automate the checkin/checkout transactions of the university user manuals and a disk copy of the final database system.
library. The second project involved the development of a Assessment was done on a group basis with one mark given to
computerised invoice entry system for a local water company. the whole group. Assessment from the two projects accounted
The major activities in the two projects involved for 80% of the overall assessment. The remaining 20% came
• the groups carrying out investigative interviews on the from a 1-hour test which tested their knowledge and
clients to elicit such things as user requirements, problems comprehension of database theory.
in existing systems leading to a need for an improved
system and also the feasibility of the new system. B Student attitudes
• Systems analysis using a CASE software tool called
ASCENT Student preferences for a particular learning environment and
• Systems design using ASCENT and Word for Windows their feelings towards such an environment were measured by
for prototyping design of menus and user interfaces self-administered questionnaires, structured interviews and
• Implementation of the system using Microsoft Access. observations by the principal researcher. The questionnaire
• Testing of the database system and making the necessary focused on issues of, the enjoyment of learning, self-esteem,
revisions and improvements. motivation, learner control, and social interactions. For the
Since the two systems were planned and designed using purpose of the current study, the enjoyment of learning was
Systems Development Life cycle approach, which involved interpreted and defined as students’ valuation and appreciation
several phases, assessment of the projects was continuous. For of the learning experience.
example, the deliverables for the early planning and
investigative phases were assessed and feedback given on V DATA GATHERING TECHNIQUES
areas needing improvement before proceeding to the systems Student attitudes to the learning environment were measured
analysis and design phases and eventually to the by self- administered questionnaires and also by observations
implementation. by the researcher. In addition to the self-administered
Throughout the two projects, which spanned the duration of a questionnaires, groups in Phase II were also interviewed on
semester, students were expected to carry out planning, their views and experiences with collaborative learning.
analysis and problem solving by means of extensive group
discussions.
VI RESULTS
IV OPERATIONALISATION OF VARIABLES The aim of the current research was to establish from the
results of the study, the following propositions. Firstly, the
In this study, the independent variable was the learning study attempted to confirm that students in Samoa enjoyed
environment which had two values in its domain, collaborative collaborative learning and preferred it to individualised
learning and individualised learning. The dependent variables, learning. Secondly, the study attempted to establish that
were student performance and student attitudes to the learning students in Samoa performed better in a collaborative
environment. In the course of the study, the independent environment.
variable (learning environment) was manipulated and Hence, for purposes of the discussion, the main areas of focus
subsequent changes in the dependent variables (student were student performance and student attitudes to learning.
attitudes to learning environment and student performance)
were gauged using questionnaires, surveys and pre and post- A Student attitudes to Learning
tests. Student attitudes to collaborative learning were gauged from
responses to the questionnaire administered at the end of the
A Student performance session and also from interviews with the researcher. Results
from the various questionnaires and interviews, showed that
In Phase II, the effectiveness of the learning environment is the majority of the students viewed collaborative learning as
student performance measured or assessed as a group using an effective instructional strategy with positive effects on
criteria such as group presentations, grades for deliverables problem solving, self esteem, motivation and enjoyment.
such as the Initial Data Model, Initial Process Model, However, despite these outcomes, the results did not show that
Feasibility study, database tables, forms, menus and reports. students preferred collaborative learning over individualised
The marks for the collaborative learning sessions were learning as the analysis of Phase 1 results did not reveal any
compared to marks from the previous Information systems significant differences in attitudes between the two treatments.
class and the difference between the two sets of grades, Discussion with the students in the interviews revealed the
analysed statistically, for significance. following:
 EFFECTIVENESS OF COLLABORATIVE LEARNING IN TEACHING INFORMATION SYSTEMS 85

Question 1 Did you enjoy working together as a team? as the group members discussed the various issues in the
Of the three groups in this second phase, the members of two project.
of the groups, enjoyed working together as a team. Two of the
members of Team 2 in the early stages of the exercise, initially Question 7 Did working in a group help stimulate thinking?
resented working in a collaborative mode, due to the inability As revealed in the interviews, all of the participants, agreed
to resolve conflicts between group members. However, their that group learning stimulated their thinking, in that the group
attitudes changed as the exercise progressed. discussions gave them new perspectives on the subject area
and brought up other valuable issues related to the topic under
Question 2 What problems, if any, did you have working discussion.
together as a group?
Members in Group 2 indicated that they had had problems in Question 8 Did working in a group give you new perspectives
scheduling times in which everyone would be available for on the subject area?
group discussions. There were also feelings of resentment All of the students agreed on the value of group learning on
where some members felt that members in the group were not broadening one’s perspectives.
contributing their share of the work. Most of the members in
the other two groups, felt that they had progressed well and
had not had problems working as a team. VII DISCUSSION
The aim of the current research was to establish from the
Question 3 Did everyone contribute equally or did some results of the study, the following propositions. Firstly, the
dominate the group? study attempted to confirm that students in Samoa enjoyed
The general feeling in Group1 and Group3 was that everyone collaborative learning and preferred it to individualised
contributed equally to the group. In Group 1, Student 1A and learning. Secondly, the study attempted to establish that
Student 1B were Accounting experts while Student 1C students in Samoa performed better in a collaborative
provided the expertise in computing applications, so there was environment. Hence, for purposes of the discussion, the main
some form of division of labour, as the members areas of focus were student attitudes to learning and student
complemented each other’s skills and shared their expert performance. Student attitudes to Collaborative learning were
knowledge. both positive and negative.
Question 4 Did working together help the student’s i) Positive attitudes to Collaborative learning
understanding of the project? The groups in which everyone participated and contributed to
All of the students agreed that collaborative learning improved team progress enjoyed the experience and found that
their understanding of the project. There were aspects of the collaborative learning led to increased social interactions
project one would not have thought of, had they not been told between the members of the group. As evident in all of the
of, by other members of the group. Discussion of some of the student interviews, students felt that solving problems and
issues in the project, led to awareness of further issues in the carrying out tasks in groups was a lot more effective than
project that they, as a group, had to resolve and address. doing it individually. From student responses in student
However, Student 2D and Student 2E in Group2, felt that interviews, it was clear that practically all of the students in
working as a group actually led to confusion of issues in their the collaborative group felt that it was an effective
group. When asked to elaborate, they explained that both instructional strategy for the following reasons :
members were convinced that their own views of the way in
which some aspect of the project should be approached were • Collaborative learning taught students teamwork
the correct one and refused to accept any other perspective. Students felt that collaborative learning taught them to work
together as a team and share both the work-load and ideas.
Question 5 Do you feel that collaborative training helps in This was consistent with the findings of researchers such as
pooling your knowledge and experience? [3] and [4] that demonstrated the positive effects of
When asked about the value of collaborative learning in collaborative learning in fostering team-work.
pooling their knowledge and experience, all of the members in
Group 1 and Group3 felt that they would not have been able to • Students enjoyed Collaborative learning
develop and implement the systems without the help from Practically all of the participants, claimed, they had enjoyed
fellow members and the lecturer. working in the collaborative mode. This was especially true in
Student 2D and 2E initially had doubts about the worth of groups where members had successfully worked together as a
group learning, but as the research progressed, they changed group and enjoyment was probably due to feelings of success.
their opinions in favour of collaborative learning. It could also be argued that enjoyment led to an increase in
intrinsic motivation. These findings were consistent with the
Question 6 Did you get helpful feedback in your group? literature and it could be argued that these feelings of
All of the students participating in the collaborative mode, enjoyment were due to increased motivation and increased
indicated that they received helpful feedback from their group social interactions. Collaborative learning, as [5] pointed out
86 MOW

was a social event where learning was fostered through peer ii) Negative attitudes to Collaborative learning
interactions and social discourse. It should be acknowledged, that student attitudes to
collaborative learning in the current study, were not all
• Collaborative learning led to increased self confidence positive. The research results also included issues in
The positive effects of collaborative learning on self collaborative learning, which the participants found
confidence and esteem have been established by findings of troublesome and aspects of the learning environment, which
studies by [7] and [9]. This was confirmed by responses of the they found disagreeable. However, in viewing all of the
majority of the participants in the collaborative treatment, who student feedback, these negative aspects were quite minimal
maintained, that collaborative learning had boosted their self- (less than 10 %) when compared to the overwhelming volume
confidence. Again this was evident mostly in groups where the of positive feedback.
collaborative mode had been successful leading to feelings of
increased self-confidence. Furthermore, the mere fact that • Lack of cooperation
most of the decisions were the product of group discussions Perhaps the most common complaint raised by the participants
and consensus, meant that students felt a lot more confident in was the lack of cooperation from some team members who
any decision making and tasks done by the group as it had the were not doing their “fair share” of the work. The lack of
approval and confirmation of other people. cooperation also resulted in the lack of group processing,
which, as recommended by key researchers ([3]; [11]) in the
• Collaborative learning promoted sharing of ideas field, was one of the essential factors for successful
Practically all of the students in the collaborative group, collaborative learning.
confirmed, that they felt that the benefits of collaborative
learning lay in the ability to share ideas through group • Slow pace of work
processing. Another complaint voiced by the students in the collaborative
mode was the slow pace of progress. It could be explained,
• Encouraged discussion of issues from various view points that apart from some students wasting time on “off task
enabling an in-depth probing of problem, task or concept. activities”, the longer time spent on tasks, was mainly due to
As pointed out by [5], knowledge is a social construct and the more time spent on discussions and increased interactions
learning process is facilitated by social interactions amongst amongst group members. However, to this end, the author
students as they collaborate with each other to actively would like to claim that this slow pace was actually beneficial,
construct their learning. In the collaborative learning groups, as the group members needed this time to carry out in-depth
the sharing of ideas and varying viewpoints through discussions, to critically reflect on the group’s progress and
discussions led to a deeper understanding and analysis of a plan of action and to resolve any conflicts and controversy
problem, task or concept as it exposed the student to concepts which may have arisen. These activities, if carried out
they had not thought about before. These supported the effectively, would have resulted in improved problem solving
findings of studies by [3], [4], [6], [7], [8] and many others, and promoted critical thinking within the group.
which showed that collaborative learning led to increased
cognitive analysis and enhanced problem solving in learning. • Resolving controversy
Another issue, brought to the fore by the author’s
• More able students provided ‘scaffolding’ for the weaker observations, was that of resolving conflict and controversy
ability ones amongst group members.
According to [12],
The concept, referred to by [10] as the “zone of proximal “Conflict exists whenever incompatible activities
development”, that the more able students provide mentoring occur...”
for the less able, was quite noticeable in the successful groups Classification of conflicts include “controversy, which occurs
in Phase II. In these groups, the more able students helped the when a person’s ideas or opinions are incompatible with
weaker students in understanding concepts. A direct result of others” ([12]), and there is a need to arrive at some consensus
this mentoring was the obvious improvement in self- or agreement. [12] have argued that controversy, when
confidence in the less able students and, more importantly the managed constructively, could lead to enhanced problem
significant improvement in achievement of the weaker ability solving, enhance the quality of reasoning and decision
students in Phase II. making and strengthen liking, respect and trust amongst
members.
In Phase II, the feelings of hostility between Student 2A and
• Increased motivation Student 2B, were due to their inability to resolve conflicts
The students also indicated that collaborative learning within the group. These students had opposing views on most
promoted increased feelings of motivation, and this was issues. However, neither was prepared to come to some
particularly true in cases where collaborative learning had agreement or work out a settlement. Controversy, when
been implemented successfully. These findings are in tune resolved constructively would prove beneficial to
with the findings of [3], and [8]. collaborative learning. However, controversy, when
 EFFECTIVENESS OF COLLABORATIVE LEARNING IN TEACHING INFORMATION SYSTEMS 87

unresolved would lead to resentment and a breakdown of

collaborative learning. Inspection of achievement of these VIII CONCLUSION
students, showed them up as having the least improvement. In
fact, for student 2A, the effect of collaborative learning on The results of Phase II of the present study confirmed that
achievement was negative. The inability to resolve conflicts students in Samoa perform well in a collaborative
led to a lack of group processing, and hence resulted in environment, provided that the essential factors for effective
ineffective collaborative learning. collaborative learning, such as positive interdependence,
individual accountability, team and social skills and group
iii) Collaborative learning versus individualised learning processing, were incorporated into the educational
On the overall, the study confirmed the social benefits of environment.
collaborative learning such as enhanced self-esteem, enhanced To conclude, it is clear from the results of the research that if
problem solving skills, and the enjoyment of learning. These collaborative learning is to be successfully incorporated into
findings were consistent with those chronicled in the research the education process, further investigation is necessary. With
literature on collaborative learning. the right conditions, and using the right procedures,
collaborative learning could prove to be an effective
iv)The effect of communal culture on attitudes to inexpensive teaching strategy that could contribute positively
collaborative learning to the education process in Samoa.
It could be argued that students enjoyed and showed positive
feelings towards collaborative learning because they were Bibliography
already familiar with such practices in everyday life. In the [1] Ochs, E. (1988). Culture and Language Development: Language
‘faa Samoa’ or the Samoan way of life, most of the activities Acqusition and language socialization in a Samoan village. Cambridge Press,
or practices are done collaboratively or collectively. The group p 71 – 78.
rather than the individual is important and groups can be the
[2] Beers, P. J., Boshuizen, H. P. A., & Kirschner, P. A. (2007). The analysis
nuclear family unit, the extended family, the church group or of negotiation of common ground in CSCL. Learning and Instruction, 17,
the village to which one belongs. Therefore, it could be argued 427–435.
that students learning in a collaborative setting enjoy it as they
are already enmeshed in these collaborative networks in their [3] Morgan, R. L., Whorton, J. E., & Gunsalus, C. (2000). A comparison of
short-term and long-term retention: Lecture combined with discussion versus
way of life and is an extension of their daily communal cooperative learning. Journal of Instructional Psychology, 27, 53–58.
practices.
[4] Van Boxtel, C. A. M., Van der Linden, J., & Kanselaar, G. (2000).
B Student performance Collaborative learning tasks and the elaboration of conceptual knowledge.
Learning and Instruction, 10, 311–330.
Students in Phase II showed a marked improvement, overall,
in their academic performance, and this was consistent with [5] Kirschner, P. A., Martens, R. L., & Strijbos, J. W. (2004). CSCL in higher
the findings of the majority of studies in collaborative learning education? A framework for designing multiple collaborative environments.
([13]; [3]; [4]; [14]. It could be argued, that the success of In P. Dillenbourg, J. W. Strijbos, P. A. Kirschner, & R. L.Martens (Eds.),
Computer-supported collaborative learning: Vol 3. What we know about
collaborative learning in Phase II, was due to the presence of
CSCL…and
all the key elements, necessary for effective collaborative implementing it in higher education (pp. 3–30). Boston, MA: Kluwer
learning. The presence of individual accountability and Academic.
feelings of personal responsibility, amongst group members,
[6] Heller, P.,Keith,R. & Anderson, S (1992). Teaching problem solving
ensured the successful implementation of collaborative
through cooperative grouping: Group versus individual problem solving.
learning, resulting in the positive effects on student American Journal of Physics Teachers, 60, p 627 – 636.
achievement. Within the groups, there were extensive
discussions as members worked together as a team to [7] Sharan,Y., & Sharan,S. (1992). Expanding cooperative learning through
group investigation. NY:Teachers College Press.
collectively devise strategies for designing successful systems.
Group members also took time aside to critically reflect and [8] Slavin, R. E. (1995). Cooperative learning: Theory, research, and practice
discuss the progress of the group. In fact, collaborative (2nd ed.). Boston: Allyn & Bacon.
learning had fostered improved social relations and friendships
[9] Repman, J. (1993). Collaborative ,computer based learning: cognitive and
amongst these group members.
affective domains, Journal of Research on Computing in Education, 9(2), p
It was evident then, from the discussions set out above, that 149 – 163.
for successful collaborative learning, certain conditions were
essential and needed to be systematically incorporated into the [10] Vygotsky, L.S (1978). Mind and society:The development of higher
psychological processes. M.Cole, V.John Steiner, S. Scribner, & E
educational process. These conditions include a commitment
Souberman (Eds & trans.)(Originally 1927) Cambridge; Harvard University
amongst members to work as a team, group processing, press.
positive interdependence and individual accountability.
[11] Smith,K.A. (1992). Cooperative/collaborative learning for active
students involvement: Essentials of Cooperative Learning. Teaching
Concerns. URL: http://www.virginia.edu/~trc/tccoop.html).
88 MOW

[12]Johnson, D. W., & Johnson, R. T (1995). Creative Controversy: [14]Springer, L., Stanne, M. E., and Donovan, S. (1998). "Effects of
Intellectual challenge in the classroom (3rd edition). Edina, MN: Interaction cooperative learning on undergraduates in science, mathematics, engineering,
Book Company. and technology: A meta-analysis." (Research Monograph No. 11). Madison:
University of Wisconsin-Madison, National Institute for Science Education,
[13] Kimber,.D. (1994) Collaborative Learning in Management education: Review of Educational Research.
Issues and Benefits, problems and solutions: A Literature Review.
 Implementation of a Curriculum Management
System at a Faculty of Medicine – Lessons Learned
from a Pilot Project
Aarti Paul Daniel Yaworski Keith McConnell Bruce Martin Dean Sandham Sat Sharma
Faculty of Faculty of Faculty of Faculty of Faculty of Faculty of
Medicine Medicine Medicine Medicine Medicine Medicine
University of University of University of University of University of University of
Manitoba Manitoba Manitoba Manitoba Manitoba Manitoba
[email protected] daniel_yaworski@ [email protected] [email protected] Dean_Sandham@c [email protected].
ba.ca hotmail.com nitoba.ca toba.ca c.umanitoba.ca ca

Abstract— Integration of electronic curricula remains highly attitudes [3] [5]. Recent advances in information technology
variable among medical schools in North America. A major have laid a strong foundation for the use of electronic
challenge has been availability and implementation of an ideal technology to deliver, support and enhance teaching, learning,
curriculum management software system (CMS). The objective and evaluation [6]. Using well-designed computer-based
of this study was to test the feasibility of using a CMS system at
the University of Manitoba (U of M). At the U of M, we
courseware, electronic learning can facilitate live teaching
implemented a multi-user software system designed to provide with streaming lectures, whiteboards, downloadable slide sets,
faculty and students with an integrated curriculum, scheduling, and discussion forums [6] [9]. The benefits of e-learning
calendar functionality, and communication tools. The goals of the include faster learning at reduced cost, increased access to
pilot project were to explore the feasibility of an electronic learning, and accountability of all participants in the learning
curriculum and to evaluate strengths and weaknesses of the process [10] [11]. Changes in health care delivery, advances
software. One block each from the 1st and the 2nd year medical in medicine, increased demands on academic faculty, and
school class were chosen for the pilot project. Data were collected transformation of health care delivery sites from tertiary care
through student and faculty surveys, student and faculty login- institutions to community-based settings have challenged
ins, student page views, and through direct communication with
various users. Most students (100%) and faculty (78%) indicated
traditional teaching time and resources [12].
a strong need for an electronic CMS. A statistically significant
increase in student log-ins (p<0.01) and page views (p<0.05) II. BACKGROUND WORK
occurred from week 1 to week 6 and then reached a plateau, Despite gaining popularity over the past decade, the use of
whereas faculty log-ins remained constant. The 1st year and 2nd
e-learning remains highly variable among medical schools and
year medical classes were dissatisfied (34% and 57%), neutral
(24% and 25%), or satisfied (42% and 18%) with the CMS, its integration has been a genuine challenge, especially when
respectively. This pilot project demonstrated that an electronic teaching clinical disciplines [6] [12]. Creation and
CMS was feasible at the U of M. In addition, we learned how to deployment of learning management systems that support e-
effectively implement such a system, and how to encourage learning are arduous and expensive, and their integration into
faculty and students to adopt the system. The lessons learned at U medical education is equally challenging. E-learning material
of M may be instructive for other faculties attempting to deploy involves several components [13] [14]. Once the content is
electronic curriculum for undergraduate medical education. developed, it must be managed, delivered, and standardized.
Content management includes all the administrative functions
Keywords- eLearning; electronic curriculum; curriculum (e.g., storing, indexing, and cataloguing) needed to make e-
management system; learning management system. learning content available to the learners [15]. Examples
include portals, repositories, digital libraries, learning-
I. INTRODUCTION management systems, search engines, schedules, and
personalized e-portfolios. A curriculum management system
Medical education aspires to provide future doctors with the includes web-based software that facilitates the delivery and
knowledge and skills to practice medicine and to develop tracking of e-learning across an institution. It may also serve
lifelong learning skills [1], [2]. The traditional model of several additional functions [16] [18]. For example, CMS can
medical education is based on transferring scientific facts to simplify and automate administrative and supervisory tasks,
students via lectures, notes and books, culminating in track learner’s achievement of competencies, and provide a
knowledge synthesis, and acquisition of skills and professional

89
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_16, © Springer Science+Business Media B.V. 2010
90 PAUL ET AL.

readily accessible and life-long repository for learning uploaded their own material, while others e-mailed it to the
resources [19]. The American Association of Medical course director, course administrator, or to us for uploading.
Colleges has suggested various approaches to implement Personal communication with the faculty and course directors
informatics skills into the medical school curriculum [1]. appeared to be the most successful method of preparing the
In 2004, the Faculty of Medicine (FOM) at the U of M curriculum for the pilot project. An implementation committee
began exploring the feasibility of acquiring a multi-user, was created to oversee the pilot roll-out. Members of the
integrated electronic curriculum management system for its committee included directors from 1st and 2nd year medical
medical school. The goal of this curriculum system was to school classes, course directors, pre-clerkship administrators,
provide medical students and faculty the best available and student representatives from each class. Feedback and
technology for knowledge transfer. Such a system was counsel was obtained from the committee and other users prior
expected to offer a simple, user-friendly interface that could be to customization of system for the U of M pilot project.
utilized by individuals with varying levels of technical skill. In (Fig. 1)
this paper, we describe our experience and the challenges we We used multiple strategies to prepare medical school
encountered at U of M during selection, preparation, faculty, students, and administrative staff for a successful
customization, implementation and evaluation of a web based launch of the pilot project. These included personal
comprehensive electronic curriculum management system. communications, presentations, and discussion at various
undergraduate administrative/education committees. Critiques
and suggestions obtained from multi-user groups were
III. METHODS
collated, analyzed, and put into practice. Hands-on training
A. Selection of the Curriculum Management System sessions were scheduled and well attended by faculty,
We began by analyzing curriculum management systems students, course directors, and their administrative staff. A
currently in use at various Canadian and American medical videoconference was arranged to communicate, share and
schools. Several candidate software systems were carefully learn from the University of Calgary’s experience with the
evaluated for functionality, architecture, support, and cost implementation. They experienced numerous adoption
efficiency. Following a preliminary review, detailed and difficulties as their curriculum gradually evolved into a
exhaustive analysis of three systems was performed. These paperless system. A focus group included 15 to 20 students
included: 1) Curriculum Information System initially from each class, CMS representatives, and academic
developed by the Med-IT Department, University of Calgary, representatives to enhance communication between the
and its intellectual property rights are currently owned by U of implementation team and the class.
M.; 2) Web Eval, developed by One 45, a commercial medical After approximately six months of preparation and
informatics company; and 3) WebCT, commercial e- planning the CMS was launched in January of 2007. We
curriculum software. targeted one block from each pre-clerkship year for the pilot
The product analysis substantiated Curriculum Information project. Block III, taught during the 1st year consists of
System (CIS®) compared to Web Eval and Web CT is a Respiratory Medicine, Cardiovascular and Otolaryngology
comprehensive software package that links scheduling, courses. Block VI, launched in March of 2007 is taught during
learning objectives and resources. Successfully implemented the 2nd year and includes Gastroenterology, Hematology, and
previously at the University of Calgary, CIS® is an integrated Dermatology courses. There were 105 students in the 1st year
electronic medical curriculum and scheduling software that class and 98 students in the 2nd year class. The launch
can be delivered via the Internet and wireless hand-held consisted of fully integrated individualized schedules linked to
PDA’s. The goals of the pilot project were to explore the the curriculum, discussion forums, news groups, calendar, and
feasibility of implementing an electronic curriculum at U of M e-mail functionality. Since this was a pilot project, paper
and to evaluate strengths and weaknesses of the CIS software. resources and student websites were maintained as before.
We envisioned a CMS that would foster and sustain lifelong Therefore, three simultaneous systems existed for the
learning for medical students by providing a solid curriculum during the pilot project: paper-based notes;
undergraduate medical education foundation for future teaching material posted on the student website; and CMS.
growth. An ideal system would be a multi-user, web based Two systems coexisted for scheduling: a printed course
collection of electronic tools designed to assist medical schedule book and CMS.
faculty, students and administrators in the management of all A pilot project evaluation committee was created to
knowledge pertinent to learning objectives defined by the evaluate the feasibility of an electronic curriculum information
curriculum. system for the FOM, U of M, and to assess strengths and
weaknesses of the CMS software. The committee contained
B. Preparation for the Pilot Project Launch representatives from 1st and 2nd year classes, academic and
Prior to the pilot project, we prepared and uploaded CMS representatives, 1st and 2nd class year program
learning resources and teaching objectives to populate the coordinators, course directors for Block III and VI, an
system. All teachers were responsible for ensuring that their education expert, and administrative coordinators. Feedback
content was current, fulfilling the teaching objectives, and was solicited from the students, faculty, course directors and
uploading content prior to the teaching activity. Most teachers administrative staff through student and faculty surveys,
 IMPLEMENTATION OF A CURRICULUM MANAGEMENT SYSTEM 91

Fig.1. A personalized interface for students and faculty opens up as the user logs in.

student and faculty log-in and page view data, and informally both 1st and 2nd year classes. Major increases (p <0.05)
through direct communication with the users. occurred between weeks 1 and 6, and the access remained
We hypothesized that a substantial increase in log-ins and stable thereafter. (Fig. 3) An independent survey conducted by
page views over time will demonstrate success, especially the students was submitted to the evaluation committee. The
since multiple systems were available. Data from week 1 was survey evaluated student satisfaction, and strengths and
compared to data from week 6. The sequential log-in and page weaknesses of the CIS® software.
view data was analyzed using chi-square analysis and
student’s t-test.
IV. RESULTS
Weekly system usage data were extracted from the system
logs and included a breakdown of various components and
modules. While faculty logins remained constant, student
logins continued to increase for both 1st and 2nd year classes. A
statistically significant increase (p <0.01) in student logins
occurred from week 1 to week 6 and then reached a plateau
(Fig. 2). Each student logged into the system approximately
once a day, including the weekends. This data is remarkable
since login was not necessary or required for the courses. Each
student received paper and other resources, including the
scheduling information and course material in advance.
Furthermore, with one login, students could access and
download all teaching material for the day. Curriculum page Fig. 2. Login data for 1st year class from week (WK) 1 to week 6 and both
views indicate that a learning resource was accessed by the 1st and 2nd year classes from week 7 to week 10. A significant increase in
user. These page views dramatically increased over time for log-ins (p <0.01) occurred from week 1 to week 6.
92 PAUL ET AL.

faculty members involved with the pilot project at the end of

the pilot project are shown. (Fig. 4 and 5)
TABLE 1
Results of an independent survey conducted by the 1st and 2nd year students.
This table shows data from student’s responses to the question “How would
you rate your overall satisfaction with the CIS?”

Fig. 3. Curriculum page views for 1st year class from week (W) 1 to week
6 and both 1st and 2nd year classes from week 7 to week 10. A significant
V. RECOMMENDATIONS FROM THE PILOT PROJECT
increase in page views (p <0.05) occurred from week 1 to week 6 and Based on analysis of the pilot data and comprehensive
week 10. consultation with all stakeholders, the evaluation committee
made several recommendations. These lessons learned at U of
A. Analyzing Students’ Responses M may assist other medical schools in their own development
It is important to analyze student evaluations and and implementation of electronic curricula.
comments in an appropriate context. The student acceptance 1) To prepare future physicians to deliver best medical
and adoption that we achieved during this pilot project met our care, the faculty and students at U of M determined
expectations. The project was conducted in the middle of an that an electronic curriculum and an integrated
academic year for the 1st year students after they had adjusted curriculum management system were high priorities.
to the paper based scheduling and curriculum system. The 2nd 2) The pilot project at U of M provided a unique
year students were generally less enthusiastic as this was their opportunity for students, faculty, and administrative
last pre-clerkship block before starting clinical rotations. staff to conceptualize, prepare, launch, troubleshoot,
Students also had ready access to the class website, an and evaluate an electronic curriculum. Extensive
unmonitored site that serves as a resource for communication preparation, free communication, thorough
and planning social events. The class website is also used as a consultation and engagement of all users from an
repository for learning materials. The student website is much early stage and throughout are required for successful
easier to navigate than the CMS which requires training and implementation.
has a learning curve to become proficient. While most students 3) The pilot project at U of M identified weaknesses of
were either supportive or neutral (Table 1), some were the software system and gaps in the wireless
resistant to change and thus were the most vocal opponents of infrastructure and hardware of the FOM. (Table 2)
the pilot project. Therefore, “cultural issues” or inherent This evaluation guided further modifications and an
“skepticism” for change definitely appeared to be an important upgrade of the CIS® software to improve interface,
factor in adoption. Such barriers during the implementation of usability, functionality and speed of the system.
electronic systems are well documented in the literature and 4) To gain faculty-wide engagement and adoption,
gradually decline before widespread acceptance and adoption duplication of the curriculum and schedule should be
[20] [21].
Feedback from the Faculty
A total of 53 instructors from 6 clinical disciplines participated
in this pilot project. Six of the course directors were intimately
involved with the roll-out and evaluation. All course directors
and many faculty members provided formal and informal
feedback on the electronic curriculum and evaluation of the
software. Faculty response was generally positive and
enthusiastic. Many course instructors used the system
extensively for scheduling functions and to update and
enhance teaching resources. Results of a survey sent to all
Fig. 4. Faculty response to the question: “I believe in the concept of an
Electronic Curriculum for the Faculty of Medicine at U of M”
 IMPLEMENTATION OF A CURRICULUM MANAGEMENT SYSTEM 93

teaching increased and the duration of the course could be

TABLE 2
Results of an independent survey conducted by the 1st and 2nd year
students. The strengths and weaknesses of the electronic curriculum
system CIS) are listed here. The data is combined for 1st and 2nd year class.

Fig. 5. Faculty response to the question: “I find it easy to log-on, navigate

and upload content on CIS”

avoided and mandatory hands-on training sessions will likely

enhance the user experience and overall success of the
electronic curriculum.
VI. DISCUSSION
Many medical schools in North America and Europe have
published their experience with deployment of electronic
curricula for either a single course or the whole curriculum.
We reviewed their experience and related it to the lessons that
we learned during the pilot project. Our experience showed
that gradual, controlled and systematic (Change Management)
rather than outright adoption is preferred as users develop
proficiency and familiarity with the on-line system.
The University of British Columbia Faculties of Medicine
and Dentistry have developed a MEDICOL (Medicine and
Dentistry Integrated Curriculum Online) site for medical
students to use during their rural family practice, a four- to six-
week experience in the summer after the second year [22]. The
system uses WebCT, tracks student progress, promotes
interactions through e-mail and bulletin boards, and delivers
self-directed learning components and multimedia learning
modules. The use of statistics indicates that over 90% of
students regularly used the MEDICOL sites and have found
them helpful.
A case study to "electrify the curriculum" at Virginia
Commonwealth University School of Medicine chose a
second-year course to serve as a model e-course and to
provide evaluation data for a two-year study [23]. E-mail reduced by half. These publications present compelling
surveys, focus groups, student diaries, and comprehensive evidence that electronic curricula and systems are viable and
end-of-course student assessments showed that students found can be successfully used to deliver medical education.
access to multiple images, interactivity, and meaningful and In spite of the enthusiasm and great potential for e-
efficient navigation within the site to be useful tools to curriculum and curriculum management systems, insufficient
enhance complex learning. Gotthardt et al. recently described literature exists to assist medical schools with implementation
a strategy of successfully implementing an e-learning strategies. Authors of one study evaluated mechanisms that
curriculum in the Netherlands [24]. The curriculum consisted may assist with implementation of an e-curriculum to enhance
of self-directed learning, an online discussion forum, and dental education [25]. Electronic Curriculum Implementation
discussion rounds. The e-content in nuclear medicine and Survey (ECIS), a 26 item questionnaire was mailed to all
radiotherapy was produced by a team of medical authors, web sixty-six North American dental schools (ten Canadian and
designers, and psychologists, and was delivered via an online fifty-six U.S. schools) during 2002-03 with a response rate of
learning management system. Online evaluation proved that 100 percent. The survey found that e-curriculum
the new curriculum was very successful, as the efficacy of implementation among North American dental schools
94 PAUL ET AL.

followed the classic innovation pattern. A few early adopting and has been successful partly because of a substantial private
institutions proceed rapidly while the majority of potential funding grant and strong leadership.
adopters make modifications slowly. The implementation
barriers identified by the survey were: lack of time, skill, and VII. CONCLUSION
incentive to develop the educational software. Another The vision of electronic learning is to foster lifelong
educational study by Riley at al. evaluated the effectiveness of learning for future physicians beginning with robust
the web-based teaching model as applied to perfusion undergraduate medical education that emphasizes strong self-
education [26]. Web-based virtual classroom (VC) learning skills and academic scholarship [30] [32]. E-learning
environments and educational management system (EMS) has been shown to be efficient, effective and has received
software were implemented independently and in addition to acceptance within the medical education community. Its
the live, interactive Internet-based audio and video success is only possible via an integrated, user-friendly,
transmission. Use of Internet-based VC and EMS software versatile and highly functional CMS. Such systems are
was a significant improvement over traditional models. currently in development and not widely available.
Interestingly, students recognized the learning efficiency of Considerable effort and resources are required for individual
on-line information but still preferred the traditional face-to- institutions to customize and implement a CMS. Our
face classroom environment. A Spanish study delivered and experience with a pilot project using an electronic curriculum
evaluated an educational web based cardiovascular course management system (CIS®) established a significant need for
prepared as adjunct educational material for the medical such a system, demonstrated that implementation is feasible,
student class [27]. Frequency of use, impact on in-class and found that adoption by faculty and students can be
attendance and student satisfaction, number of visits and expected. Successful implementation of e-learning in medical
quality of the website were assessed. The quality score rating education mandates a thorough consultation, effective
was a mean of 7.7 (scale: 1-10); 93.4% students attended the communication and engagement of all users.
class, and 88.2% of the students evaluated the web as a useful
or very useful addition to medical teaching. VIII. ACKNOWLEDGMENT
Kerfoot et al. conducted a multi-institutional randomized We would like to acknowledge and thank class of 2011
controlled trial to investigate the impact of an adjuvant web- students, Blocks III and VI instructors and administrative staff
based teaching program on medical student learning during of Undergraduate Medical Education Office (Pat McCullough,
clinical rotations [28]. Web-based teaching cases were Pre-Clerkship Program Administrator; Mark Torchia, Med I
developed covering four core urologic topics. Students were Coordinator; and Vieanna Beck, Administrative Assistant) at
block-randomized to receive web-based teaching on two of the the Faculty of Medicine, University of Manitoba for their
four topics. Before and after a designated duration at each superb enthusiasm and terrific support.
institution, students completed a validated 28-item web-based
test covering all four topics. Compared to controls, web-based REFERENCES:
teaching significantly increased test scores in the four topics at [1] Effective use of educational technology in medical education. AAMC
each medical school (p <0.001, mixed analysis of variance), 2007. https://aamc.org/ Accessed on August 8, 2008.
corresponding to an effect size of 1.52 (95% confidence [2] D. Sandham, A. Paul, S. Sharma. Integration of e-curriculum into
medical education. Gravitas 2008; 41:4-5.
interval [CI], 1.23-1.80). Learning efficiency increased three-
[3] Educating doctors to provide high quality medical care - a vision for
fold by web-based teaching (effect size 1.16; 95% CI 1.13- medical education in the United States. AAMC 2004. https://aamc.org/
1.19). Furthermore, students who were tested a median of 4.8 Accessed on August 8, 2008.
months later and those who received web-based teaching [4] B. Barzansky, S.I. Etzel. Educational programs in US medical schools,
demonstrated significantly higher scores than those who 2002-2003.
received non-web-based teaching (p = .007, paired t-test). This [5] H.B. Slotnick. How doctors learn: education and leaning across the
randomized controlled trial provides convincing evidence that medical-school-to- practice trajectory. Acad med 2001; 76: 1013-26.
web-based teaching can significantly and durably improve [6] J.P. Ward, J. Gordon, M.J. Field, H.P. Lehmann. Communication and
information technology in medical education. The Lancet. 2001;
medical student learning. 357:792–796.
Our experience developing and implementing an e- [7] P. Devitt, E. Palmer. Computer-aided learning: an overvalued
curriculum at U of M is similar to the journey undertaken by educational resource? Med Edu 1999; 33:136-139.
McGill University [2] [29]. David Fleiszer and Nancy Posel [8] T. Greenhalgh. Computer-assisted learning in undergraduate medical
were the leaders in development of an online, multimedia- education. BMJ. 2001; 322:40–45.
enhanced undergraduate medical curriculum. The electronic [9] C.L. Friend, C.L. Cole. Learner control in computer-based instruction: a
current literature review. Ed Tech. 1990; November: 47–40.
curriculum aimed to include: (1) learner-centered, self-
[10] D.J. Steele, J.E.J. Palensky, T.G. Lynch, N.L. Lacy, S.W. Duffy.
directed, and guided learning; (2) inherent interactivity of the Learning preferences, computer attitudes, and student evaluation of
technology; (3) a powerful means for integrated learning; (4) computerized instruction. Med Educ. 2002; 36:225–232.
ability to address a variety of learning styles experienced by [11] C.J. Bland, S. Starnaman, L.Wersal, L. Moorhead-Rosenberg, et al.
the students; and (5) opportunities for horizontal and vertical Curricular change in medical schools: how to succeed. Acad Med 2000;
curricular integration. The experience at McGill is ongoing 75:575–94.
[12] M. Vogel, D.F. Wood. Love it or hate it? Medical students’ attitudes to
computer-aided learning. Med Educ. 2002; 36:214–215.
 IMPLEMENTATION OF A CURRICULUM MANAGEMENT SYSTEM 95

[13] M.L. Rice, E.K. Wilson, W. Bagley. Transforming learning with [23] J.D. Hoban, J.B. Schlesinger, R.P. Fairman, M.M. Grimes. Electrifying a
technology: lessons from the field. J Tech Teacher Educ 2001; 9(2):211– medical school course: a case study. Teach Learn Med. 2003
30. Spring;15(2):140-6.
[14] B.M. Karnath, W. Thornton, M. DasCarlos. Pilot study of computer- [24] M. Gotthardt, M.J. Siegert, A. Schlieck, S. Schneider, A. Kohnert, M.W.
based self-teaching in cardiac auscultation. Med Educ. 2003; 37:1048– Gross, et al. How to successfully implement e-learning for both students
1049. and teachers. Acad Radiol. 2006 Mar; 13(3):379-90.
[15] A.J. Lipman, R.M. Sade, A.C. Glotzbach, C.J. Lancaster, M.F. Marshell. [25] W.D. Hendricson, F. Panagakos, E. Eisenberg, J. McDonald, G. Guest,
The incremental value of internet-based instruction as an adjunct to P. Jones, et al. Electronic curriculum implementation at North American
clasroom instruction: a propsective randomized study. Acad Med. 2001; dental schools. J Dent Educ. 2004 Oct; 68(10):1041-57.
76:1060–1064. [26] J.B. Riley, J.W. Austin, D.W. Holt, B.E. Searles, E.M. Darling. Internet-
[16] B.L. Ingerman. Form follows function: establishing the necessary based virtual classroom and educational management software enhance
infrastructure. In: Barone CA, Hagner PR, eds. Technology-enhanced students’ didactic and clinical experiences in perfusion education
teaching and learning. San Francisco: Jossey-Bass, 2001. programs. J Extra Corpor Technol. 2004 Sep; 36(3):235-9.
[17] J. Daniel. Computer-Aided Instruction on the World Wide Web: The [27] X. Gómez-Arbonés, A. Ferreira, M. Piqué, J. Roca, J. Tomás, J.L.
Third Generation. J of Econ Ed. 1999. pp. 163–174. Frutos, et al. A cardiological web as an adjunct to medical teaching:
[18] S. Chodorow. Educators must take the electronic revolution seriously. prospective analysis. Medical Teacher 2004; 26:187-191.
Acad Med. 1996; 71:221–226. [28] B.P. Kerfoot, H. Baker, T.L. Jackson, W.C. Hulbert, D.D. Federman,
[19] H.K. Muhsin. The use of e-learning techniques to improve the medical R.D. Oates, et al. A multi-institutional randomized controlled trial of
education. ICTTA 2008. http://ieexplore.ieee.org/ Accessed on August adjuvant web-based teaching to medical students. Acad Med. 2006 Mar;
2008. 81(3):224-30.
[20] R.M. Palloff, K. Pratt. Lessons from the cyberspace classroom: the [29] D.M. Fleiszer, N.H. Posel. Development of an undergraduate medical
realities of online teaching. San Francisco: Jossey-Bass, 2001. curriculum: the McGill experience. Acad Med. 2003 Mar; 78(3):265-9.
[21] H.S. Wentzer. Technology development and implementation in the [30] S. Schultze-Mosgau, T. Zielinski, J. Lochner. Web-based, virtual course
public health institutions and pedagogical task for the learning units as a didactic concept for medical teaching. Med Teach. 2004 Jun;
organization. Stud Health Tech Informatics 2000; 77:264–70. 26(4):336-42.
[22] M. Broudo, C. Walsh. MEDICOL: online learning in medicine and [31] J.G. Ruiz, M.J. Mintzer, R.M. Leipzig. The impact of e-learning in
dentistry. Acad Med. 2002 Sep;77(9):926-7. medical education. Acad Med 2006: 81:207-212.
[32] K. Masters, T. Gibbs. The spiral curriculum: implications for online
learning. BMC Med Educ. 2007 Dec 21; 7:52-55.
 Use of Ontology for Reusing Web Repositories
for eLearning
Punam Bedi, Hema Banati, Anjali Thukral

Abstract— eLearning scenario has witnessed a shift from framework that provides latest updated learning material,
instructional to adaptive learning. The focus is now moving from semantically searched from web repositories for the entered
closed to open scenario. However, learning in such situation query.
relegates to the task of retrieving as the information is largely
scattered across the web. Searching such voluminous information The paper is organized as follows: Section II presents the
is practically infeasible and leads to incomplete or irrelevant background work, briefly defining eLearning, advantages of
results. Hence, a need is felt to automatically integrate the vast Semantic Web (SW) for it and ontology. Section III describes
pool of information present in varying forms on web. This paper the framework and algorithm for reusing web repositories for
proposes a framework that searches for information semantically eLearning. Finally section IV discuss future work and
across all existing web repositories for a given query. The concludes the paper.
approach utilizes WSD (word sense disambiguation) algorithm to
semantically analyze and expand the query. A personalized
II. BACKGROUND WORK
knowledge base for learning purpose is subsequently created by
deploying agents to compile information from various learning eLearning is defined with eight dimensional characteristics
resources. The integrated organization of knowledge provides a as: Pull: Learner determines agenda (delivery), Reactionary:
single point reference for the posted query. It thus increases the Responds to problem at hand (responsiveness), Non-linear:
probability of the learner obtaining complete and relevant Allows direct access to knowledge in whatever sequence
information from a single reference. This framework therefore makes sense to the situation at hand (access), Symmetric:
allows a learner to concentrate on learning than the searching Learning occurs as an integrated activity (symmetry),
process thus enhancing the learning experience. Continuous: Learning runs in parallel to business tasks and
never stops (modality), Distributed: Content comes from the
Keywords-eLearning; eLearning; web repositories; query interaction of the participants and the educators (authority),
expansion; software agents; personalized ontology. Personalized: Content is determined by the individual user’s
needs and aims to satisfy the needs of every user
I. INTRODUCTION (personalization) and Dynamic: Content changes constantly
through user input, experiences, new practices, business rules
In the past practice traditional systems like LMS, LCMS [1, and heuristics (adaptivity) [6,7]). eLearning characteristics can
2] and other intelligent systems like AWBES [3] have played effectively be achieved through the new WWW architecture,
an important role in the evolution of eLearning [4]. These Semantic Web [8] (SW). Research [6, 7, 9, 10, 11, and 12]
systems emphasis on providing an integrated environment for a shows the benefits of using semantic web as the platform for
learner, but within a separate and closed scenario, viz within an eLearning. Conceptually the SW represents a layered
educational organization or within some paid association. This architecture (Fig 1), where each level provides different degree
restriction significantly narrows the educational inputs, of expressiveness. The bottom most layer allow user to provide
depriving a learner from vast pool of information thus a controlled vocabulary and namespaces. Next layer, Resource
precluding from the efficiency of learning [5]. Learning in open Description Framework (RDF) represents information about
scenario is also long-winded. The presence of voluminous and resources in the World Wide Web. It consists of a triplet –
scattered information in varying formats deviates and hinders resource, property and value. Ontology vocabulary, the next
the learning process of a learner. A need is hence felt for a upper layer, adds more semantics to RDF by allowing a better
system that can automatically and dynamically organize the specification of constraints on concepts. The rules for explicit
information from all possible resources to a well formed inference are established by Logic layer which are then
structure. This structured learning material helps a learner for a executed by the Proof layer. The top most layer, Trust provide
better understanding of concept. This paper, therefore mechanisms to imbibe trust in a given proof [13]. The
addresses above mentioned need and proposes an eLearning semantic web as whole, acts as a decentralized knowledge bank
that provides meaning to a learner's search and thus allows
Manuscript submission date: 26th October 2009.
extracting most appropriate learning material.
Punam Bedi,( [email protected]), Hema Banati,([email protected])
Anjali Thukral ([email protected])
Dept. of Computer Science, University of Delhi, India

97
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_17, © Springer Science+Business Media B.V. 2010
98 BEDI ET AL.

Trust
a large scale [18] to facilitate a vast group of learners.
Standards were established to generalize the structure for such
Proof
metadata as Dublin Core [19] and LOM (Learning Object

Digital Signature
Logic Metadata) [20]. However, there were issues in using LOM.
Ontology Vocabulary They represent static repositories and suffer from various
shortcomings [21]. The ontology based metadata provided a
RDF + rdfschema
perfect blend of eLearning systems and ITS (Intelligent
XML + NS + xmlschema Tutoring Systems) [22] in closed scenario, overcoming most of
Unicode URI the earlier shortcomings. Subsequently, the ontology layer was
identified as a standard layer of the knowledge based systems
[23]. Existing work of open learning systems either provides
Figure 1: Layered Architecture of Semantic Web links to the World Wide Web documents, based on refined
query searches or exclusively build on semantic web. Semantic
The middle layer and the backbone of Semantic web, web conveys meaning to the content whereas existing WWW
Ontology Layer, forms knowledge about a certain domain by has evolved as a powerful information portal.
providing relevant concepts and relations between them. Often
cited definition of ontology is “an explicit specification of a
conceptualization” [14] i.e., an ontology defines (specifies) the III. FRAMEWORK
concepts, relationships, and other distinctions that are relevant This paper proposes an open learning framework that
for modeling a domain. The specification takes a form of provides the leverage of both worlds. Unlike other existing
representational vocabulary (classes, relations, and so forth) systems, it dynamically integrates all available resources on
which provides meaning for the vocabulary and formal web to construct ontologies. This integration is established on
constraints on its coherent use [15]. the posted query. Words contained in the query are
Mathematically, a core Ontology is defined as a structure, [16] semantically analyzed for sense based query expansion.
Expanding query in this way improves the probability of
O := (C,≤C , R,,≤ R ) retrieving relevant information drastically [24]. Thus, the
system first formulates the query and then integrates resources
consisting of: to form an exhaustive and updated ontology.
i) two disjoint sets C and R whose elements are called This framework is based on three types of ontologies:
concept identifiers and relation identifiers, respectively.
• NLO (Natural Language Ontology): It contains lexical
ii) a partial order ≤ C on C , called concept hierarchy or relations between the language concepts. They are
taxonomy, large in size and do not require frequent updates. They
iii)a function  : R → C called signature, and
+ represent the background knowledge of the system to
understand the query formed in the natural language.
iv)a partial order ≤ R on R , called relation hierarchy, where
The framework uses NLO, WordNet to expand the
r1 ≤ R r2 implies | (r1 )| = | (r2 )| and original query.
 i ( (r1 )) ≤ C  i ( (r2 )) for each 1 ≤ i ≤ | (r1 )| • DO (Domain Ontology): It captures knowledge of one
particular domain, e.g., operating system, computer
and Knowledge Base is defined as a structure [16]: etc. These ontologies provide detailed description of
KB := (C KB , RKB , I,C ,R ) the domain concepts. In this framework it belongs to
the horizontal upper layer of the two level ontology
consisting of: [25] structures. (Fig 2).
• OI (Ontology Instances): They are the main resource
i) two sets C KB and RKB , of knowledge presented in the web and on the learner's
ii) a set I whose elements are called instance identifiers (or local machine. These instances consist of interlinked
instances or objects for short), web resources (includes web pages etc.) forming a
iii)a function C : C KB →  (I ) called concept instantiation, horizontal hierarchy (lower layer of two level
organization - Fig 2) among themselves and also
( )
iv)a function R : RKB →  I called relation instantiation.
+
vertically linked to their related concepts in the
Ontologies have the potential to construct a dynamic and domain ontology. These are generated automatically
semantic structure of e-content that can be shared, reused and and updated frequently by software agents. DO and OI
enhanced with trust for learning purposes. Since the invention forms the local knowledge base.
of semantic web in 2000, many researchers have proposed The framework (illustrated in Fig: 2) initializes the process by
various architectures and frameworks that use this technology expanding query through query engine. The expanded query is
to enhance the learning experience of learners [17]. Most of then used to extract contents from web repositories. This
these frameworks initially started with the use of metadata. retrieved information is ontologically organized to form
Efforts were made to convert eLearning courses to metadata on
 USE OF ONTOLOGY FOR REUSING WEB REPOSITORIES FOR ELEARNING 99

Multimedia
Relational Ontology files
databases

Human
Text (doc,
s/w agents expert
pdf, html)

IGA Natural Language

Concept
Ontology

Query Engine
Image Query (Qo)
file Formalized
Audio query (Qf)
file Movie
PA Learner
OA Domain Ontology,
Two level organization of domain
ontology and ontology instances Ontology Instances

Compiled
learning

Synthesizer
material
Query based
ontology extraction

Figure 2: Framework- reusing web repositories for eLearning using ontology

knowledge base. The output is subsequently re-framed to a where C x is the best possible configuration of senses for Qo .
presentational format by the Synthesizer.
One of the possible expansions for the above query can be seen
Remaining section explains the process of query expansion as:
and content extraction followed by algorithmic summarization
of the complete process of information retrieval. Q f = {medium, Ethernet, coaxial cable, radio frequency
A. Query expansion signal, wireless local area network}.
The query engine (ref Fig: 2) lexically extract words from The words, medium and Ethernet have been semantically
the query for the expansion. As an example, let the query be: analyzed to deduce new words that are used as medium for
“medium used in Ethernet”. It generates a set of words from Ethernet. Though, the words like co-axial cable etc. does not
original query as, exist in the original query, still the learner benefits by obtaining
the intended information related to these supplementary words.
Qo = {medium, Ethernet}
In general, B. Content extraction from web repositories
Web repositories contain learning material in varying
Qo = { w1 , w2 ,..., wk } formats. IGA (Information Gathering Agent) manages the
searching process to gather information from all possible
where, wk are the concepts extracted from the entered query. resources. This information is organized semantically by OA
The query engine formalizes the query using a mapping (Ontology Agent) to update local personalized knowledge base.
OA deals with each source format using specific procedure to
function,  . The function uses WordNet 1 and Word Sense give information an ontological structure. Following are the
Disambiguation (WSD) algorithm [24] to expand the query. different formats handled by OA (ref. Fig. 2).
The expanded query consists of semantically analyzed set of
words that results from the maximum number of intersecting i) Text documents: The automatic generation of ontology from
nodes in the semantic networks, generated from WordNet text (ontology population) [26] is achieved with KAON
(Karlsruhe ontology) based tool TextToOnto [16]. The
synonym sets (synsets) of wk , k = 1, .. , n. Thus, resulting in a agent inputs text as a corpus and generates ontology using
formulated query, this tool. Generated ontology is later pruned or enhanced
through the domain expert, if needed. The automatically
Q f = { w1 , w2 ,..., wk , C x } generated ontology does not produce satisfactory results,
but nevertheless provide some basic structure of concept
hierarchy. Generation of quality ontologies without the
human intervention is still a topic of research.
1 http://wordnet.princeton.edu/
100 BEDI ET AL.

ii) Multimedia files: Ontology instances are used to organize  : Qo → Q f is formed where,  is a mapping from
multimedia files (audio files, video files, presentations,
images etc.). These files are arranged in ontological form the original query, Qo to a formalized query,
and at the same time linked to its related concept in Domain
Ontology (DO). Q f consisting of a set of semantically expanded words
iii)Human experts: Human experts' knowledge can be (explained in section III/A).
communicated through e-mail, blogs of FAQs (Frequently c) PA then, submits Q f , a formalized query to Ontology
Asked Questions), or chatting. This is achieved by linking
useful urls in the ontology instances and then linking them Agent, OA.
to related concepts in DO. Software agents [27] can be iii)OA performs two tasks simultaneously:
trained to take care of such conversations on behalf of the a) First, it searches the most relevant ontology in its local
passive learner. knowledge base consisting of domain ontologies and
iv)Ontology: Reusing existing ontologies offer a much ontology instances.
cheaper alternative than building new ones from scratch b) Secondly, it instructs Information Gathering Agent,
[28]. It provides one of the valuable repositories on web IGA to explore learning material in the World Wide
and reduces efforts and time to retrieve relevant content. Web and semantic web using Q f (explained in section
However, the problem of determining relevance of the III/B).
ontologies is not much different from finding a relevant c) IGA returns to OA with:
searched document. A brilliant approach for automatic
ontology construction from the existing ontologies on the 1 ) a set of top n ranked links (URLs), Ln after
web was proposed [28]. It extracts the relevant parts of the searching WWW and
top ranked (ranked according to some criteria) ontologies 2 ) top m ranked ontologies, Om after searching
and merge those parts to acquire the richest domain
semantic web.
ontology. The ontologies on the web are searched using the
iv)Steps i) and ii) are repeated by PA till the learner is active
semantic web search engines like Swoogle or Ontosearch.
or wishes to update its knowledge base passively.
This approach requires further exploration of areas such as
Consequently, PA keeps the learner updated by
ontology ranking, segmentation, mapping, merging and
communicating with OA, and OA subsequently keep its
evaluation.
knowledge base updated by frequently communicating with
v) Software Agents: Software agents are used by a system for
IGA.
personalized search and interaction with almost each form
of web repository. As the future vision, one may find s/w v) OA update its knowledge base, KB with Ln and Om to
agents on the web, similar to the human learner. It will be a finally generate the result.
normal scenario to have web full of s/w agents vi)Result is provided to the learner in one or both of the two
communicating to each other and negotiating for accessing forms:
relevant content for its owner. The challenge is to form a a) a set of ontologies ranked according to the relevance,
force of such good behaviored servants on web! b) top ranked ontology, forwarded to the 'Synthesizer', in
vi)Relational databases: Web repository also exists in the order to generate a readable document with all its
form of database. Majority of data on old systems consist of instances as links for the learner.
databases. Such resources of eLearning cannot be The above algorithm represents a complete process of
neglected. This framework maps the relational schemata of reusing information by extracting only relevant content and
eLearning databases found on web, to ontology according providing single point reference to the compiled information
to their underlying intrinsic relationships [29]. Tables, for the learner. By using this algorithm, the framework creates
Fields and records are mapped to the ontology as class an open scenario for learning, where a learner automatically
(concept), property and instances. They are further adjusted gets the relevant and dynamically updated information related
to get accommodated in the structure of proposed ontology. to the required concept without getting deviated from the flow
All of the retrieved information from web repositories by of learning. This allows a learner to concentrate more on the
above methods is organized in two level ontological structures concept understanding instead searching for the relevant
to form a learner's personalized knowledge base. This content, which effectively enhances the learning experience.
knowledge base is then used further to answer queries related
to same concept. The following algorithm generates a IV. CONCLUSION AND FUTURE WORK
personalized knowledge base for a given query.
A framework has been proposed that reduces the chances of
i) A learner enters a query in natural language, Qo . loosing out relevant content available on web significantly to
ii) A learner's personalized interface agent (PA) explores create a personalized ontology for learning purpose with the
query entered by learner to infer the context and semantic. additional vision of agent communication. It takes a query
a) Through query engine, Qo query is formalized by PA. from learner, semantically expands it and gathers content from
web repositories with the assistance of multiple software
b) The query engine extracts most appropriate words by agents. These agents work for the active learner as well as
applying Natural Language Ontology (WordNet) and
searches content for the passive learner. Moreover, a
WSD algorithm [24]. Thus a mapping,
personalized agent is provided to assist the learner at various
 USE OF ONTOLOGY FOR REUSING WEB REPOSITORIES FOR ELEARNING 101

learning points. At present, the framework has been partially [14] Gruber, T. R., “A Translation Approach to Portable Ontology
Specifications”, Knowledge Acquisition, 5(2):199-220, 1993.
implemented that uses JADE and a tool TextToOnto for
[15] Thomas Gruber, “Ontology (Computer Science) - definition in
extracting text and transforming it to ontology automatically. Encyclopedia of Database Systems”, Ling Liu, Tamer M Özsu (Ed),
In future, other modules will be implemented and explored Encyclopedia of Database System (2008)
further to provide a more relevant ontology generation for the [16] Erol Bozsak, Marc Ehrig, Siegfried Handschuh, Andreas Hotho,
learner's knowledge base. A more formalized agent Alexander Maedche, Boris Motik, Daniel Oberle, Christoph Schmitz,
communication on the web and setting up a descent behavior Steffen Staab, Ljiljana Stojanovic, Nenad Stojanovic, Rudi Studer, Gerd
Stumme, York Sure, Julien Tane, Raphael Volz, Valentin Zacharias,
will also be considered. “KAON – Towards a Large Scale Semantic Web”. Proc. of the 3rd Intl.
Conf. on E-Commerce and Web Technologies (EC-Web 2002), pages
REFERENCES: 304–313, 2002.
[1] S.R. Robbins, “The evolution of the Learning Content Management [17] Janez Brank, Marko Grobelnik, Dunja Mladeni , “A Survey Of
System”, ASTD's Source for E-learning, 2002. Ontology Evaluation Techniques”, In Proceedings of the Conference on
[2] Brandon Hall Research Article, 2007 [Online]: hhtp://www.brandon- Data Mining and Data Warehouses (SiKDD) 2005
hall.com/ [18] Jan Brase and Wolfgang Nejdl, “Ontologies and Metadata for
[3] Peter Brusilovsky, “KnowledgeTree: A Distributed Architecture for eLearning”, pages 579–598. Springer Verlag, 2003.
Adaptive E-Learning”, ACM 1-58113-912-8/04/0005, May 17-22, [19] The Dublin Core Metadata Initiative, http://dublincore.org
New York, USA. [20] Learning Technology Standards committee of the IEEE: Draft Standard
[4] Anjali Thukral, Richa Sharma, Hema Banati, Punam Bedi, , “Modeling for Learning Object Metadata, July 2002, IEEE 1484.12.1-2002,
Roles in E-learning: MoRel”, ACVIT'07 - International Conference on http://ltsc.ieee.org/wg12/files/LOM_1484_12_1_v1_Final_Draft.pdf
Advances in Computer Vision and Information Technology,612-620, [21] Amal Zouaq and Roger Nkambou, “Building Domain Ontologies from
Aurangabad, November 28-30, 2007. Text for Educational Purposes”, IEEE Transactions on Learning
[5] Michael G, Dimitry V.,Pavel, Denis, “Intelligent Educational Technologies, vol. 1, no.1, January-March 2008
Environments Based on the Semantic Web Technologies”, Proceeding, [22] C. Brooks, J. Greer, E. Melis, and C. Ullrich, “Combining ITS and
IEEE International Conference on Artificial Intelligent Systems eLearning Technologies: Opportunities and Challenges,” Proc. Eighth
(ICAI'02) 0-7695-1733-1/02, 2002. Int’l Conf. Intelligent Tutoring Systems (ITS ’06), pp. 278-287, 2006.
[6] Drucker, P. ,“Need to Know - Integrating e-Learning with High Velocity [23] Neches, R., Fikes, R. E., Finin, T., Gruber, T. R., Patil, R., Senator, T.,
Value Chains”. Delphi Group White Paper. 2000 www.delphigroup.com & Swartout, W. R. “Enabling technology for knowledge sharing”. AI
[7] Yas A. Alsultanny, “eLearning System Overview Based on Semantic Magazine, 12(3):16-36, 1991.
Web”, The Electronic journal of e-Learning, Volume 4 Issue 2 (111- [24] Navigli, Roberto and Velardi, Paola, “An Analysis of Ontology-based
118) 2000. Query Expansion Strategies”, In: Proceedings of Workshop on Adaptive
[8] T. Berners-Lee, J. Hendler, O. Lassila , “The Semantic Web”, Scientific Text Extraction and Mining at the 14th European Conference on
American, May 2001, pp. 28-37. Machine Learning, 2003.
[9] Ljiljana Stojanovic, Steffen Staab, Rudi Studer, “eLearning based on the [25] Antonio De Nicola, Michele Missikoff, and Frederica Schiappelli,
semantic web”, WebNet2001 - World Conference on the WWW and “Towards an Ontological Support for eLearning Courses”, R. Meersman
Internet. et al. (Eds.): OTM Workshops 2004, LNCS 3292, pp. 773–777, 2004.
[10] Rosic, M.; Stankov, S.; Glavinic, V., “Application of semantic web and [26] Alexander Maedche and Steffen Staab, “Ontology Learning for the
personal agents in distance education system”, Electrotechnical Semantic Web”, IEEE Intelligent Systems and Their Applications, Vol.
Conference, MELECON 2002. 11th Mediterranean,Volume , Issue , 16, No. 2. (2001), pp. 72-79, 2001.
2002 Page(s): 542 – 546, 2002. [27] James Hendler, "Agents and the Semantic Web," IEEE Intelligent
[11] Hatem, Muna S.,Ramadan, Haider A.,Neagu, Daniel C., “e-Learning Systems, vol. 16, no. 2, pp. 30-37, Mar./Apr. 2001,
based on context oriented Semantic Web”, Journal of Computer Science doi:10.1109/5254.920597, 2001.
1(4):500-504,2005. [28] Harith Alani, “Position Paper: Ontology Construction from Online
[12] Sylvain Dehors, Catherine Faron-Zucker, “Reusing Learning Resources Ontologies”, WWW 2006, May 23–26, 2006, Edinburgh, Scotland.
based on Semantic Web Technologies”, ICALT 2006: 859-863. ACM 1-59593-323-9/06/0005, 2006.
[13] Bittencourt I.I., Costa, E., Soares, E. and Pedro, A., “Towards a New [29] Zhaohui Wu, Yuxin Mao, and Huajun Chen, “Subontology-Based
Generation of Web-based Educational Systems: The Convergence Resource Management for Web-Based e-Learning”, IEEE Transactions
between Artificial and Human Agents”, IEEE multidisciplinary on Knowledge and Data Engineering, vol. 21, no. 6, June 2009.
education Magazine, vol. 3, March 2008.
 Actual and Future Trends of the E-education
and Its Costs Specification in the Czech Republic
M. Hedvicakova, P. Maresova
Department of Economics
Faculty of Informatics and Management, University of Hradec Kralove
Rokitanskeho 62, 500 03 Hradec Kralove, Czech Republic

Abstract – The contribution is focused on the specifics of distance We cannot forget the school system, where a great majority of
education in the Czech Republic. First, reasons of emergence of both universities and high schools, and to a lesser extent also
distance education in the Czech Republic and abroad are basic schools, takes advantage of various education and
analyzed. Further, two possible directions of development of e- instruction software.
education in the Czech Republic are outlined, where the first
direction represents an expansion of capacities of the existing
full-time courses. The second possible direction consists of
customization of educational courses according to specific needs II. FORMS OF STUDIES IN UNIVERSITIES
of students. The contribution includes definitions of individual IN CZECH REPUBLIC
participants of the educational process, typology of costs and
revenues and break-even point modeling. Currently three ways can be applied in the process of
instruction: present lessons, i.e. traditional, face-to-face
teaching and learning; distance education supported by ICT,
I. INTRODUCTION e.g. e-learning, virtual learning environment; and blended
learning, i.e. combination of present and distance elements.
With the arrival of new technologies and varied demands in
In the present form of instruction in economic subjects
society, universities must continuously seek new ways to
teacher faces two main problems: students´motivation and
combine advantages of modern technologies with
interest, and the level of knowledge. The former one can be
requirements of target customer groups. Requirements can be
strongly influenced by the teacher, his/her personality, image,
for example: demographic, locality, time or costs etc. Virtual
professional qualities, computer literacy, etc. And it results
universities offer the way to satisfy these requirements while
into ways how to solve the latter problem. Numerous suitable
continuing to multiply advantages of information and
examples, applications from real life, new technologies
communication technologies and thus they ensure a greater
supporting “boring“ calculations, and last but not least
return of realised investments and thereby also the growth of
didactic aspects – these are the means which can help. If the
a university as such.
teacher is able to use appropriate methods and forms to
There are already many virtual universities in the world.
explain the topic, present examples, practice them, motivate
Especially US universities have become pioneers of this type
students to follow-up activities, organise team and project
of education for the reason of their strong financial position
work – all the mentioned aspects may support the required
and a high level of technological equipment. Also e-education
features mentioned above. Implementation of new
in Australia has broken through with success. Interest in
technologies in the process of instruction calls for both
distance programs has been growing in these continents for
special equipment of the classroom and for the teacher who is
the reason of high commuting distance to reach education.
able to work with them in an appropriate way. Modern
State institutions lead by the European Union put a great
software programmes neither work nor bring increasing
accent on strengthening and changing of the approach to
interest and effectiveness by themselves. Only such a teacher,
education. This can be also taken advantage of by regional
who is a professional in both the subject field and
universities, which can flexibly respond to the demand for job
methodology, evokes the required activities from students.
positions in the labour market. They can either focus on
Having a PC, data projector etc. in the classroom, using
retraining courses for employment bureaus or to create
PowerPoint presentations, links to Internet sources, it is a
“customized” courses for commercial organizations.
common standard these days. Various types of software are
Supranational organizations and large companies have
able to present graphs and their animations, calculate values
already been using e-learning education methods for training
and structure them to tables, everything with the aim to make
of their employees. With regard to the number of employees,
the process of instruction easier for both students and
they have become aware of advantages of on-line courses and
teachers. Teacher’s ability to use these tools appropriately and
they try to use them as their competitive advantage.
support all activities by personal interest and knowledge,
Currently we can observe that e-learning education has
these are the qualities that count for much.
reached all areas, industry as well as energetic, financial
sector, healthcare, telephone market, state sector and army.

103
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_18, © Springer Science+Business Media B.V. 2010
104 HEDVICAKOVA AND MARESOVA

It can be said the same goes for the distance education, but III. TWO PATHS FOR UNIVERSITIES
twice more. As there is no face-to-face contact between the
teacher and students, the organisation is more demanding on One of the paths for universities is the above mentioned
both sides, teacher’s and students´. Although the distance distance studies as another form of education. E-learning can
form of education is applied for tens of years, its main be applied as an effective support of all forms of studies. The
development came in last few years because of immense advantage for the existing “stone universities“ is elimination
implementation of information and communication of space and staff inadequacy of today's education. In
technologies into all spheres of life. The adequate use of the consideration of the fact that there are no time or distance
technologies must be emphasized, just because of the absence barriers, interest in this form of studies has also been growing
of direct contact. The quality of all used tools and methods, among new groups of students (employed students,
including constant motivation, strongly influences students´ handicapped, women on maternity leave, prisoners and the
results and success. Professional teams of teachers, IT like). People have also been choosing this form of studies for
workers, economists and other specialists participate in the reason of interactivity of courses.
successful organisation and running the distance education. The advantage for the university itself lies in a strong
Blended learning combines the best of the above mentioned background, high-quality professional staff and tradition of a
forms, i.e. present lessons are supported by ICT, self-study university as an educational institution. Another benefit for
activities and sources provided by the Internet are frequently the university is reinforcement of its image as a holder of new
used, etc. [2] modern technologies and a wide spectre of courses.
As regards newly emerging virtual universities, the
Unlike other states with a larger population or larger universities which get involved in the project exchange the
distance barriers (USA, Australia etc.), where the tradition of subjects not offered by their faculty and thereby expand the
taking advantage of various education methods is greater, the circle of their students.
market with virtual education is still only developing in the The second path for universities is to create commercial e-
Czech Republic. learning courses. These are courses customized for individual
Only few Czech universities offer a compact study program companies or based on needs of the market or labour bureaus.
by means of the e-education method. These are for example A greater accent is put on economical aspects in these
the University of Economics, Prague, offering a follow-up courses. Another effect is interconnection of universities with
master's two-year program Economics and Management see needs of the given region and therefore again reinforcement
[3], or VŠB – Technical University of Ostrava, Faculty of of the image of university as a core of erudition.
Economics, offering the study field Quantitative Management
Support [4] in its follow-up master's studies. Combined
programs can be found in many universities including UHK. IV. FRAMEWORKS FOR COSTING
In our republic, the transition to this new type of education
has been stimulated above all by insufficient capacities of If we want to calculate costs of a course, first we have to
universities for full-time studies, economic conditions of define key types of costs.
students (studying while employed), pressure of the European
Union on the growth of the number of university educated 1. Categorisation of E-learning Costs
inhabitants etc. Another important motive is increasing In the process of comparing costs and revenues in the
attractiveness of studies by combining study programs of distance and traditional education, a clear definition of the
individual universities, modernity of the topic of information basic terms is necessary, i.e. what they consist of. Not the
society etc. whole virtual university is evaluated, but only one distance
Also university networks have been emerging educational course.
experimentally (e.g. project of Inter-university studies IUS at
the universities: University of Hradec Kralove, University of E-learning study costs
Western Bohemia in Pilsen and University of Tomas Bata in The standard structure for cost and revenue classification is
Zlin, Technical University of Ostrava) [5]. Or Erasmus used as it is in the common sphere of commerce, i.e. the
Virtual Economics & Management Studies Exchange variable and fixed costs will be counted.
(EVENE) – the international project EVENE is supported
from resources of the European Union within the program a) Fixed costs
eLearning and it provides students with the possibility to • Designing study materials – when creating an e-learning
study at foreign universities as well. [6] Nine international course, at the very beginning study materials must be
universities are involved in the project. prepared, including their electronic form in the e-
environment. This is a non-recurring action.
• Authors´ royalties for creating study materials in the text
form – authors must be evaluated in terms of finance.
• Printing study materials
• Costs of designing an e-course
 ACTUAL AND FUTURE TRENDS OF THE E-EDUCATION AND ITS COSTS SPECIFICATION 105

c) Total costs
The total costs cover variable and fixed costs.
TC = FC + VC

TC TC

VC

FC
Fig. 1: Fixed costs

Stair-step function fixed costs

The level of fixed costs depends on such criteria as a Q
number of students, frequency of tutorials, etc. It means stair- Fig. 4. Total costs
step function costs must be added to classical fixed costs, see
the following examples: d) Average costs
- Requirements for SW and HW infrastructure of the Furthermore, expenses per one student are calculated,
distance education depend on the number of competitive which is an aliquot part of the total costs, i.e. the total amount
working students. divided by number of students (N).
- Financial means necessary for tutorials depend on the AC = (FC + VC) / N
number of students and frequency of tutorials.
e) Average fixed and variable costs
FsC Average fixed and variable costs are counted in the same
way:
AFC = FC / N
AVC = VC / N

2. Profits
The traditional and distance course revenue varies and
consists of two items:
a)Contribution on each student (either from a state or
Q public institution supporting education).
Fig. 2. Stair-step fixed costs b) Contribution from each student (some
universities require full or partial fees from students).
b) Variable costs Total of these sums creates the income per student.
• Virtual environment user license
• Running an e-course – tutors and administrators of each
group are evaluated 3. The Break-Even Point
• Consumer material costs Finding the rate of return comes from a principle of
• Rental service costs paid for tutorial rooms comparing costs and profits and finding a number of students,
• Printing study materials who have to be in the course so that the course is profitable.
• Staff costs In economical formulation it is the break-even point. When
costs and profits are equal, the more students above the
minimal number we have, the higher profit we will receive.

To be able to find the point we must formulate cost and

profit functions.

4. The basic cost function

The cost function reflects variable and fixed costs.

Fixed Costs:
Fig. 3. Variable costs F1 - designing study materials
106 HEDVICAKOVA AND MARESOVA

F2 – authors royalties The process of providing certificates is expressed as a fixed

F3 – print item.
F4 – creating e-course F5 – certificate costs are added to the function.

Variable Costs: TC = (F1 + F2 + F3 + F4 + F5) + (Fs1 + Fs 2 +

V1 – virtual learning environment license
V2 - running an e-course Fs3 ) + (V1 + V2 + V3 + V4 + V5 + V6 + V7 ) ∗ N
V3 – consumer material The costs function for short course (3)
V4 – rent for tutorial rooms and office rooms
V5 – rental services for tutorial rooms (including HW, web One-term course (not longer than 6 months)
camera, data projector, flip board, internet…) If the course lasts 1 – 6 months, tutorials must be
V6 – indirect expense organised. The costs for 3 tutorials are included: an
V7 – staff costs introductory, middle, final one.
Staff costs consist of: Zlamalova [9], Rumble [12] and other authors emphasize
- Staff costs for designing modules the necessity of introductory tutorials for courses longer than
- Staff costs for creating supportive study materials one month.
- Staff costs for teaching the module (tutor’s salary) This item is a fixed cost and includes tutor’s salary, fare,
- Staff costs for instruction support (study department) rent for classrooms, SW and HW, refreshments etc., and F6 –
- Staff costs for management tutorial costs, are added, including F5 – Certificate.
- Staff costs for examination board, etc.
TC = (F1 + F2 + F3 + F4 + F5) + (Fs1 + Fs 2 +
N – Number of participants in a course
Fs3 ) + (V1 + V2 + V3 + V4 + V5 + V6 + V7 ) ∗ N
TC = (F1 + F2 + F3 + F4) + The costs function for one-term course (4)

(V1 + V2 + V3 + V4 + V5 + V6 + V7 ) * N Full study (Bachelor’s degree programme and Master’s

The basic costs function (1) degree programme)
The full study is the most important and represents the
Other elements could be the costs which were not most expenditures.
displayed here, or are still unknown, and marginal to the This type of study includes not only tutorials but also
educational institution. summer school (one week per 6 moths) to form those
demanded skills which cannot be formed via distance
The costs function taking into account step function fixed instruction. The frequency of tutorials and/or summer schools
costs is expressed by variable costs and the item “summer school”
Fixed costs are on the same level without consideration of is added to the costs. Its coefficient depends on the support of
the number of students. It changes by step. E.g. when the special aids, laboratories, time view etc. and results in:
number of students increases by 25 students, another
classroom for tutorials must be rented, another HW and new TC = (F1+ F2 + F3 + F5 + F6)
SW licenses bought.
Other costs are added to the cost function (1): + (Fs1+ Fs 2 + Fs3 )
Fs1 = step function fixed costs for tutorials + v ∗ (V1 + V2 + V3 + V4 + V5 + V6 + V7 )/ * N
Fs2 = step function fixed costs for SW
The Cost function for full study (5)
Fs3 = step function fixed costs for HW
It results in a new function:
Furthermore the costs covering the final phase of study, i.e.
TC = (F1 + F2 + F3 + F4) + (Fs1 + Fs 2 + Fs 3 ) final exam and certificate ceremony, are added. The function
+ (V1 + V2 + V3 + V4 + V5 + V6 + V7 ) ∗ N includes F7 function – final study fixed costs.
The new costs function (2)
TC = (F1+ F2 + F3 + F5 + F6 + F7)
The cost function in a time-period + (Fs1+ Fs 2 + Fs3 )
The point of time has to be taken into account in the
situation. The costs increase in relation to the time-period. + v ∗ (V1 + V2 + V3 + V4 + V5 + V6 + V7 )/ * N
Three time periods are used for displaying the costs. The Cost function for full study (6)

Short course (not longer than 14 days)

In a short course no tutorials are taken into account.
 ACTUAL AND FUTURE TRENDS OF THE E-EDUCATION AND ITS COSTS SPECIFICATION 107

5. The revenue function No less important is the image of the educational

As presented above, it also respects the revenue institution/university. Guarantee of a high-quality education
particularity. at a prestigious educational institution will also have an
R1 – contribution on each student impact on the price of the course.
R2 – Contribution from each student
R3 – revenue from grants It is further also necessary to determine a minimum number
of students, on which will depend whether the course will be
N – Number of students realised. This number will differ according to the type of the
educational institution and also according to the range of its
TR = (R1 + R2 ) ∗ N + R3 offer. [11]
The revenue function (7)
V. THE PROCESS OF MODELLING OF THE BREAK-EVEN POINT

The above mentioned analyses show that finding the break-

6. The break-even point
even point (point of equality of costs and profits) is a question
The break-even point introduces the place of the
with many variables.
equilibrium between costs and revenues, i.e.:
- Various kinds of costs,
- Various types of costs (fixed, variable),
TR=TC
- Various types of courses,
- Various lengths of duration of courses.
or
In practice, we could hardly find two identical courses at
(R1+ R2 ) ∗ N + R3 = (F1+ F2 + F3 + F4) universities at this moment, which could be placed into the
+ (V1+V2 +V3 +V4 +V5 +V6 +V7 ) ∗ N same group and for which the same figures would apply.
Nevertheless, there is still a need (which has been increasing
The Break-event function (8) continuously in my opinion) to identify costs and profits of an
e-learning course as precisely as possible, before the
realisation of the course is initiated (in fact, it is a certain
This is the general equation (8) of the break-even point. It form of a business plan, whether to initiate the realisation of
arises from precondition that the whole study is a distance the course or whether to cancel it already in the phase of
study and there are no personal sessions. If there are any preparation).
tutorials or certificate ceremony present in a course, the Currently there is a new trend, that an increasing number of
appropriate costs must be included to the function. universities have been buying already prepared courses from
prestigious educational institutions. As regards these courses,
(R1+ R2 ) ∗ N + R3 = it is then much easier to express the amount of their costs.
(F1+ F2 + F3 + F5 + F6 + F7) I consider the above mentioned equations as a starting
point for construction of an “interactive” model, that could be
+ (Fs1+ Fs 2 + Fs3 ) – after certain “user modifications” - applied in the general
environment of traditional educational centres
+ v ∗ (V1 +V2 +V3 +V4 +V5 +V6 +V7 )/ * N As for me, user modifications mean the possibility to enter,
The Break-event function with tutorial and diploma (9) before the calculation of the break-even point, a group of data
into the model that describe costs of the immediate
The function (9) contains tutorials during the term, summer environment of the course such as e.g.:
school and certificate ceremony. Other items of the - administrative requirements
educational organisation can be included. - physical spaces
The break-even point depends not only on costs but also on - technical background
the amount of determined fees for a course, that is for - requirements of tutors and people directly or indirectly
example in the case of a low fee, 120 students at least have to participating in the education (salary costs).
be in the course, but in the case of a high fee, e.g. only 25
students are necessary. Another dependency of the break-even In my opinion, interactivity is a certain guide (some
point is that each course is subject to different costs (e.g. software program), that would ask in steps (or by one-time
craft, artistic, music courses require a small number of filling-in of a table) about basic variable facts related to the
students in the group and will be thus subject to higher costs). course that is being calculated.
Also competition in the market has been constantly This model would presuppose a certain database of realised
growing, which is reflected in the resulting price of the e-learning courses (of course with their exact classification
course. into the above mentioned groups) and on the basis of answers,
108 HEDVICAKOVA AND MARESOVA

it would evaluate the financial requirements and it would division but a dynamic model which could be adjusted to the
recommend the amount of “tuition” for the given course. conditions of an institution.
Another possible element of this questioning would be Working in a virtual learning environment brings
searching for the most similar courses and enabling contact advantages to universities, e.g. decreasing costs on operation,
with parent educational centres that realised these courses to travelling expenses, accommodation, effective training,
share their experiences, so that we can avoid problems and continuous updating, etc.
become aware of what can be done better in the case of the Even though the distance and virtual study is running in
given type of the course. some large and rich states (Great Britain, France, Germany,
This all would lead to a relatively exact determination of Ireland), it is financially demanding and cannot work without
financial requirements of the course and thus of the considerable state grants and support. The cost calculations
corresponding needs of financial resources. cover author’s remuneration, salaries, distribution costs,
rental service costs, etc. Starting costs which may include
The main objective of such an interactive model is to establishing (building) administrative and study centres,
increase markedly the effectively of the financial resources preparing LMS, are very high. That is the reason why the
expended on the distance education and to achieve such a virtual study is supported from several sources (foreign
state, when a maximum of collected resources would be supportive programmers, tuition, production of teaching aids
invested into an effective functioning of the educational for other customers, selling the study materials, etc.) [7], [8],
system and a minimum of resources would be expended on [11].
the bureaucratic machinery, that represents an “unnecessary
evil”. An increasingly higher number of educational courses
have been appearing that, after the end of the subsidy REFERENCES
program, “suddenly loose” their financial benefit for the
educational institution (the course is loss-making after the end [1] T. ARCHIBALD, M. NIELSEN, T. TORY NEIWERT, K. OLIVEK,
Y-S. LEE, L. MARLER, Virtual Universities: A Current Snapshot,
of the subsidy period) and represents a financial burden in the Utah State University URL: http://www.kwadeoliver.com/it-
form of negative effectively for the institution. Correct portfolio/research-writing/Virtual%20Universities.pdf [online], [2006-
setting of the costs and profit side of the course already in its 11-8].
beginnings is a necessary key factor for a long-term effective [2] M. HEDVICAKOVA, I. SIMONOVA, Contemporary Ways to
Teaching Economic Subjects, 38th IGIP Conference, September 6 - 9,
functioning of the distance education. [11] 2009, Graz, Austria – in press
[3] The University of Economics, Prague, URL:
http://www.vse.cz/kategorie.php3?IDkat=722,
VI. CONCLUSIONS http://distancni.fph.vse.cz/ [online], [2008-2-4].
[4] VSB TUO, Technical University of Ostrava, URL:
Having found the break-even point, the number of students http://moodle.vsb.cz/moodle/206, [online], [2008-2-2].
[5] Project UIS, URL: http://www.uhk.cz/fim/studium/1188, [online],
is set to create and run a profitable course. The effectiveness [2009-2-2].
of any course can be calculated and compared by using this [6] Project EVENE, URL: http://www.uhk.cz/fim/evene/2102 [online],
simple method. [2008-3-11].
Granted finances were not taken into account in the [7] H. Zlámalová, Úvod do distanního vzd lávání, CODV UP Olomouc,
Andragogika, 2001, ISBN 80-244-0276-9
example, but they are often the principal starting point which [8] Bakešová, M.., Koudelka, F..: Ekonomika distanního vzdlávání
enables the work. (Economic of Distance Education), CODV Olomouc 2001, ISBN 80-
The contribution shows that any profitable and effective 244-0274-2
distance module requires great numbers of participants, [9] M. KADAVOVÁ, Virtual Education Relation to Economic Profit of
Educational Subjects, In E-Activities: Networking the World,
according to foreign researches: at least 300 students to cover Proceedings of the 6th WSEAS International Conference on E-
the costs (including fees and payments for printed and Activities (E-Learning, E-Communities, E-Commerce, E-
multimedia study materials). Correct and general information Management, E-Marketing, E-Governance, Tele-Working / E-
on the provided possibilities is essential for the system ACTIVITIES ´07), Puberto De La Cruz, Tenerife, Canary Islands,
Spain, December 14-16, 2007, s. 384 – 388, ISBN: 978-960-6766-22-
approach to virtual university development. 8, ISSN: 1790-5117
An advantage of distance courses, in contrast with the [10] M. KADAVOVA, A. SLABY, F. MALY, Key Factors Involving the
classic courses, is the fact that average variable costs, but Design of the System of Virtual University, In 7th WSEAS Int. Conf.
above all average fixed costs, are decreasing in the long term. on APPLIED COMPUTER & APPLIED COMPUTATIONAL
SCIENCE (ACACOS '08), Hangzhou, China, April 6-8, 2008,
With increase in the number of participants of the course, Published by WSEAS Press, s.678 – 683, ISBN: 978-960-6766-49-7,
inputs such as the number of teachers do not have to be ISSN: 1790-5117
increased in the long term: it is not necessary to rent new [11] M. KADAVOVA, (M. HEDVICAKOVA) Identifying the Break-Even
classrooms, it is not necessary to print other materials. All Point in Distance Courses, In WSEAS TRANSACTIONS on
ADVANCES in ENGINEERING EDUCATION, Issue 5, Volume 5,
materials are created in an e-learning course that can be easily May 2008, WSEAS Press, s. 282 – 294, ISSN: 1790-1979
updated and students can print current information as needed. [12] G. RUMBLE, The costs and costing of networked learning, JALN
The division of costs and their amount differ in various volume 5, Issue 2 – September 2001, Available at http://www.sloan-
institutions. It is not possible to create and present a static c.org/publications/jaln/v5n2/v5n2_rumble.asp [online], [2008-1-5].
 Modern Improvements
in the Digital Logic Laboratory
K. M. Noga, M. Radwanski
Gdynia Maritime University
Morska str. 83
81-225 Gdynia, Poland
[email protected]
[email protected]

Abstract-This paper presents examples of application for implemented in PLDs. Altera offers MAX+plus Baseline and
several models of physical objects controlled by programmable Quartus II packets, Xilinx offers WebPack ISE packet.
logic devices (PLDs) to digital logic laboratory course. During
laboratory sessions students create control systems for models and Besides, both companies offer education starter kits for
realize particular assignments with application of PLDs. With the academic use and release handbooks and guides on their
availability of new electronic components, equipment in websites [2, 3].
laboratory has to change. Exercises based on standard digital Popularity of PLDs forced changes in digital logic teaching
integrated circuits made in TTL or CMOS technology were curricula. Last years digital logic course on Electrical
removed from teaching curriculum. Now the purpose of
laboratory sessions is to familiarize students with programming in Engineering Faculty of Gdynia Maritime University (GMU)
VHDL language and software environments designed for PLD. was conducted in form of lectures (30 hours), exercises (30
hours) and laboratory sessions (30 hours) during first and
I. INTRODUCTION
second year of studies. In new schedule another 15 hours of
Necessity of adding PLDs into digital logic education lectures and 15 hours of laboratory sessions intended only for
curriculum appeared together with growing number of PLDs were added. Special starter kits and models made in
implementations and popularity of these devices. Digital Department of Ship Automation and Altera DE2 Development
control of various devices is common today. Control using and Education Board [4] are used during laboratory classes.
PLDs is new technical solution supplementary to used till now Models described in this paper are connected to these kits. In
control systems based on microprocessors and programmable [5] authors depicted first simple models used in digital logic
logic controllers (PLC). In comparison to microprocessor laboratory, newer models are more complex, interactive and
systems, programmable devices contribute simplification of are closer to reality.
some control tasks and technical realization of control system. II. LABORATORY STATIONS AND MODELS
For example, modern PLDs offer lot of universal inputs and
outputs, where propagation time of signals is exactly fixed. A. Basic station with CPLD device
Microcontrollers have limited amount of inputs and outputs
(ports) where computation time is hard to determine, especially
with utilization of high-level programming languages.
Programmable logic controllers are the best solution in control
of mechanical devices in heavy industry.
PLDs find more and more practical applications. Today they
are generally used in data transmission as coders and decoders
of transmitted signals. The new solution is implementation of
PLDs in control of power electronics devices such as voltage
converters, inverters and rectifiers. Control of these devices
requires realization of specific functions in fixed time.
Research and development concerning this implementation are
conducted in many universities and research centers [1].
Altera and Xilinx corporations are the biggest manufacturers
of PLDs. The most important kinds of existing PLDs are:
Complex Programmable Logic Devices (CPLD) and Field Fig. 1. Basic CPLD starter kit
Programmable Logic Arrays (FPGA). Both companies offer
wide range of mentioned kinds of devices, and regularly The first station allowing running of programmable devices
introduce to market newer models with improved capabilities. available in laboratory of digital logic at Department of Ship
The manufacturers of PLDs offer also free of charge basic Automation was laboratory kit ZL1PLD from BTC. ZL1PLD
versions of computer software intended to develop projects kit was equipped with CPLD devices EPM7128 from Altera

109
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_19, © Springer Science+Business Media B.V. 2010
110 NOGA AND RADWANSKI

and XC95108 from Xilinx. There aren’t universal inputs and of control electronics are placed. Near the printed circuit board
outputs available on board, so connecting external circuits and is placed IDC-40PIN socket for communication with DE2
devices to this kit is impossible. ZL1PLD kit has more board. Entire model is shown on Fig. 4.
technical flaws, like unavailable clock pins of CPLD. By these
reasons new extended kit (Fig. 1) with Altera EPM7128 device
was designed. This kit utilizes almost all capabilities of CPLD
device.
B. Robotic arm
Mechanical construction of robotic arm (Fig. 2) is based on
human anatomy, it contains three pivotal joints [6]. One of
them is placed in the mount and allows full circle rotation,
second allows vertical deviation at the base, third one allows
bending at half length of the arm. At the end of the arm is
placed mechanical grab, which can lift load weight up to 1 kg.
Control circuit of the robotic arm consists of three units: Altera
DE2 board, Atmega16-16PU microcontroller and four DC
bridge controllers.

Fig 3. Model of dwelling house diagram

Movable parts of doors and windows are driven by worm-
gears. Stepper motors power opening motion works.
Movement stop sensors are activated when window frame is
closed or fully opened.

Fig 2. Complete station with robotic arm

Students during laboratory session can control this model in

combinatorial mode using switches placed on DE2 board or
program sequence of movements that can move various objects
into selected places. The program for the FPGA device placed
on DE2 board should be written in VHDL (Very-high-speed-
integrated-circuit Hardware Description Language).
C. Model of dwelling house Fig 4. Model of dwelling house – rear view
Model of dwelling house together with devices installed Lighting systems consists of five light sources: chandeliers
inside is part of laboratory station designed for testing student in hall, living room, loft and two lamps in front of the house
projects implemented in Altera DE2 board. Model of house near the door. Chandelier in hall is turned on by manual switch
was built for showing possibilities of basic actuators control, it or by movement sensor after detection of person in this room.
includes circuits opening automatic doors and windows and The light is turned on for 5-second period and is turned off
powering lights [7]. The user with DE2 board controls all when no another movement is detected. All devices in the
devices in the house. Additionally thief alarm with movement model can be controlled manually with use of switches and
and open doors and windows sensors is available. Air buttons on DE2 board.
temperature sensor indicates alarm signal when temperature is
higher then acceptable limit selected on setting dial. Sample projects
The measurements of ground floor are 60 by 30 by 31 These are some of the student projects of the control systems
centimeters and 60 by 30 by 24 centimeters loft (Fig. 3). The made during laboratory classes within digital logic course:
ground floor is divided on two rooms: hall and living room. In • assign selected switches to open windows,
the living room power supply, temperature setting dial and part • windows should close automatically after 8 seconds,
 MODERN IMPROVEMENTS IN THE DIGITAL LOGIC LABORATORY 111

• open the door and switch on light in hall when movement Eleven signals connected to CPLD are used to control this
detector is activated, close the door and switch of light 6 model. Four of them power the engines and set directions of
seconds after last movement, wheels rotation. Next four signals are wired with rotation
• activate alarm when opened windows reach extreme sensors and three last are connected to barrier sensors. Engines
positions. can be turned on and of at any moment with change of
The purpose of exercises is to acquaint students with Altera direction without pauses. Rechargeable battery Sunnyway
DE2 board, Quartus II environment and VHDL. SW613 (6V, 1.3Ah) placed behind control unit allows to
D. Model of Mobile platform operate model up to one-hour period. Data send to PC are used
Mobile platform laboratory station (Fig. 5) consists of to draw movement trajectory with special software. Connecting
vehicle model (1) with control circuit, radio receiver with characteristic direction-changing points formats trajectory plot.
RS232 interface (3) and ZL11PRG (2) PLD programmer. The Position of the vehicle is transmitted to PC in fixed moments in
vehicle communicates with PC through radio receiver, RS232 time or when barrier is detected. Data obtained from radio
interface and sends position data to application developed in receiver are decoded and stored by software in movement
Borland Delphi environment [8]. The model is based on Xilinx history file. Trajectory is stored in bitmap file at the maximum
CPLD programmed with utilization of WebPack ISE resolution of 600 by 600 pixels. Data are transmitted with 2400
environment. bps bitrate in frames with 8 data bits and 1 stop bit.

Sample project
Student should create program that allows to move the
model along fixed trajectory. Current position has to be
recognized by signals in feedback from rotation sensors.
Turning points should be transmitted by radio to vehicle and
tracked position data should be stored in the file and visible on
the screen. Student during this exercise becomes familiar with
use of radio data modems for communication between PC and
movable robot.
E. Multipurpose vehicle
Multipurpose remote controlled vehicle with AVR
microcontroller and CPLD was built in Department of Ship
Fig. 5. Complete laboratory station
Automation too [9]. Additionally vehicle can be controlled by
Schematic diagram of the model is shown on Fig. 6. radio with remote desktop (Fig. 7). Vehicle can run on almost
Underbody structure is made from acrylic glass with three- any surface and may be used to penetrate hard to access or
wheel motion system. Control electronics and battery pack is unsafe places. Load capacity about 5 kg and moderate size (36
placed on top of the platform. Three barrier sensors are placed x 23 x 8 cm) of the robot makes it perfect platform for another
on front bar. Rear wheel isn’t driven; it works as bearing point control and measuring devices. The main control circuit
only. Separate autonomous servomotors drive two front consists of EPM7128 CPLD from Altera and ATmega8-16PV
wheels, than platform can move in every direction. Distance is microcontroller from Atmel. CPLD controls engines, proximity
taken using wheel rotation. Rotation sensors placed on wheel sensors and indicators, microcontroller controls
axes are made from two optocouplers. communication through radio data modem (Fig. 8).

Left barrier
Left drived sensor
wheel

Left rotation
sensor
Middle barrier
sensor
Right rotation
Free wheel sensor

Right drived
wheel Right barrier
sensor
Fig. 7. Vehicle with remote desktop controller
Fig. 6 Diagram of mobile platform
112 NOGA AND RADWANSKI

Radio data modem with antenna, wireless camera, lights,

ultrasonic sensors, temperature sensor, main power switch,
battery charge socket, power and charge LED are mounted on
the housing of vehicle.

433 MHz 2.4 GHz

MEASUREMENT MODEM CAMERA

MODULE ATC863 +9V
+8V
+9V RS232

MICROCONTROLLER
Temp
MODULE +5V POWER
ADC0 PC0
ATmega8-16PV MODULE

D0-D7
+5V
PIN4/RX

CPLD MODULE +12V/24V

LEFT EPM7128SCL84-15N RIGHT
MOTOR MOTOR
PIN74 Lights
PIN77 Buzzer
PIN75 Camera
PIN58,60 PK1,2 +12/24V
LIGHTS
BUZZER

Fig. 8. Block diagram of electronic part Fig. 9. Quadruped walking robot

Vehicle motion can be controlled by portable desktop too. Control circuit of the robot contains ATMega8 and ATtiny
Desktop is based on ATmega8 microcontroller working with 2313 microcontrollers and EPM7128SLC84 CPLD. Robot’s
ATC-863 radio modem, LCD and keyboard. ATC-863 radio outfit contains sound, light, and distance sensors, battery
modem includes RS-232 and RS-485 interfaces allowing direct charge indicator, buzzer and 2x16 LCD (Fig. 10).
connection with other standard electronics devices containing
UART (Universal Asynchronous Receiver and Transmitter)
port.

Sample project
The vehicle should find a gap between barriers and scour
room behind in autonomic mode. The task for student is to
create program for CPLD, which uses proximity sensors to
determine location of free space and allows sending pictures
from camera. The pictures should be taken in hidden room
with use of vehicle rotation to show the room in various
directions.
F. Quadruped Robot
Quadruped Robot is another model designed and build in
Department of Ship Automation (Fig. 9). Each leg contains
two joints driven by separate servomotors [10]. Data Fig. 10. Block diagram of control modules
concerning angle of flexion are transmitted in control line of Eight digital servomotors HX12K from HexTronik are used,
the servomotor from ATmega8 microcontroller. Two kinds of two in each leg. One moves leg in horizontal plane, second in
steps for boost of motion capability of robot were designed. In vertical. Servo controller uses simple 3-bytes communication
first one robot lay between steps on the bottom side of body, in protocol. Data are transmitted at standard UART bitrate 9600
the second one robot stands on legs during entire walking bps. Maximum resolution of flexure depends on minimal
cycle. length of pulse on output of ATtiny2313 microcontroller and
equals 7.5°.
Motion control system works in three modes. In the default
one running after power-on movements are autonomous
 MODERN IMPROVEMENTS IN THE DIGITAL LOGIC LABORATORY 113

according to the program implemented in programmable located inside the mockup. Gantry crane model contains
device. In the remote mode robot’s movements are controlled running tracks, the tower, jib, jib ties, trolley with hoisting
by user’s orders given with utilization of PC software interface. system and hoisting block with electromagnet. Model is
Third one is service mode allowing setting initial position of powered by three electric motors. DC motor moves the tower
servomotors. on running tracks, bipolar stepper motor placed inside the jib
Robot indicates it’s own status with buzzer, LEDs and LCD. moves the trolley, another servomotor inside trolley lifts and
Status signals are generated by ATtiny2313 microcontroller lovers hoisting system.
programmed with Bascom AVR environment. Radio modem Limitary sensors turn off motors when moving part reaches
CC1000PP modules are used for communication between extreme position and send alert signal to FPGA device. Stepper
robot and PC. Communication is based on FSK (Frequency motor is driven by controller based on ATtiny2313
Shift Keying) modulation, data are coded using Manchester microcontroller programmed in C language. Optical sensors
code. Communication protocol allows bi-directional data detect positions of tower on running tracks and trolley on the
transmission in half-duplex mode with error suppression at jib and send sequence of pulses to FPGA device. Using this
bitrate 9600 bps. data it is possible to define current position of movable parts
and to generate signals controlling next moves.
Sample exercise This model is based on the blueprints of Baltic Container
Using VHDL and MAX+plus II or Quartus II environment Terminal in Gdynia [12] and is made in 1:140 scale. Basic
student should prepare program describing sequence of construction of the model has measurements 165 x 145 x 530
movements with status indication. Buzzer, LEDs and LCD mm, jib is 880 mm long.
should be used to display robot state, like distance to obstacle,
intensity of light or low battery level. Sample exercise
The most typical student task is to program sequence of
G. Gantry crane
actions allowing loading and unloading containers from the
Cargo loading is one of the most important issues in
ship. Status of the model should be displayed on 7-segment
maritime transport. Laboratory station with gantry crane model
displays and LEDs included on DE2 board.
(Fig. 11) allows to familiarize students not only with FPGA
programming but also with logistic issues. Gantry crane model H. Bank vault alarm system
is controlled by Altera DE2 board with Cyclone II FPGA To fulfill requirements of didactic occupations interesting
device. model of bank vault alarm system was built too. Model
controlled by DE2 board equipage consists of professional
sensors, GSM cell phone, door opening drive and alarm system
(Fig. 12).

Fig. 12. Station with bank vault model

Fig. 11. Laboratory station with gantry crane model
Signals from vibration sensor, microwave sensor, PIR sensor
Gantry crane model moves alongside the wharf, a trolley (detects changes in infrared radiation), limitary sensors,
moves along the jib and hoisting mechanism contains external keyboard and signals connected to actuators (stepper
electromagnet allowing lifting and carrying cargo container motor, beacon light, hooter, GSM cell phone and fan) are
models between wharf and ship model [11]. Control and power connected through printed circuit board with amplifiers and
system printed circuit boards and driving mechanism are separators to 40-pin socked compatible with Altera DE2 board
114 NOGA AND RADWANSKI

[13]. The door is opened and closed by stepper motor with the pen. EPM7128 device and basic CPLD starter kit fulfill
smooth regulation of speed. role of the controller (Fig. 14). Students can draw simple
graphics (letters or logos) containing segments of straight line,
Sample exercise there is no possibility of drawing angles. Resolution of the
Using Quartus II environment and VHDL the designer drawing equals to 16 x 16 points. Control circuit can be entered
should develop a project that supports vault control system, as logical diagram with standard TTL 74xxx devices or as text
e.g. allows access after entering proper code from the file written in VHDL.
keyboard. In case of double error in code system should III. CONCLUSIONS
generate alarm signals and send alert SMS through cellular
network to selected phone number. Models and boards with programmable logic devices
presented above vary standard laboratory sessions and allow
I. Control of stepper motors watching directly results of control process. Flexibility of
Stepper motors are in common use in automatics. The PLDs gives us possibilities to build various kinds of models.
method of control of stepper motors forces use of digital logic Knowledge about control of DC, AC and stepper motors, using
in practice. Laboratory station (Fig. 13) contains basic CPLD signals generated by sensors and communication between
kit and two separate motors. Use of programmable logic allows devices is required in professional career of electrical engineer.
control in all work modes of motors (full step, half step, Diversity of laboratory stations allows accurate assessment
microstepping, one-phase, two-phase and multi-phase). of student’s work and knowledge from the wide range of
engineering. Students have to be familiar not only with digital
logic, but must know basics of mechanics, electrotechnics,
electronics, data transmission and signal processing.
Developing interactive project of control system for complex
laboratory station is effortful, but students receive guides
containing several examples of projects. Hard-working and
ambitious persons have a chance to prove their abilities by
presenting complex project with optional solutions.
Fig. 13. CPLD kit with stepper motors
REFERENCES
Students can create projects of control systems that allow
[1] M. Baszynski, “A model of the three-phase bridge rectifier with
generating sequence steps in selected direction or generating sinusoidal source current using FPGA implementation”, Przeglad
fixed number of rotations. Source of the project can be entered Elektrotechniczny, vol. 3/2009, pp. 36-41.
as logical diagram or written in VHDL in MAX+plus II or [2] Altera Corporation website: http://www.altera.com.
[3] Xilinx Inc. website: http://www.xilinx.com.
Quartus II environments. This station is used during basic and [4] Altera Corporation, DE2 Development and Education Board User
advanced digital logic course. Manual, electronic version, 2006.
[5] K. M. Noga and M. Radwaski, “Our Stations for Teaching
J. Control of the plotter Programmable Devices”, Innovations in E-learning, Instruction
Technology, Assessment and Engineering Education, pp. 268-273,
Springer 2008.
[6] D. Kupiec amd S. Murawski, Budowa modelu ramienia robota oraz
implementacja algorytmu sterowania w srodowisku Quartus, 1st degree
thesis, GMU, Department of Ship Automation, 2008, in Polish.
[7] L. Meissinger and S. Krawczyk, Zastosowanie ukladow
programowalnych do sterowania modelem domu mieszkalnego – budowa
i oprogramowanie stanowiska laboratoryjnego, 1st degree thesis, GMU,
Department of Ship Automation, 2008, in Polish.
[8] P. Ptach and, S. Sieranski, Sterowanie wybranym modelem po
okreslonej trajektorii z wykorzystaniem ukladow programowalnych, MA
thesis, GMU, Department of Ship Automation, 2008, in Polish.
[9] K. Gorski and S. Orzechowski, Projekt i budowa wielozadaniowego
pojazdu samobieznego sterowanego mikrokontrolerem i ukladami
programowalnymi, 1st degree thesis, GMU, Department of Ship
Automation, 2009, in Polish.
Fig. 14. Laboratory station with plotter. [10] P. Wozniak and M. Wysocki, Projekt oraz budowa kroczacego robota
sterowanego ukladami programowalnymi, MA thesis, GMU, Department
Plotter is one of the oldest devices purposed to print of Ship Automation, 2009, in Polish.
[11] M. Witek and P. Witek, Projekt, budowa oraz sterowanie modelem bazy
computer graphics. Simple method of control requires only kontenerowej z zastosowaniem ukladow programowalnych, 1st degree
basic knowledge from the range of digital logic. Printing task thesis, GMU, Department of Ship Automation, 2008, in Polish.
is based on sending sequence of two 4-bit numbers to the [12] Baltic Container Terminal in Gdynia website: http://en.bct.gdynia.pl/
[13] W. Laskowski and T. Taczynski, Model systemu alarmowego skarbca z
plotter. These numbers determine position of the pen on X-Y wykorzystaniem zestawu edukacyjnego Altera DE-2 i jezyka VHDL, MA
plane, one additional bit represents writing or high position of thesis, GMU, Department of Ship Automation, 2009, in Polish.
 e-Cheating and Calculator-Technology
A Preliminary Study into Casual Implications of Calculator-Technology Usage on
Students’ Attitude Towards e-Cheating

Zeenath Reza Khan

Instructor
Faculty of Computer Science and Engineering
University of Wollongong in Dubai
Email: [email protected]
PO Box 20183, Dubai UAE

Abstract daily life, yet a technology that has increased student ability
in all science fields, in and out of classrooms, and into offices.
Across the globe, educational institutions are adopting e-learning
tools into their curricula to cater to the ever-growing demand for This paper has been written as a subsequent resultant of an
technology inside the classrooms. Where traditional blackboards
on-going research that is being carried out by the author on
have been replaced by virtual whiteboards and library books by
the casual implications of readily-available technologies on
online resources, teachers are trying hard to cope with the growing
competition from technology outside the classroom that has students’ attitude towards e-cheating. At the start of the study,
captured their students’ attention and imagination, with more it looked as though enough research had been conducted on
technology, perhaps increasing student dependency. various aspects of e-cheating. However, as the literature
review has highlighted, distinct gaps seem to exist in the
However, the author’s previous research has shown that engaging literature that highlight the absence of studies showing the
e-learning tools to enhance student learning experience is not impact of certain technologies besides the Internet on
enough as it has negative impact on students’ attitude towards e-
students’ attitude towards e-cheating [8]. One such
cheating, disillusioning their awareness to cyber ethics. So, how do
technology is that of calculators that aid in computations and
teachers ensure that the methods of teaching and the tools
therewith truly enhance student learning and not affect their sense estimations.
of ethics?
This paper looks at the ease with which students can access
As a part of an on-going research, this paper highlights the calculators and how the usage has affected their attitude
exponential usage of calculators in the classrooms and their towards e-cheating.
impact on student’ attitude towards e-cheating, and recommends
further studies to measure the correlation of calculator II. CALCULATORS NOW AND THEN
dependency in the classroom to student attitude towards e-cheating
and professionalism at workplaces. . The first instrument to help making calculations easy and fast
was the abacus. This has wooden beds that are moved along a
I. INTRODUCTION wire or thin wooden stick to represent numbers [9]. The
young French mathematician Blaise Pascal invented the first
Technology in education is not a new phenomenon. It has
adding machine in 1642, a device driven by gears and capable
been around for decades and has become a vital part of
of performing mechanical addition and subtraction [1].
teaching and learning in the twenty-first century. Inside and
out of classrooms, technology usage has gained popularity to A revolution in calculating machines took place between the
feverish heights. Almost any and every technology is now early 1960s and the late 1970s. It was during this time period
used as a part of e-learning in order to enhance the overall that electronics for calculators was at the forefront of
experience for students and teachers [6]. Calculators are one electronics research. ‘Calculators evolved from large,
such technology that users have taken for granted as a part of expensive, mechanical machines to cheap, electronic, credit

115
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_20, © Springer Science+Business Media B.V. 2010
116 KHAN

card sized devices. The development of micro-electronics for in the process of researching for this paper, the author has
calculators was an important phase in the history of found many websites and other sources that seem to entice
technology, which included the development of the students into using calculators in the exam halls, for their
microprocessor.’ [2] term papers and homework in order to achieve higher grades.
Sites such as Online 2 College and Teachapolis that is
III. CALCULATORS – BOON OR BANE supposed to be a ‘virtual city for teachers’ are free to visit by
any user sitting on a PC connected to the Internet. With such
Calculators are very much a part of technology that have
topics as ‘Using Technology to Cheat • Calculator:
upgraded over the last decades to become part of e-learning.
programmable calculators can hold text, formulas, even
As mentioned in the previous study, calculators have been a
pictures’ [3] it is giving students hints on how to use the
breakthrough way before computers had become common
calculators in something other than simply aiding in quick
place at every home [8]. It was definitely a technology above
calculations. Another article by Andrew Kantor in
the use of booklets with pre-calculated tables and slide rulers
USAToday.com relents about ‘add-on memory [that] lets you
[10] As highlighted before, in 1965, the first pocket calculator
store software, turning your calculator into a pocket
was introduced to the market; by 1974 it had achieved
notebook… the company [Texas Instruments] is happy to
providing four functions with LED screen, and although it
point out that the TI-83 Plus is "allowed for use on the PSAT,
was well over a decade before school children had their own
SAT I, SAT II Math IC and IIC, AP Chemistry exam, AP
pocket calculators, the technology advanced rapidly [10]
Physics exam, and AP Calculus exam.’ [4]. In 2005, Texas
Now-a-days, calculators range from simple scientific to
Instruments (a company that pioneered in the development of
graphic to programmable with large amounts of memory
calculator technology in the 1970s) recalled thousands of
space, data wires to allow sharing of information, formulas
calculators issued to students in the United States of America,
and so on among friends.
after a student discovered that pressing certain keys converts
However, studies have shown that where calculators make it decimals into fractions [5]. As schools expected students to
easy for students and adults to make quick calculations, they have the knowledge and skill to perform this task, the
are ‘becoming a mental crutch for students, rather than a tool calculators being used for the same purpose was labeled as
that promotes higher order learning’ [11]. It has been seen cheating.
that most often than not academic institutions ban the use of
Research into existing literature has shown that calculator
certain types of calculators in examinations to ensure students
technology has definitely had an impact on students’ attitude
are able to work out problems upon their ability rather than
towards e-cheating. However, the author has found little or no
with the aid of technology. Such technology in the classroom
evidence of actual studies carried out to establish a casual
is feared to ‘result in an over-reliance on technology to
relationship between the two.
provide solutions, thereby stifling a student’s educational and
creative growth’ [11]. The author has found little or no V. THE STUDY
literature to show the actual effects of allowing calculators in
the classroom. In many schools and universities, teachers are Based on the literature reviewed, and the corresponding study
on high alert in examination halls, keeping an eye out for being carried out on readily-available technologies and their
programmable calculators that students can bring in with effects on students’ attitude towards e-cheating, the following
uploaded formulas and pre-sketched graphs that would study was proposed: investigate if there are any casual
constitute cheating. But, there does not seem to be any study implications of allowing calculator-technology in the
to actually register if there is any casual implications of classroom to students’ attitude towards e-cheating.
allowing high-end calculators, or any computational devices
that are affordable and readily available in stationary shops, To carry out this study, the sample size chosen consisted of
on students’ attitude towards cheating [11]. about 100 students of mixed ethnic and educational
(curricula) backgrounds. Of these, 28 were discarded as they
IV. IS IT AFFECTING STUDENTS’ ATTITUDE TOWARDS E- had not completed the surveys. A survey model was
CHEATING? developed to gather student feedback on various topics related
to academic integrity and specifically to the usage of
Although the author has found no direct studies showing
calculator and other technology. Students interviewed were
affects of calculators on students’ attitude towards e-cheating;
 E-CHEATING AND CALCULATOR-TECHNOLOGY 117

chosen in random in terms of age, gender and year of VII. CONCLUSION

program. The survey model was developed on a 5-point
Likert-scale and the questions ranged from testing students’ ‘Discovering, evolving, enhancing and adopting are all
knowledge on academic integrity, to their understanding of reasons why man moved through so many different and
calculators as e-learning tools and if they were using varied eras from ice age to the information age’ [6]. However,
calculator-based technology in the classrooms, particularly to over dependence on certain tools and technology can and
cheat. The results were tabulated using weighted average (ref have had adverse effects on the human kind before. Education
Appendix A) for each category of question and summarized and academic integrity are no different. Where
as illustrated in Appendix B. A concurrent survey was carried teachers/parents are in a lead role to introduce and encourage
out on various schools to collect data on the usage of student learning through whatever means available, be it
calculators in the classrooms from 2006 – 2009 alone. Table 1 traditional classrooms to electronic whiteboards, power point
in Appendix B shows the results. projectors to WebCT, these adults are also faced with the
dilemma on the age-old concern of academic integrity, of
VI. RESULTS AND DISCUSSION testing students of what they know rather than what they can
use.
From the study, it can be seen that when agreeing on
concepts, students have a fairly good idea of what is right and Although this is a preliminary study, and a branch-out of an
what is wrong. When asked ‘Academic Integrity can be on-going study on the effects of readily-available technology
defined as an adherence to a code of academic values’, 67% on students’ attitude towards e-cheating, the author believes it
of the students agree to this philosophy, just as 51% agree that has shed some light on the possible casual effects of increased
‘”E-cheating” or electronic cheating can be defined as using usage of calculators in the classroom by students. Where
information technology (IT) to aid in the process of cheating calculators have indeed saved students precious time in
in a class’. However, as has been established by previous solving complicated calculations quickly and easily, they
studies [6] when it comes to applying these same concepts, seem to have been added to the ever-growing list of
students deter from their ‘prior knowledge’ [7]. A whopping technologies that students use in order to destroy academic
40.8% have agreed to the fact that they cheat on exams, 5.6% integrity.
using minimized text (cheat sheets), 1.4% using mobile
phones, 1.4% using memory sticks and an astonishing 13.8% As it is beyond the scope of this paper, based on a previous
using programmable calculators. Upon further interviews for study by the author, it is recommended that further
clarifications, students suggested calculators as cheap and fast investigation be carried out to measure the impact of
ways of carrying large sums of information into exam halls calculators on students’ academic achievements, and their
and using infra red to pass on the information to friends. ultimate usage in the work place; and study if then there is
any casual implications towards e-cheating.
As illustrated by the graph below, the usage of calculators by
students for various subjects have increased substantially over REFERENCES
the three years (2006 – 2008 inclusive). At this rate of
dependence, 13.8% students agreeing to use programmable [1] Ament, P. (2007) Hand-held calculators. Troy MI. The Great Idea
Finder. (Online). Available url: http://www.ideafinder.com
calculators for cheating on exams is not a surprising
percentage. [2] Tout, N. (2009). Vintage Calculators. About.com. (Online). Available
URL:http://inventors.about.com/gi/dynamic/offsite.htm?site=http://www.vint
When compared to other technologies mentioned in the agecalculators.com/
survey, students seem to access calculators more frequently as
mode of cheating tool on exams than iPods, mobile phones, [3] Bramucci, R. S. (2003). How to Cheat: techniques used by cheaters.
(Online). Available URL: http://www.teachopolis.org/justice/cheating/
PDAs, pagers and such. cheating_how_to.htm
Calculator usage 2006 - 2008
[4] Kantor, A. (2008). Cheating goes high-tech with common-place tools.
7000 USAToday.com. (Online). Available URL: http://www.usatoday.com/tech/
6000 columnist/andrewkantor/2004-05-21-kantor_x.htm
5000
Series3
2008
Students

4000
Series2
3000
2007
Series1
2000
2006
1000

0
s
t

e
ry

ts
h

cs
s
ec

lu

nc
at
at

tic

ist

un
i
cu
bj

ys
tM
M

na
is

em

co
Su

Ph
at
Ca
sic

Fi
ee

Ac
St

Ch
Ba

cr s
Di

Subjects

Fig 1: Calculator usage 2006 - 2008

118 KHAN

[5] Block, R. (2005). Texas Instrument calculators help students cheat. For instance, for the question ‘It is okay to download MP3 or movies from
(Online). Available URL: http://www.engadget.com/2005/06/09/texas- peer-to-peer websites’, the weights start from 1 to 5 for each of the Likert
instrument-calculators-help-students-cheat/ items ‘strongly agree’, ‘agree’, neither agree nor disagree’, ‘disagree’,
‘strongly disagree’ respectively. This is because if a respondent chooses
[6] Khan, Z. R. (2006). E-learning techniques in IT programs – are they ‘strongly agree’, he/she demonstrates he/she has chosen a wrong answer. For
helping my sense of cyber ethics?. Conference paper presented at the IEEE another question, ‘Cheating can be defined as violating accepted standards or
Second International Joint Conferences on Computer, Information, and rules’, the scale is reversed such that if a respondent chooses ‘strongly agree’,
Systems Sciences, and Engineering. December 4-14. the response receives 5 point on the scale thus showing he/she has chosen the
right answer. Once the weights have been placed on each response, the
[7] Khan, Z., Al Qaimari, G. & Samuel, S. D. 2007, 'Professionalism and average is calculated by totaling response for each question and diving by the
ethics: is education the bridge?', in G. R. Lowry & R. L. Turner (eds), total number of respondents for each question (72 students) in each case.
Information Systems and Technology Education: From the University to the
Workplace, Information Science Reference, New York.

[8] Khan, Z. R. and Samuel, S. D. (2009). E-cheating, technologies and APPENDIX B

online sources: a critical review of existing literature. Conference
Proceeding. 9th Global Conference on Business and Economics. Cambridge
University, UK. TABLE 1: CALCULATOR USAGE OVER THREE YEARS FOR A VARIETY OF
SUBJECTS.

[9] Allsands.com History of Calculator. Accessed October 24, 2009.

(Online). Available URL: Calculator Usage
http://www.allsands.com/History/Objects/historycalculat_sih_gn.htm
Year 2006-2008
[10] Calculator.org (2009). A First Calculator. Flow Simulation Ltd. [Online]
Accessed March 2009. Available URL:
http://www.calculator.org/firstcalc.html Subject 2006 2007 2008

[11] Bsmarte. (2006). Managing Technology: Is technology in the classroom Basic Math 468 401 650
good or bad?. Bsmarte.com [Online] Accessed February 2009. Available
Discreet Math 238 265 300
URL: http://www.bsmarte.com/issues/managingtechnology.pdf
Calculus 210 365 365
APPENDIX A Statistics 200 200 255
Physics 165 165 165
Calculating the weighted average:
Chemistry 100 55 290
In Table 1, for each Likert Item, there is a ‘weight’ placed depending on how Accounts 110 100 300
close the choice is to being right. Finance 100 100 144

TABLE 2: ACADEMIC INTEGRITY

Neither
Strongly agree or Strongly
Answer Options agree Agree disagree Disagree disagree

Academic Integrity is the same as Ethics 2 25 33 10 1 230.00

Weighted
Weight 5 4 3 2 1 3.19 Average
Academic Integrity can be defined as an adherence to a code of
academic values (Webster Dictionary) 10 48 13 1 0 283.00
Weighted
Weight 5 4 3 2 1 3.93 Average
Cheating can be defined as an act of deception for profit to yourself 27 27 10 5 1 284.00
Weighted
Weight 5 4 3 2 1 3.94 Average

Cheating can be defined as violating accepted standards or rules 27 31 10 2 2 295.00

Weighted
Weight 5 4 3 2 1 4.10 Average
“E-cheating” or electronic cheating can be defined as using information
technology (IT) to aid in the process of cheating in a class (King and
Case, 2007) 14 37 16 2 2 272.00
Weighted
Weight 5 4 3 2 1 3.78 Average
 E-CHEATING AND CALCULATOR-TECHNOLOGY 119

TABLE 3: CONCEPT APPLICATION

Neither
Strongly Strongly
Answer Options Agree agree or Disagree
agree disagree
disagree

it is okay :
(a) to share information among friends during tests or exams 5 10 16 29 12 249.00

Weighted
Weight 1 2 3 4 5 3.46 Average
(b) to copy from the website that has the required information for an
3 17 18 28 6
assignment 233.00

Weighted
Weight 1 2 3 4 5 3.24 Average
(c) to copy from a text book that has the required information for an
4 21 22 20 5
assignment 217.00

Weighted
Weight 1 2 3 4 5 3.01 Average
(d) to write in the information from what someone else says for an
1 13 19 33 6
assignment 246.00

Weighted
Weight 1 2 3 4 5 3.42 Average
(e) to copy from another friend who has the information for an
2 7 15 37 10
assignment 259.00

Weighted
Weight 1 2 3 4 5 3.60 Average
All of the above points a-e but with due citations and reference list 28 30 7 4 0 289.00

Weighted
Weight 5 4 3 2 1 4.01 Average
It is okay to install a copy write software given to me by a friend 9 25 24 10 3 186.00

Weighted
Weight 1 2 3 4 5 2.58 Average
It is okay to download MP3 or movies from peer-to-peer websites 18 35 9 7 3 158.00

Weighted
Weight 1 2 3 4 5 2.19 Average
It is cool to buy pirated movies from vendors on the streets for AED
13 21 15 17 5
5/- instead of the original for more than AED40/- 193.00

Weighted
Weight 1 2 3 4 5 2.68 Average
‘Anything goes’ is a sure attitude to success 4 9 30 22 4 220.00

Weighted
Weight 1 2 3 4 5 3.06 Average

TABLE 4: WHAT I PRACTICE

Cheated on exams: 40.8% (29 out of 72) have answered YES

Cheated on exams using technology

Yes No
such as : %
A) Mobile phone wireless
1 71
connectivity 1.4
B) PDAs 0 72 0
C) Programmable calculators 10 68 13.8
D) Pagers 1 71 1.4
E) Minimized text using photocopy
4 68
machine to make cheat sheets 5.5
F) iPOD 0 72 0
G) e-dictionaries 0 72 0
H) Memory sticks for online exams 1 71 1.4
 Perspectives on Critical Thinking through Online
Discussion Forums in Engineering Mathematics
1
S. M. Jacob
Swinburne University of Technology (Sarawak Campus), Malaysia.
2
H. K. Sam
University Malaysia Sarawak, Malaysia.
1
[email protected] and [email protected]

Abstract-This study set out to examine the online involves constructing the situation and supporting the
mathematical communication and measure the progress of critical reasoning behind a solution. It focuses on the process of
thinking (CT) skills of students that occurred in collaborative
groups, during two collaborative problem solving sessions via learning rather than just attaining information. It involves
online Discussion Forums (DFs), in a first year Engineering discovering how to analyze, synthesize, make judgments, and
Mathematics unit which spans 14 weeks. It was found that the create and apply new knowledge to real-world situations. A
calculated CT scores have improved from the first to the second simple standard definition of critical thinking adopted in this
DF session. The CT scores were connected with moderation study is “the ability to use acquired knowledge in flexible and
techniques associated with Socratic Questioning and reflection
scaffolds, during the first session, but not in the second. Students’ meaningful ways, through understanding the problem or issue,
perceptions on the DFs were analyzed and was found that evaluating evidence, considering multiple perspectives, and
majority of the students found the DF sessions quite challenging taking a position” [3], p. 275.
to their thinking skills and profitable for developing their
problem solving skills. B. Socratic Questioning
Yang, Newby & Bill [4] has used Socratic Questioning to
I. INTRODUCTION promote critical thinking skills in asynchronous discussion
forums. A powerful method of cultivating critical thinking in a
Critical thinking is both a process and an outcome [1]. As an
group, Socratic Questioning uses the technical strategy of
outcome, it is best understood from an individual
regressive abstraction and develops a rigorous inquiry into the
perspective—that is, the acquisition of deep and meaningful
ideas, concepts and values held by people with regard to the
understanding as well as content-specific critical inquiry
topic. The questions fall under the six categories of Socratic
abilities, skills, and dispositions. Judging the quality of critical
questioning prompts [5], namely, questions of clarification,
thinking as an outcome within a specific educational context is
questions that probe assumptions, questions that probe reasons
the responsibility of a teacher as the pedagogical and content
and evidences, questions about viewpoints or perspectives,
expert. As a product, critical thinking is, perhaps, best judged
questions about the question and questions that probe
through individual educational assignments. Critical thinking
implications and consequences. Instead of directly providing
(CT) and mathematical problem solving have existed side by
statements as clues for students to advance in their process of
side and have attracted research over the decades.
solving the problem, questions are posed with a reasoning
The research questions answered through this study, done on
approach, from which the students are led to thinking into the
online Discussion Forums (DFs) for a batch of Engineering
different aspects/stages of the problem.
Mathematics students were: (1) What was the level of CT skills
of the participants, individually and group wise, in the online C. Asynchronous Discussion Forums and CT
DFs? (2) What was the effect the moderation techniques have There are two approaches [6] to teach CT using content
on the level of CT skills in the online discussion forums? (3) disciplines: (1) The “infusion” approach where critical thinking
What were the participants' perceptions about the online DFs skills are taught manifestly using the discipline’s content; and
and the impact of the DFs on their critical thinking skills? (2) The “embedded” approach, where critical thinking skills
are taught in indirect ways without spelling it out to students.
II. REVIEW OF LITERATURE
The asynchronous discussion forums are seen as an attempt to
A. Critical Thinking and problem solving try the embedded approach. Discussion forums were identified
Erwin [2] defines problem solving as a step-by-step process by [7] as one of the effective forms of assessments of critical
of defining the problem, searching for information, and testing thinking. Perceptions of asynchronous communication tools
hypotheses with the understanding that there are a limited have been reported by [8] and [9]. In their study on cross-age
number of solutions. Critical thinking is a broader term peer tutors’ contributions in asynchronous discussion groups,
describing reasoning in an open-ended manner, with an [10] have commented that students’ learning experiences and
unlimited number of solutions. The critical thinking process perceptions about the scaffolding offered on asynchronous

121
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_21, © Springer Science+Business Media B.V. 2010
122 JACOB AND SAM

discussion forums might nourish a prospective qualitative “Modern Engineering Mathematics” [13]. During the course of
design. the discussion, the instructor moderated, or, scaffolded and
III. RESEARCH STUDY DETAILS modeled the process through Socratic questioning.
The proposed model shown in Fig. 1 was based on the
The study involved the first semester Engineering following principles: (1) Learning is a dialogical process in
Mathematics unit of the Bachelor of Engineering programme which communities of practitioners socially negotiate the
offered by the Swinburne University of Technology (Sarawak meaning of phenomena [14],[15],16]; (2) views of knowledge
Campus) during the year 2009. Out of the total number of 115 are constructed by, rather than transmitted to, the learner [17];
students, 60 participants were chosen by voluntary consent. (3) ill-structured problems and Socratic questioning are stimuli
The number of participants was fixed at 60, since the study was [4],[5] for critical thinking; and (4) asynchronous discussion
an in-depth study involving a lot of variables, especially the forum is a collaborative mind tool [17] which contributes to the
analysis of the online forum postings of individual students. A development of critical thinking and effective communication.
similar study on online forums in mathematics done by Kosiak The model integrates the principles of the model presented by
[11] has used around 50 participants. The participants filled in [15] for the analysis of collaborative knowledge building in
the Consent Form agreed by the Swinburne University Ethics asynchronous discussions.
Committee, as an agreement to their participation in the study.
The participants consisted of a mixture of Malaysian (83%)
and international (17%) students. The 10 international students,
out of the total of 60 participants, comprised students from
Indonesia (1), Iran (1), Iraq (1), Kenya (2), Mainland China (2),
Saudi Arabia (2), and Sri Lanka (1). Generally the students of
this unit are predominantly male. The sample reflects this
typical scenario – it was an uneven mixture of males (85%)
and females (15%). The first year university students go
through a transition stage from school into university. Hence
they need a lot of support in terms of coping up with new
learning environments and taking ownership of their learning.
The online mathematical forums were used as a means of
establishing good rapport between the students and the lecturer,
and between the students themselves. All these students were
new to the experience of using discussion forums in the
educational setting. Figure 1. A proposed model for designing an online mathematical discussion
The unit was supported by the online Learning Management forum environment.
System (LMS) called the Blackboard Learning System
(http://www.blackboard.com), or BBLS. The Discussion The model conceived an ill structured problem or, its
Forums were available on the BBLS. Two problem solving problem space as the focus of the discussion. As seen in Fig. 1,
sessions were planned on these Discussion Forums - that is, ill the fundamental unit of the whole activity is the problem space
structured problems were posted at two different times (in [18], [19] constructed from the ill structured problem by the
Week 4, DF1 and in Week 10, DF2) during the 14 week long students. The creation of a problem space involves mapping
course. The goal of these sessions was for students to work the problem statement from the ill structured problem onto
collaboratively to solve and understand the ideas on these ill- prior knowledge and constructing a personal interpretation of
defined problems [12]. The 60 students were divided into 11 the problem. The process of solving the problem would follow
groups of 5 or 6. The first forum carried a weightage of 7 %, the iterative and cyclical process [19] that ill-structured
and the second carried a weightage of 8%. 15% of the course problems follow. The process would expect the participants to
work marks was assigned for the Discussion forums to articulate goals, verify the problem, relate problem goals to
encourage students into active participation. The instructor problem domain, clarify alternative perspectives, generate
encouraged these collaborative sessions by moderating problem solutions, gather evidence to support/reject positions,
(scaffolding) the discussion thread in order to stimulate both construct arguments, implement and monitor solution. In the
mathematical learning and mathematical understanding. In this process, the participants generated clarification questions,
study, during the first three weeks of the unit, when the assessment questions, inferences, strategies (viz., the four
discussion forum sessions have not yet begun, the students phases of critical thinking stated in the model shown in table 2)
were given access to examples of ill structured problems and and took part in active and collaborative learning. The
solutions. The ill structured problems were chosen by the moderator would pose scaffolding questions [12] to clarify key
lecturer, who is also the moderator of the forum, from a mathematical concepts and to advance/direct the discussions
standard book of engineering application problems called towards the solution.
 PERSPECTIVES ON CRITICAL THINKING THROUGH ONLINE DISCUSSION FORUMS 123

Data was collected in this study using two modes: (1) The section 3 consisting of one open ended question asking for
Discussion Forum postings; (2) Survey Questionnaires- qualitative comments.
perceptions on discussion forums and critical thinking
questionnaire. TABLE II
TABLE I MODEL FOR IDENTIFYING CRITICAL THINKING IN MATHEMATICAL PROBLEM
SCORING TABLE FOR CT CATEGORIES SOLVING SESSIONS OVER DISCUSSION FORUMS

Clarification
Category No. of postings Weightage
Formulates the problem precisely and clearly.
Clarification 1-2 1
Analyses, Identifies one or Identifies Defines or
3-4 1.5 negotiates or more underlying relationships criticizes the
>4 2 discusses the assumptions in among the definition of
Assessment 1-2 3 scope of the the parts of the different parts relevant terms
3-4 3.5 problem problem of the problem
>4 4 Assessment
Inference 1-2 5 Raises vital questions and problems within the problem.
3-4 5.5 Gathers and Provides or asks Make value
>4 6 assesses relevant for reasons that judgment on the
information. proffered assessment
Strategies 1-2 7 evidence is valid criteria or
3-4 7.5 or relevant. argument or
>4 8 situation.
Inference
An ill structured problem was posted on the discussion Reasons out based on relevant criteria and standards
forum by the instructor, students were briefed in detail about Makes Arrives at well Makes Frames
appropriate thought out generalizations relationships
the sessions and given the deadline of one week to solve. The deductions from conclusions from relevant among the
postings were downloaded by the instructor after the sessions discussed results. different parts
were over. The postings were coded into the four categories, results. of the problem.
using the indicators in the model as guides. In cases where Strategies
Thinks and suggests open mindedly within alternative systems of
more than one critical thinking process appeared within a thought.
posting, only one code was associated, which seemed to be the Propose specific Discuss possible Evaluate Predicts
most important in the context. steps to lead to steps. possible steps. outcomes of
The model for content analysis of the postings in the current the solution. proposed
steps.
study was adapted from two models- the model proposed and
tested by Perkins and Murphy [20], and the framework for
assessing critical thinking developed by Paul and Elder [21]. IV. RESEARCH STUDY RESULTS
The new model is shown in Table II [22], with the indicators
The first Discussion Forum, DF1 occurred during Week 4,
and description to each phase/category. The higher levels of
and the second Discussion Forum, DF2 occurred during Week
critical thinking are associated with the stages of Inference and
10 of the 14-week mathematics unit. During the course of the
Strategies.
one week participation, the total number of posted messages in
Also a weighted critical thinking score (CT1 and CT2) was
the discussion forum sessions- namely DF1 and DF2 was 616
associated with every student for DF1 and DF2 respectively,
and 539 respectively, as it occurred in the 11 groups. Fig. 2
based on the classification of the postings, as shown in Table I.
shows the mean number of postings across the 11 groups. As
The score has been developed for the purpose of quantitative
seen from the Fig. 2, participation varied among groups and
analysis based on the CT model. The table was developed with
individuals. The descriptive statistics in table III are compiled
the following principles: (1) The limit of the number of
from DF1 and DF2, hence each variable carries a ‘1’ or ‘2’ in
postings has been set, assuming the average number of
its name, to denote it comes from DF1 or DF2. The arithmetic
postings of a student, per category, for one discussion forum
means in table 3 show that the number of postings was more in
was around three. (2) Higher weights are associated with
the lower phases of critical thinking, namely clarification and
Inference and Strategies, to indicate the higher levels of critical
assessment, than in the higher phases. But the standard
thinking. (3) The maximum weightage of one category differs
deviation (S.D.) and the quartiles (Q1, Q2 and Q3) reveal
by one from the minimum weightage of the next category.
substantial variability in the levels of critical thinking of the
Hence the maximum score possible is 20, if a student has more
students. The quartiles give a better picture of the distribution
than four postings in all categories.
of the variables, since the distribution was not normal.
The perceptions of students were measured using the Critical
Research Question 1a: What is the level of CT of
Thinking and Discussion Forum Questionnaire (CTDFQ). The
individuals in the online discussion forums? Shapiro-Wilk’s
questionnaire had three sections – section 1 on CT, consisting
test for normality done on the variables – the number of
of 4 questions, section 2 on DFs, consisting of 9 questions, and
124 JACOB AND SAM

postings in the four CT categories, and the CT model scores results showed that CT skills measured in the mathematics
(CT1 and CT2), showed that the data values are not normally problem solving sessions were very varied. But, generally CT
distributed (p < 0.05) for all these variables. But the parametric skills were exhibited in the lower phases of critical thinking,
calculations and tests were valid on the data set, since the throughout the two discussion forums. A Pearson correlation
sample size of 60 is large (n >30) enough for the normal proved significant between CT1 and CT2 ((Pearson correlation
assumption of distribution of the data. = 0.644, p < 0.01) at 1% significance level. Or, the CT model
The distribution of the data for the CT categories and the CT scores for the students move linearly between DF1 and DF2.
model scores looked very widely distributed among the batch As students progressed from DF1 to DF2, the students’ CT
of students. One fourth of the students posted no postings in skills exhibited in the problem solving sessions over the DF
the higher phases of CT – namely, inference and strategies, but also showed an improvement.
another one fourth posted more than 4 postings.

Figure 2. Line chart showing the mean total number of postings across the
groups in DF1 and DF2.

TABLE III
DESCRIPTIVE STATISTICS OF THE 60 STUDENTS
Category Mean S.D. Q1 Q2 Q3

Clarification1 2.77 2.580 1 2 4

Assessment1 2.93 2.622 1 2 4.75
Inference1 0.85 1.400 0 0 1
Figure 3. Box plots showing the CT1 and CT2 scores across the 11groups.
Strategies1 0.50 0.948 0 0 1
Unclassified1 1.33 1.980 0 0 2 Research Question 1b: What is the level of CT of groups
in the online discussion forums? The variations were
Clasiification2 2.62 1.958 1 2.5 4
noticeable within the groups, much more than between the
Assessment2 2.50 1.827 1 3 4 groups. In DF1, the CT model scores looked roughly similar
Inference2 0.77 0.909 0 0.5 1 for all the 11 groups – lying between 2 and 18. But there were
Strategies2 0.43 0.810 0 0 1
exceptional students in Groups 2 and 4. The CT model scores
in DF2 show much variation within the group members, than
Unclassified2 0.81 1.181 0 0 1 between the groups. There was at least one student who scored
CT1 8.33 6.045 4 7 12 well above the rest of the group members in Groups 1, 3, 7,
CT2 8.76 5.112 4.5 9.5 14.25 and 11. Also there was a low scorer who was way below the
rest of the group members in Groups 3 and 7. A MANOVA
was employed to test for differences in the four phases of CT
The median (Q2) shows that half of the students posted
and the CT model score between the different groups in DF1
fewer (more) than 2 postings in the lower phases of CT. But
and DF2. Taking these phases as the combined dependent
half of the students have just one or more postings in the higher
variables, the test was found significant between the groups in
phases of CT. The CT scores increase from DF1 to DF2, as
DF1 (F = 1.498, p < 0.05) and DF2 (F = 1.460, p<0.05). But
seen from the median scores of 7 and 9.5 respectively. The
univariate tests revealed that the significant differences were
 PERSPECTIVES ON CRITICAL THINKING THROUGH ONLINE DISCUSSION FORUMS 125

found for the phases, namely clarification (F = 2.743, p < 0.01) DF1 to DF2. The students have positively agreed to the fact
and assessment (F = 2.593, p < 0.01) in DF1 and only for that their attitude to mathematics has been improved as a result
assessment (F = 2.173, p < 0.05) in DF2. A one-way ANOVA of the DFs. The students have disagreed to the comment that
done for the CT model scores, CT1 (F = 1.647) and CT2 (F = moderation did not play an important role.
0.975) between the different groups showed the differences TABLE IV
were not significant at 5% significance level. SAMPLE COMMENTS OF STUDENTS
Research Question 2: What is the effect of moderation Positive comments Negative comments
techniques on the level of CT of individuals in the online
discussion forums? MANOVA was employed to test the Questions connected with daily
Unsystematic posting of threads
life applications boosted my
effects of moderation (scaffold) on the participant’s phases of appreciation towards mathematics.
causes discomfort.
CT and the CT model scores, in DF1 and DF2. The moderation
types were: Type 1- Question scaffolds (Socratic questioning) I get to see different opinions over All group members do not think
only and Type 2- Question scaffolds (Socratic questioning) one problem. Thus I find out more and post at the same pace; So
than one way to approach the some group members might be
plus reflection scaffolds.6 out of the 11 collaborative groups, given problem. discouraged to post any further.
had the type 1 scaffold, and the rest of the groups had the type
2 scaffold. Thus the moderation type, which is a combination It is an easier way to get marks for
Not all members are active.
of Socratic questioning and reflection scaffolds, produced Besides we sometimes could not
the course work.
care less about the Disc. Forums.
significant differences in student postings in the first forum
DF1. The differences were noticed in the phases of It’s good to have it online; Saves No active participation from the
clarification, assessment and strategies. But in DF2, students time/hassle free. group members.
seemed to have become more independent that the postings
didn’t vary much between the two groups. Or, moderation We get idea from others, even if I
But two Discussion Forum
seemed to have less effect as time progressed and students don’t know how to solve the
sessions is too much.
problem.
were used to the pattern of discussions over the DFs.
Research Question 3: What are the student perceptions You explore and think outside the Lack of participation from group
on CT and DFs as they experienced during the 14 week box. It gives a wider picture than members is like a cold blanket,
course? Section 1, covering CT, consists of 3 questions rated what you see and hear in the and disables us from exchanging
classroom. views.
on a 4-point Likert scale: Strongly Agree (1) to Strongly
Disagree (4). The 4-point scale was chosen in favour of the
The questions are great and help
usual 5-point scale since the researcher does not believe in engineering students to see how
I am not a fan of team work for
“neutral” people [23] walking about with no positive or problem solving. I prefer to work
simple mathematics could be used
individually than in a team.
negative response towards the issue in hand. There was a in applications.
fourth question asking for the student’s self evaluation of
his/her own CT rated on a scale of Excellent (4) to Poor (1). It makes me excited, and I will My group members did not co-
not give up without getting the operate well. It looks like I am
Student perceptions have favoured the fact that CT was answers; I like critical thinking. discussing with the lecturer.
understood as a result of doing the particular mathematics unit.
The time frame is too short. I need
Group work was also seen as helping CT skills. The argument It increases my group work skills,
time for solving any question
critical thinking skills. Can share
that CT skills became better from DF1 to DF2 showed a mean knowledge with friends.
because the main problem with
closer to 3. Majority of the students were echoing the fact that me is how to begin.
their CT skills (as they understand) became better from Forum
1 to Forum 2. It is noticed from the results of research question But they have agreed that the moderator modeled by posing
1 that the CT model scores also reflected the same result- there appropriate discussions. Nearly 40% of the students sounded
was seen a general progress in CT skills from DF1 to DF2.The that they participated simply because of the course work marks
students’ rating of their own CT was an average (the mean allotted for participation in the DF sessions, and that
value is close to 2) on a scale of 1-4. Majority of the students participation was a demanding task. This was reflected in some
saw themselves as belonging to the class of average critical of the negative comments to the open questions listed later. But
thinkers. the majority of students have participated with a more positive
Section 2 of the CTDFQ dealt with the DFs and the problem mind, and enjoyed the online sessions which made them think
solving sessions. There were 9 questions in this section – 5 out of the box. The rating of the DF sessions on the Black
positively worded statements, 3 negatively worded statements Board was also a pleasant mixture of ‘Good’ and ‘Average’.
and 1 question to rate the DF problem solving sessions But the mean score pointed to the fact that the sessions were
activated on the Blackboard Learning System. Majority of the rated good.
responses were positively favouring the DF sessions as an Section 3 consisted of an open ended question asking for
excellent way to promote CT. They did agree to the fact that positive and negative comments on the experience of students
the discussion and pace of the DF sessions became better from with the DFs.
126 JACOB AND SAM

Table IV shows a set of positive and negative comments [5] Y-T.C Yang and H-A. Chou, “Beyond critical thinking skills:
Investigating the relationship between critical thinking skills and
made by a few students. It was clear from the responses that dispositions through different online instructional strategies,” British
these students have never been exposed to such problem Journal of Educational Technology, vol. 39 (4), pp. 666-684, 2008.
solving sessions in mathematics. Team work skills, innovation, [6] H. Innabi, H. and O.E. Sheikh, “The Change In Mathematics Teachers’
Perceptions Of Critical Thinking After 15 Years Of Educational Reform
sharing with friends, and the fun of being free to think beyond In Jordan,” Educational Studies in Mathematics, vol. 64, pp. 45 – 68,
limitations were seen as the positive sides of the DF sessions. 2006.
But, being first semester students who were new to the [7] G. E. Perry, “Critical Thinking and Web-Based Education in Maryland
Community Colleges: How the Medium Promotes Development”.
university environment, some of them felt that the sessions Unpublished doctoral dissertation, University of West Virginia, in Bers,
required a lot from them on top of their packed schedules. Also T. H., Assessing Critical Thinking in Community Colleges, New
lack of response from their team mates, preference for Directions For Community Colleges, vol. 130, Wiley periodicals, Inc.,
pp. 15-25, 2005.
individual work, short time frame were the reasons for the [8] M. Guzdial, P. Ludovice, M. Realff, T. Morley, K. Caroll, and A. Ladak,
negativity of the students. The challenge of collaborative learning in engineering and math.,
Frontiers in Education Conference, 2001, 31st Annual, pp. T3B-24-29,
V. CONCLUSIONS 2001. Retrieved January 2008 from http://coweb.cc.gatech.edu/cs1.
[9] T. Karsenti, “Teacher Education and Technology: Strengths and
The idea of measuring critical thinking (CT) skills of students Weaknesses of Two Communication Tools,” Proceedings of the 2007
through the online Discussion Forums, proved to be very useful to Computer Science and IT Education Conference, 2007. Retrieved
the instructor, especially in the context of a large class (more than January 2008 from http://csited.org/2007/83KarsCSITEd.pdf.
[10] M.D. Smet, H.V. Keer, and M. Valcke, “Cross-age peer tutors in
50 students), where it is difficult to assess the thinking skills, asynchronous discussion groups: A study of the evolution in tutor
personally or one to one. The instructor could classify the support,” Instructional Science, 37(1), pp. 87-105, 2009.
students’ postings based on their thinking skills and thus give [11] J. J. Kosiak, “Using Asynchronous Discussions To Facilitate
Collaborative Problem Solving In College Algebra”, Doctoral
feedback to the students. In general, the majority of the postings Dissertation, Montana State University, 2004.
fell into the lower categories of CT, as was the case noticed by [12] M-S. Chiu, “Approaches to the teaching of creative and non-creative
Lee [1] and Kosiak [11]. But the CT skills have improved from mathematical problems,” International Journal of Science and
mathematics education, vol. 7(1), pp. 55-79, 2009.
the first session to the second, as shown by the CT model scores. [13] G. James, Advanced Engineering Mathematics (3rd Edition). Pearson
But there were students who were ahead of the others, considered Prentice Hall, 2004.
as outliers. The moderation process was very much necessary in [14] M. Mabrito, “A Study of Synchronous Versus Asynchronous
Collaboration in an Online Business Writing Class,” The American
the beginning stages, but during the second session, students Journal of Distance Education, vol. 20(2), pp. 93-107, 2006.
seemed to be confident to move on by asking, assessing and [15] S. Schrire, "Knowledge building in asynchronous discussion groups:
discussing among the group members. The students perceived the going beyond quantitative analysis", Computers & Education, vol. 46(1),
pp. 49 – 70, 2006.
forums as a boon to work on their CT, problem solving skills and [16] L. S. Vygotsky, Mind in society: The development of higher
majority of the students appreciated the innovation and challenge physiological processes. Cambridge, MA: Harvard University Press.
offered by the DF sessions. Students also gave the researchers [17] S. Schrire, “Interaction and cognition in asynchronous computer
conferencing,” Instructional Science, vol. 32, pp. 475-502, 2004.
enough feedback to work on. [18] A. E. Campbell, V. M. Adams, and G. E. Davis, “Cognitive Demands
and Second-Language Learners: A Framework for Analyzing
REFERENCES Mathematics Instructional Contexts,” Mathematical Thinking & Learning,
[1] K.S. Lee, “Effects of Individual versus Online Collaborative Case Study vol. 9(1), pp. 3-30, 2007.
Learning Strategies on Critical Thinking of Undergraduate Students. [19] S. Sendag, and H.F.Odabasi, “Effects of an online problem based
Doctoral Dissertation,” The University of Texas at Austin, 2004. learning course on content knowledge acquisition and critical thinking
[2] T. D. Erwin, “Definitions and assessment methods for critical skills,” Computers & Education, vol. 53, pp. 132-141.
thinking,problem solving, and writing,” in T. Bers, Assessing critical [20] C. Perkins and E. Murphy, “Identifying and measuring individual
thinking in community colleges, New Directions for Community Colleges, engagement in critical thinking in online discussions: An exploratory
vol. 130, pp. 15-25, 2005. case study,” Educational Technology and Society, vol. 9(1), pp. 298-307,
[3] S. VanderStoep and P. Pintrich, Learning to learn: The skill and will of 2006.
college success. Upper Saddle River, NJ: Prentice Hall, 2003. [21] R. Paul and L. Elder, Critical Thinking: Learn the Tools the Best
[4] Y-T.C.Yang, T.J. Newby, and R.L. Bill, “Using Socratic Questioning to Thinkers Use. Prentice Hall, 2006.
promote critical thinking skills in asynchronous discussion forums in [22] S. M. Jacob and H.K. Sam. “Effect of Online Discussion Forums to
distance learning environments,” American Journal of Distance Improve Mathematics Problem Solving and Critical Thinking – A
Education, vol. 19(3), pp. 163-181, 2005. Comparative Study,” Proceedings of the 5th International Conference on
Innovations in Information Technology (Innovations 2008), United Arab
Emirates, ISBN 978-1-4244-3397-1, pp. 366-370, 2008.
[23] M.G. Lodico, D.T. Spaulding, and K. H. Voegtle. Methods in
Educational Research: From Theory to Practice [E-Book], 2006.
 Digital System Description Knowledge Assessment
K. Jelemenská, P. iák, M. Jurikovi and P. Pištek
Institute of Computer Systems and Networks, Faculty of Informatics and Information Technologies
Slovak University of Technology Bratislava
Ilkoviova 3
842 16 Bratislava, Slovakia
[email protected]

Abstract— The paper presents a knowledge assessment approach worse when distant learning is considered, where teachers have
developed for support of Digital system description course. In the do not have the possibility to check out the students
course the students should comprehend the methods and understanding of the solutions.
techniques used in digital system design, and gain the skills in The second part of the course assessment is usually formed
digital systems modelling using hardware description languages by an exam, sometime supplemented by some midterm tests.
(HDL). For many years the course exams used to be done only in The tests and exams are often performed in a traditional way –
pen and paper form. The students hated writing HDL programs using pen and paper. Although this type of exam gives
on paper and the teachers disliked correcting the programs on minimum chance for cheating, there are many reasons why this
paper as well. What is more, this type of exam has not allowed
type of exam is not suitable for practical skills assessment.
the verification of the student’s ability to debug the created
First, the students have no chance to check the correctness of
model, which is the substantial part of the student’s practical
their programs or HDL models before handing them out for
skills. That is why, some years ago, we concentrated our work on
evaluation. Secondly, this type of exam does not allow
designing a knowledge assessment system that will enable reliable
evaluation of practical skills in the area of digital system
verification of the student’s ability to debug the created
description without substantial teacher involvement. Several
program/model, which is a substantial part of student’s
applications have been developed and tested in the course. One of practical skills. Finally, hand written works are very difficult to
the best will be presented here. read, correct, and evaluate. It is very demanding and time
consuming work for teachers to evaluate programs/models
written on paper and the result is often uncertain. In order to
I. INTRODUCTION make the course assessment process more objective, fair and
effective, there was an urgent need for redesign of the
Knowledge and skills assessment has always been problem assessment. One of the solutions assumes that the substantial
in the courses devoted to programming languages, whether it part of the course assessment will be shifted from the
was assembly or higher level programming languages [3,4]. assignments to the skills based midterm tests. At the same time,
The problem is very similar in the case of hardware new technology will be developed to perform the exam and
description languages (HDL). To master the programming or midterm tests. The new applications, specially designed for this
modeling language, the students should solve various problems purpose, will enable the testing of not only the knowledge, but
using the language. It is useless to learn language constructs or also the skills used, all in an automated manner.
exact syntax by heart without writing any program. That is why
the programming courses assessment process often adopt the
II. LEARNING MANAGEMENT SYSTEMS CAPABILITIES
following schema.
To ease the work needed for knowledge and skills
The lab exercises are based on more or less individually
assessment in the course, various Content Management
allocated assignments that the students can solve at school or at
Systems (CMS), Learning Management Systems (LMS),
home. The solutions are evaluated by teachers and usually form
Computer-Based Training (CBT) or Web-Based Training
a part of the overall course assessment. Unfortunately, there are
(WBT) systems can be used which support automated
always some students who try to find the easiest way to get the
knowledge and/or assignments assessment. Nowadays there
lab work results (e.g. search for a complete solution on the
are a vast number of such systems, but most of them are unable
internet, copy the solutions from their classmates, etc.). The lab
to evaluate a student‘s skills using an online test.
teachers usually have no chance to find out whether the
students really did the programming work themselves, or not. Various systems supporting knowledge testing and/or
They can just check up, whether the students understand their assignment assessment have been reviewed in [2]. The
solutions and whether they are able to perform some minor emphasis was placed on supported questions, test types and
changes in them. But it is not an easy task to do. Another automated assessment. The existing systems can be divided
problem is that different teachers might use various scales for into two groups.
assignments evaluation. As a result, many students do not get Test Creation Tools: These tools represent standalone
the required knowledge and practical skills, even though they solutions. The tests created using these tools can not usually be
get the highest evaluation for lab assignments. This part of the connected to learning content and must be used en bloc. The
course assessment is often neither objective nor realistic created tests conform to e-learning standards and therefore can
although requiring a lot of teachers’ work. The situation is even be integrated to LMS. Another advantage is that these systems

127
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_22, © Springer Science+Business Media B.V. 2010
128 JELEMENSKÁ ET AL.

TABLE I. III. VHDL MOODLE MODULE DESIGN

ASSIGNMENT AND QUESTION TYPES SUPPORT IN REVIEWED SYSTEMS
A. Functional Requirements

Macromedia Authorware 7
There are three types of users in the module – student

Toolbook Instructor
teacher and system.

Program Autor

HotPotatoes
WebToTest
1) Student: Student is the user with minimum rights. He or

QUIZIT

Moodle
Dokeos
eDoceo

ATutor

Ilias
she can do only actions allowed by users with higher rights
and therefore cannot modify functionality of the system.
Authentication – Students must log in to the system using
the Moodle built in authentication system. They will have
access to the Digital system description course, where a VHDL
Assignment management - - + - - - + - + - +
test is created as a special activity.
Question types Source code editing – A test consists of test assignment and
Program + - - - - - - - - - - VHDL source code editing windows, which can contain a
program core created by a teacher. Comfortable work during
Multi-choice single the test will be provided by VHDL syntax highlighting, similar
- + + - + + + + + + +
answer
to that used in ModelSim.
Multi-choice multi
- + + - + + + + + + + Source code compilation – Students can compile their source
answer codes unlimited number of times. The compilation errors are
True/False - + + - + + + - + + + returned to students by the system.
Gap filling - + + - + + + + + + +
Source code simulation – The compiled model can then be
simulated, using the test bench, prepared by the student or by
Matching - + + - - - - + + + - the teacher. The student should receive the signals waveforms
Slider - - + - - - - - - - - as an output.
Arrange objects - - + - - - - - + + - Source code submission – The students have the possibility
Place object
to upload their partial solutions to the server any time they
- - + - - + - - - + -
wish. After the final submission, the system will further
Identify correct picture process the student’s solution to get it evaluated automatically.
- - - - + + - - + - -
region The final submission of the source code is activated
Identify correct object - - - - - + - - - - - automatically in case the time allocated for the test elapses.
Crossword - - - - - - - - - + - Results review – Students can see the results of all their
+ =YES - = NO VHDL tests in the separated module.
2) Teacher: The teacher can administrate his/her teaching
module and adapt it to his/her own needs (i.e. adding teaching
are better elaborated and support more question types, and test aids, activities, tests, creating forums etc.).
creation is easier and flexible. The VHDL test will be created adding the special new
Learning Management Systems (LMS): These systems activity into the course. There are several activity attributes that
support e-learning and provide assignment management can be set during this process.
subsystems as well as testing subsystems. All considered LMS Instance name – Using this name the teacher can locate and
support the e-learning standards SCORM, ADL and IMS. download or review the results of the respective activity.
The capabilities of the reviewed systems are summarized in General instructions – These instructions are optional and
Table I. As we can see most of the reviewed systems support can be used to display any remarks made concerning the test.
several types of test questions but only WebToTest enables Start of the test – This attribute can be used to schedule the
students to edit, compile, and debug a program and submit it test at any date and time. The test will not be accessible to
using a special form. We were not able to find any system students before this time.
supporting skills based assessment in the area of HDL
modelling. This fact inspired us to develop several skills Test duration – The teacher has to set up the time that the
assessment systems supporting languages like VHDL and students can spend working on the test. After this time the final
SystemC. source code submission, followed by test evaluation is
activated automatically.
The standalone testing application presented in [5,12] and Test assignment – The teacher creates a text form of
the Drag&Drop Moodle module [13] that have been used for assignment saying what the students should design.
student’s knowledge and skills assessment in the last academic
year are two examples of them. The special VHDL testing Program core - The teacher can copy parts of the solution to
module, presented in this paper, is one of the latest solutions, the core which will be displayed in student’s VHDL source
suitable for online skills based examination in the area of code editing window. Typically at least the entity declaration of
digital systems modelling using VHDL. The module integrated the designed circuit is provided to make the student’s design
into the Moodle LMS is already a part of the set of assessment consistent with the reference. Also, the teacher’s test bench
tools used in the Digital system description course. can be provided here, in case it should be made visible to
students. Although this part is voluntary, if the teacher doesn’t
 DIGITAL SYSTEM DESCRIPTION KNOWLEDGE ASSESSMENT 129

use it, the system will not usually be able to do automated

evaluation.
Reference test bench – This test bench will be used to
replace the student’s one to make the final simulation after the
final source code submission.
Reference solution – The complete solution including
reference test bench should be provided here. This solution is
used to generate the reference waveform that will be used for
automated evaluation. The reference solution can also be used
to make some source code comparisons.
Other parameters of the test include maximum number of
points, difference penalty, syntactical correctness gain, and the
figure to be displayed.
After the test, the teacher can use another module to review
the solutions, eventually to alter their evaluation. There is also
the possibility to download all the solutions of the selected test
for further revisions or archiving.
3) System: Automated tasks are realized by the system. The
rights must carefully be chosen to prevent system malfunction.
Source code storage – The system must guarantee the
student’s solution storage any time it is required to do so. The
source code must be stored correctly even though it may
contain special characters such as apostrophes, etc.
Source code compilation – Source code is compiled each Fig.1. The system architecture
time a student requires it. The actual compilation is done
invoking the vcom compiler, an integrated part of ModelSim.
ModelSim is placed on the server. It is called by a PHP
Source code simulation – The compiled model is simulated application using a console mode and provided source code as
each time a student requires it. Simulation is done invoking the an input. ModelSim processes the provided input and returns
vsim simulator, an integrated part of ModelSim. the output to the PHP application. Using this solution the
Final source code submission – This process is activated by problem of server overloading might arise in the case of too
the system either automatically in case the time allocated to the many users. We can then try processing the requirements in
test elapses, or interactively by the student’s command. As part parallel using, for example, a connected cluster of processors.
of this process the student’s test bench will be replaced by the
teacher’s, and the test will be automatically evaluated, TABLE II.
comparing the resulting waveform to the reference. For each TABLE MDL_VHDL: STORES DATA FOR MODULE INSTANCES
difference the assigned penalty is subtracted from maximum
number of points. In case the source code can be compiled, but Name Description
there are too many differences, a syntactical correctness gain is
returned as a result. In case the source code is not syntactically ID Identifier of module instance
correct, it can still be compared to the reference solution to find Course Identifier of course where instance belongs
some partial similarities.
Name Name of a course where instance belongs
B. System Architecture
Nadpis Assessment title
A client-server architecture model was used as shown in Fig.1.
The LMS Moodle, together with the new module, are installed Intro Opening instructions and test directives
on the server. This web-based application requires a web server Zaciatok_testu Time in epoch format when the test will start
– in this case the Apache HTTP Server is used. A database
server is also required. Although several types of databases can Dlzka_testu Test duration in minutes
be used, the MySQL database is probably best supported. On
Text_zadania Text of assessment
the client side, a web browser is used to communicate with the
server using a specially developed user interface. Kostra_kodu Source code scheme in VHDL
The most important module functions are source code syntax Testovacia_vzorka
Test sample which will be used for
verification and design simulation. To compile and simulate compilation and simulation
VHDL code, the commercial software ModelSim from Mentor Name of the picture which will be used
Obrazok
Graphics Company is used. The compiler and the simulator can during test
be launched from the command line and therefore initiated Pocet_bodov Maximum number of points
from another application. The vcom output - compilation errors Bodova_zrazka Points discount for one difference
- can be redirected into the file. The output of vsim is a .wlf
Base number of points which are given to
format file, containing simulation results. Body_kompilacia
student in the case of successful compilation
130 JELEMENSKÁ ET AL.

IV. VHDL MOODLE MODULE IMPLEMENTATION

System implementation uses the MySQL database. This
database is used by the Moodle system as a storage place for all
data except files, stored on disc. The VHDL testing module
uses data from this database as well. However, some new
database tables had to be created to store data associated with
this module.
A. VHDL Testing Module
Two new database tables were created in order to store data
associated with this module: mdl_vhdl in Table II. stores data
for module instances, and mdl_vhdl_odovzdane (shown in
Table III.) stores the source code of particular students. The
fields identify the specific test and student, as well as the full
source code and results of the evaluation. The first names and
surnames of students are stored for test archiving purposes –
even if the Moodle user database is cleared, the complete tests
are still available.
B. Module for Reviewing Submitted Assignments
This module also needs to have its own table, where all
created instances are stored. The table is called mdl_prehlad
and contains only 3 items: ID, Course, and Name.
No other tables are necessary since the module uses data from
the mdl_vhdl_odovzdane table, which holds all of the vhdl code
submitted by students.
C. Shell Script for Compilation and Simulation
To compile and simulate the VHDL model a shell script is
called from PHP application. It consists of three main branches.
1) The compilation branch creates a new subdirectory in the
directory /tmp/data/, named using user ID. In the subdirectory,
the library work and file source.vhd are created. Thereafter, the
file is compiled with the vcom compiler. The compiler output is
returned to PHP application.

TABLE III.
TABLE MDL_VHDL_ODOVZDANE: STORES SUBMITTED SOLUTIONS

Name Description

ID Identifier of submission

Course Identifier of course which was used for examination

Fig.2. The correction algorithm
Test Test name which was submit
2) The simulation branch is a bit more complicated. At first,
User_id Identifier of user who submitted the test the existence of library work is checked. If it is there, the
simulation .do script is dynamically created. This script is sent
Zadanie Assignment of the test which was solved by user
to the vsim simulator and causes the test bench simulation to be
Kod Code which was sent by user run. It is necessary to name the test bench “TESTBNCH”. If
simulation runs without errors, the reference and student’s
Cas_zacatia Time in epoch format when the test was started waveforms (simulation outputs) are compared. The comparison
Cas_odovzd
Time in epoch format when the test was submitted .do script is dynamically created for this purpose and sent to the
ania vsim simulator. The outputs of the simulation and the
Hodnotenie Points evaluation of student comparison are returned to the application.
Result Results for compilation and simulation 3) The final evaluation branch cleans up all the temporary
files created during the test. Then the file source.vhd is created
Meno First name of student. once more, using the reference test bench in this case. The file
Priezvisko Student’s surname. is compiled and simulated analogously. The outputs are then
returned to the PHP application where they are processed and
 DIGITAL SYSTEM DESCRIPTION KNOWLEDGE ASSESSMENT 131

the results are calculated. The algorithm for the evaluation of similar to that used by ModelSim (used during the term), is
student solutions is given in Fig. 2. used to make the editor familiar to students.
D. Test Creation Shell Script
After the instance of the testing activity is created by the
teacher, a reference waveform in the form of .wlf file is
automatically generated. A shell script is used for this purpose,
which is called after the test creation form is sent. This script
compiles and simulates the reference solution entered by the
teacher. The output of the simulation, a .wlf file, is then stored
together with the other files associated with this testing activity
instance (like e.g. image). They are all stored in the directory
moodledata/courseid/instanceid.

V. VHDL MOODLE MODULE USER INTERFACE

A. New Instance of VHDL Test Creation

The screenshot given in Fig. 3 represents a part of the test
creation form. This form is only accessible to the teacher or
administrator. The teacher sets all the test attributes here - for
example, duration of test, test bench, reference solution, etc.
B. VHDL Test Elaboration
The screenshot shown in Fig. 4 illustrates the student
interface to the designed module. Shown at the top of the
screen are the remaining time, the test assignment text, and an
optional scheme. Under these parts is the VHDL source code
editing window, where the provided core should be completed
using the student’s VHDL code. VHDL syntax highlighting,
Fig. 3. The teacher screen – Test creation

Fig.4. The student screen – Test elaboration

132 JELEMENSKÁ ET AL.

VI. RESULTS AND EXPERIENCE disposal a source code editor with syntax highlighting, a
We started to use the new testing module in the course VHDL language compiler, and simulator outputs. The
Digital systems description to evaluate students’ practical skills application gives the teacher a means to comfortably manage
during midterm tests. In the course, the students will pass three the testing system. Finally, the students have the opportunity to
midterm tests and one final exam. On each midterm test, the verify their designs before sending them for evaluation.
students are divided into three groups based on the testing
environment - TabletPC, Drag&Drop, or VHDL Moodle test. ACKNOWLEDGMENT
Each student will pass each type of test once. However, the test The support by Slovak Science Grant Agency (VEGA
assignment is the same, regardless of the test type. This 1/0649/09 “Security and reliability in distributed computer
constraint ensures that the severity of all the types of the test systems and mobile computer networks“) and HP Technology
should be more or less the same. Fig. 5 shows the results of the for Teaching grant 2007 “Mobile Education Center” is
first and the second midterm test - the distribution of Average gratefully acknowledged.
Test Attainment. As we can see, the students reached the best
attainment in the Drag&Drop test. The reason is that in this REFERENCES
case the correct solution is already given in the test and the [1] M. Mayer, “Distance Learning Support for Digital Systems Description
students just have to find the correct statements and place them Course,” master theses, FIIT STU Bratislava, 2009.
correctly into the VHDL model. [2] P. Polako, “Learning Support for Specification and Description
Languages Course,” master theses, FIIT STU Bratislava, 2007, pp. 86
[3] Ch. Douce, D. Livingstone and J. Orwell, “Automatic Test-Based
Assesssment of Programming,” ACM Journal of Educational Resources
in Computing,Vol. 5, No 2, September 2005
[4] C. A. Higgins, G. Gray, Pavlas Symeonidis and A. Tsintsifas,
“Automated Assessment and Experience of Teaching Programming,”
ACM Journal of Educational Resources in Computing,Vol. 5, No 3,
September 2005
[5] Š. Beleš, K. Jelemenská, M. Jurikovi, P. Pištek, T. Polák, J. Zeman and
M. Žucha, “Mobile Testing System,” Košice : Elfa, 2008, In: ICETA
Fig. 5. The first and the second midterm test results 2008 : 6th International Conference on Emerging e-Learning
Technologies and Applications. High Tatras, Slovakia, September 11-
After the test students were asked several key questions. The 13, 2008. - Košice : Elfa, 2008. - ISBN 978-80-8086-089-9. - pp. 133-
answers to these questions are shown in Fig. 6. As we can see, 138
most of the students think that the Drag&Drop test, composing [6] Š. Beleš, M. Jurikovi, P. Pištek, T. Polák, J. Zeman, and M. Žucha –
a model from given statements, is the best way of testing “Learning Support for Specification and Description Languages,” Team
practical skills. Project, FIIT STU Bratislava, 2008, pp. 83
[7] J. Kytka, “Learning Support for Specification and Description
Languages Course,” master theses, FIIT STU Bratislava, 2006, pp. 85
[8] N. Pollock, J. Cornford, “Theory and Practice of the Virtual University,
report on UK universities use of new technologies,”
http://www.ariadne.ac.uk/, Jun-2000
[9] Bruggeman, J., Long, H., M., Misra, C.: Effective IT Security Practices
(ID: EDU06165), Presented at EDUCAUSE Annual Conferences
(10/09/2006),
[10] Brown, M., B., Smith, J., M., Windham, C.: From Today's CMS to
Tomorrow's Learning Support System, Presented at EDUCAUSE
Annual Conferences (10/11/2006), WINNER: 2006 EDUCAUSE
Fig. 6. Survey results Catalyst Award.
[11] Hemmi, A., Pollock, N., Schwarz, C.: If not the Virtual University then
what? Co-producing E-learning and Configuring its Users, 8th European
VII. CONCLUSION Conference on Media in Higher Education, (Best Overall Paper), GMW,
University of Duisburg, 16-19 September 2003.
The first results prove that the kind of online test described [12] K. Jelemenská, M. Jurikovi, P. Pištek and P. iák, “Automated Skills
previously provides more realistic image of student knowledge Assessment in Mobile Environment,” in: Research, Reflections and
of digital systems modelling, compared to tests based on the Innovations in Integrating ICT in Education, Vol.1, V.International
types of questions commonly used in online tests. Because the Conference on Multimedia and Information & Communication
Technologies in Education (m-ICTE 2009), Lisbon, Portugal : Vol. 1. -
VHDL Moodle test requires creativity by students, the test Badajoz : Formatex, 2009. - ISBN 978-84-692-1789-4. - pp. 487-491
results are worse compared to the Drag&Drop test. [13] K. Jelemenská, E. Tomalová, and P. iák, Pavel, “Knowledge
The presented testing solution brings number of advantages. Assessment Improvement in Moodle Environment,” in: Research,
First, there is a substantial reduction in the demanding and time Reflections and Innovations in Integrating ICT in Education, Vol. 2,
consuming work of the teacher, relative to paper exams V.International Conference on Multimedia and Information &
Communications Technologies in Education (m-ICTE 2009, Lisbon,
correction. At the same time, the level of difficulty of the exam Portugal. - Badajoz : Formatex, 2009. - ISBN 978-84-692-1790-0. - pp.
is preserved. Second, it enables the teacher to check the skill of 979-983
students in the area of model debugging, which was previously [14] J. Geni, “Knowledge Assessment – Practical Example in Testing,” In:
not possible. All participants in the examination have at their Journal of Communication and Computer, Vol. 6, 8, 2009, pp. 65-69,
ISSN 1548-7709
 Wikibooks as Tools for Promoting Constructivist
Learning in Higher Education: Findings from a Case Study
Ilias Karasavvidis
Department of Preschool Education
School of the Humanities
University of Thessaly
[email protected]

Abstract - The present paper focuses on wiki-related pedagogies all sorts of content. In fact, most of the typical web 2.0 tools do
in higher education and examines the potential of wikibooks in a not only rely on user participation but would actually be incon-
blended learning context. A wikibook was integrated as a com- ceivable without it. For example, web sites such as Wikipedia
pulsory class assignment in an undergraduate course in a pre- and YouTube would not exist without user-generated articles
school education department. Twenty seven students partici-
pated in the study which investigated (a) the extent of student
and videos respectively. Consequently, the first generation of the
participation in the wikibook and (b) students' perspectives on web, now called Web 1.0, has been transformed into a new, par-
the contribution of the wikibook to their learning. In stark con- ticipatory, read-write web which is known as Web 2.0. The most
trast to findings reported by other studies, results indicated that distinctive feature of Web 2.0 is that it relies on the wisdom of
the rate of student participation was very high as suggested by crowds [3] which in turn leads to crowdsourcing and wikinomics
indices such as page views, page edits, and number of students [4].
involved per page edit. Moreover, the students rated positively
wikibooks' contribution to the learning of the course concepts B. Web 2.0 and E-Learning 2.0
and the active mode of learning required. However, the students The impact of Web 2.0 technologies has been substantial [5].
also mentioned significant problems with the wikibook activity E-learning is one of the areas of education where Web 2.0 has
such as difficulty understanding certain sections and concerns
with the validity of their writings. The paper is concluded with a
had a major impact [6]. Terms such as Education 2.0, School
discussion of the findings and implications for further research. 2.0, Classroom 2.0, and Learning 2.0 are becoming common-
place.
Keywords – Wiki, Wikibook, Web 2.0, Collaborative The importance of Web 2.0 for enhancing learning in higher
Learning education is being increasingly recognized. Its significance lies
in that it can address some of the problems which characterized
I. INTRODUCTION several E-learning implementations. One of the most often
voiced criticisms is related to the adopted e-learning pedagogy.
A. The second generation of the world wide web More specifically, it has been argued that many E-learning envi-
Web 2.0 has attracted considerable attention over the past ronments are content-driven and are being mostly used to deliver
few years. Currently, the emphasis is on collective intelli- course materials on-line ([7], [8], [9]). Such transmissionist ap-
gence, the wisdom of the crowds, wikinomics, distributed proaches, which are dominant in E-learning, lead to the replica-
asynchronous collaboration, social networks and networking, tion of traditional modes of teaching and learning on-line [10].
social software, massively multiplayer on-line games, and As opposed to traditional E-learning, the role of students in
virtual worlds. According to [1], Web 2.0 can be defined in a the new E-learning 2.0 era is more active and participatory. The
number of ways. On a technical level, Web 2.0 tools such as students are not consumers of material which has been compiled
blogs, wikis, podcasts, social bookmarking, photo sharing, by instructors ([9], [5]). One of the major tenets of Constructiv-
instant messaging, VoIP applications, RSS feeds, mashups etc ism, the current dominant learning paradigm in higher education,
afford more possibilities for communication and collaboration is that learning can only be achieved through the active student
compared to Web 1.0 tools. For example, wikis have consi- participation in the learning process ([11], [12], [13]). This ac-
derable advantages over threaded discussions [2] and blogs tive participation by learners is perfectly compatible with the
have more affordances compared to Content Management notion of active participation by users in the context of Web 2.0.
Systems (CMS). Regardless of technical differences, howev- Thus, the affordances of Web 2.0 technologies are in line with
er, what sets apart Web 2.0 from the first generation of the the active approach to learning which constructivism requires,
Web is the role users play. which constitutes the added value of E-learning 2.0 from a learn-
In the first generation of the world wide web users were ing perspective. Web 2.0 technologies can help transform tradi-
simply browsing information which was provided on web tional higher education practices which are based on objectivist
pages. In the case of Web 2.0 the users are not simply con- notions of learning and are characterized by learning metaphors
sumers of information available on web sites. Instead, given that are transmisionist in nature.
the architecture of web 2.0, the users are actively contributing

133
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_23, © Springer Science+Business Media B.V. 2010
134 KARASAVVIDIS

C. Wikis ically, it has been reported that wikis can facilitate group work
Much like all other Web 2.0 tools, wikis have attracted [16], collaboration [17], collaborative learning [18], inquiry
widespread attention. ''WikiWiki'' is a Hawaian word that learning and co-reflection [19], and peer assessment [20]. Re-
means quick. According to [14], a wiki is a “freely expanda- garding performance gains, there is some evidence to suggest
ble collection of interlinked webpages, a hypertext system for that the use of wikis can lead to improved course performance
storing and modifying information – a database, where each (see e.g. [21], [22]) and improved writing and critical thinking
page is easily edited by any user with a forms-capable Web skills [23]. Finally, students appear to have positive attitudes
browser client” (p. 14). Essentially, a wiki system constitutes towards wiki use ([24], [25]) and find their wiki experiences to
a collection of webpages with information which can, in prin- be valuable for their learning ([22]; [17]).
ciple, be edited by anyone. Users can add content to existing Despite positive findings, significant challenges have been
pages, restructure the information on existing pages as well as identified: (a) minimal student participation ([26], [27], [28],
create new pages and create links between existing and new [21]), (b) limited on-line interaction ([23]; [29], [30], [22]), (c)
pages. The fact that everyone can create, edit and delete a preference for individual work [17], and (d) negative student
page means that wikis have a very open-ended nature. Cun- perceptions of wiki's contribution to their learning ([26]; [30];
ningham developed the first wiki, WikiWikiWeb, in 1995 and [22]). While this brief literature review does suggest that wikis
there are more than 200 clones today. hold great promise for higher education, one of the critical issues
The popularity of wikis is due to their simple and flexible which remain to be addressed is related to pedagogy. There is a
nature. Firstly, wikis are characterized by a non-linear hyper- knowledge gap regarding how wikis can be best incorporated
text structure: a wiki enables the creation of associative into the teaching and learning practices of higher education. The
hypertexts where each page contains links to other pages and evidence from existing research clearly suggests that the suc-
the user can choose which links to follow. Secondly, wikis cessful wiki integration into a course requires mildly put a tho-
are characterized by ease of access since all a user needs to rough redesign ([21], [28]).
use a wiki is a web browser. As a web application the wiki E. Focus of the study
does not require any specific browser or operating system
software. Thirdly, the technical skills required for participa- The present study aimed to fill this knowledge gap in wiki-
tion are minimal as users do not have to learn HTML to create related pedagogies in higher education. The specific focus in this
a webpage. Thanks to wiki markup, a simplified version of study is wikibooks, one of the many possible uses of wikis in
HTML, and the built-in WYSIWYG editors, the burden for university courses.
creating and formatting a webpage is significantly reduced. Wikibooks can be considered as class assignments whose
Finally, a wiki system also provides other functions such as main objective is to engage students in the collective writing of
(a) tracking of edits through a logging and notification sys- an electronic textbook. This effectively combines the active ap-
tem, (b) comparison between different versions of edits, (c) proach to learning which is required by constructivism and the
rollbacks to earlier versions, (d) discussions per wiki page and unique potential of wikis to be used both (a) as process tools to
(e) customizable read and edit access rights as well as pro- facilitate collaboration and (b) product tools to enable the crea-
tected pages. tion of a shared artifact such as a textbook.
Wikipedia, the prime example of wiki success, constitutes The potential of wikibooks has been researched in a number
an indication of how popular wikis have become. of recent studies in higher education settings (see e.g. [31], [22],
[32], [33], [34]). The typical use of wikis in those studies in-
D. Wikis in Education volves the writing of textbooks by students as part of the stan-
Wikis can be used to support a wide range of activities in- dard course assignments. As a rule, the findings reported by
cluding but not limited to documentation, reporting, project those studies are favourable in terms of academic performance,
management, on-line glossaries and dictionaries, discussion level of student engagement, and student attitudes. It should be
groups, and information systems. Wikis can be used to pro- stressed, however, that the aforementioned studies are either
mote learning, collaboration, communication, interaction, case studies or quasi experimental ones and, as such, are mostly
sharing, meaning making, and reflection. [15] outline three exploratory. Consequently the evidence is far from conclusive.
main types of uses for wiki: In fact, despite positive findings significant problems also
(a) distributing information: the wiki becomes the medium emerged. More specifically, some students in the [22] study ar-
to disseminate information as well as the place to store the gued that they did not learn anything new from the wikibook
information activity because they only wrote what they already knew. Along
(b) collaborative artifact creation: text is the main artifact the same lines, [34] argued that it is naïve to anticipate that the
produced, be it in the form of glossary or other formats wikibook activity would necessarily lead to collaborative writing
(c) discussion and review activities: anchored discussion, and learning without taking any other measures.
project case libraries, and peer review. The present study examines the potential of wikibooks in ter-
Even though wikis can be very important for education, tiary education in a blended learning context. A wikibook activi-
there is a knowledge gap in the literature. Research on wiki ty was incorporated in an undergraduate course in a preschool
applications shows that wikis can be very useful tools for education department as part of the typical course assignments.
supporting learning in higher education settings. More specif- Given the wikibook use in such a context, the study aimed to
address the following research questions:
 WIKIBOOKS AS TOOLS FOR PROMOTING CONSTRUCTIVIST LEARNING 135

(a) what is the rate of student participation in the wiki- ments. Using the wiki log files, a number of measures were
book? computed: total number of pages created, number of page views,
(b) what are students' perceptions of their experiences with number of user views per page, number of edits, and the number
the wikibook? of users involved per page edit. The two survey instruments
were developed ad hoc for the purposes of the study. The first
II. METHOD instrument measured student background knowledge and atti-
tudes towards ICT. The second instrument measured student
A. Subjects
views about their wikibook experience. It contained 15 state-
27 students participated in the study. The students enrolled ments on several aspects of the wikibook and a typical 5-point
in an undergraduate course on the educational applications of Likert scale was used. Qualitative data collection involved 8
the Internet which was taught by the author in his parent insti- semi-structured interviews which aimed to provide a richer un-
tution. The course was elective for fourth semester students derstanding of how students viewed the wikibook.
and this cohort of students constituted a convenience sample.
H. Procedure
B. Course
The students were introduced to the wiki at the beginning of
The course aimed to introduce (a) common internet tech- the course. All students received an extensive introduction in
nologies and services and (b) a set of pedagogical principles using the wiki: in the sandbox provided students experimented
which could support educational uses of the internet. The with all technical aspects of the wiki such as creating new pages,
course comprised two main components: technical and peda- revising existing ones, linking, rolling back, and discussing. Af-
gogical. The main objective of the technical component was ter each class lecture or lab session the students selected a set of
to introduce students to information, communication, file, and concepts which they deemed both important and interesting,
application services. The main objective of the pedagogical searched and gathered relevant information, wrote up the text
component was to use the internet services to support a broad and contributed the information to the wikibook.
range of teaching and learning activities. In the beginning of the course the students were surveyed re-
Given that wikis were among the Web 2.0 tools introduced garding background knowledge of and attitudes towards ICT. At
in the context of the technical component and that the new the end of the course the students were surveyed about their
learning approaches were also presented in the context of the learning experiences with the wiki and eight students were ran-
pedagogical principles, the wikibook afforded a unique op- domly selected for interviews.
portunity to put theory into practice. The students could not
only familiarize themselves with the wiki technology and the III. ANALYSIS
associated pedagogical ideas but also experience it firsthand. To address the first research question, data analysis involved
F. The Wikibook assignment frequency counts and descriptive statistics. To address the
second research question descriptive statistics indices were
The wikibook constituted the main course assignment
computed for the two surveys while the interviews were content-
which was compulsory for all students. This assignment re- analyzed to determine the main themes the students reported.
quired students to (a) gather information on a number of top-
ics from various sources (e.g. printed textbooks, electronic IV. RESULTS
textbooks, wikibooks, web sites etc), (b) process the informa-
tion (e.g. restructure, summarize, compare, contrast etc), and A. Rate of student participation
(c) write the information to a wiki page thereby contributing The students created 66 pages which were viewed 6909 times.
to an existing chapter or sub-chapter. The contents page (i.e. the wiki start page which is the equiva-
Regarding participation, no specific guidelines were pro- lent of the table of contents) is given in table 1.
vided: the students could freely decide on how much informa-
tion to contribute to the wikibook. Moreover, there were no Internet
specific requirements set for covering a minimal number of World Wide Web
topics or chapters: the students could contribute information The social impact of the Internet
File Sharing
to topics which they found interesting. To encourage partici- Secure Data Transfer
pation and avoid social loafing, the wikibook assignment Networks
amounted up to 60% of the total course credit. Distance Education
MediaWiki (http://www.mediawiki.org), the software that pow- Hacking & Cracking
Portals
ers Wikipedia, was the software used in the study. To avoid Operating Systems
spam, vandalism, and all sorts of extraneous distractions, the Important Internet Applications
wikibook was used in a closed-study group. Internet & Schools
Table 1: The chapters of the wikibook
G. Measures
For the purposes of the present study both quantitative and The main wiki statistics are presented in table 2. As can be
qualitative measures were employed. Quantitative measures seen in the table, some pages were more popular than others. In
included the wiki system log files and two survey instru- fact, the wiki contents page amounted to 88% of the total num-
136 KARASAVVIDIS

ber of views. On average, each page was read 104 times and For the most part the pages were reasonably interlinked. The
the variability was very high. Log file analysis, however, in- contents page (initial wiki page) included links to 12 wiki pages
dicated that 55 out of the total 66 pages were created by one through which the majority of all other pages were accessible.
student and were characterized by a small number of views Only 9 pages (13,64%) were not linked to any other pages in the
(typically 1-3). While technically these constituted wiki pag- wiki.
es, they contained little information -usually an image or icon
and a couple of sentences describing the theme of the page. B. Student perceptions
As this student case clearly represents an outlier, the descrip- The student responses to the questionnaire administered at the
tive statistics reported in table 2 should be interpreted with end of the course are given in table 3.
caution. To address this issue, we examined the page views
for the 13 most viewed pages, that is, the contents page (main Variable Mean SD
wiki entry page) and the 12 other pages listed in that page. a 3,85 0,66
This subset of wiki pages accounted for 88,57% of the total ease of use
page views and did not include any of the wiki pages created a 4,07 0,92
by the aforementioned student. After this correction, the mean ease of access
general b 4,22 0,58
number of reads increased by a factor of five (M=470,69,
SD=578,97). This finding clearly suggests a very high rate of overall impression
access. c 3,89 0,58
The wiki pages were edited 1065 times. Again, the statis- understanding contribution
tics reported in table 2 are subject to the same distortion con- d 4,19 0,40
sidering that most of the 55 pages that one student created had writing contribution
not been edited at all. Following the same procedure adopted d 3,81 0,62
above to deal with this outlier the descriptive statistics were reading contribution
d 3,59 0,64
recalculated. The contents page and the pages listed as the contribution
edits made contribution
main wikibook contents were edited 757 times (71,08% of the d 3,59 0,80
total edits recorded by the system). The average number of
edits received contribution
edits per page was 58,23 which is considered to be very high
d 3,59 0,80
and indicates that all pages underwent extensive revisions. It
discussion contribution
should be noted, however, that even for this subset of wiki
pages the variability was rather high (SD=75,20). e 1,85 0,91
The number of users involved per page edit is an important extent of revisions made
edits
measure as it shows how many students were involved in the e 1,74 0,86
authoring of a wiki page. Table 2 shows that 2,7 students extent of revisions received
were involved per page edit. However, as discussed above, a. Very difficult-Very easy b. Very negative-Very positive. c. Not at all
this figure is heavily biased because 83% of the pages were important – Very important. d. Strongly disagree-Strongly agree. e. Mi-
created by a single student. After correcting for this outlier, nor-Major
the mean number of students involved per page edit rises to
8,69 (SD=6,21). This suggests that the wiki pages were essen- Table 3: Descriptive statistics of students' views regarding the wikibook
tially co-authored as on average more than 8 students were
involved in the writing of every wiki page. This finding is As table 3 shows, the students did not experience any difficul-
very promising as the number of joint edits reflects the degree ties in using or accessing the wiki and rated their overall wiki
of on-line student collaboration. Finally, in addition to obtain- experience very positively (4,22). In fact, the mean for this vari-
ing evidence of on-line collaboration among students, table 2 able reflected the highest positive rating in the questionnaire.
also shows that the levels of individual student participation Moreover, the students evaluated positively wiki's contribution
were also very high: the average number of edits per user is to their understanding of the course concepts. While students
about 40 which suggests that students' contributions were rated positively the four aspects of wiki work, the contribution
extensive. of writing to learning clearly stands out. This is quite interesting
as the students appeared to favor writing over reading, editing,
Variable Total min max Mean SD
and discussing. Finally, table 3 shows that students reported that
they neither made extensive edits to the writings of other stu-
Page Views 6909 1 2045 104,68 308,8 dents nor were their texts extensively edited by fellow students.
Page Edits 1065 1 262 14,3 39,1 The students were also asked to rate the most difficult aspect
Users per Page n.a. 1 22 2,77 4,03 of wiki-related work. The final rating is as follows: writing:
Edit 33%, reading: 11%, revising: 44%, discussing: 11%. Thus, re-
Edits per User 1065 3 184 39,44 44,94 vising was found to be the most difficult aspect of the wikibook
Table 2: Descriptive statistics for the main Wikibook variables
assignment.
The qualitative analysis of the interviews indicated that while
students expressed positive views about the wiki, they also men-
 WIKIBOOKS AS TOOLS FOR PROMOTING CONSTRUCTIVIST LEARNING 137

tioned considerable difficulties. An outline of the pros and where the students collaborate on-line and create an electronic
cons is presented in table 4. textbook. A number of recent studies have began to explore how
wikibooks can be used in the context of higher education ([31],
Pros Cons [22], [32], [33], [34]). Although the findings from those explora-
Facilitation of conceptual under- first to write
tory studies are rather positive, a number of important issues
standing as the first contributors usually related to pedagogy remain unexplored. The present study aimed
the wiki task helped students master summarized the main points pre- to fill this knowledge gap regarding wikibook applications in
factual information sented in class, the remaining blended learning contexts. The study examined both the extent
students reported difficulties in
Information gathering & ex- finding what to write
of student participation in the wikibook activity and students'
change perceptions of the wikibook activity.
the wiki task necessitated informa- extent of contributions Regarding the first research question, the findings of the study
tion gathering for the students because no specific guidelines indicate that the students created several pages which were both
were given, the students were
group work unable to decide how much infor-
viewed and edited several times. As the analysis indicated, each
the wiki task required a collabora- mation in each book section was wiki page was viewed on average 470 times. Moreover, each
tive effort enough page underwent extensive edits as it was edited on average 58
times. Finally, more than 8 students were on average involved in
revising structuring
the task afforded the opportunity to the students considered the open-
the editing per wiki page. As a whole, these findings are very
revise not only one's texts but also ness of the wiki as somewhat limit- promising as most wiki-related studies typically report that stu-
the texts contributed by others ing and requested more structure dent participation in the wikis is limited ([26], [27], [28], [21])
(e.g. providing a detailed table of and the same holds for on-line interaction ([23], [29], [30], [22]).
writing contents)
the students considered the wiki
Thus, unlike former studies, the outcomes of the present one
task to be a genuine text-production understanding clearly suggest that a compulsory wikibook activity for which
process and not a simple reproduc- the students complained that some students can get a considerable portion of the total course credit
tive one sections of the wikibook were can effectively alleviate the problem of limited student participa-
unreadable which obfuscated un-
mode of learning derstanding
tion which is reported in the literature.
the wiki task exclusively relied on Regarding the second research question, the quantitative anal-
student contributions and this pro- validity of interpretations ysis indicated that the students rated their wikibook experience
moted a more active approach to the students were uncertain about very positively as the mean of 4,22 on a 5 point scale suggests.
learning in the context of the course the validity of the information
which they added to the wikibook
As for the contribution of wikibook to their learning of the
course concepts, the students gave higher ratings to writing and
technical issues reading than to the remaining aspects.
some students mentioned that they The qualitative analysis of the student interviews revealed a
were sometimes unable to commit
the edits they had made to a wiki
number of positive and negative points of the wikibook activity.
page (probably due to concurrent On the one hand, the students favored the wikibook activity for a
edits) number of reasons. It should be noted that all aspects of wiki-
Table 4: Students' perceptions of the pros and cons of the wikibook activi- book which were viewed as favorable by the students are in line
ty with the theory and research regarding the potential contribution
of wikis to learning. As far as the contribution of the wikibook
V. DISCUSSION
to the learning of the course concepts is concerned, the facilita-
The impact of Web 2.0 on many domains of human activi- tion of conceptual understanding and the active mode of learning
ty has been substantial [5]. One of the areas where this impact that the wikibook required stand out from a learning point of
is most evident is education and more particularly E-learning view. On the other hand, the students also expressed considera-
[6]. Considering that many web-based education and e- ble difficulties. Two of the problems students reported, namely
learning approaches have tended to replicate traditional mod- understanding and validity of interpretations, deserve closer fur-
es of teaching and learning which are transmisionist in nature, ther examination because they are related to learning and, to a
Web 2.0 is expected to enhance learning in higher education large extent, contradict the facilitation of conceptual understand-
because, by default, it requires the active participation of us- ing that the students mentioned as one of the most important
ers. Unlike Web 1.0 tools, the corresponding Web 2.0 tools wiki contributions. The remaining difficulties identified are
would not be at all possible without user participation. This more procedural in nature and can be more easily addressed (e.g.
specific feature is a perfect match to the active learner partici- devising specific rules for wiki contributions or providing more
pation which according to Constructivism characterizes all structured versions of the task such as a detailed list of contents
human learning. Thus, Web 2.0 holds great promise for pro- with chapters and sub-chapters). To a large extent, the outcomes
motive active learning in higher education. Wikis are amongst of the qualitative analysis replicate other literature findings re-
the Web 2.0 tools which have great potential for contributing garding the difficulties students experience when using wikis
to learning because they are uniquely suited to support both (e.g. [26], [30], [22], [35]). Thus, the fact that students face im-
process and product approaches. There are many types of portant challenges when engaged in wiki activities remains and
wiki uses in education and the specific focus in this paper was should be addressed in future research.
wikibooks. Wikibooks can be used as class assignments
138 KARASAVVIDIS

References [19] Yukawa, J. (2006) Co-Reflection in on-line Learning: Collaborative Criti-

cal Thinking as Narrative. International Journal of Computer-Supported
Collaborative Learning, 1(2), 203-228.
[1] O'Reilly, T. (2005). What Is Web 2.0. Design Patterns and Business [20] Xiao, Y. & Lucking, R. (2008). The impact of two types of peer assessment
Models for the Next Generation of Software. Retrieved March 20, 2009 on students' performance and satisfaction within a Wiki environment. In-
from ternet and Higher Education, 11(3-4), 186-193.
http://www.oreillynet.com/pub/a/oreilly/tim/news/2005/09/30/what-is- [21] Rick, J. & Guzdial, M. (2006). Situating CoWeb: a scholarship of applica-
web-20.html tion. Computer-Supported Collaborative Learning, 1(1), 89–115.
[2] West, J.A. & West, M.L. (2009). Using wikis for on-line collaboration: [22] Ravid, G., Kalman, Y.M. & Rafaeli, S. (2008). Wikibooks in higher educa-
the power of the read-write web. San Fransisco: Jossey-Bass. tion: Empowerment through on-line distributed collaboration. Computers
[3] Surowiecki, J. (2004). The Wisdom of Crowds: Why the Many Are in Human Behavior, 24(5), 1913–1928.
Smarter Than the Few and How Collective Wisdom Shapes Business, [23] Wheeler, S. Yeomans, P. & Wheeler, D. (2008). The good, the bad and the
Economies, Societies and Nations. Anchor. wiki: Evaluating student-generated content for collaborative learning.
[4] Tapscott, D. & Williams, A.D. (2006). Wikinomics: how mass collabo- British Journal of Educational Technology, 39(6), 987–995.
ration changes everything. NY: Penguin. [24] Robertson, I. (2008). Learners’ attitudes to wiki technology in problem
[5] Mason. R. & Rennie, F. (2008). E-learning and social networking based, blended learning for vocational teacher education. Australasian
handbook: resources for higher education. London: Routledge. Journal of Educational Technology, 24(4), 425-441.
[6] Downes, S. (2005). E-learning 2.0. eLearn Magazine. Association for [25] Theng, Y-L, Li, Y., Lim, E-P., Wang, Z., Goh, D., Chang, C-H., Chatter-
Computing Machinery, Inc. Retrieved March 20, 2009, from jea, K., & Zhang, J. (2006). Understanding User Perceptions on Usefulness
http://elearnmag.org/subpage.cfm?section=articles&article=29-1. and Usability of an Integrated Wiki-G-Portal. Proceedings of the 9th Inter-
[7] Naidu, S. (2003). Designing Instruction for e-Learning Environments. national Conference on Asian Digital Libraries (ICADL) (pp. 27-30). Kyo-
In M. G. Moore & W.G. Anderson (Eds). Handbook of Distance Edu- to, Japan.
cation (pp. 349-365). Mahwah, New Jersey: LEA. [26] Carr, T. Morrison, A., Cox, G., & Deacon, A. (2007). Weathering wikis:
[8] Littlejohn, A., Cook, J., Campbell, L., Sclater, N., Currier, S. & Davis, Net-based learning meets political science in a South African university.
H. (2007). Managing educational recourses. In G. Conole & M. Oliver Computers and Composition, 24(3), 266–284.
(Eds). Contemporary perspectives in e-learning research: themes, me- [27] Choy, S.O. & Ng, K.C. (2007). Implementing wiki software for supple-
thods and impact on practice (pp. 134-146). London: Routledge. menting on-line learning. Australasian Journal of Educational Technology,
[9] Palloff, R.M. & Pratt, K. (2007). Building on-line learning communi- 23(2), 209-226.
ties: effective strategies for the virtual classroom. San Fransisco: John [28] Cole, M. (2008). Using Wiki technology to support student engagement:
Wiley & sons. Lessons from the trenches. Computers & Education, 52(1), 141-146 .
[10] Laurillard, D. (2007). Preface. In H. Beetham, & R. Sharpe (Eds.), [29] Ma, W.W-K. & Yuen, A.H-K. (2007). Learning News Writing Using
Rethinking pedagogy for a digital age: designing and delivering e- Emergent Collaborative Writing Technology Wiki. In J. Fong, F. L. Wang
learning. London: Routlege. (Eds). Blended Learning. Workshop on Blended Learning (pp. 303-314):
[11] Jonassen, D. (1991). Objectivism vs. Constructivism. Educational Edinburgh, United Kingdom: Pearson.
Technology Research and Development, 39(3), 5-1. [30] Ma, W.W.K. & Yuen, A.H.K. (2008). A Qualitative Analysis on Colla-
[12] Jonassen, D. (1999). Designing constructivist learning environments. In borative Learning Experience of Student Journalists Using Wiki. In J.
C. M. Reigeluth (Ed.), Instructional-design theories and models: A new Fong, Reggie, K. Wang, Fu Lee, W.F. (Eds.). Hybrid Learning and Educa-
paradigm of instructional theory (Vol. 2, pp. 215-239). Mahwah, NJ: tion. Proceedings of the first International Conference on Hybrid Learning
Lawrence Erlbaum Associates, Inc. (pp. 103-114). Lecture Notes in Computer Science 5169. Berlin: Springer-
[13] Jonassen, D., Cernusca, D., & Ionas, G. (2007). Constructivism and Verlag.
instructional design: The emergence of the learning sciences and design [31] Bold, M. (2006). Use of wikis in graduate course work. [Electronic ver-
research. In R. A. Reiser & J. V. Dempsey (Eds.), Trends and issues in sion]. Journal of Interactive Learning Research, 17(1), 5-14.
instructional design and technology (2nd ed., pp. 45-52). Upper Saddle [32] Kidd, J., O'Shea, P., Baker, P., Kaufman, J. & Allen, D. (2008). Student-
River, NJ: Pearson. authored Wikibooks: Textbooks of the Future?. In K. McFerrin et al.
[14] Leuf, B., & Cunningham, W. (2001). The Wiki Way. Addison-Wesley (Eds.), Proceedings of Society for Information Technology and Teacher
Professional. Education International Conference (pp. 2644-2647). Chesapeake, VA:
[15] Guzdial, M. Rick, J. & Kehoe, C. (2001). Beyond Adoption to Inven- AACE. Retrieved from http://www.editlib.org/p/27619.
tion: Teacher-Created Collaborative Activities in Higher Education. The [33] Lai, F. & Chang, C. (2008). Wikibook Writing for Collaborative Learning.
Journal of the Learning Sciences, 10(3), 265–279. In G. Richards (Ed.), Proceedings of World Conference on E-Learning in
[16] Choy, S.O. & Ng, K.C. (2007). Implementing wiki software for sup- Corporate, Government, Healthcare, and Higher Education (pp. 977-982).
plementing on-line learning. Australasian Journal of Educational Chesapeake, VA: AACE. Retrieved from http://www.editlib.org/p/29735.
Technology, 23(2), 209-226. [34] Lin, H. & Kelsey, K.D. (2009). Building a Networked Environment in
[17] Elgort, I., Smith, A.G. & Toland, J. (2008). Is wiki an effective plat- Wikis: The Evolving Phases of Collaborative Learning in a Wikibook
form for group course work? Australasian Journal of Educational Project. Journal of Educational Computing Research, 40(2), 145 – 169.
Technology, 24(2), 195-210. [35] Karasavvidis, I. (in press). Wiki uses in higher education: exploring bar-
[18] Minocha, S. & Thomas, P. G. (2007). Collaborative Learning in a Wiki riers to sucessful implementation. Interactive Learning Environments.
Environment: Experiences from a software engineering course. New
Review of Hypermedia and Multimedia, 13(2), 187-209.
 The Impact of the Context of an Information
System on Strategic Information Quality - The
Institute of Technology Sector in Ireland
Laurence P. Elwood* and Owen V. Foley**
*School of Business, Galway Mayo Institute of Technology, Ireland. Email: [email protected]
**School of Business, Galway Mayo Institute of Technology, Ireland. Email: [email protected]

Abstract— Information Quality has become an ever The workstation model was predominant in the late 70s
increasing problem for many organizations relying on and early 80s [2], followed by widespread deployment of
information systems for many of their core business the client server model [3]. The rapid expansion of
activities. Despite advancements in technology and internet services soon followed [1, 3]. This expansion has
information quality investment the problem continues to continued with the pervasive availability of mobile
grow. The context of the information systems over the last devices and wireless technologies [1, 6, 7]. Virtualization
number of years has also evolved. This has resulted in a has also made an impact especially with the widespread
diverse number of architectures accessing multiple implementation of unplanned solutions [8].
information systems. The context of the deployment of an
The significant changes and increase in complexity of
information system in a complex environment and its
IS context has in many cases occurred independently of
associated information quality problems has yet to be fully
the underlying databases accessed [1]. Applications may
examined by researchers. Our research addresses this
have been designed, built and tested with a mature
shortfall by examining the relationship between the context
of the information system and information quality. The
software development methodology for a particular
research is conducted within the context of strategic
context, yet within a very short period of time it may be
information quality within the higher education system in accessed and predominantly employed from a different
Ireland specifically the Institute of Technology Sector. A context [9]. Data models have also evolved [2],
mixed methodology approach is employed to tackle the nonetheless a considerable number of IS in use today have
research question of the impact of the context of an data models designed prior to the contexts that are
information system on strategic information quality. employed to access them. There are as a result multiple
Research to date has examined different information accesses from diverse and complex contexts.
systems’ context factors and their effect upon information The quality of information that strategic decision
quality dimensions both objectively and subjectively. Thus makers require is heavily dependent on these IS. An
far our research puts forward a novel information quality examination of the Institute of Technology Sector in
methodology that is context related; that is it takes the user, Ireland reveals that Management Information Systems,
task and environment into account. An initial experiment Decision Support Systems and User driven Office
has been conducted. Results of preliminary experiments Information Systems are the primary source of this
indicate that context affects the perception of information information. In light of the evolving nature of both the
quality. technology and the higher education sector it is imperative
to measure the impact of these contexts on the strategic
Keywords ü Information Quality; Higher Education; information quality (IQ).
Strategy; Information Systems

I. INTRODUCTION II. HIGHER EDUCATION IN IRELAND

Over the last two decades Information Systems (IS) Ireland’s higher education providers can be
contexts have radically altered [1-3]. The relatively characterized across a university and institute of
simple mainframe – workstation configuration has technology divide. The principal differentiator across both
evolved and is accompanied or replaced in its entirety. sectors is the applied orientation of the institute of
Contexts have become more complex. A typical IS is technology sector, which may be likened to the
most likely accessed via a number of different tiers, fachhochschulen in Germany. The Institute of Technology
applications and platforms. Therefore the context has sector was established in 1969 (when colleges were
changed from a homogeneous one of a single database and referred to as Regional Technical Colleges) following the
access method to a heterogeneous and complex one earlier recommendations of the “Mulcahy Report” [10].
resulting in many databases and access methods. This Prior to the Regional Technical Colleges Act [11], these
landscape will continue to evolve with the ever increasing colleges had limited autonomy and were controlled by the
deployment of Service Oriented Architectures (SOA) [4]. local education authority, the Vocational Education
The changing context of the IS comprises a number of Committee. The Regional Technical Colleges Act [11]
factors; including the user role, application tasks, access expanded the mission of the Regional Colleges permitting
devices and associated services. Lee [5] describes this as them for the first time to engage in research, participate in
the relationship between contents and environments. collaborative research and the commercial exploitation of

139
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_24, © Springer Science+Business Media B.V. 2010
140 ELWOOD AND FOLEY

research. Teaching nonetheless continues to be seen as the research, regional development, links with industry, social
primary mission of the institute of technology sector, with inclusion and meeting skills needs” [14]. The response of
research as an interesting, but ultimately costly, adjunct. the institute of technology sector to these emerging
The Regional Technical Colleges Act [11] also established metrics of performance, demanded nationally, yet
the following bodies; Governing Body, Academic Council assessed locally, is the introduction of frameworks of
and the designation attaching to the principal officer of appropriately designed management information systems
each institute, namely Director (re-designated to that of which seek to;
Institute President under the Institutes of Technology Act Improve organisational effectiveness (the determination
[12]. Consequently, three distinct groupings now exist in of unit costs, the quality, retention and destination of
each college: (1) The Governing Body, which oversees the graduates, including attrition and retention reporting.).
work of the College Director, (2) the College’s senior
Understand and minimise the risks facing the respective
management team, headed by the College Director, which institute of technology (the generation of strategy
oversees the day to day management of the College, and
supportive data, the prioritisation of data characterised by
(3) the College’s Academic Council, charged with reliability and timeliness to inform management decision
protecting, maintaining and developing the academic
making).
standards of the courses and activities of the college. Two
of these bodies; Governing Body and the senior Comply with statutory requirements and corporate
management team, are essentially managerial in focus, governance requirements (the generation of statutorily
requiring data which facilitates the discharge of their required financial information and annual reports.).
managerial duties. The mid 1990s saw further pressure on
Ireland’s higher education institutes to act as levers in III. OVERVIEW OF INFORMATION QUALITY
helping Ireland’s economy transition to a knowledge Currently many of the IQ frameworks are employed
economy. The White Paper “Charting Our Education post deployment of the IS. A number have been
Future” [13] signposted a role whereby all higher implemented as a direct result of IQ problems,
education institutes were given responsibility, through highlighting accuracy or completeness problems in the
research, for the development of new ideas, new main. The use of IQ frameworks allows for an
knowledge and the application of existing knowledge. examination of the environment without the need for prior
This devolution of accountability and transparency to all knowledge of the development methodology [17].
higher education institutes has also served to heighten the The area of IQ as a field of research has developed
requirement for timely, accurate and relevant information considerably within the past decade, predominantly
within Ireland’s higher education institutes. stemming from reliance by enterprises at all levels on IS
This altered higher education landscape, coupled with for the completion of their core business activities. IS
Ireland’s economic transformation, provided the backdrop context has also evolved with widespread deployment of
to the OECD Review of Higher Education in Ireland in web services. Enterprises are now increasingly
2004 [14]. Whilst lauding Ireland’s success in increasing characterized having mission critical systems based on
participation rates, and its growing recognition of the diverse contexts, yet have no measure of the quality of the
importance of research to sustainable economic growth, information that they derive their business and profit from.
the review nonetheless highlighted that Ireland’s tertiary This is a wide and varied problem affecting nearly every
education system is at “a significant point of departure” major type of business activity and enterprise. Al Hakim
Over the next decade higher education in Ireland will [6] provides numerous examples of these anomalies from
develop against the framework for higher education many areas, outlining the reason along with the particular
outlined in the Programme for Government [15] and the IQ dimension affected. The examples indicate how the
National Development Plan [16]. The policy objectives generation of information from disparate sources can
outlined for Ireland’s higher education system is to be at impact on many aspects of an enterprise, often not even
the top rank of performance within the OECD [14]. A considered when the IS was initially designed. The impact
common feature pervading most of these policy of these IQ problems has prompted researchers to examine
documents is the increasing use of the funding whip-hand such aspects as IQ dimensions, IQ assessment and IQ
to encourage the higher education system to both agree to management.
and implement centrally determined policy initiatives.
IV. DESIGN AND DEVELOP AN IQ METHODOLOGY
More specifically the following development needs are
specified as key metrics in the delivery of this policy An examination of the literature [5, 6, 17-22] suggests
objective; (1) increased participation and improved access, that the issue of context has not been adequately
(2) greater flexibility of course offerings to meet diverse addressed. Consequently we believe that accessing the
student population needs in a lifelong learning context, (3) same information from diverse contexts will be
enhancement of the quality of teaching and learning, (4) problematic and inconsistent with existing frameworks
significantly increase the number of PhD numbers and and IQ measurement.
research activity, and (5) effective provision for Our construction of the IQ methodology examines and
technology transfer [14]. The Higher Education Authority builds on work completed by Knight & Burn [20] in the
(HEA) envisions a critical role for the Institute of area of algorithm choice for Internet search engines and
Technology sector, that is broadly in line with the OECD Naumann and Rolker’s [23] analysis of information
[14] recommendations, and which recognises: “that the sources. Our dimension selection were chosen by
Institutes of Technology and the Universities have applying Leungs [22] Importance, Urgency and Cost
developed differently and to establish greater clarity on metric. The methodology proposed is flexible in its
the diverse role of Institutes of Technology and the application and caters for the employment of other
Universities in such areas as, teaching and learning, usability metrics such as Doll and Torkzadeh [24] The
 THE IMPACT OF THE CONTEXT OF AN INFORMATION SYSTEM 141

development of the methodology can be summarised as

follows
Identification of the user(s)
Identification of the tasks that make up the system Context Services
End User Role IS Environment
Identification of the dimensions to be assessed Access
Device Feedback/
Application
Prioritisation of the dimensions to be assessed Task Monitor

Assessment of the dimensions selected. User

Feedback

Context-specific IQ Measurement
In order to overcome these context deficiencies we Ser DB
Context-specific
propose a novel IQ methodology that takes the context of IQ Requirements
the IS into account, allowing us to analyse the perception Assessment
Objective
Ri
IQ Dimensions Technique RRi
of the same information from different contexts. Knight Subjective Ri i
& Burn [20] point out that despite the sizeable body of
literature relatively few researchers have tackled the Priority Testing
difficult task of quantifying some of the conceptual Survey
Dimension
Instrument
definitions. Naumann and Rolker [23] further suggest that Dimensions Score
Application

the subjective view of quality makes the actual assessment

of IQ dimensions difficult. The unplanned nature of the Figure 2. Application of the Methodology
evolution with respect to context has the potential to
further complicate matters. VI. AN EXPERIMENT
The methodology is now illustrated with an example of Our initial experiment was conducted over a three
an IS used by strategic decision makers in the institute of month period, involving forty eight participants and three
technology sector which is being examined by our IS at Galway Mayo Institute of Technology. The
research. The primary challenge in the application of the experiment involved the control of one independent
methodology is the analysis of the context and selection of variable namely the access device. Each of the groups
appropriate IQ dimensions. were assigned tasks particular to their profile and IS. The
tasks within each group were conducted from three
different contexts namely workstation, web, and mobile.
Consistency; The first requirement of the methodology is to analyse the
guidelines and
reference models precedes consists of software services identified and selected. This at present
Meta Model
- creates Design
is a binary test and examines the availability of the
Activity
Component uses service. Prior to the conduct of the full experiment further
consists of precedes analysis will be carried out to examine the possibility of
Result performs applying a more detailed checking algorithm.
contains Document
The application of the methodology first checked the
defines availability of the three software services identified. In
Tool
uses
Technique Role
the event of service availability the objective integrity
analysers were initiated. The results of this analysis were
stored in an IQ table-space similar to an audit table-space
[28]. The subjective survey instrument was run in
Role Design Activity Result Document Technique Tool
conjunction with the objective integrity analysers. An
Who performs? What? With the result? Applying Using tool?
technique? outline of the user group, task and associated environment
was followed by a dimension score for the objective
Figure 1. Methodology Description measure and service availability indication. The
subjective survey result then followed.
V. IMPLEMENTATION OF THE METHODOLOGY
VII. ANALYSIS OF RESULTS
Grounded IS research to date in the field of IQ has
concentrated on intrinsic dimensions. The seminal The results charts and statistical tests were completed
research with respect to context has focused on user for four dimensions identified by our methodology. We
opinion or satisfaction [21, 25]. will now describe in detail the process of analysis for one
of these dimensions (free of error). The same method and
The most commonly used IS research methodologies approach was employed in examining all four dimensions.
have been collated by Galliers [26] and extended by
Minger [27]; adopting an interpretive and empirical The initial step in the analysis was the binary test of
approach. The primary goal of our research is to examine services. In the pilot study all the services were present
the relationship between the context of the IS and IQ therefore the subjective and objective tests were applied to
therefore an empirical approach appears the most all participants. In order to test the level of significance
appropriate to the research question. This approach of the remainder of the results it was necessary to apply an
allows for an identification of the precise relationship appropriate statistical technique to the data gathered. In
between variables via an experiment design, which can be the pilot study we sought to ascertain if there was a
conducted by laboratory experiments, field experiments or relationship between the context of the IS (Web,
a combination of both. Workstation or Mobile) and the level of IQ. As IQ is a
multidimensional concept it was necessary to do this at a
dimension level.
142 ELWOOD AND FOLEY

The first statistical test completed, illustrated in figure 3

below, is a line plot of the mean between the objective
integrity analysis and the subjective survey instrument.
Line plots are used to illustrate the frequency of an events
occurrence [29]. The pattern on the graph plots the
relationship between the objective results for free of error
as measured at the database against each of the individual
contexts. The graph indicates a relationship between the
objective measure and each of the contexts. The objective
value is a constant at 89% and the graph illustrates that the
subjective measures of the individual contexts follow a
distinct pattern. The results clearly indicate that the Figure 5. Accessibility V Completeness
subjective view of IQ was best from the workstation with
less satisfaction from the web and less again from the
mobile. In other words the context of the IS access device
affects the users subjective view of the IQ. The initial line
plot of the mean warrants further examination.

Figure 6. Accessibility V Free of Error

Figure 3. Plot - Objective V Subjective Dimension – Free-of-Error

VIII. ESTABLISH HOW THE PERCEPTION OF IQ IS

AFFECTED BY THE CONTEXT OF THE IS.
A variation in the context of the IS we contended
impacts on IQ. The context was controlled by
manipulating the service level. Similar to the process for
Figure 7. Accessibility V Consistency
objective one we outline this for one dimension in the
pilot study and intend to use the same process to complete The initial results as outlined in figures four to seven
for all dimensions in the main experiment. indicated that there was a positive linear correlation
We have chosen the service of systems accessibility. between system accessibility level and timeliness. As the
This service is set for various levels and the affect on IQ is accessibility levels to the IS increase, the timeliness
measured as per the measurement instruments outlined. dimension also improved. Our research identified a
Descriptive statistics are applied in the pilot experiment. relationship between accessibility level and completeness
This part of the experiment involved varying the levels (figure five). At low levels of accessibility completeness
of systems access and measuring the corresponding IQ is also low, however the relationship does not hold as
dimensions that were identified for the framework. levels of accessibility improve. At the higher level this
relationship was again observed. This fall off in the
relationship requires further examination. The scatter plot
for both consistency and free-of-error did not display any
relationship. We deduced from our experiment that the
methodology is sensitive to varying IS context.
Our research also discovered a further significant
finding particular to the study of IQ. Many researchers
have proposed ways to aggregate single measures of IQ
dimensions, often underlying a weighted aggregate of
single values for IQ dimensions [25]. Although, recently
some researchers have attempted to propose IQ value
curves and trade-offs by analysing the potential impacts of
IQ [30], many researchers propose to measure the overall
impact of IQ as weighted aggregate. A principle measure
Figure 4. Accessibility V Timeliness
of the weighed sum of all the criteria (IQCi) is illustrated
below
 THE IMPACT OF THE CONTEXT OF AN INFORMATION SYSTEM 143

N REFERENCES
IQ = ¦ α i IQCi where ∀αi :0 ≤ α
N
i ≤ 1

i =1
¦i =1
α i = 1 [1] Adelman, S., L. Moss, and M. Abai, Data Strategy. 2005:
Addison-Wesley.
Equation 1 Aggregate measure of information quality
[2] Date, C., Databases in Depth. 2005: O Reilly.
The aggregated value should define the quality level [3] Sangwan, R., et al., Global Software Development Handbook.
that characterizes information source. The approach to 2007: Auerbach Publications.
use the average as an aggregation functions may not be [4] Rosenblum, M. and T. Garfinkel, Virtual machine monitors:
current technology and future trends. IEEE Computer, 2005.
suitable among heterogeneous dimensions since 38(5): p. 39-47.
dependencies introduces bias that negatively affect the
[5] Lee, Y., et al., Journey to Data Quality. 2006: MIT.
reliability of the assessment procedure. Change in the
[6] AlHakim, L., Challenges of Managing Information Quality in
context (accessibility service) will have an impact on other Service Organizations. 2007: Idea Group Pub.
dimensions in a framework and as a consequence the [7] Fisher, C., et al., Introduction to Information Quality. 2006: MIT.
aggregate score. The research provides a methodology to [8] Gilen, A., B. Waldman, and J. Humphreys, Technology
measure this impact. This it is argued should be Assessment: The Impact of Virtualization Software on Operating
completed prior to applying the aggregate, because it will Environments. 2006.
provide a more accurate result with respect to IQ. [9] AlHakim, L., Information Quality Management Theory and
Applications 2007.
IX. LIMITATIONS [10] Mulcahy, Report of the Steering Committee on Technical
Although our research revealed interesting results, this Education. 1967.
was a pilot study concentrating only on a subset of [11] Regional Technical Colleges act. 1992: Ireland.
dimensions and functionality used by strategic decision [12] Institutes of Technology Act. 2006: Ireland.
makers in higher education. However we believe that the [13] Education, D.o., Changing our Education Future, Education,
principles and methodology proposed have the potential Editor. 1994, Stationary Office.
for wider application across IS in higher education. The [14] OECD, Review of Higher Education in Ireland. 2004.
application of the full experiment will address this [15] Taoiseach, D.o.A., Programme for Government 2007-2012,
D.o.A. Taoiseach, Editor. 2007.
limitation. Correlation has been used to examine the
relationship between IQ dimensions. Analysis of [16] Taoiseach, D.o.A., National Development Plan 2007-2013,
Taoiseach, Editor. 2007, Stationary Office.
Variance will be completed for all sections of the research,
[17] Batini, C. and M. Scannapieco, Data Quality Concepts,
thus allowing the general application of our findings. Methodologies and Techniques. 2006: Idea Group Publishing.
Finally the research to date has only employed [18] Eppler, M. and P. Muenzenmayer. Measuring Information in the
experiments along with survey instruments as its research Web Context: A survey of state of the art instruments and an
approach. The other data collection methods which will application methodology. in 5th International Conference on
form an integral part of the research such as diaries, semi Information Quality. 2000.
structured interviews and focus groups will be expanded [19] Helfert, M. and M. Ge. A Review of Information Quality Research.
and form part of the next phase of the project. in 11th International Conference on Information Quality. 2006.
[20] Knight, S. and J. Burn, Developing a Framework for Assessing
Information Quality on the World Wide Web. Informing Science,
2005. 8: p. 159-172.
X. CONCLUSIONS AND FUTURE WORK [21] Lee, Y., et al., AIMQ: a methodology for information quality
This research contributes to the analysis of IS context assessment. Information and Management, 2002. 40: p. 133-146.
and IQ. In recent times companies have invested heavily [22] Leung, N., Quality Metrics for Intranet Application. Information
in IQ programmes in an effort to improve IQ [9]. Our and Management, 2001. 38: p. 137-152.
research demonstrates that a relationship does exist [23] Naumann, F. and C. Rolker. Assessment Methods for Information
Quality Criteria. in 5th International Conference on Infprmation
between IS context and other dimensions in an IQ QualityIntetnational Conference. 2000.
framework. The traditional method of measuring IQ [24] Doll, W. and G. Torkzadeh, The Measurement of End User
dimensions will require re-examination as relationships Computing Satisfaction. MIS Quaterly, 1988. 12(2): p. 259-274.
between the context and other dimensions become more [25] Wang, R. and D. Strong, Beyond Accuracy: What Data Quality
clearly established. By demonstrating clearly that the Means to Data Consumers. Journal of Management Information
context of the IS affects the quality of the information Systems, 1996. 12: p. 5-34.
produced, strategic decision makers can proactively allow [26] Galliers, R., Information Systems Research: Issues, Methods and
for this when using information generated by these IS. Practical Guidelines. 1992: Blackwell Scientific.
This research also contributes to the field of IQ research [27] Minger, J., The Paucity of Multimethod Research: A Review of the
by providing a methodology and test environment that can IS Literature. 2001, Warwick Business School.
be employed in a context related manner. Upon [28] Greenwald, R., R. Stackowiak, and R. Stern, Oracle Essentials:
refinement this methodology has the potential to allow Oracle Database 11g. 4th ed. 2004: O Reilly.
organisations to measure the impact of introducing new [29] Field, A. and G. Hole, How to Design and Report Experiments.
2006.
contexts post the development of an IS.
[30] Stvilia, B., et al., A Framework for Information Quality
Assessment. Journal of the American Society for Information
Science and Technology, 2007. 58(12): p. 1720-1733.
 Can Mobile Computing Help Reverse the Decline in
Informatics Studies?
D.Vitkute-Adzgauskiene, V.V.Fomin
Dept. of Informatics, Vytautas Magnus University
Vileikos 8,
LT-44404 Kaunas, LITHUANIA

Abstract-- In this work we analyse the factors leading to there are also “global” factors explaining the decreasing
the decline of popularity of traditiona l Computer Science / popularity of the Informatics in Europe at large and the U.S.
Informatics study programmes as observed in Europe and the To name a few, the transformation of the hardware (HW) and
U.S. We further examine the possible directions for improving software (SW) markets, job market trends, transformation of
the Informatics curricula to reverse the downturn trend. Finally,
we report on the first positive experience of including Mobile
the computing paradigm, the effect of the dot-com bubble
Computing related topics into the Informatics study programme burst, inadequate curricula for Informatics education at the
with resulting increase in students’ interest. school level, etc.
In this work, we reflect on the factors affecting the
I. INTRODUCTION downturn in Informatics enrollments. We analyze and
synthesize data from Lithuania, EU, and the U.S., aiming to
The last decade saw rapid growth of Internet usage. Our understand how the universities can react to the decline in
young generation today is usually referred to as the “Internet enrollments, and, specifically, whether inclusion of Mobile
generation”, having a spectrum of uses for the Internet, from a Computing in the study programmes can help re-instigate
virtual library, to a place for social networking, to study students’ interest in the Informatics studies.
room, to gaming room, etc.
Along with the Internet becoming an inseparable part of II. THE RISE OF THE MOBILE SOCIETY
our daily activities, phenomena of the “Mobile” came to
witness much stronger presence in the life of the world’s In many countries, mobile communications have entirely
population. With an estimated 20 percent of the world’s changed the overall national picture of ICT usage. Especially,
population being Internet users in 2007, there were twice as it is true of countries, which had lower personal computer and
many mobile phone subscribers, and there is a plenty of room fixed Internet penetration before the advance of the mobile
for growth, except maybe for the developed countries, where era – such is the case for most of the former Soviet Republics,
there is already a 100 percent penetration rate [1]. With Lithuania being one of these.
mobile phones having become a commodity, it shouldn’t Today, Lithuania is a highly saturated mobile phone user
come as a surprise that mobile broadband is becoming a country – its mobile penetration by the end of Q2‘2008 was
popular way for accessing Internet, and short messaging and 148% [2], i.e., every citizen on average had 1.5 mobile
mobile data services are being used to accomplish everyday subscriptions in his/her disposition. Besides the population,
tasks. the number of mobile subscriptions is increasing in machine-
Despite the visible proliferation of ICT (Information and to-machine applications (e.g., remote control or security).
Communication Technologies) in our everyday lives and Demographic categorization reveals that 88 percent of all
business, surprisingly, during the last decade universities in population, falling into the 16–74 age group, are mobile
Lithuania and globally saw a substantial downturn in phone users [3]. For the 16-24 year group this number is even
Computer Science (CS, referred to as “Informatics” in higher – 98%, which let us safely assume that 100% of
Europe) and Management Information Systems (MIS) university students are mobile subscribers.
enrollments. In the United States the trend started earlier, with While fixed Internet user numbers were also growing rather
enrollments in CS falling by half from 60.000 in 1998 to rapidly in Lithuania during the last decade, showing an
30.000 in 2006/7.1 In Lithuania, we saw the same 50 percent impressive 835% growth during 2000-2008, the Internet
decreases in admission numbers for Informatics students from penetration reached only 59 percent of population2 - much
2004 to 2008. While the Lithuanian case can be partly lower than that for mobile.
explained by highly hyperbolized admission numbers in The relatively low fixed Internet penetration may be the
2001-2002 (2-2,5 times higher, compared to 1999) due to the reason why Lithuanian mobile subscribers show high usage of
State policy and Government orders, as well as by the different mobile phone enabled services, as compared to other
possibilities to study abroad after Lithuania joined the EU, countries. For SMS, for example, statistical Lithuanian user

2
As of March 31, 2009, according to the Internet World Stats,
1
CRA Taulbee Survey, http://www.cra.org/statistics/ http://www.internetworldstats.com/stats9.htm.

145
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_25, © Springer Science+Business Media B.V. 2010
146 VITKUTE-ADZGAUSKIENE AND FOMIN

sends 156 SMS per month – more than two times the number steady decline in the enrollments to Informatics programmes
for the French subscribers, and mobile data services are used around the world. When analyzing possible reasons for the
by half of the Lithuanian population [2]. decline, we may single out three major factors:
Globally as well, mobile Internet is becoming a daily 1) Changes in the environment, including technology
activity for more and more people. In the U.S., for example, changes, transformation of the industry, job market’s
the number of people accessing news and information on their image.
mobile devices more than doubled during the last year [4]. 2) Inadequate Informatics education at schools.
Mobile service usage closely follows the habits of modern 3) Conservatism of Informatics study programmes.
society. With the growing popularity of social networking
services, access to these services from mobile devices has A. Transformation of the industry
grown over 152 percent in Western Europe from Q3’2007 to At least since the dot-com bubble burst in 2001, Computer
Q3’2008. For the U.S., this category saw an impressive 427% Science (CS) / Informatics enrollments saw a downturn in
increase, while access to news and infotainment services enrollments, which continues to date. Although slowly
increased by 100-190 percent [5]. For the popular Facebook recovering since, this time a more fundamental challenge may
service, the mobile access rate came to stand even with the deepen the “enrollments crisis” – that of the transformation of
fixed Internet access rate. the managerial and industrial paradigms. This time the new
Interesting to note, the use of mobile services kept to Lithuania programme of “Business Informatics”3, and
increasing even during the economic downturn. The top U.S. world-established concept and Business Schools’ study
sites saw doubled use from wireless pre-paid users over the program of Management Information Systems (MIS) is also
last year [6], indicating that people consider mobile under the pressure. The global trend of flattening of enterprise
broadband a convenient and cost saving access channel to resulted in management functions becoming increasingly
everyday (Internet) services. horizontal. This, in turn, resulted in elimination of traditional
We can find many examples showing that the mobile usage Ford-style vertical integration with its multi-divisional
growth is dependent on the availability of convenient mobile management system. Consequently, where there used to be
services for everyday use. Mobile parking services in Estonia Chief-X-Officers for every division X, in the new enterprise
and Lithuania are a good example – the services gives a paradigm the Chief Information Officers (CIOs) found
possibility of paying for parking, extending parking time, themselves in a situation where they “couldn’t communicate
getting corresponding notifications through SMS or mobile with the top management group, were ineffective change
Internet services. In Lithuania, mobile payment for parking agents, or could not contribute to business strategy” [7].
grew to about 30% of all parking transactions in just 3 years. Transformation of the industrial landscape couldn’t have
For Tallinn, the capital city of Estonia, the number is 63%. left unaffected the job market, and student enrollments in
When looking for what constitutes the “convenience” in these managerial programmes. Secondary school graduates, when
services, in both cases we find that designers of the services deciding on future professions, cannot be encouraged by
were successful in accommodating the habits and preferences numbers showing that CS graduates are up to 10 times less
of local population (reminders, payment schemes, interfaces, likely to be employed, than graduates from any other subject.
etc.) into the service offering. Other successful non-Internet The employment stats published last year by The Guardian
mobile service uses examples exist, too, such as SMS-based [8] place CS graduates the last in the list of 20 different
purchase of public transportation tickets, vending machine degree programmes.
payments, etc. The recent data from the U.S. present a somewhat more
The trends, that we notice already, show that mobile phone, optimistic picture for the future of Informatics than that in
possessed today by the absolute majority of active society UK. For the first time in six years, the number of CS majors
members, enables fast and convenient access to mobile enrolled at U.S. universities increased by 8.1 percent [9]. IT-
services. If the necessary services are available, it is rather related jobs are reported to be still in demand and expected to
easy for a user to start using them, and, moreover, to enter the stay at this level or even experience growth in the years to
state of “can’t do without”. However, the growth potential is come. Especially this is true for those specialities, that require
still high in this area, and challenges exist. With aging more complex IT and interdisciplinary skills, e.g. IT
population in Europe, and the emergence of “smartphones” as management consulting [10].
mobile Internet access device, the majority of mobile phone
users aren’t familiar with many advanced mobile phone B. Inadequate education in schools
features, and would require an adequate training to take The prospects of the job opportunities as drawn by mass-
advantage of the devices’ functionality, as it is getting closer media in Europe are “complemented” by rather narrow scope
and closer to a computer functionality. of CS/Informatics teaching at secondary schools. This
tendency is true both for the U.S. and Europe. For example, a
III. THE DECLINE OF INFORMATICS AT UNIVERSITIES
3
Despite the growing dependence on ICT in daily lives, It is probably correct to say that “Business Informatics” is the
closest proxy of Management Information Systems programme in
work, and communication, for the last decade there was a Lithuania.
 CAN MOBILE COMPUTING HELP REVERSE THE DECLINE IN INFORMATICS STUDIES? 147

study by Blum and Cortina [11], as quoted in recent report (ECSS) held in 2005-6, where heads of computer science
“Informatics Europe” [12], identified the following reasons departments from all over the European region joined forces
for the decline in enrollment in CS courses and programs at to define and promote common policies and study common
the high school level in the US: issues. In August 2007 Informatics Europe published a report
• Dot-com bubble burst/not enough high paying careers “Student Enrollment and Image of the Informatics Discipline“
• Increased amount of required courses and tests [12]. The report aimed at answering the following questions,
• Elective status/competition with many other electives among other:
• Lack of understanding of CS by guidance counselors and • Why is student enrollment a problem in
administrators Informatics/Computing, at least in many of the (Western)
• Ill-prepared teachers/lack of interest European countries?
• Difficulty of material/not ‘fun enough’ • Why is enrollment by female students lagging behind?
• Irrelevance to students What are the reasons of it, and what can be done about it?
• Lack of meaningful curriculum and standards. • Do potential students have the right image of Informatics
The “Informatics Europe” report [12] further concludes as a field of study, as a science, as a profession?
that the situation in Europe is not much better but vastly • How should the field be positioned so it is “clearly” as
differs from country to country: “the problem is not only the attractive and challenging as many other disciplines and
very limited time for Informatics in the school schedules as perhaps even more so?
compared to e.g. mathematics or physics, but also the lack of Their main recommendations on this matter were the
good standards for content (‘what should we teach’), following:
textbooks (‘how do we describe it’) and didactics (‘how • Offer multidisciplinary and cross-disciplinary programs
should we teach it’) at this level” [12], p.10. Similar to the • Fix computing science’s image
situation in the U.S., there is a lack of teachers for Informatics • Increase women’s enrollment in CS
in Europe, too. • Train high school computing science teachers
Besides the narrow scope of Informatics programmes in • Make CS courses fun.
schools, there is reported to be a pressure on teachers of
Informatics from their fellow teachers, nurtured by fear that IV. DIRECTIONS OF THE REVISION OF INFORMATICS
with the rise of the Internet and ICT, Informatics will present PROGRAMMES
a threat to the established teaching subjects, and also a threat
to the teachers of these [12], p.11. Some of the recommendations on the revival of Informatics
have actually been implemented already for a while in what
C. Conservatism of university Informatics programmes came to be known as a School of Information (SI / i-School).
While the two aforementioned categories of factors With some 60 SIs established worldwide up to date, the
contributing to the decline of Informatics programmes are enrollments of SIs in the U.S. grew 50% since 2001, from
“out of control” for the universities, the revision of study 15.000 to over 20.000.
programs can and should be done to tackle the trend. The concept of the School of Information suggests the
Looking at the CS/Informatics study programmes at directions towards which the CS/MIS disciplines should be
different universities, we see that they hardly make use of the re-oriented. Specifically, there must be a shift from isolated to
potential of Informatics to cover broad field of study, based integrated, from technology-centric or community-centric to
on the view of Informatics as the “Science of Information”. synthesized inter-relational approaches to informational
The computational aspect in Informatics study programmes studies.
is usually addressed to the largest extent, taking into account Having examined the mission statements of a dozen of the
the large share of the programmes covered by programming, leading North American SIs, the aims and problem areas of
algorithm design, algorithm analysis, data management, the study programmes of different SIs can be synthesized in
operating systems, applied computing disciplines. Cognitive two broad categories, these of information-centric and
aspects are also to a certain extent present in the majority of technology-centric (see Figure 1 and Figure 2). The former
the programmes, covered by the mathematical logics, problematization is not that distant from the traditional
artificial intelligence, neural network disciplines. However, Informatics studies, and hence the lessons of SIs can be
the social aspect issue, becoming more and more important in applied to the revival of the discipline.
modern society, is clearly underestimated.
We can observe numerous attempts, trying to overcome the A. What Can Be Done At The School Level
crisis in Informatics studies. A European initiative aiming at While universities cannot exert direct control over the
taking a timely account of the transformation of the field, is Informatics’ teaching at schools, certain success in
“Informatics Europe“ association4, which was created as a stimulating pupils’ interest to the studies of Informatics can
result of the first two European Computer Science Summits be achieved with effort from the universities. Making
Informatics studies “more fun” starting from the school level
could be a way out of crisis. This could be done first of all by
4
http://www.informatics-europe.org/about.php
148 VITKUTE-ADZGAUSKIENE AND FOMIN

adequately exploiting the new trends inside computer science When analyzing the Lithuanian context, we can see, that
itself: “An unprecedented change in our information there are attempts to both add transdiciplinary focus and
processing practices [that] has resulted from the introduction include Mobile Computing topics in the Informatics study
programmes. Most cases observed are related to introduction
• Illuminate the role of information or development of transdisciplinary Informatics programmes,
namely “Business informatics” or “Management
Human
needs • Understand the transformation
• Synthesize the human-centered and
technological perspectives Informatics”, combining knowledge in the Informatics and
Information
• Illuminate both computational and cognitive
processes
Business Management fields. Completely new BSc and MSc
Social/ org.
structures
IS / IT • Encompass community and communication programmes in Business Informatics were established at
• Expand human capabilities through
information Vytautas Magnus University (VMU) and at Kaunas
• Turn data to information, information to Humanitarian Faculty of Vilnius University more than a
knowledge
decade ago. Mykolas Riomeris University added a “Business
Fig.1. Information-centric approach Informatics” BSc programme in 2008, after analyzing its
perspectives in the context of other similar programs in
• Understand how technology interacts with
Business
business processes, policy, and strategy
Lithuania [14].
The admission numbers for the new business informatics
processes
• How information impacts our lives,
organizations, cultures and societies
• Shape policies that influence how people seek, programs show they were popular among students [15], For
IS / IT use, and share information and create example, “Business Informatics” MSc programme at
Strategy Policy knowledge
Vytautas Magnus University collects twice as many students,
compared to the “Applied Informatics” MSc programme. The
overall admissions, however, continue to decline.
Besides the introduction of new specializations in the
Fig.2. Technology-centric approach
Informatics study programmes, new course modules, related
to new business and societal trends can be developed. Here,
and use of the World Wide Web… [is] being accompanied by Mobile Computing comes to play the central role. When
technological developments which are in turn educating talking about mobile communication or ubiquitous computing
successive generations as a by product of social practice”[13]. issues in the university CS curricula, usually two basic
Informatics today is different. It “changes every science opinions prevail. The first of them is stating, that mobile
that it touches”, “it leads the way to many promising new environment is just a subset of a broader electronic
directions in research”. And this new image of Informatics environment, therefore this topic shouldn’t be considered
must be promoted starting at the highschool / college level, separately. Another opinion places the mobile computing in a
and supported by new educational curricula for universities. special position, that of boosting the attractiveness and up-to-
Reports from the U.S. give promising signs on how the task date characteristics of Informatics study programmes at the
of the raising popularity of Computer Science and universities.
Information Systems can be dealt with. One of the reasons for These two approaches to certain extent are reflected in the
the growing interest in computer science degrees in the U.S. situation we can see at different universities, offering
is teens’ excitement about social media and mobile Informatics study programs. Some have special programmes
technologies. The perception that computer science is cool is related to mobile and ubiquitous computing, others have
spawned by all the interesting things on the Web. The iPhone special mobile-related modules included in standard CS
and Web 2.0 reinforces the excitement, and that attracts the programmes, while many remain conservative regarding this
best students [9]. Such excitement could not be not shared by point, maybe reflecting only selected topics in corresponding
the European teens. However, there may be some directing course modules, such as “network management”, “Internet
required on behalf of Informatics’ teachers at schools. technologies”, etc.
Looking at the cases when completely new mobile-oriented
B. How Inclusion of Mobile Computing in Study Programmes study programmes are introduced, we mostly find technically-
Can Help Reverse the Decline oriented programmes, which are usually categorized as
Taking into account the transformations in the belonging to electronics or electrical engineering fields.
environment, the rise of the mobile society, and the Another observed trend here is the orientation towards
excitement of teenagers about the social media and mobile, application design, related to mobile and ubiquitous
we can expect, that by including the topics of Mobile computing – for example, the MSc Program of the Lancaster
Computing in the curricula, universities could increase the university in UK “Mobile and ubiquitous computing”
popularity of Informatics studies. Such an assumption can be (http://www.comp.lancs.ac.uk/study/pg/ubi/). In this case, two
to a large extent justified by the success of the i-School modern computer science trends are combined – the new
concept, teaching programmes of which specifically target the ways of communicating and accessing the Internet, brought
aspect of the interaction of technology with society. Inclusion by the mobile computing revolution, and moving away from
of Mobile Computing would also reflect the desktop PCs to everyday digital devices and smart
recommendations on the emphasizing the transdisciplinarity environments. Such an approach can truly make CS teaching
in Informatics programmes [13], p.5.
 CAN MOBILE COMPUTING HELP REVERSE THE DECLINE IN INFORMATICS STUDIES? 149

programmes more attractive, as it deals with new ways of V. CONCLUSIONS

working, staying in touch, entertaining, and, consequently,
with new promising industry and business models. In this work we show, that the decline in CS enrollments
Talking about mobile-related modules, included into that is being observed globally can be ascribed to the overly-
existing study programmes, the majority of them are related conservative curricula of the traditional Informatics/CS
either to mobile technology introduction, or to the programmes, which do not reflect the role of mobile
development of mobile applications. An example of such a computing and social media in the modern society.
course could be “Application development for mobile The topic of Mobile Computing in CS curricula usually is
devices” from Pittsburg University, U.S.A under-valuated, given the global trend of the growing in scope
(http://www.ischool.pitt.edu/bsis/course-of-study/course- and depth mobile usage. There are large opportunities of
descriptions.php). We can also notice first attempts include including mobile topics in university CS curricula, such as the
mobile courses, related to the social aspects, as, for example, increase of attractiveness of study programmes and CS in
the “Mobile Social Networking” module, provided by the general, and the possibility to educate specialists with more
Department of Continuing Education at the Oxford University complex IT and interdisciplinary skills, as well as qualified IT
in UK (http://www.conted.ox.ac.uk/courses/ users, thus stimulating the overall knowledge society
details.php?id=O07C831H6J). development process.
Looking at the Lithuanian context, we see, that the The hypothesis of increasing attractiveness of study
moderate approach of including separate mobile-related programmes with mobile-related modules or topics covered,
modules or module topics in the existing Informatics is to a certain extent proven by current examples of
programmes prevails - examples include “Mobile Lithuanian Informatics studies, namely by the increased
Technologies” at Kaunas University of Technology, “Mobile interest in the mobile topics from the electives’ and final
Communication Technologies” at Vilnius Gediminas thesis perspective, as well as regarding student attitudes
Technical University, “Mobile and Wireless towards the whole improved “Business Informatics” MSc
Communications” and “Mobile Application Development” at programme at VMU.
Vytautas Magnus University (VMU). Also, mobile-related Two main approaches can be used for including Mobile
topics, covering also social aspects, are included in different Computing issues in university programmes – either the
VMU “Business Informatics” MSc Programme modules, such introduction of completely new programmes dedicated to
as “E-commerce and ICT infrastructures”, “Knowledge mobile topics, or the improvement of existing programmes
society ICT strategies”, etc. with mobile-related modules or topics, the choice dependent
The strategy of whether to choose a more revolutionary on the specific country context of study regulation and quality
approach of designing new programmes, or a more moderate assessment.
approach of improving the existing programmes clearly
depends on the country-specific environment, related to the REFERENCES
registration and assessment of study programmes, legal base
of study regulation, etc. Moderate approach taken by the [1] ITU. (2009, 2008-07-15). Market information and
Lithuanian universities is to a certain extent dependent on the statistics (stat): Global ICT developments. 1997-2007.
rather complicated current new programme assessment Retrieved Jun 18, 2009, from http://www.itu.int/
procedure and rather limiting normative documents for ITU-D/ict/statistics/ict/index.html
Informatics studies. [2] Comm. Regulatory Authority (2009). Lithuanian
When evaluating the impact of including Mobile Telecommunication Sector in 2008 (in Lithuanian).
Computing topics in the VMU Informatics study Retrieved June 22, 2009, from http://www.rrt.lt/
programmes, we can name the following outcomes: index.php?-1501980636
• Elective mobile-related modules attract 2-3 times larger [3] Department of Statistics to the Government of the
student numbers, than other elective modules offered in Republic of Lithuania (2008). Information technology
parallel. usage in households (in Lithuanian). Retrieved June 22,
• The number of students, choosing mobile-related topics 2009, from http://www.stat.gov.lt/lt/news/view/?id=2737.
for their final thesis, increased 2-3 times (about 20% of [4] comScore Inc. (2009). Mobile internet becoming a daily
BSc Thesis works in 2009 covering mobile aspects). activity for many. Retrieved June 18, 2009, from
There are signs of more positive attitude towards the whole http://www.comscore.com/Press_Events/Press_Releases/
MSc programme of “Business informatics” in 2009, the 2009/3/Daily_Mobile_Internet_Usage_Grows
hypothesis based on student responses to different surveys, [5] comScore Inc. (2009b). Mobile social networking
and quite a number of students expressing their wish to driving growth of the mobile internet in europe,
transfer from other programmes. However, the situation will according to comscore. Retrieved June 18, 2009, from
be clarified only after the end of admissions. http://www.comscore.com/Press_Events/Press_Releases/
2009/1/Mobile_Social_Networking_Europe/(language)/
eng-US
150 VITKUTE-ADZGAUSKIENE AND FOMIN

[6] comScore Inc. (2009c). Prepaid wireless services attract [11] Blum, L., & Cortina, T. J. (2007). Cs4hs: An outreach
increased online interest during economic downturn. program for high school cs teachers. Paper presented at
Retrieved June 18, 2009, from the The 38th SIGCSE Technical Symposium on
http://www.comscore.com/Press_Events/Press_Releases/ Computer Science Education.
2009/6/Prepaid_Wireless_Services_Attract_Increased_O [12] Leeuwen, J. v., & Tanca, L. (2008). Student enrollment
nline_Interest_During_Economic_Downturn and image of the informatics discipline (No. IE-2008-
[7] Andersen, E. (2007, January 24). Why don't cios become 01). Zurich: The Association of Informatics /
ceos? Retrieved May 11, 2009, from Computing Departments and IT Research Institutes in
http://www.espen.com/archives/2007/01/why_doesnt_ci Europe.
os.html [13] White, S. A., & Irons, A. D. (2006). The shape of
[8] The Guardian. (2008, Friday 8 August). Unemployment informatics: A UK perspective. In A. McGettrick & G.
by degree, 2004-07. Percentage of uk and eu domiciled McAlister (Eds.), Informatics Education Europe.
first degree qualifiers assumed to be unemployed by Montpellier, France.
subject area, 2002-03 to 2006-07. The Guardian. [14] Dzemydiene, D. (2008). The need and perspective for
[9] Marsan, C. D. (2009, March 30). Does a computer an interdisciplinary informatics study programme (in
science degree matter anymore? Retrieved May 11, Lithuanian). Mykolas Romeris University.
2009,from [15] Gruževskis, B., Blažien, I., Pocius, A., Dikius, G.,
http://www.networkworld.com/news/2009/040609-hot- Bagdonas, A., Jonauskis, M., et al. (2008). Systematic
tech-skills.html assessment of the correspondence between supply and
[10] Dawson, C. (2009, March 18). Is computer science the demand of high quality (MSc) specialists (in
one smart major in a recession? Retrieved May 11, Lithuanian). Vilnius.
2009, from
http://education.zdnet.com/?p=2271&tag=rbxccnbzd1
 Towards an Academic Social Network for Bologna Process
João C. Ferreira, Porfírio P. Filipe

{jferreira,pfilipe}@deetc.isel.ipl.pt
GuIAA – DEETC - Instituto Superior de Engenharia de Lisboa, Lisboa, Portugal

Abstract — The Bologna Process aimed to build a multi-faceted skills. In this scenario, the student proposals
European Higher Education Area promoting student’s for individual studies plan can be submitted and approved
mobility. The adoption of Bologna Declaration directives by the desired educational institution. The proposed system
requires a self management distributed approach to deal with offers a graphical user interface that is autonomous from a
student’s mobility, allowing frequent updates in institutions particular institution. The data acquisition and data
rules or legislation. This paper suggests a computational system submission can be implemented in multiple ways such as
architecture, which follows a social network design. A set of
web services or simple by file transfer protocols.
structured annotations is proposed in order to organize the
user’s information. For instance, when the user is a student its
annotations are organized into an academic record. The
academic record data is used to discover interests, namely II. BOLOGNA DECLARATION
mobility interests, among students that belongs the academic
network. These ideas have been applied into a demonstrator
As the Confederation of EU Rectors’ and the Association of
that includes a mobility simulator to compare and show the
student’s academic evolution.
European Universities (2000) explains it (CRE et al, 2000
p.3-4):
I. INTRODUCTION: BACKGROUND AND MOTIVATION “The Bologna declaration reflects a search for a
common European answer to common European problems.
With Bologna Process (BP), the social and cultural The process originates from the recognition that in spite of
developments impose a revision of the graduate and their valuable differences, European higher education
postgraduate education systems. Students can personalize systems are facing common internal and external challenges
their studies from a diversity of options. related to the growth and diversification of higher
Furthermore, study programmes have become modular, education, the employability of graduates, the shortage of
giving students the chance of more mobility during their skills in key areas, the expansion of private and
studies. Organizational and legal efforts have been taken to transnational education, etc. The Declaration recognises the
guarantee high compatibility of study programmes and value of coordinated reforms, compatible systems and
modules, e.g. the so called BP in Europe [1]. Resulting from common action.
this process students can personalized their studies and can The action program set out in the Declaration is based
compose different educational modules at any institutions to on a clearly defined common goal, a deadline and a set of
a highly personalized curriculum. This process will change specified objectives:
the role of educational institutions, such as universities [2]. A clearly defined common goal: to create a European
They are to be considered as components of a common, space for higher education in order to enhance the
integrated educational space. employability and mobility of citizens and to increase the
The traditional tasks like offering lectures, seminars, international competitiveness of European higher
materials, and infrastructure for education will be education;
supplemented by new tasks, such as personalized A deadline: the European space for higher education
educational programmes. This scenario requires new should be completed in 2010;
information systems and data integration among the A set of specified objectives:
different institutions. - The adoption of a common framework of readable and
In order to support this new academic reality the current comparable degrees, “also through the implementation of
work proposes a self-management distributed approach to the Diploma Supplement”;
deal with student’s mobility, allowing frequent updates in - The introduction of undergraduate and postgraduate
institutions rules or legislation. The management strategy is levels in all countries, with first degrees no shorter than 3
based on annotations organized in academic information years and relevant to the labour market;
records, which extend the student profile. - ECTS-compatible credit systems also covering lifelong
Such system should help current and prospective students to learning activities;
search for suitable course plans or single course units in - A European dimension in quality assurance, with
order to compose a personal curriculum by analyzing their comparable criteria and methods;
- The elimination of remaining obstacles to the free
mobility of students (as well as trainees and graduates) and

151
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_26, © Springer Science+Business Media B.V. 2010
152 FERREIRA AND FILIPE

teachers (as well as researchers and higher education descriptions supplementing Web Services available in a
administrators).” network of cooperating organizations is a remarkable
alternative to known standardisation issues. Instead of
III. THE SOCIAL NETWORK developing an inflexible standard API and producing a huge
The Social Web represents a set of web sites and overhead concerning its support and evolution, the service
applications in which user participation is the primary driver provider agree to reference the same domain ontology in
of value. The principal of such systems is well described by their Web Service descriptions. The ability of machine read
Tim O'Reilly [3] and this development is around Web 2.0 such descriptions, makes possible the automatic search of
[4] principals. Examples of these types of initiatives are required services. Therefore the interoperability of services
Wikipedia, MySpace, YouTube, Flickr, Del.icio.us, within the targeted network would be achieved. In this work,
Facebook, and Technorati. Discussions of the Social Web one will basically developed reference ontology and the
often uses the phrase "collective intelligence" or "wisdom of generation of a set of agent applications needed for the
crowds" to refer to the value created by the collective search, invocation and orchestration of Web Services
contributions of all these people writing articles for offered by educational institutions.
Wikipedia, sharing tagged photos on Flickr, sharing Semantic web architecture is a functional, non-fixed
bookmarks on del.icio.us. Knowledge sharing have great architecture [5]. Barnes-Lee defined three distinct levels that
potential due to the diversity of information available, incrementally introduced expressive primitives: metadata
diversity of individual cultures and the low costs of layer, schema layer and logical layer [6]. XML and XML
gathering and computing over big network (Internet). schema define syntax, but mean nothing about the data that
Currently we have millions of persons offering their it describes. That means that some standards must be built
knowledge online, which means that the information is on top of XML that will describe semantics of data. This
stored, searchable, and easily shared, so the big challenge is conduct to RDF and a general model in metadata layer RDF
to find the right match between what is available online and schema, which provides mechanisms for defining the
the approach to take the right/useful information from this relationships between properties declared and others
data. A collective intelligence/knowledge can result from the resources. To enable services for the semantic web, one has
data collected from all those people that are aggregated and the logical layer on top. This layer introduces ontology
transformed to create new knowledge. This collaborative languages, that are based on meta-modeling architecture
approach had the spam problem and other fraudulent sources defined in lower layer. This enables the use of tools with
in the mix, but the majority of users have a proper behavior generic reasoning support, independent of the specific
and can handle these disrupt behavior [3]. problem domain. Examples of these languages are OIL
So, our proposal is collaborative system based on a social (Ontology Inference Layer), DAML (DARPA Agent
network design that is able to connect not only at your own Markup Language), OWL and AOBP. OWL is a semantic
institution but to others within the European Higher markup language for publishing and sharing ontologies in
Education Space. This collaboration is achieved by users the web. OWL is developed as a vocabulary extension of
annotations divided into two main description fields: (1) RDF and is derived from DAML and OIL [7, 8].
semantic annotation, descriptions using terms from system
ontology; (2) descriptions performed by free descriptions
Logical Layer
terms introduced by the users. All these stakeholders can OWL
AOBP
provide semantic annotation by a specific metadata Schema Layer RDF Schema Semantics

generation and usage schema aiming to enable new RDF

information access methods and to enhance existing ones. A Metadata Layer

XML/XML Schema
Syntax
web tool provides access to ontological space, where users
can choose predefined terms, see description at section 6. A
Figure 1: OWL and AOBP in the semantic Web
synergy can be established between students, which are the
Architecture.
producers, and the consumers of information and machines
in the role of enablers (store, search, combine, inference). In
V. BOLOGNA ONTOLOGY
the communication process people create knowledge and
machines can help on this process, giving tools to help the The Ontology for Bologna Process consists of four major
process of communication and information sharing. So our subjects: (1) Course Plan; (2) Academic Record; (3)
approach will address the knowledge offered by millions of Individual Studies Plan; and (4) Diploma Supplement as
people (experts and also novice users) with a background identified in Figure 2. The Academic Record can be
infrastructure for storing, searching and easily sharing of processed to produce a student Individual Studies Plan
information. according to a determined Course Plan. When the student
completes the requirements of Individual Studies Plan a
IV. SEMANTIC WEB AND OWL Diploma Supplement can be released. Skills as defined in
Bologna process are the basis transformation unit.
The required Interoperability can be achieved by exploiting
of Semantic Web Services. Machine readable “semantic”
 TOWARDS AN ACADEMIC SOCIAL NETWORK FOR BOLOGNA PROCESS 153

Plan aggregates all the Course Units that figure in the

Course Plan indicating also the skills achieved when the
student is approved in the Course Unit.
Student C. Individual Studies Plan
Institution
Individual Studies Plan, illustrated on Figure 5, is similar to
a Course Plan representing the already achieved ECTS,
Course Plan Academic Record IndividualStudiesPlan DiplomaSupplement
according the course units in the student’s academic record,
1
and the amount of credits by scientific areas by scientific
1 1 1 1 1
1 areas considering a specific course plan.
Transformation Rules

Transformation Rules

Figure 2: General overview

A. Academic Record
The Academic Record (AR), referenced in Figure 3, can
aggregate information items in several languages, namely:
(1) Extra Curricular Course; (2) Qualification; (3) A Figure 4: Course Plan diagram
Contract of Studies (is celebrated between the student and
the institutions to formalize the terms of the temporary
mobility); (4) Work Experience; and (5) Student Data.
The AR includes all student paths during the formation
period. There is no need of often cross checks between AR
and the information that is kept private by the process
stakeholders. For each, information unit updated in the AR,
the student should send, if needed, a confirmation document
to the institution where he belongs. For instance, if the
student completes a professional formation he should update
his AR and after he will deliver the right certificate to his
higher institution education that holds the AR. When the
students do course change or intuition transfer, a copy of the Figure 5: Individual Studies Plan diagram
AR is produced and must be detained by the destination D. Diploma Supplement
institution.
Diploma Supplement follows the model developed by the
European Commission, Council of Europe and
UNESCO/CEPES. The purpose of the supplement is to
provide sufficient independent data to improve the
international “transparency” and fair academic and
professional recognition of qualifications (diplomas,
degrees, certificates etc.). It is designed to provide a
description of the nature, level, context, content and status of
the studies that were pursued and successfully completed by
the individual named on the original qualification to which
this supplement is appended. It should be free from any
Figure 3: Academic Record diagram value-judgments, equivalence statements or suggestions
about recognition. Information in all eight sections should
be provided. Where information is not provided, an
B. Course Plan explanation should give the reason why. Diploma Suplement
The Course Plan characterizes a specific graduation plan by diagram is a detailed description of the activities developed
the means of describing the Figure 4: (1) Curricular by the student in the graduated process based on the follows
Structure; and (2) the Studies Plan. attributes: employee, student data, course unit, qualification,
The Curricular Structure represents the distribution of ECTS scientific area and associated language.
credits, minimum and maximum, of each scientific area. The This UML representation can transform to a RDF; (2)
goal of the Curricular Structure is to define the amount of missing values are replaced with default values when
ECTS, because of the optional course units. The Studies possible, and (3) the RDF is validated against the metadata
154 FERREIRA AND FILIPE

schema and other validation rules. Each metadata producer A. Enhancing an Existing CMS with Bologna Ontology
gets a report of warnings, errors and other problems that Most content providers in SIR use a CMS for authoring,
were encountered during harvesting and validating the publishing and archiving content on their website. A typical
content. If some parts or all of the metadata is unacceptable CMS supports creation of textual metadata about
due to serious errors, the metadata is discarded until documents, such as title and publication time, but not
necessary corrections are made. Otherwise, the metadata is classified annotations. This would require that the system
added to and published in the Semantic Information has functionalities supporting ontology-based annotation
Repository (SIR), see section 6. work, e.g., concept search for finding the relevant concepts
(identified with URIs), concept visualization for showing the
concept to the user, and storing concepts along other
information about the documents. The CMS should also be
able to export the metadata, preferably in RDF format, to be
used by semantic web applications.
Figure 6: Diploma of Supplement diagram Currently, classification systems are typically shared by
downloading them, and each application must separately
E. Processing Rules implement the classified defined schema. To avoid
The skills concept <www.disco-tools.eu> allows the duplicated work and costs, and to ensure that the classified
establishment of a semantic link between: (1) Extra systems are always up-to-date, we argue that one should not
Curricular Course; (2) Qualification; (3) A Contract of only share the classified system, but also the functionalities
Studies; and (4) Work Experience. for using them by centralized mash-up component services.
A specific course plan is completed when the hosting Such services, e.g. Google Maps, have been found very
institution recognizes that the student has achieved all the useful and cost-effective in Web 2.0 applications for
skills that belonging to a course plan. integrating new functionalities with existing systems.
The skills is a matching unit that allows comparison We have applied the idea of using mash-ups to provide
information items that aggregates its owns skills. A trial classified system services for the content producers of SIR
example is the comparison of course unit against course creating the ONKI Ontology Server framework [9]. ONKI
unit. However it is possible to compare course units against provides ontological functionalities, such as concept
work experience or qualifications. This matching can be searching, browsing, disambiguation, and fetching, as ready-
based on historical tables, recommended tables or even to-use mash-up components that communicate
requires human supervision. asynchronously by AJAX (or Web Service technologies)
with the shared ontology server. The service integration can
VI. BOLOGNA APPLICATION SUITE be easily done by changing only slightly the user-interface
component at the client side. For example, in the case of
AJAX and HTML pages, only a short snippet of JavaScript
code must be added to the web page for using the ONKI
services.
The main functionality addressed by the ONKI UI
components is concept finding and fetching. For finding a
desired annotation concept, ONKI provides text search with
semantic auto completion [10]. This means that when the
annotator is typing in a string, say in an HTML input field of
a CMS system, the system dynamically responds after each
Figure 7: BAS architecture system. input character by showing the matching concepts on the
The Bologna Application Suite, is our proposal for a Web ONKI server. By selecting a concept from the result list, the
2.0 system environment and major components are concept’s URI, label or other information is fetched to the
described at Figure 7. A major challenge in the distributed client application.
content implementation of Bologna process is how to Concept browsing can also be used for concept fetching.
facilitate the cost-effective production of descriptive, Study plan only changed in specific time periods, so the
semantically correct high quality metadata. In SIR three content change can be handling in a better way. In this
ways of creating metadata are considered and supported: (1) model, the user pushes a button on the client application that
Enhancing existing web content management systems opens a separate ONKI Browser window in which
(CMS) with ontology mash-up service components for annotation concepts can be searched for and browsed. For
producing semantic metadata; (2) Using a browser-based each concept entry, the browser shows a Fetch concept
metadata editor for annotating web content; (3) Automatic button which, when pressed, transfers the current concept
conversion of metadata. These approaches are outlined information to the client application. ONKI supports
below. multilingual ontologies, has a multilingual user-interface,
supports loading multiple ontologies, and can be configured
extensively.
 TOWARDS AN ACADEMIC SOCIAL NETWORK FOR BOLOGNA PROCESS 155

ONKI is implemented as a Java Servlet application running System is running for 3 months and we are now getting the
on Apache Tomcat. It uses the Jena semantic web first results, with the start of a new academic year. The
framework for handling RDF content, the DirectWeb System prove to use well past experience, i.e. once a course
Remoting (DWR) library for implementing the AJAX unit equivalence is performed from the course x to the y, the
functionalities, the Dojo JavaScript toolkit, and the Lucene others request are automatic processed. This reality helps a
text search engine. lot because most changes occur from a small percentage of
10 courses. At this moment we have register a community of
B. Browser-based Metadata Editor
100 users (small number) but we think this number will be
Some SIR content providers can not add mash-up ontology increase with the usage of the system.
support to their CMS due to technical or economical The users when access the system, they need to choose the
reasons. Furthermore, some content providers do not even institution through a menu, see Figure 8. They choose first
have a CMS or they may not have access to the CMS that the country, then the city and finally the institution. Since
contains the content, e.g., if the content originates from a we did not developed the security module the application is
third party. To support metadata productions in these cases, only available in ISEL network, but the BAS system will be
we have created a centralized browser-based annotation prepared for a European usage. Bologna Process makes
editor SAHA [11] for annotating web pages based on sense if a big network is constructed around it. Login at
RDF/OWL-based ontology as a schema (collection of BAS can be performed by a Figure 9 menu. If they are new
classes with a set of properties, described on section 4). An users they can create new information or import academic
annotation in Saha is an instance of a schema’s class that register from other institution.
describes some web resource and is being linked to it using
the resource’s URL (in same cases, URI). We make the
distinction between the annotation of a document (e.g. a web
page) and the description of some other resource (e.g. a
person) that is somehow related to the document being
annotated. In addition to containing classes used to annotate
documents (annotation classes), an annotation schema used
with Saha can also contain reference classes for describing
resources other than documents. In other words, an
annotation schema forms a basis for the local knowledge
base (KB) that contains descriptions of different kinds of
resources that may or may not exist on the web. Instances of Figure 8: BAS System Menu to choose the academic
the reference classes are used as values of properties in institution.
annotations. The schema element fields in SAHA can be
connected with ONKI ontology service components,
providing concept finding and fetching services to the
annotator, as discussed above. In practice, an annotation
project is Jena’s ontology model stored in a database. A Figure 9: User login menu: (1) for a new user; (2) to
model is comprised of the annotation schema and the import the Academic Register from other institution.
instances of the schema’s classes. It can be serialized to Due to space limitation is not possible to give all details
RDF/XML in order to use the annotations in external about the application. For a complete description see [12],
applications. we will same detail about collaboration issues, as illustrated
C. Automatic Conversion on Figure 10 and 11. In the Figure 10, is the menu process
The third content producing method in SIR is automatic associated with the creation of a new qualification. The
conversion of original data to SIR metadata. This method is number two is another collaboration tag where you can
currently used in cases where metadata exists in a CMS, but make a note to a course simply by being classified and
it is in an incompatible format, does not contain ontological selected the desired course.
annotations (URIs) and/or some minor information is Users can create annotations related with transfer process or
missing in the metadata. simply comments about the subject. In Figure 11 we show
an annotation process regarding course unit equivalence.
VII. CASE STUDY Predefine information field’s helps the reuse of the
information by the users.
We perform a real case implementation of our BAS system
to a polytechnic institute of Lisbon (ISEL). ISEL have more A. Simulation of Course transfer
than 5000 students distributed by 7 engineering degree Students simulate a transfer or change of course.
(licenciatura in Portuguese) and 6 Masters. Bologna reform A prerequisite to perform this operation is the creation of an
was introduced at ISEL in the academic year of 2006/2007. individual study plan. To perform the simulation simply
The BAS system was proposed to help academic process select the individual plan of study desired travel destination
regarding students changes (inside institution and outside). and the location will be generated an XML document
156 FERREIRA AND FILIPE

created with all the information of the simulation. The classified system rather than keywords; (3) The problem of
results are represented in a Simulation Menu (Figure 12), outdated and missing links is eased by providing the end-
where the student can see the results of equivalences if he user with semantic recommendations that change
applies that change. dynamically as content is modified; (4) Content aggregation
is facilitated by end-user facets that collect distributed but
related information from different primary sources; (5)
Quality of content is maintained by including only
trustworthy organizations as content producers (academic
institutions); (6) share suggestions and comments of
students regarding the courses.
At the same time, the problems Bologna Mobility process:
(1) Redundancy of content creation can be minimized by the
possibility of aggregating cross-portal content; (2) Reusing
the global content repository is feasible. By using a central
SIR, external applications, can reuse the content; (3) Internal
and external link management problems are eased by the
dynamic semantic recommendation system of the portal and
the content aggregation mechanisms; (4) The tedious content
Figure 10: Creating a new qualification. indexing task is supported cost-effectively by shared
ontology service components; (5) Metadata quality can be
enhanced by providing indexers with ontology services by
which appropriate indexing concepts can be found and
correctly entered into the system.

REFERENCES
[1] CRE et al, 2000. Confederation of EU Rectors’ Conferences and the
Association of European Universities (CRE), 2000. The Bologna
Declaration: an explanation.
[2] B..J. Krämer (2003), CUBER für die Ausbildung á la carte, in: G.
Simonis (ed.), Lernort Bologna Declaration Communiqué of the
meeting of European Ministers in charge of Higher
Education,Prague,May 19th 2001, ec.europa.eu/education/policies/
Figure 11: Example of a anotation performed by a educ/bologna/bologna.pdf.
student using the descriptors available on the ontological [3] T. O'Reilly. What Is Web 2.0: Design Patterns and Business Models
space. for the Next Generation of Software, September 30, 2005.
http://www.oreillynet.com/pub/a/oreilly/tim/news/ 2005/09/30/what-
is-web-20.html>.
[4] Web 2.0 Conferences. http://www.web2summit.com/ Wensing, M.,
Grol, R. and Smits, A. (1994) Quality judgements by patients on
general practice care: a literature analysis. Social Science and
Medicine 38, 45±53.
[5] Tim Berners-Lee, Weaving the Web, Orion Business Books, London,
1999.
[6] van Harmelen, F., et al, “OWL Web Ontology Language Reference”,
W3C Working Draft, http://www.w3.org/TR/2003/WD-owl-ref-
20030331/, 2003.
[7] Miller, J., Mukerji, J. (eds.), “MDA Guide Version 1.0”, OMG
Document:omg/2003-05-01,
[8] Paulo Trigo and Helder Coelho. Agent inferencing meets the semantic
web. In Progress in Artificial Intelligence, EPIA-09, volume 5816 of
Lecture Notes in Artificial Intelligence (LNAI), pages 497-507.
Springer-Verlag, 2009.
[9] Handschuh, S., Staab, S. and Ciravogna, F., S-CREAM -- Semi-
Figure 12: Simulation Menu automatic CREAtion of Metadata in SAAKM from Web Pages in First
International Conference on Knowledge Capture, (2001).
VIII. CONCLUSIONS [10] Hyvönen, E., Viljanen, K., Mäkelä, E., et al.: Elements of a national
semantic content infrastructure—case Finland on the semantic web.
This paper addressed the needs of Bologna Mobility In: Proceedings of the ICSC 2007, IEEE Press (2007).
process, with a proposed of BAS system in a Web 2.0 [11] Valkeapää, O., Alm, O., Hyvönen, E.: Efficient content creation on the
semantic web using metadata schemas with domain ontology services
environment to: (1) Support information exchange; (2) (system description). In: Proceedings of the ESWC 2007, Springer–
Content finding is supported by cross-portal semantic Verlag, Berlin (2007).
search, based on hierarchical structure imposed by a [12] Paulos, J. and Costa, R. Bologna Application Suite. Final Year project
available at www.deetc.isel.ipl.pt/matematica/jf/bas.pdf.
 Supporting Knowledge Discovery in an eLearning
Environment Having Social Components
Paola Monachesi and Thomas Markus Vlad Posea and Stefan Trausan-Matu Petya Osenova and Kiril Simov
Utrecht University Politehnica University of Bucharest Bulgarian Academy of Sciences, Sofia,
The Netherlands Romania Bulgaria

Abstract – One of the goals of the “Language Technology for knowledge, as concluded in [1]. In our specific case we use an
LifeLong Learning” project is the creation of an appropriate ontology in the domain of computing, which is employed to
methodology to support both formal and informal learning. provide deep annotation of the learning materials that should
Services are being developed that are based on the interaction facilitate their understanding and reuse.
between a formal representation of (domain) knowledge in the
form of an ontology created by experts and a social component
However, the ultimate goal of the LTfLL project is to
which complements it, that is tags and social networks. It is complement the formal knowledge represented by ontologies
expected that this combination will improve learner interaction, developed by domain experts with the informal knowledge
knowledge discovery as well as knowledge co-construction. emerging from social tagging in order to provide more flexible
and personalized ontologies that will include also the
I. INTRODUCTION knowledge of communities of users. The enhanced ontology
will be connected to the social network of the learners through
In a Lifelong Learning context, learners access and process the tags that they provide, improving thus the possibility of
information in an autonomous way. They might rely on formal retrieving appropriate material and allowing learners to connect
learning, that is they might focus on textual material approved to other people who can have the function of learning mates
by content providers in the context of a course developed by an and/or tutors.In this way, it will be possibleto provide a more
organization or institution. They might also want to rely on personalized learning experience able to fulfill the needs of
informal learning, that is on (non-)textual content available different types of learners.
through the web which is uploaded and accepted by the The paper is organized as follows. In section II, we discuss
community of learners and not necessarily by a content the differences between formal and informal learning. Section
provider of an institution. III presents the design of the Common Semantic Framework,
One of the objectives of the Language Technologies for while section IV focuses on its implementation. In section V,
LifeLong Learning (LTfLL)1 project is to develop services that we present an initial evaluation of the system while VI
facilitate learners and tutors in accessing formal and informal introduces some concluding remarks.
knowledge sources in the context of a learning task. More
specifically, the aim is to support learner’s interaction in order II. FORMAL VS. INFORMAL LEARNING
to facilitate knowledge discovery and knowledge co-creation.
To this end, a Common Semantic Framework (CSF) is being As already pointed out above, in the LTfLL project, we are
developed. The CSF allows the stakeholders to identify, developing services that facilitate learners and tutors in
retrieve and exchange the learning material. More specifically, accessing knowledge sources within the context of a certain
it supports formal learning by addressing course material which learning task. The learning process is considered an attempt to
includes textbooks, articles, slides as well as informal learning make the implicit knowledge which is encoded in the learning
which we identify with (non-)textual material emerging from materials, explicit.
social media applications. Within the CSF, communication is In the LTfLL project, we aim at supporting both types of
facilitated through the use of social networks and new learning - formal and informal one. Under ‘formal learning’ we
communities of learners are established through the understand the process, in which the learners follow a specific
recommendations provided by the system. In order to provide curriculum. For each topic in the curriculum, there is a set of
recommendations, the user’s profile, his interests, his learning materials, provided by the tutor. In contrast to this, the
preferences, his network and obviously the learning task are curriculum in the informal learning process is not obligatory,
taken into account. It is through the access to formal and and the role of the tutor either is absent or is not obvious. The
informal material that new knowledge will originate. learners exploit different sources to succeed in their learning
Knowledge repositories are employed to achieve the above goal, very often relying on the information, available on the
goal. An appropriate way to retrieve formal content might be web, and on the social knowledge sharing networks.
by means of an ontology which can support the learner in the From the perspective of formal learning (i.e. guided,
learning path, facilitate (multilingual) retrieval and reuse of monitored one), learners rely mainly on material which has
content as well as mediate access to various sources of been prepared by the tutor on the specific topic and is
addressed by the curriculum. Learners are expected to access
1 certain pieces of pre-stored information, which would lead
http://www.ltfll-project.org/

157
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_27, © Springer Science+Business Media B.V. 2010
158 MONACHESI ET AL.

them to the required knowledge. In the informal learning logging each day to Facebook.com.3
process, learners have to locate knowledge sources on the web, The communities that emerge through social media
then to select the relevant ones, and finally to investigate each applications such as Delicious, Flickr or YouTube provide two
of them. During this process, learners often need guidance and important elements that can be integrated in the Common
thus they might profit from finding appropriate people (peers) Semantic Framework, in order to support informal learning.
on the web. They provide the “knowledge of the masses” in the form of
The two learning paradigms are complementary from the tags that are being used to annotate resources. On the other
lifelong learner’s perspective. In the LTfLL project, we are hand, the structure of these communities which can be
developing two sets of services that would support both - the represented through a social network can be employed to
formal style of learning and the informal one. These services recommend content and peers. The social networks
are integrated within the Common Semantic Framework. automatically introduce a notion of trust in the search results in
relation to the user's social network.
III. COMMON SEMANTIC FRAMEWORK: DESIGN The knowledge produced by communities in the form of tags
can be employed to enrich existing domain ontologies semi-
One of the aims of the LTfLL project is to build an automatically. More specifically, we have merged the dynamic
infrastructure for knowledge sharing which we call Common knowledge provided by users/learners through tagging with the
Semantic Framework (CSF). It allows for identification, formal knowledge provided by the ontologies by adding
retrieval, exchange and recommendation of relevant learning tags/concepts (or instances of concepts) and relationships
objects (LOs) and of peers. It is ontology driven allowing thus between concepts in the domain ontology. In the CSF, the
for a formalization of the knowledge arising from the various connection between words/tags and concepts is established by
stages of the learning life-cycle. Domain ontologies offer means of language-specific lexicons, where each lexicon
useful support in a learning path since they provide a specifies one or more lexicalizations in one language for each
formalization of the knowledge of a domain approved by an concept. DBpedia [5] is used as a knowledge base to resolve
expert. In addition, ontologies can support formal learning the extracted lexicalization to unambiguous existing concepts.
through the annotation of the learning objects [1]. DBpedia doesn’t provide much information about relations
More specifically, with respect to formal learning, we are among concepts when compared to specially crafted domain
developing a Formal Learning Support System (FLSS) that ontologies, but compensates this shortcoming with the huge
supports the individual work of the learners and tutors in number of available concepts.
manipulating the learning objects, including the knowledge There is an important side effect of this ontology enrichment
which is implicitly encoded in them and in adding their own process: if tags given by learners or emerging from social
descriptions to these learning materials. In particular, an media applications are related to the concepts present in the
annotation service has been implemented which is meant as a ontology, we manage to include not only the expert view of the
way to make the implicit knowledge attested in learning domain, but also the learner’s perspective. In fact, domain
objects, explicit. Thus, through the annotation service, we aim ontologies developed by knowledge engineers might be too
at relating the knowledge, encoded in the text, to the formally static, incomplete or might provide a formalization that does
represented knowledge in a domain ontology. not correspond to the representation of the domain knowledge
Despite the possible limitations of the ontologies, they available to the learner which might be more easily expressed
provide a highly structured way of interlinking different by the tagging emerging from communities of peers via
knowledge sources. From the point of view of the learning available social media applications.
process itself, the domain ontology can be viewed as a way to It is important to help the learner manage the content that is
produce a concise representation of the most important created daily around him in his learning communities. The
concepts in a given domain. Consequently, the learner has to knowledge produced inside the learner’s network is very
interact with them in order to assess his current level of valuable to him. According to the Communities of Practice
familiarity with the domain or to use them to acquire new (CoP) theory [6], a learner acquires knowledge, as he moves
knowledge. from the periphery to the center of a CoP.
While in formal learning, courses and learning objects play a In order to facilitate the informal learning experience, we
relevant role, in informal learning, communities arising from automatically monitor the learners’ activities and his peers on
social media applications are becoming more and more the social network applications. The data that the learner and
dominant, but are poorly integrated in current learning his network create or annotate is indexed in a semantic
practices. Learners prefer to use the internet to search for an repository. This data can then be used to offer learning support
answer rather than asking a tutor or a colleague [3]. Most of the which exceeds that which is currently offered through keyword
documentation for a given learning task is found on the internet search or simple recommendations.
and most learners do not check if the information is accurate or In this way, the CSF supports self-organization and the
reliable [4]. Recently, the usage of the internet has shifted emergence of collaborative knowledge and classification. In
towards social networking applications. 1 billion movies are addition, we aim at connecting learners to other learners in an
seen everyday on YouTube.com2 and 150 million users are
3
http://www.facebook.com/press/info.php?statistics
2
http://youtube global.blogspot.com/2009/10/y000000000utube.html
 SUPPORTING KNOWLEDGE DISCOVERY IN AN ELEARNING ENVIRONMENT 159

appropriate way. To this end, the content the learner is

searching and selecting can be used as a trigger to get him in
touch with other users that employ similar content or
annotations. Furthermore, the system monitors the changes that
appear in his network with respect to content and to users over
time and will provide recommendations for how he could
update his network (by adding new peers or removing old
ones). Managing the social network is especially important for
novices that need to create their own CoP with people with a
similar interest. Alternatively, if a learner is already part of a
CoP, he needs to keep up with the changes or new information
in his domain(s) of interest. Most of the learner's attention will
probably go to content produced by peers that are highly
relevant and trustworthy. Therefore, it is important that the
system maximally exploits these relations.
To summarize: In the CSF we establish an obvious link
between the network of users, tagging and the resources (cf.
also [7]). We further extend this theoretical model by
automatically linking and enriching existing domain ontologies
in order to structure the heterogeneous information present in
social networks, (deeply) annotated resources, and concepts.
The knowledge rich ontologies integrated with tagging provide
compelling advantages for a wide range of tasks such as
recommendation, community building and discovery learning.
Figure 1. The Common Semantic Framework implementation
IV. COMMON SEMANTIC FRAMEWORK: IMPLEMENTATION The crawler extracts information from a user’s social
network such as indexing bookmarks that the user is posting on
The Common Semantic Framework implementation is based Delicious, videos posted on YouTube, slides posted on
on a Service Oriented Architecture (SOA) that represents the Slideshare together with the tags used to classify the resources
backbone of the system. It includes five components: data and information about the social connections developed inside
aggregation, ontology enrichment, multimedia document these web sites. The data is converted into semantic
annotation, search and visualization. The services that link information using ontologies like SIOC, SCOT, FOAF.
these components have a strong emphasis on semantics by The data extracted by the crawler can be interpreted as a
employing shared URI's for all the various entities such as folksonomy, which is a hypergraph describing the information
concepts, resources, users and annotations. Furthermore the about users, resources and tags as specified in [7]. The storage
information exchanged through the Web services carries of the folksonomy in a repository allows the system to provide
semantic meta-data, as most of the data used in the CSF is various search and recommendation services. These services
stored in RDF and OWL. can provide references to users, resources, tags, concepts and
As can be seen in figure 1, the communication between the annotations that the learner can use in his learning activities.
modules is implemented using web services and the core of the The advantage of using a semantic repository and commonly
CSF is represented by the semantic repository. The repository used vocabularies for the purpose of data integration is that we
adopts RDF/OWL semantic vocabularies like SIOC, FOAF, obtain a highly integrated dataset containing semantically
Dublin Core and SKOS to represent information extracted from compatible data from different internet sources. These data can
the social networking applications and employed by all the be used to enrich an existing domain ontology for providing
modules. The other modules, that is ontology enrichment and improved search services that can exploit the structured
search, use web services to extract information from the knowledge formalized within an ontology as well as the
repository in order to provide the required information. The knowledge coming from the participating communities.
final results are converted to data that can be visualized It should be noted that the repository also contains a
through a widget inside a learning or social networking collection of learning material collected from the web which
platform (e.g. Moodle or Elgg). has been semantically annotated.
Here, the various components are described in more detail.
Ontology enrichment with social tags
D a t a Ag g r e ga t i o n The purpose of this module is to benefit from the
The data aggregation module consists of a crawler and the information extracted from social media applications for the
services that link it to the semantic repository. The crawler uses enrichment of existing domain ontologies. To this end, we
APIs provided by an increasing number of social networking determine which concepts and relations are the best candidates
applications to get information about users, resources and tags. to be added to the given domain ontologies. In the CSF we
take, as starting point, the ontology on computing that was
160 MONACHESI ET AL.

developed in the "Language Technology for eLearning" project we have implemented two types of search, that is ontology
(cf. [1]). It contains 1002 domain concepts, 169 concepts from based search that exploits the structured information attested in
OntoWordNet and 105 concepts from DOLCE Ultralite. the ontology as well as search based on tags and social
Candidates are generated by applying similarity measures or networks that exploits the social aspects of the knowledge
other sorting mechanisms to derive socially relevant tags. provided. Our plan is to integrate these two types of search into
Implemented examples include coocurrence-based and cosine- a recommendation system. It should exploit the strong features
based measures [9]. This analysis drives lookups and selective of both types of search, that is ontological structure and
integration from DBpedia for the domain ontology enrichment. community based knowledge (i.e tags and social networks).
The various web services provide additional lexicon entries, Ontology based search
socially relevant specific classes linked to more abstract classes The semantic search can be triggered by a list of keywords,
already present in the ontology and additional relationships provided by the user. The query string is analyzed on the basis
discovered between already existing domain concepts. The of a lexicon, mapped to the domain ontology which is enriched
system provides an automatically enriched ontology which with tags. The result of this analysis is a set of concepts from
contains the vocabulary of the Community of Practice that the the ontology. The concepts, related to the keywords, are
user is part of. More specifically, the resulting ontology additionally processed on the basis of the information from the
integrates the socially relevant concepts, but with the structure ontology by adding related concepts, that is query expansion.
of an expert view domain ontology. Extraction methods Query expansion is done via reasoning over the ontology. The
exclusively focused on deriving ontology-like structures from most typical case is the addition of the sub-concepts to the ones
folksonomies cannot provide such a high quality of results due extracted from the ontology on the basis of the keywords. The
to unavailability of the implicit knowledge in folksonomies initial set of concepts is interpreted conjunctively and the
which has been made explicit in domain ontologies. addition from the query expansion is added disjunctively to it.
Multimedia Document Annotation The query, formed in this way, is evaluated over the
The aim of the multimedia document annotation is to repository of annotated documents. The result is a list of
annotate explicitly learning material with concepts from the documents and in the case of the annotated ones it is a set of
domain ontology previously mentioned. The purpose of this pointers to parts of the document which are considered as
annotation is to allow semantic search, based on the ontology. relevant to the query - sentences, paragraphs and pictures. The
The usage of the domain ontology in the search provides possibility of ontology browsing is also offered to the learner.
abstraction from the concrete wording of. The annotation is More specifically, the enriched ontology can be exploited by
done in the document. This type of annotation allows the user the learner to get an overview of the domain specifically
to be directed to a particular part of the document. tailored to some search query, relevant user or specific
Within the CSF both text annotation and image annotation document. This allows the learner to discover resources which
are possible. The text annotation is implemented as a language do not match previously known keywords by supporting
pipe performing tokenization, POS tagging, lemmatization, discovery of new concepts and associated resources through a
semantic annotation and coreferential annotation. The first visualization of the domain.
three steps rely on available state-of-the-art tools. The semantic Search based on social networks and tags
annotation is performed by a concept annotation grammar In addition to ontology based search and browsing we have
constructed on the basis of an ontology-to-text relation as implemented an alternative type of search based on the
described in [13] and [14]. The coreference annotation module structure of the social network and the tags provided to
provides mechanisms to make the concept annotation more resources.
precise and with a better coverage of text (15% improvement). More specifically, in order to return trusted results, the
The image annotation is performed manually. It is search service focuses on documents created or bookmarked by
complementary to the text annotation. The reason to include it the learner’s friends and close contacts. The results found in
is that very often the pictures and diagrams, etc, present in the this way also contain the name of the users who created or
learning documents contain important information for the recommended them. Being in the learner’s network, he can
learner. The user interface consists of a document browser, now contact them and ask for more information or simply trust
image editor and ontology browser. The user selects an image the document returned more as it has the guarantee of a peer
from the document browser. The image is opened within the recommendation. The resources found this way will most
image editor. Then the user can select a region of the image likely be in the learner’s zone of proximal development [10].
and assign one or more concepts to the region. The The search algorithms can also identify the most relevant
interpretation of this type of annotation is that the region peers for a given task. The learner can interact using chat or e-
depicts an instance of the corresponding concept. mail with the peer to ask further questions and the peer might
The document annotation is stored in the semantic repository very well be helpful. He is already part of the learner’s social
together with links to the documents which will be used for network and we suppose that this means there is already a
document search by the ontology search engine. connection established between them.
Search The search services use the FolkRank algorithm [8]. This
The search service takes the learners’ query as input and algorithm is used because it can retrieve any type of content
analyzes the information stored in the semantic repository in that has been previously tagged.
order to provide relevant and trusted results. At the moment, A similar tag-based algorithm is used to search documents
 SUPPORTING KNOWLEDGE DISCOVERY IN AN ELEARNING ENVIRONMENT 161

on the basis of the tags provided by users [11]. The algorithm In the first use case, the scenario employed for the validation
computes results based on tag co-occurences, on tag-resource focused on providing support to tutors in the creation of
and on tag-user affinities. These affinities are computed based courses in the IT domain. The idea was to evaluate the
on the frequency of the user-resource and tag-user pairs. The efficiency of the computer-assisting tool when creating a
tag co-occurence is the most used measure for tag similarity course on a certain topic. This evaluation focused on the
and it measures how often 2 tags are used together. After performance of the various functionalities, reflected in users’
computing the co-occurence matrix and the affinities matrices, opinions. The outcome would give feedback to the developers
the algorithm clusters tags and computes an affinity between a on how to improve the supporting system. The evaluated
user and a cluster of tags. Finally, using the affinity between functionalities were the following: relevance of the learning
user and the tag clusters the algorithm returns a number of material corpus wrt a domain topic; relevance of the retrieved
resources of interest. material; suitability of the ontology for structuring a lecture in
Visualization the chosen domain; possible visualization of the interface;
The visualization module offers a graph based view of either combination between ontology browsing and semantic search.
a social network of users and resources centered around a given There were two target groups: five teachers and three
learner or a part of a domain ontology. managers (vice-deans at the Faculty of Slavonic Languages,
The social network view can present the whole network as it Sofia University). Our assumptions were as follows: the
exists in the repository or it can show only the part of the combination among semantic search, ontology browsing and a
network returned by a search query. The search results are concept map based visualization would support the tutors in the
presented as a network of users and resources attached to them. processes of finding the relevant materials, structuring the
The user can thus see which path links him to a specific user or course and representing the content in a better way. Our
resource on the given search topic. In this way, the user can hypothesis was that, on the one hand, the tutors might differ in
identify the best means to contact a user for further questions their preferences of using the functionalities. For some tutors,
and can also rapidly identify who are the most competent users search would be more intuitive, while others might prefer
on a specific topic. The unified relations represented structure (in our case - ontology). On the other hand, the
graphically are from peers across different social networking managers would look at the facilities from a different
applications because our system monitors and integrates all the perspective, such as how this supporting system would
social media applications that the learner is using. communicate with the well-known Moodle, for example. Two
The ontology based graph allows a user to quickly gain methodological formats were adopted: the think-aloud strategy
insight into the important relations which exist between domain for the tutors, and the interview for the managers. The think-
concepts. The ontology visualization is a hyperbolic aloud format included seven questions, related to the specific
hypergraph that contains domain concepts with their preferred functionalities (search, browsing, visualization) as well as the
lexicalisation and (taxonomic) relations to other domain general usability and problematic issues of the system. The
concepts that are drawn as labeled edges. The hyperbolic interview format included three questions, which were related
geometry allows the user to focus on the important topics while to the usability and management issues related to the adoption
still retaining a sense of the larger surrounding context. The of the system.
user can discover, previously unknown, concepts which would The results confirmed our hypotheses. Additionally, four
not have shown up either in the social network or through a teachers reflected in their answers the fact that the
keyword search, by interactively browsing the ontology. functionalities are more efficient when operating in some
Together the visualization services provide a way for the interconnected mode. Regarding the semantic vs. text search
user to navigate through the knowledge sources, focus on three tutors gave preference to the semantic one, while two
specific details, add new information, take notes, etc. were more cautious and balanced between the two.
The visualization can be customized to be included in a Visualization was the most commented functionality, because
number of widgets to show specific information like the teachers had different ideas on how it should be controlled.
visualizing a part of the ontology, the social network, a The main worry of the managers was the requirement of a
definition, relevant peers, etc. These widgets can easily be training course to be able to use the CSF.
recombined or disabled in order to provide a learning In the second use case, we have focused on the support
environment that best fits the user. provided by an ontology enhanced with tags in comparison
with clusters of related tags extracted from Delicious (lacking
V. EVALUATION ontological structure) in the context of a learning task. The
underlying assumption being that conceptualization can guide
The Common Semantic Framework was evaluated on the learners in finding the relevant information to carry out a
basis of three use cases: learning task (a quiz in our case). The hypothesis was that
1. tutor support in the creation of a new course; learners might differ with respect to the way they look for
2. comparison between browsing of an ontology information depending on whether they are beginners or more
enriched with social tags and a cluster of related tags advanced learners. While beginners might profit from the
in order to solve a quiz; informal way in which knowledge is expressed through tagging
3. recommendation of content and peers on the basis of a more advanced learners might profit from the way knowledge
social network and tags. is structured in an ontology. The experiment was carried out
with six beginners and six advanced learners, all with an
162 MONACHESI ET AL.

academic background. Responses were elicited by means of a VI. CONCLUSIONS

questionnaire that contained questions about which elements
learners used to provide an answer to the quiz (i.e concepts, One of the goals of the LTfLL project is to develop services
relations, documents, related tags, structure). that facilitate learners and tutors in accessing formal and
The responses to the questionnaire by beginners showed that informal knowledge sources in the context of a learning task.
the enriched ontology can be a valid support to answer certain To this end, a Common Semantic Framework is being
types of questions, because the relations between the concepts developed that provides recommendations on the basis of the
are given in an explicit way. However, beginners rely mainly user profile, his interests, his preferences, his network and the
on documents to find the relevant information both in the case learning task. An advantage of this system with respect to
of the enhanced ontology and in the case of the cluster of standard search (e.g. Google) is that the strong semantic
related tags. This attitude is probably influenced by the way
component behind the search and recommendation system as
search occurs in standard search engines.
well as the integration of social network analysis will improve
On the other hand, advanced learners indicated that they
learner interaction, knowledge discovery as well as knowledge
were able to find the answer to the quiz quickly by using the
co-construction.
enriched ontology and they also used less documents to find
the answers. They gave the structure of the ontology a higher REFERENCES
rating than the beginners. Interestingly, advanced learners were
more positive about the structure of the tag visualization than [1] P. Monachesi, L. Lemnitzer, K. Simov, “Language Technology for
the beginners. This is probably due to their background eLearning”, in Proceedings of EC-TEL 2006, in Innovative Approaches for
knowledge that enabled them to interpret the graph better. We Learning and Knowledge Sharing, LNCS 0302-9743, 2006, pp. 667-672
refer to [12] for additional details on the results of the [2] L. Lemnitzer, K. Simov, P. Osenova, E. Mossel and P. Monachesi, “Using
experiment and their interpretation. a domain ontology and semantic search in an eLearning environment”, in
Proceedings of The Third International Joint Conferences on Computer,
In the third use case, we have focused on how we can Information, and Systems Sciences, and Engineering. (CISSE 2007).
support the activity of a group of learners (six) and one tutor Springer-Verlag. Berlin Heidelberg, 2007
that are using social networking applications for learning [3] V. Posea, S. Trausan-Matu, and V. Cristea, “Online evaluation of
purposes. The tutor looked for a way to recommend resources collaborative learning platforms”, in Proceedings of the 1st International
for the learners in a time efficient way while the learners Workshop on Collaborative Open Environments for Project-Centered
needed trust-worthy documentation on a given topic. The idea Learning, COOPER-2007, CEUR Workshop Proceedings 309, 2008
of the experiment was to see if the combined usage of social [4] L. Graham , P. Takis Metaxas, ““Of course it's true; I saw it on the
Internet!”: In critical thinking in the Internet era”, Communications of the
networking environments together with our tag based search
ACM, vol. 46, May 2003, p.70-75
could be valuable for their learning experience.
[5] S. Auer, C. Bizer, J. Lehmann, G. Kobilarov, R. Cyganiak and Z. Ives,
The scenario was the following: the tutor and the learners “DBpedia: A nucleus for a web of open data”, in Lecture Notes in
provided only their public usernames of their social networking Computer Science 4825, Aberer et al., Eds. Springer, 2007
sites to our system and continued using the platforms as before. [6] E. Wenger, Communities of Practice: Learning, Meaning, and Identity,
The crawling service gathered information about them, their Cambridge: Cambridge University Press, 1998
network of friends and the resources present in these networks. [7] P. Mika, “Ontologies are us: A unified model of social networks and
The learners searched for documents of interest to them and the semantics”. Journal of Web Semantics 5 (1), page 5-15, 2007
system returned documents written or recommended by their [8] A. Hotho, R. Jäschke, C. Schmitz, and G. Stumme, “FolkRank: A Ranking
peers ordered by the relevance of the content. Among the Algorithm for Folksonomies”, in Proceedings FGIR 2006, 2006
results were some documents bookmarked or written by the [9] C. Cattuto, D. Benz, A. Hotho, and G. Stumme, "Semantic analysis of tag
tutor. The results contained the name of the document together similarity measures in collaborative tagging systems," May 2008. [Online].
Available: http://arxiv.org/abs/0805.2045
with the name of the peer that recommended the document.
[10] L.S. Vygotsky, Mind and society: The development of higher
The learners trusted the resources retrieved this way more than
psychological processes, Cambridge, MA: Harvard University Press, 1978
the ones retrieved on a similar Google search due to the fact
[11] S. Niwa, T. Doi, S. Honiden, “Web Page Recommender System based on
that they knew and trusted the people behind those resources. Folksonomy Mining for ITNG ‘06 Submissions”, In Proceedings of the
For the tutor it was even easier, as his task consisted only of Third International Conference on Information Technology: New
creating and bookmarking resources. As the learners were Generations (ITNG'06), 2006, pp.388-393
already connected with the tutor, his resources were part of [12] P. Monachesi, T. Markus, E. Mossel, “Ontology Enrichment with Social
their search results. That means they could get answers from Tags for eLearning”, in Proceedings of EC-TEL 2009, Learning in the
their tutor without actually bothering him. Synergies of Multiple Disciplines. LNCS 1611-3349, 2009, pp. 385-390
The only inconvenience with this scenario for our little [13] P. Osenova, K. Simov, E. Mossel “Language Resources for Semantic
Document Annotation and Crosslingual Retrieval”. In: Proc. of LREC
experiment group was that the tutor was not comfortable with 2008, ELRA.
all of his resources being accessible to the learners. These
[14] K. Simov and P. Osenova “Language Resources and Tools for Ontology-
privacy issues have to be taken care of in the next version of Based Semantic Annotation”. OntoLex 2008 Workshop at LREC 2008, pp.
the platform. 9-13
 Computer Numerical Control, Mass Customization
and Architectural Education in an Emergent Economy
Neander F. Silva1
Faculdade de Arquitetura e Urbanismo, Universidade de Brasília, Brasil
Alan H. Bridges2
Department of Architecture, Strathclyde University, United Kingdom
Ecilamar M. Lima3
Faculdade de Arquitetura e Urbanismo, Universidade de Brasília, Brasil

Abstract – The objective of this paper is to demonstrate that

digital fabrication has had little impact in the Brazilian II. RESEARCH PROBLEM
architecture so far, as compared to other countries, not because
of an alleged unavailability of CNC technology in this country’s These resources are being incorporated, though slowly, to
construction industry, but for other reasons that must be architectural design and building construction in several
investigated. We show in this paper the results of a preliminary
countries [6]. These technologies are often referred to as
CAD/CAM systems (Computer-aided Design / Computer-
survey carried out in the region of Brasilia that reveals a
Aided Manufacturing) and also as CNC (Computer
significant presence of CNC technology in the local construction
Numerical Control) processes [7].
industry which points to new opportunities for innovation in the
The incorporation of digital fabrication to building design
architectural field. We will also show that some experiments in
and construction processes is more slow and recent in Brazil
architectural education involving rapid prototyping and digital than in other countries.
fabrication may provide a lead towards the solution of this It has been popularly believed that such resources are not
problem. suitable to the reality of an emergent economy, such as the
Brazilian one, because they would be more expensive than
I. INTRODUCTION conventional processes. However, it has been already
demonstrated by several authors that this does not result in a
Computer resources that allow the computerized
more costly architecture [5][8][9].
manufacturing of artifacts from three-dimensional virtual It has also been popularly believed that digital fabrication
models have been in use routinely by the car and aircraft is yet not viable in Brazil due to the lack of CNC machinery
industries for several years. We will refer to such resources
in the construction industry. We will demonstrate in this
from now on as digital fabrication. paper that this is not true and that other reasons must be
These digital resources, if incorporated to the process of considered such as the AEC (Architecture, Engineering and
the built environment production, allow a fundamental Construction) education system. We will also show some
paradigm change to take place in contemporary architecture. initial results of teaching mass customization at early
The construction industry has relied so far in the mass architectural education.
production of standardized components. The produced
elements are generic material which will be customized later
in the life cycle of the product. The mass produced III. HYPOTHESIS
components are classified into defined categories and We believe that the incipience of digital fabrication in the
produced in a limited array of forms and sizes. They are then Brazilian AEC industry is not the result of a possible lack of
stored and catalogued until they eventually, if sold, end up in technological resources in this country. It is important to
a combination of elements in a factory or as a part of a stress that such resources are available and large in use by the
building in the construction site [1]. Brazilian aircraft and car industries. It is even more important
As a new paradigm, the mass customization provided by to stress that digital fabrication systems are already available,
such digital resources, allows that construction components for example, in factories of steel structures and metal frames
may be produced for specific purposes, to become singular in this country [10].
elements in unique contexts of specific buildings. The savings We believe the present main reason for the incipient use
obtained in the automation of this process mean that the costs of digital fabrication in building design and construction in
of singular components are hardly different from the old Brazil is due to lack of information rather than to an alleged
standardized ones [2][3][4][5]. technological unavailability. This lack of information on its
(1) [email protected], [email protected], http://lecomp.fau.unb.br turn is due to our architectural education being still strongly
(2) [email protected], http://www.strath.ac.uk/architecture based on the mass standardization paradigm. The weak
(3) [email protected], http://lecomp.fau.unb.br connections between architectural schools and the

163
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_28, © Springer Science+Business Media B.V. 2010
164 SILVA ET AL.

construction industry in Brazil also contribute to the lack of Figure 2 below shows the percentages of companies that
information about fabrication processes available in this have only conventional technologies and those with CNC
country. machines. The percentage of firms with CNC equipment is
Therefore, the objective of this paper is to demonstrate significant and supports our hypothesis. It is important to
that the thesis of CNC technology unavailability as the main stress that 0.9 CNC machines per company in the surveyed
cause of it not being used in architecture is unsustainable. We geographic area is close to the national average detected by
will also show that some experiments in architectural [11]. This result is particularly important considering that the
education involving rapid prototyping and digital fabrication Federal District in Brazil is an area predominantly
may provide a lead towards the solution of this problem. administrative where only in the last 20 years significant
industrial activities have emerged.
IV. RESEARCH METHOD AND RESULTS
According to the 3rd Brazilian Inventory of Metal
Forming and Cutting Machines [11] the total number of
equipments in this area increased in 2.7% from 2006 to 2008.
During the same period, the number of CNC machines
increased in 44.4%. This suggests a technological potential in
fast expansion which can be better explored.
A. CNC availability in Federal District construction industry
Most of the data in the aforementioned Inventory (96,5%)
was gathered from the south and southeast regions of Brazil.
In this paper we show the results of a survey with the
construction industry in the specific region of the Brazilian
Federal District. Companies of small, medium and large size
and with different functions in the construction industry were Figure 2. Percentage of companies with CNC technologies.
(Source: authors of this paper)
surveyed.
Figure 1 shows the percentage of participation of different Figure 3 shows the percentage of companies that have
types of industries in the total of firms surveyed. Several CNC technologies taking into consideration only metal
companies were approached to establish a reasonably frames and structures.
representative sample. Fourteen of those firms accepted to
participate in the survey. The metal structure companies are
the majority in the sample. This is due to the fact that this
sector is the most dynamic of the construction industry in the
process of incorporating CNC technologies. This choice was
made under the assumption that this sector should influence
others in the long term. As an example, the steel structure
companies, through the manufacturing of metal molds, should
influence concrete construction systems.

Figure 3. Companies of metal frames and structures with CNC technologies.

( Source: authors of this paper)

Figure 4 below shows the amount of CNC machines

according the type of equipment. We use here a classification
of four groups, i.e., 2D cutting, addition, subtraction and
forming, in a similar way to that proposed by [12] and [13].
The 2D cutting CNC machines are the most common, but
CNC forming (bending and stamping) is also significant.

Figure 1. Types of companies surveyed.

(Source: authors of this paper)
 COMPUTER NUMERICAL CONTROL, MASS CUSTOMIZATION AND ARCHITECTURAL EDUCATION 165

Figure 7 below shows a sculpture in steel sheets designed

by the artist Darlan Rosa (http://darlanrosa.com/portu/jk.htm)
and produced with CNC machines by the metallurgic
metalúrgica Ferro e Aço Badaruco at the Federal District
(http://www.badaruco.com.br). This example is important
because it shows the same formal freedom that has already
been explored in architectural design in other countries but
remains unexplored in Brazil.

Figure 4. Percentage of CNC machines per type.

( Source: authors of this paper)

Figures 5 and 6 show oxyacetylene torches of 2D CNC

machines cutting metal sheets at the Mega Forte Metalurgica
and Ferro Aço Badaruco (http://www.badaruco.com.br),
respectively, at the Federal District of Brazil.

Figure 7. Sculpture of Darlan Rosa created using CNC technologies by

Ferro e Aço Badaruco, Brasília
(Source: photo taken by authors of this paper)

B. CNC in Architectural Education in Federal District

It seems clear that the reason for the slow pace of CNC
integration into the architectural design process in Brazil is
not due to a lack of technological resources. The above
mentioned survey has shown quite the contrary. The reason
must reside elsewhere. We believe that the prevailing
architectural education in our country must have an important
role to play in this problem, to say the least.
Figure 5. 2D CNC cutting at Mega Forte, Brasília.
(Source: photo taken by authors of this paper)
Therefore, we have carried out several educational
experiments at the Faculty of Architecture and Urban Design,
University of Brasilia, Brazil, over the last two years. Those
experiments were developed in the context of the second
semester of the first year of the architectural design studio
education. The focus of this semester is architectural language
and expression with an emphasis on formal analysis and
synthesis. The adopted design themes were an entrance porch
and an exhibition pavilion.
One of our major objectives was to promote an
understanding of the shifting paradigm in architectural design
and construction at the beginning of this 21st century. We
wanted the students to understand, as earlier as possible in
their studies, the process of moving from mass
standardization to mass customization in order to break free
from the boundaries of orthogonal and serial architecture and
to start exploring richer forms.
In order to achieve those goals we gradually introduced
the paradigm of mass standardization (and its relation to the
Figure 6. 2D CNC cutting at Ferro e Aço Badaruco, Brasília. modern movement), the late 20th century paradigm of mass
(Source: photo taken by authors of this paper)
166 SILVA ET AL.

customization (and the benefits it may provide to the rising of potential. However, one of the surprising results was that
new forms in architecture) and rapid prototyping. many students started designing taking into consideration the
The experiment was developed during four semesters and possibilities of the aforementioned paradigm shift even before
rapid prototyping was introduced at the end of the third one, we introduced rapid prototyping and they had started seeing
on a limited range. Most students were requested to design its practical potential.
using handmade scaled models. The design projects with
more complex geometry were selected for rapid prototyping V. CONCLUSIONS
in a three-axe milling machine.
Figure 8 shows a rendered view of one of the students The results of our survey show that thesis of technological
design project for the exhibition pavilion. unavailability is not sustainable and that the presence of CNC
technology in the Federal District of Brazil is already
significant, although its distribution may be considered
heterogeneous.
Our experiments with CNC at early architectural
education show that the educational issue is a strong
candidate for the problem of slow incorporation of digital
fabrication and mass customization into architectural design
practice in Brazil.
The implications are huge for the future of the Brazilian
architecture considering the possibility of incorporating the
mass customization paradigm and thus allowing for complex
and non standardized components to be produced within
reasonable economic constraints.
Figure 8. Rendered view of the exhibition pavilion of the student Nágila
Ramos
(Rendered with formZ RenderZone Plus) ACKNOWLEMENTS
Figure 9 shows the prototype of roof of the design project We would like to thank the management of Gravia
shown above, just before being removed from the sustaining (http://www.gravia.com/), Mega Forte e Ferro e Aço
frame. Badaruco (http://www.badaruco.com.br) for kindly
participating in survey described in this paper.

REFERENCES
[1] Polette, Doug and Landers, Jack M. “Construction Systems”,
Goodheart-Willcox Company, Inc, Publishers, Illinois, 1995, pp. 45-
49.
[2] Kieran, Stephen and Timberlake, James, “Refabricating
Architecture”, McGraw-Hill, New York, 2004, pp. 131-153.
[3] Kolarevic, Branko: Digital Production, in Branko Kolarevic,
Architecture in the Digital Age – Design and Manufacturing, Taylor
& Francis, New York, pp. 31-53, 2005.
[4] Schodek, D.; Bechthold, M.; Griggs, K.; Kao, K. M. & STeinberg, M.
“Digital Design and Manufacturing – CAD/CAM Applications in
Architecture and Design”, John Wiley & Sons, New Jersey, 2005,
pp. 339-344.
[5] Franken, Bernhard, “Real as Data”, in in Branko Kolarevic,
Figure 9. Rapid prototype of the exhibition pavilion of the student Nágila Architecture in the Digital Age – Design and Manufacturing, Taylor
Ramos & Francis, New York, p. 138, 2005.
(Milled in a Roland Modela MDX-40 three-axes milling machine) [6] Stacey, Michael; Beesley, Philip & Hui, Vincent, “Digital
Fabricators”, University of Waterloo School of Architecture Press,
The above example was one of design project specially 2004, p. 6.
selected for rapid prototyping. [7] Schodek, D.; Bechthold, M.; Griggs, K.; Kao, K. M. & STeinberg, M.
“Digital Design and Manufacturing – CAD/CAM Applications in
We observed that the more theoretical background we Architecture and Design”, John Wiley & Sons, New Jersey, 2005,
gave to the students about digital fabrication and mass pp. 3-4, 237-296.
customization and the more information we provided about [8] Kolarevic, Branko: Digital Fabrication: Manufacturing Architecture
the existence of digital fabricators in our region, the more in the Information Age, in ACADIA 2001, Buffalo, New York, 2001,
they felt confident to endeavor outside the boundaries of p. 275.
[9] Schodek, D.; Bechthold, M.; Griggs, K.; Kao, K. M. & STeinberg, M.
orthogonal and serial architecture.
“Digital Design and Manufacturing – CAD/CAM Applications in
The presence of CNC equipment in our school was Architecture and Design”, John Wiley & Sons, New Jersey, 2005,
important to convince the students of this technology’s p. 341.
 COMPUTER NUMERICAL CONTROL, MASS CUSTOMIZATION AND ARCHITECTURAL EDUCATION 167

[10] Cunha, Edgard L., Gonçalves, José R., Souza, José R. A., Silva, [12] Schodek, D.; Bechthold, M.; Griggs, K.; Kao, K. M. & STeinberg, M.
Sílvio P.(Editores): Corte e Conformação de Metais, Editora Aranda, “Digital Design and Manufacturing – CAD/CAM Applications in
São Paulo, 2008, número 44, pp. 70-91. Architecture and Design”, John Wiley & Sons, New Jersey, 2005, pp.
[11] Cunha, Edgard L., Gonçalves, José R., Souza, José R. A., Silva, 3-4, pp. 239-244.
Sílvio P.(Editores): Corte e Conformação de Metais, Editora Aranda, [13] Kolarevic, Branko: Digital Fabrication: Manufacturing Architecture
São Paulo, 2008, número 44, p. 71. in the Information Age, in ACADIA 2001, Buffalo, New York, 2001,
pp. 268-277.
 Application of Project Based Study
in the Learning Process: A Case Study
Ashfaque A Chowdhury, Mohammad G Rasul and M. Masud K. Khan

Jennings [4] reported on a case where students were asked to

Abstract— The study is a step toward studying the advanced expand on the professional dimensions of the implications of
technology specially to learn the control system of CNC lathe the problems they solved as a part of class room teaching and
machine. This paper describes the development of a closed loop the outcomes of the solutions to engineering problems.
control system with encoder mounting as real-time control device Navaz et al. [5] introduced a new approach in teaching
for a conventional lathe. A personal computer is used for Thermal/Fluid Sciences to undergraduate students through
calculation as well as for controlling the motion of the drives. The
servomotors are used for controlling the movement of carriage. A
inclusion of new and advanced technologies into the
program is developed using Turbo C for evaluating the overall curriculum using latest computational and experimental
system. The developed control system is applied to evaluate work methods. Musto [6] described the outcomes of the introduction
piece dimension in real time with the integration of production of a project based approach in a numerical methods course
process. Encoders are used to collect real-time machining data. It where the subject concepts were related to a term-long design
is found that the errors are less than 0.005 mm of turning project. Student teams were required to report on the design
example part on X and Z axes. The average error is methodology and the results through formal design reports and
approximately 0.002 mm. a presentation. The integrated project based approach assisted
the students to enhance their enthusiasm and understanding of
Index Terms— CNC, Lathe, real-time control system. the course materials [6].
Mustoe and Croft [7] reported the increased emphasis on
open ended design exercises and case studies in engineering
I. INTRODUCTION curricula. Gomes [8] adopted case studies in modern

L EARNING document methods and supporting

informational materials in the higher education can be
improved for varying disciplines by engaging students to
engineering teachings to improve integration of cross
disciplinary education and to develop a professional outlook in
the young engineers. Feisel and Rosa [9] identified the lack of
unravel real life problems through project based investigation. coherent learning objectives in engineering laboratory
The aim of engineering education is to develop skilled exercises and demonstrated the impact of this shortcoming in
manpower to solve real life engineering problems and to meet limiting the effectiveness of these exercises. Feisel and Rosa
society’s needs with advances in knowledge. Gray [1] [9] encouraged engineering students to seek knowledge
investigated the integration of case study based investigation beyond the classroom theories. DeBartolo and Robinson [10]
in engineering courses to establish a practice where demonstrated the successful implementation of design and
undergraduate engineering students exercise their growing experimentation in engineering curriculum to develop the
technical knowledge and analytical capability in the context of students’ concepts on design, build and prove the functionality
real engineering problem. The applied principals of technical of the designed project.
investigations were generic in nature. Harris [2] demonstrated The integration of project based investigation into the fluid
the implementation of the project based design methodologies mechanics course described in this paper has not only
in an undergraduate course by integrating the relevant theories provided the students an insight into the fundamentals of fluid
with hands on design projects. It not only benefits the students mechanics, but also developed students’ abilities to carry out
but also changes the instructors to integrate theory into professional tasks in real life. The study has also reported that
practice. Holt [3] reported that students often receive little the suitability of project based study along with conventional
practice in closing the gap between engineering necessity and lecture classes in fluid mechanics course.
technical knowledge. This lacking may lead to mistakes by the
students in treating the engineering problems in practice.
II. SUMMARY OF THE COURSE
Manuscript received November 2, 2009. This work was supported by the
Department of Mechanical and Chemical Engineering, Islamic University of In the Department of Sustainability and Infrastructure at
Technology. CQUniversity, teaching of fluid mechanics course is offered
A. A. Chowdhury is with the CQ University, Rockhampton, Qld 4701,
Australia (corresponding author phone: +61 4 1176 7886; fax: +61 7 4930
through dedicated course work, laboratory experiments and
9382; e-mail: [email protected]). group design experiments. The core components of the course
M. G. Rasul is with CQ University, Rockhampton, Qld 4701, Australia (e- consists of a collection of weekly modules and contains the
mail: [email protected]). general characteristics of fluids, manometry, fluid statics,
M M K Khan is with CQ University, Rockhampton, Qld 4701, Australia
(e-mail: [email protected])

169
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_29, © Springer Science+Business Media B.V. 2010
170 CHOWDHURY ET AL.

analysis of incompressible flow in pipes, analysis of buoyancy The graduate attributes mapping was prepared to obtain an
and stability of floating bodies and measurement of fluid flow. indication of how the fluid mechanics course contributed to
It introduced the methods of analysing fluid systems using the achieve course learning outcomes and graduate attributes and
conservation of mass and momentum equations, Navier to understand better how attributes are developed as students’
Stoke’s equation combined with the concept of a control progress from one year to next year. This is done by indicating
volume. To solve the problems in fluid mechanics course, the graduate attributes targeted by each course learning
students familiarised with the theories of incompressible flows outcome. The indicative marks are placed under the headings
in pipe systems, Euler’s equation, Bernoulli’s equation and the of Teaching, Practice and Assessment to illustrate the course
use of similitude and modelling principles and techniques. emphasis (low/medium/high).
Students were required to act professionally in presenting
information, communicating, working and learning
individually and in peer learning teams to achieve the learning V. ASSESSMENTS
outcome of the fluid mechanics course. Initially the fluid
To pass the course students had to attempt all the
mechanics course was offered to on-campus students. Later,
assignments, laboratory reports and complete all examinations
the existing course was adapted for Interactive System-wide
and obtain a grade of at least 50% and a passing grade in the
Learning (ISL) delivery on other regional campuses. An
formal exam. Assessments for the course were on the basis of
online lecture delivery was developed for flexible course
laboratory reports (20% weighting), a fluid mechanics project
offerings with a one week on-campus mandatory residential
plan (20% weighting) and investigation (20% weighting),
school requirement. The restructured course materials can be
workbook submission (pass/fail in competency) and formal
used both as supporting teaching materials and as reference
examination (40% weighting).
materials in project work for research and self study.
A. Laboratory Exercises
There were three components in laboratory exercises – a
III. STUDENTS LEARNING ACTIVITIES formal lab report, video lab questions and virtual lab design
At the beginning of the course, all students were given an and experimentation. The first two components were group
overview of the course to make clear the expectations about assessment items and each group were required to submit a
how the course would operate. Flex students were introduced report on the assessments. Every group member contributed
to online learning practice and were assessed on online to the conduct, preparation and write up of the laboratory
learning skills required for this program. Students were report. The objective of involving students in video lab
encouraged to study and respond to the review questions and questions and virtual lab activities is to develop higher order
solve exercise problems. They were also familiarised with the thinking from the problem statements and thus a deep
real fluid systems and flow phenomena through the laboratory understanding of the knowledge on the subject matters.
exercises. Students were required to study the text and other Practical videos in most chapters of weekly schedules were
course materials as a guide to solve the workbook problems, uploaded on the course delivery site.
virtual laboratory activities and other assessment items. The students were asked to solve the related practical
Students were able to discuss their difficulties with problems questions demonstrating the steps with related background
in the discussion list or email the course lecturer. Students knowledge for a virtual lab problem. In the design and
were expected to follow a self-directed study schedule which experimentation component, the students were encouraged to
met the required deadline. The weekly schedule of activities design an experiment in fluid mechanics from any of the
for on campus students were Lectures (two hours per week), topics covered in the weekly modules that the students would
Tutorials (two hours per week), Laboratory (two hours per like to improve further in future. Students were required to
week). A residential school was conducted for flexible and submit a report on the selected experimentation covering the
other regional campus students after the midterm vacation. necessity for the investigation, a description of the work, the
Tele-tutorials over Skype and/or local tutorials were arranged parameters requiring observation and the safety procedures.
for off-campus students. Students were also required to complete the entire laboratory
exercise including the drawing of graphs and final answers on
issues such as calibration of obstruction meter, flow in pipes,
IV. ADDRESSING OF GRADUATE ATTRIBUTES stability of a floating body and velocity profile and pressure
losses in a smooth pipe.
In the design and development process of a technical course
for the engineering students, there are several objectives need B. Engineering Fluids Project
to consider such as the goals, resources, course learning The aim of an engineering fluids project is to ensure that
outcome, and last but not the least the graduate attributes. The the future engineers will comply with the principles of
targeted graduate attributes were Science and Engineering sustainability though improving their research ability and
knowledge, Effective communications, Technical competency, knowledge of fluid mechanics. The projects related to basic
Problem solving and systems approach, Functionality in a principles of fluid mechanics assist the students to enhance
team, Social, cultural, global and environmental awareness, their learning experiences and to achieve the graduate
knowledge on sustainable design and development, attributes through a team based design approach utilising the
professionalism and ethics and lifelong learning. concept of sustainability to meet the needs of present and
 APPLICATION OF PROJECT BASED STUDY IN THE LEARNING PROCESS 171

future generations. The students experienced a productive show the application and understanding of the subject matter
pedagogy in fluid mechanics through improved knowledge & and to use the correct procedures and calculations in their
understanding. workbook answers. Attainments of competency attracted extra
Several project options were available for the students. In a marks to move to a higher grade when short by a few marks.
project proposal, on part of land assumed, the right side was Students were also required to sit for closed book final exam.
assumed to construct an engineering complex. Based on the
knowledge in the course, a group of students were responsible
to design the supply of drinking water and the management of VI. ASSESSMENTS CRITERIA
waste water facilities for the engineering workshop. Another Engineering fluid projects and workbooks were evaluated
two groups of students were assigned to design the same for a based on the evidence of the application and integration of
shopping complex and a cinema theatre. There were several knowledge and understanding of the material studied and
steps suggested in the course profile for the students to follow communication and justification of the subject matters. It was
and report to the campus tutors on the progress of the project. kept in consideration that students demonstrated accurate use
In the first two to four weeks, students were required to of engineering terms, collection and organisation of
define problems and project scope through the fundamentals information in the various forms of presentation. It includes
of fluid mechanics learnt, and then proposes a sustainable well structured solutions, use of engineering mathematical
concept design complying with course learning outcomes. In reasoning and proof to develop logical arguments, recognition
week five, students were required to present the progress in of the effects of assumptions used evaluation of the validity of
groups identifying the contribution of each member. In the arguments and justification of procedures.
next three to four weeks, students prepared the preliminary The laboratory components of the assessment items were
sustainable design with specifications. Often students revisited assessed based on reporting of key elements required to
the preliminary plan and refined the concept design with the undertake the laboratory sessions, clarity of expressions,
feedback from course tutors and lecturer. Later in the term, appropriate referencing of sources, accurate use of fluid
students finalised and documented the sustainable design with mechanics knowledge, presentation of mathematical
a strategy to implement in future. It was required to build a formulation, clarity and logical presentation of ideas and
physical prototype design to have a nice memory of the young arguments by means of data analysis and synthesis. To
engineers’ efforts in future. evaluate the ability of the students on the application and
The amount of traditional homework problems assigned integration of knowledge and understanding - evidence of
was reduced approximately by half. The design project was interpretation, analysis, identification of assumptions,
assigned to groups of four or five students at the start of the selection and use of symbols of strategies and procedures
course. Handing out the project immediately at the outset of while showing initiative to solve the problem, were
the course, where students are largely unfamiliar with the considered. In some cases, it was found that the problem has
material required for the completion of the project, renders the not been solved, however substantial progress towards a
learning process goal driven. This approach is in support of rational solution and shows evidence of appropriate
the life-long learning scenario for which students ought to be interpretation, analysis, identification of assumptions and
prepared and where the learning typically occurs on a needs selection of strategies and procedures. Additionally, the
basis in an interactive and often collaborative learning mode. problem has been solved however, there is a lack of
The submissions of two written progress reports were appropriate evidence was also reported.
required in order to guide students through the design tasks
and enforce progress throughout the entire semester. In
addition, a total of two class periods throughout the semester VII. STUDENTS PERFORMANCE
were allotted for a progress presentation and a final
presentation by each student team. It was found that the Course history showed that, after introduction of the
project based learning concepts enhanced students’ learning integration of project based investigation in fluid mechanics
experiences and assisted in the achievement of key attributes. course, on average (based on three years data), about 4.6% of
The students experienced a dynamic pedagogy in fluid the total enrolled students withdrew from the course after
mechanics through improved content quality, connectedness to enrolment and the failing rate dropped to 6.5%. Most of the
real world problems, a supportive team learning environment students performed very well and their overall performance
and recognition of differences between the team members. was quite impressive. On an average, 28.7% students received
high distinction and 30.6% students received distinction in the
C. Other Assessment Items course. As shown in Fig 1, the comparisons of student
To improve the research capability and gain knowledge and performances over 2007 and 2008 demonstrate that the failing
improve understanding in fluid mechanics, students were rate of the students in the course was reduced from 14.3
asked to prepare a teaching and learning journal in fluid percent to 5.4 percent. The blended assessment mechanism
mechanics through the lectures, tutorials, experiments and incorporating lab components, project works, workbook and
project work. Students were expected to solve higher order final exam assisted students to understand the subject matter
problems and open answers which require researching on the and to develop a higher level of skills in an advanced
weekly topics. The workbook contributed to the overall grade engineering topic like fluid mechanics.
of the course as it supplemented the assignment work and
confirmed students overall competency. They were required to
172 CHOWDHURY ET AL.

offering of the course. In 2009, based on student comments

and peer assessment, the course was progressively restructured
to maintain the relevance of the content and satisfy the
demand and need of the industry and community. Dr Philip
Morgan’s (the University of Newcastle) flowchart (Fig 2) was
followed to incorporate the feedback from the past students,
tutors, lecturer and course coordinator. To collect the feedback
received from the student, course evaluation and focus group
interviews were conducted. In course evaluation the students
where asked about the active learning (actively engaged and
participated in collaborative learning), assessment
(satisfactorily demonstration of the acquisition of the
Fig. 1 Indication of students’ performance in the course knowledge, skills and attitudes needed), expectation (to learn
to satisfactorily complete this course materials), relevance of
the course materials (satisfactory development the appropriate
knowledge, skills and attitudes to future professional career)
VIII. FURTHER DEVELOPMENTS and overall satisfaction (positive learning experience). Some
Integration of project based study based on practical feedbacks were received from open ended comments by the
engineering problems in mechanical engineering courses is students in the emails. The summary of the results of the
required to ensure the professional and engineering context of feedback and recommendations have been articulated in the
applied science and to develop subsequent learning course improvement flowchart. In has been reported that the
methods. The program review committee meeting at engineering projects relevant to the theories not only solve
CQUniversity suggested the Program Director the course professional cases but also encourage the students to involve
outcome, students’ feedback, and peer assessments (tutors and in an encouraging manner.
course coordinator) for potential improvement of the next

Fig. 2 Flowchart of course improvement process

 APPLICATION OF PROJECT BASED STUDY IN THE LEARNING PROCESS 173

IX. CONCLUDING REMARKS [4] A. Jennings, “Viewpoint: the need to upgrade experimental learning,”
International Journal of Engineering Education, vol. 14, pp. 83-88,
It is intended to teach the fluid mechanics course by 1998.
creating a learning centred environment which will encourage [5] H. K. Navaz, B. S. Henderson, R. M. Berg, and S. M. Nekcoei, “A new
the students for creative and critical thinking and lifelong approach to teaching undergradaute thermal/fluid sciences - courses in
applied computational fluid dynamics and compressible flow,”
learning. It is necessary to prepare undergraduate engineering International Journal of Mechanical Engineering Education, vol. 30,
students to design, and analyse real life problems. The paper pp. 35-49, 2002.
describes an integrated assessment methodology which was [6] J. C. Musto, “A project based approach to teaching numerical methods,”
implemented in a fluid mechanics course and demonstrates its International Journal of Mechanical Engineering Education, vol. 30,
pp. 233-247, 2002.
effectiveness in facilitating engineering undergraduate [7] L. R., Mustoe, and A. C. Croft, “Motivating engineering students by
students to become efficient problem solvers of real life using modern case studies,” International Journal of Engineering
problems. The success of the assessment procedures were Education, vol. 15, pp. 469-476, 1999.
evaluated based on the successful completion of the course, [8] V. Gomes, “Consolidation of engineering education through industrial
case studies,” International Journal of Engineering Education, vol. 18,
students’ grades, students’ withdrawal rate from the course pp. 479-484, 2002.
and the students’ overall perception of benefits of the course [9] L. D. Feisel, and A. J. Rosa, “The role of the laboratory in undergradaute
studied. Integrated assessment methods assisted with engineering education,” Journal of Engineering Education, vol. ,
strengthening previous and prerequisite learning requirements. pp. 121 -130, 2005.
[10] E. DeBartolo, and R. Robinson, “A freshman engineering curriculum
Therefore, benefits the subsequent learning process is achieve integrating design and experimentation,” International Journal of
through excellence in research and teaching. The assessment Mechanical Engineering Education, vol. 35, pp. 91-107, 2007.
items offered students experience of real life problems in a
professional manner. Further improvement of the course is in Ashfaque Ahmed Chowdhury obtained his B.Sc. in Mechanical
Engineering degree from Islamic University of Technology, Bangladesh in
progress considering the feedback from all the stakeholders of 2003. In 2008, he completd Master of Engineering degree at Central
the course. Queensland University, Australia. Since 2004, he worked as a lecturer in the
Department of Mechanical and Chemical Engineering at Islamic University of
Technology and in the College of Engineering and Built Environment, Central
ACKNOWLEDGMENT Queensland University. His current research interest include intelligent
control system, manufacturing technologies, computer numeric control
The authors gratefully acknowledge the contributions of Dr manufacturing, computer-aided design and manufacturing, building energy
Ahmed Elgezawy and Mr Fons Nouwens who generously simulation, energy conservation and management, low energy cooling
technologies.
assisted in the development and implementation of the course
M G Rasul is a Senior Lecturer of Mechanical Engineering at the College
in the faculty. Special appreciation goes to Mr Tim of Engineering and Built Environment, Central Queensland University,
Mcsweeney for his remark to improve the manuscript. The Australia. He received his PhD and MEng degree from The University of
authors also acknowledge the support of the Learning and Queensland, Australia and Asian Institute of Technology, Thailand
respectively. He is specialized and an expert in energy technologies, process
Teaching Education Research Centre (Special Interest Group –
industries, thermo-fluid engineering including combustion and process
Future Engineering Education Directions), which is engineering, and light metals. He has published over 100 research articles in
established and supported under CQUniversity’s Research reputed journals and conferences on a wide range of research interests.
Centres Program. M. Masud K. Khan is an Associate Professor and Head of Department of
Infrastructures in the Faculty of Sciences, Engineering and Health, Central
Queensland University, Rockhampton, Queensland 4702, Australia. He received
REFERENCES his MS (Mech) with 1st class Honours from Moscow Institute of Petrochemical
[1] T. Gray, “Integration of case study technical investigations in and Gas Industry in 1982. Subsequently he worked with the oil industry for 2
honours/masters engineering courses,” International Journal of years. He obtained his PhD in engineering from the University
Mechanical Engineering Education, vol. 34, pp. 121-130, 2006. of Sydney in 1990. His teaching, research and consulting activities are in the
[2] D. Harris, “A Case for Project Based Design Education,” International area of non-Newtonian fluid mechanics, thermofluid engineering and rheology
Journal of Engineering Education, vol. 17, pp. 367-369, 2001. of industrial fluids and polymeric materials. He has authored more than 95
[3] J. E. Holt, “On the nature of mechanical engineering - an engineering refereed journal, conference and technical papers. He is a member of American
ethos,” International Journal of Engineering Education, vol. 12, Society of Rheology, Australian Society of Rheology, Engineers Australia and
pp. 332-338, 1996. the Australasian Association for Engineering Education (AAEE).
 Interdisciplinary Automation and Control
in a Programmable Logic Controller (PLC)
Laboratory
Jack Toporovsky, Christine Hempowicz, and Tarek M. Sobh
School of Engineering, University of Bridgeport
[email protected], [email protected], and [email protected]

Abstract – The University of Bridgeport’s School II. PROBLEM STATEMENT

of Engineering created a new PLC Industrial
Control Lab to accommodate lab-based, hands- The University of Bridgeport’s School of
on training using different types of Engineering (UBSOE) identified the need for a
Programmable Logic Controllers (PLCs). The Programmable Logic Controller (PLC) Industrial
Lab and associated courses were designed to Control Lab to support laboratory-based courses in
meet ABET’s EC2000 standards and prepare which students could be educated in technological
students for the workforce. processes that are part of the field of automation
control. Course curricula should support student
I. INTRODUCTION/BACKGROUND outcomes outlined in EC2000’s Criterion 3. Program
Outcomes and Assessment, specifically 3a-e, j, and
At the beginning of the new millennium, k. Further, course curricula should be
engineering education programs were challenged to interdisciplinary such that electrical engineering and
meet the latest set of standards established by mechanical engineering students could enroll in the
ABET, known as Engineering Criteria 2000 same course. Finally, students should have the
(EC2000). EC2000 emphasized outputs, as opportunity to learn on leading PLC models in order
expressed in 11 learning outcomes [1]. In particular, to be best prepared to work in the private sector. The
upon program completion, graduates should have lab should be equipped, then, with Allen-Bradley
acquired the knowledge and skills to solve and Mitsubishi PLCs since the former is widely used
problems, using techniques and tools of engineering in the U.S. and the latter is widely used throughout
practice in a collaborative setting. Asia.
This approach to engineering education is not
entirely new. Early engineering education focused III. METHODOLOGY
on skill acquisition through apprenticeships,
practical training, and the like, until the end of UBSOE’s existing Interdisciplinary Robotics,
World War I [2]. The emphasis on engineering Intelligent Sensing, and Control (RISC) Laboratory
science and mathematics in engineering curricula was expanded to incorporate the new PLC Industrial
was not introduced until the mid-1950s. This had the Control Lab. The new PLC lab was equipped with
effect of producing engineers who could analyze a electrical cabinets and control panels outfitted with
problem but were less prepared to design and apply five Allen-Bradley and four Mitsubishi PLCs [5]-[12].
solutions [2]. EC2000 was, in part, a response to the Two courses were developed: EE 463: Industrial
need for engineering graduates to be able to Controls and EE 464: Programmable Logical Controls.
seamlessly transfer classroom and laboratory In EE 463, students learn basic concepts in
knowledge and practice to the workplace. pneumatic controls and different types of sensors
Engineering graduates must be “job ready” and
practiced in working as part of an interdisciplinary
team [3] and [4].

175
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_30, © Springer Science+Business Media B.V. 2010
176 TOPOROVSKY ET AL.

used widely in industry, robotics, and vision A team of PLC students designed an integration of
concepts. different brands of control components into one
In EE 464, students learn basic and advanced application. The components are: SLC5/03 Allen-
concepts and applications for PLC programming. In Bradley PLC, Network Card 1761-NET-ENI,
addition, students learn to integrate various Mitsubishi Variable Frequency Drive, and Cutler
components with PLCs, such as sensors, switches, Hammer PM1700 Touch Screen. For this project,
vision cameras, variable speed drives, etc. students were involved in wiring, communications
Instructional design for each course was more cable design, equipment programming, and
focused on inquiry and problem-based learning troubleshooting.
within the context of classroom community, and less Students produced the layout design and wiring
focused on lecture-style instruction. All are diagrams for the project to integrate the SLC5/03
supported in the literature [13] and [14]. PLC with the Mitsubishi Variable Frequency Drive
(VFD) and Cutler Hammer Touch Screen, and to
A. Equipment Design and Integration establish a network to the PLC. The design layouts
In the PLC Industrial Control Lab, all control and wiring diagrams were created using AutoCAD.
cabinets share the same local network with different Students were able to build a control cabinet with
types of PLCs and Human Machine Interfaces output to an AC motor, programming of the PLC,
(HMIs), so students are able to communicate to any VFD, touch screen, etc. (See Figure 1.)
of them from the class computer stations.

Fig. 1. Any computer station can establish communication to the SLC5/03 PLC through the network hub switch and through the internet
adapter (# 1746ENIW). At the same time, an AC motor (3-phase AC) runs commands from the PLC, transferring from the relay module
to the Mitsubishi Variable Frequency Drive (VFD/FR/S500). The value of the frequency is converted to an analog voltage (0 to 5 volts)
from the VFD; it is then read by an analog card. In this way, a closed-loop communication link is established between the PLC and the
VFD.
 INTERDISCIPLINARY AUTOMATION AND CONTROL IN A PLC LABORATORY 177

During the project, students learned how to use the

software FR Configurator to run the motor with
different parameter sets and observe the effects.
After the students decided which parameters were
needed for the operation, they downloaded them into
the VFD.
At the end of the project, different touch screen
recipes for various applications had been created.
RECEIPE 1
Selector switch # 1 should be at the ON position.
Press the start push button on the touch screen and
select a value for the motor’s speed. (The motor will
run at one of three predetermined speeds.) The
motor speed control parameter (expressed as
voltage) is shown on the touch screen. Also shown
is the motor’s speed in RPM’s.
RECEIPE 2
Selector switch # 1 should in the OFF position. In
this mode, the motor will run continuously, first at a
low speed (typically 50 to 100 RPM’s), then at a
medium speed (approximately 100 to 300 RPM’s),
then at a high speed (300 to 500 RPM’s), then back
to low speed, medium speed, etc. The analog voltage
drive speed and RPM are displayed on the touch Fig. 2. Robotic arm is shown, controlling a conveyor line,
screen. with optical, proximity, and capacitive sensors. A Mitsubishi
PLC is located on a shelf below the arm and to the right, and
a Mitsubishi RV-M1 robotic unit is on the same shelf, to the
RECIPE 3 left.
This mode is designed to control the jogging
operation by pressing the jog-forward or jog-reverse the Mitsubishi robotic arm Roboware™ software
pushbutton or touching the jog-forward/reverse icon and FX series Mitsubishi PLC. See Figure 2.
on the touch screen. The motor will rotate forward The major task of the project was to establish the
or reverse, depending on which jog button is communication and movements of the Robotics
pressed. The analog voltage drive speed and RPM Arm using the Mitsubishi Movemaster RV-M1
are displayed on the touch screen. series robotics kit. When the photo eye sensor
detects the metal or plastic object on the loading
RECEIPE 4
Establish motor control by reading the value area, the pneumatic cylinder then pushes the object
potentiometer on the front panel. The speed can be to the part-detecting sensor (proximity sensor) (See
increased from 0 RPM to a maximum of 500 RPM Figure 3). If the object is plastic, the conveyor starts
(as controlled by the PLC) by turning the and stops at “position one,” which is closer to the
potentiometer knob clockwise. To perform this test, sensor. The robotic arm will move to pick up the
the VFD parameter # 79 should have a value of 4. plastic object and place it into “box one”. If the
object is metal, the conveyor stops when it is
B. PLC Programming and Robot Integration detected by the track sensor (see-through eye sensor)
The first author worked with a team of PLC on the conveyor belt. The robotic arm moves to that
students to design an automation conveyor line to position to pick up the object and place into “box
implement closed-loop communications between two”.
178 TOPOROVSKY ET AL.

(1) The least input increment of the turning angle

is 0.1, e.g. specify 15.2 for 15.2.
(2) The open/close state of the hand does not
change before and after the movement. Error
mode II occurs before the axis motion if any
turning angle entry exceeds the robot
operational space.
(3) The default turning angle is 0.

The position data table for pick and place scenario

loaded into robotic amplifier can be found in Table
2.
TABLE 2

Robotic Arm Position Data (PD)

1 +254.6 +427.2 +174.6 -46.9 +32.2
4 -259.0 +185.1 +212.0 -87.0 +32.2
5 -256.6 +187.2 +236.0 -87.0 +32.2
6 -259.0 +185.1 +199.0 -87.0 +32.2
7 -259.0 +185.1 +235.0 -87.0 +32.2
8 -259.0 +188.1 +192.0 -87.0 +32.2
Fig. 3. Controlling the belt movements (Y6), part present (X12), red light 9 -259.0 +185.1 +263.0 -87.0 +32.2
indicator (Y5) and ready signal from robot unit (X0), etc, presented in the 10 -317.4 -24.5 +263.0 -86.9 +32.2
ladder diagram. Constant communication between the robot controller and 11 -324.5 -25.1 +43.1 -89.0 +32.2
Mitsubishi PLC is provided through executing this program and running 12 -324.5 -25.1 +36.1 -89.0 +32.2
Roboware™ software. 13 -324.5 -25.1 +44.1 -89.0 +32.2
14 +299.1 +182.9 +233.8 -81.0 +145.1
15 +297.9 +179.1 +196.4 -86.0 +145.1
The inputs and outputs are on the conveyor line, 16 +302.5 +184.8 +258.3 -76.9 +145.1
operated using the GX-Developer (Mitsubishi 17 +22.2 +353.8 +258.3 -76.9 +174.7
18 +23.3 +371.4 +58.8 -92.0 +174.7
software). See Table 1. 19 +28.8 +397.2 +44.9 -87.0 +174.7
20 -14.4 +281.1 +515.6 -49.0 +179.9
TABLE 1

PROGRAMMING OF PLC AND ROBOT

C. Assessment
Input Address Output Address Assessment of student understanding and ability
Capacitive X10 Conveyor Y00
Proximity Sensor Motor to implement a PLC’s interaction with sensors,
motors, and other objects were performed for both
Part Load Sensor X11 Gear Motor Y01
(prox. sensor) short-term and long-term assessment. (1) Short-term
assessment took place within the timeframe of an
Photo Eye X12 Red Light Y05
Sensor (loading individual course. Each student was assessed by
area) performance on eight homework assignments (20
Track Sensor X13 Air Y06
(end of the Cylinder percent of course grade), an in-class mid-term exam
conveyor) (30 percent of course grade) and an individual,
open-ended project, in which each student was
The command line to enter some commands such as expected to incorporate codes, commands, and
MJ (Move Joint) is as follows: design functions (50 percent of course grade). (2)
Long-term assessment is conducted after course
Function: Turns each joint the specified angle from completion and post-graduation. Each student is
the current position. (Articulated interpolation) surveyed periodically to determine how each
component of the lab course has helped the graduate
in his or her professional employment and career.
 INTERDISCIPLINARY AUTOMATION AND CONTROL IN A PLC LABORATORY 179

IV. RESULTS Teaching, 2002, pp. Available:

http://www4.ncsu.edu/unity/lockers/users/f/felde
Students who successfully complete a laboratory r/public/Papers/Scholarship_chapter.pdf
course are able to: understand and implement [3] D. Richter, “Infusing an interdisciplinary
different types of software packages for the purpose automation experience in engineering
of robot control (such as GX-Developer or technology education,” in Proceedings of the
Roboware™ for Mitsubishi products or RS-500 and 2007 American Society for Engineering
RS-5000 and Factory-Talk for Allen-Bradley Education Annual Conference & Exposition,
products); and program Human Machine Interface Honolulu, HI.
(HMI) screens to integrate them into PLC control [4] T.J. Stiller and G.R. Johnson, “Constituent
automation lines. After graduation, they can perform influences on engineering curricula,” in
the tasks listed above and communicate intelligently Proceedings of the 2004 American Society for
with members of a team of designers working on Engineering Education Annual Conference &
automating any industrial process. Exposition, Salt Lake City, UT.
[5] C. Mirman, “Educating future engineers requires
V. CONCLUSIONS AND FUTURE WORK teamwork,” Automation Notebook, Fall Issue 12,
2008, [Online]. Available:
UBSOE’s PLC Industrial Control Lab will http://automationnotebook.com/2008_issue_12/f
continue to provide critical laboratory education for eaturestory_issue12_2008.html
graduate engineering students. Further, the authors [6] E-Designer for the E1000-series Reference
envision that the Lab will assist local companies in Manual, Mitsubishi Electric, MA 00759A, 2005-
hardware design and manufacturing and provide the 03.
necessary software programming for different [7] MicroLogix 1100 Programmable Controllers,
applications in process automation. Bulletin 1763, Instruction Set Reference Manual,
Rockwell Automation, Inc., Publication 1763-
ACKNOWLEDGEMENT RM001B-EN-P, April 2007.
[8] FX Series Programmable Controllers
The authors acknowledge the equipment Programming Manual, Mitsubishi Electric,
contributions by Gibson Engineering Company, Manual number JY992D88101, Manual revision
Incorporated of Norwood, MA and HESCO of D, April 2003.
Rocky Hill, CT. [9] Roboware™ Programming/Control Software for
Mitsubishi Robots Ver 2.xx User Manual,
REFERENCES KakeWare¤, RV-M1 & RV-M2 Models.
[10] Melsoft FR Configurator Instruction Manual
[1] L.R. Lattuca, P.T. Terenzini, and J.F. Volkwein. VFD Software, Mitsubishi Electric, FR-SW2-
(March, 2006). Engineering Change: A Study of SETUP-WE.
the Impact of EC2000. ABET, Inc. Baltimore, [11] Logix 5000 Controllers Common Procedures
MD. [Online]. Available: Programming Manual, Rockwell Automation,
http://www.abet.org/Linked%20Documents- Inc., Publication 1756-PM001I-EN-P, January
UPDATE/White%20Papers/Engineering%20Ch 2007.
ange.pdf [12] SLC 500 Instruction Set Reference Manual,
[2] P.C. Wankat, R.M. Felder, K.A. Smith, and F.S. Rockwell Automation, Inc., Publication 1747-
Oreovicz, “The scholarship of teaching and RM011E-EN-P, January 2006.
learning in engineering,” in Disciplinary Styles [13] K.C. Bower, T.W. Mays, and C.M. Miller,
in the scholarship of teaching and learning: “Small group, self-directed problem based
Exploring common ground, M.T. Huber and S. learning development in a traditional
Morreale, Eds., Washington, DC: AAHE/ engineering program,” in 34th ASEE/IEEE
Carnegie Foundation for the Advancement of
180 TOPOROVSKY ET AL.

Frontiers in Education Conference, Savannah, engineering educators, in 37th ASEE/IEEE

GA, 2004. Frontiers in Education Conference, Milwaukee,
[14] S.M. Lord and M.M. Camacho, “Effective WI, 2007.
teaching practices: Preliminary analysis of
 Stereo Spectral Imaging System for Plant Health
Characterization
S. C. Yoon1 and C. N. Thai
University of Georgia
Biological and Agricultural Engineering Department
Athens GA 30602-4435
E-mail: [email protected]
Web site: http://www.engr.uga.edu/~mvteachr
1
S. C. Yoon is now at USDA, ARS, Richard B. Russell Research Center, Athens, GA 30604-5677,
[email protected]

Abstract - Three-dimensional (3D) measurements of whole plants of plants with 3D spatial information, spectral information can
may provide detailed structure information about plant growth also biochemically characterize plants growth and health. For
patterns and also complement existing X-ray systems for the example, the normalized difference vegetation index (NDVI)
below-ground part. In addition to this structural characterization using two spectral reflectance responses at red and near-
of plants, spectral information may also biochemically
characterize plants' health. A stereo vision technique is a cost-
infrared (NIR) spectral regions has been widely used in
effective and rapid imaging technique for measuring and remote sensing as an index to estimate various vegetation
reconstructing 3D structures. The Normalized Difference properties including chlorophyll concentration in leaves [6],
Vegetation Index (NDVI) requiring measurements of two leaf area index [7], biomass and plant vigor [8], plant
spectral wavelengths in the NIR and red spectral regions has productivity [9], and stress [10]. We integrated both stereo
been widely used in remote sensing as an index to estimate vision and NDVI techniques and developed a dual-band stereo
various vegetation properties including chlorophyll spectral imaging system for chlorophyll and biomass
concentration in leaves, leaf area index, biomass, and plant quantification of plants in 3D space.
productivity. We integrated both stereo vision and NDVI Our main objective was to develop a SSI system that
techniques and developed a stereo spectral imaging (SSI) system
for chlorophyll and biomass quantification of plants in 3D space.
could monitor whole plant canopies in 3D. The specific goals
We used a stereo vision camera system and custom designed a were to reconstruct plant surfaces in 3D space via stereo
dual-wavelength filter system at 690 nm and 750 nm to develop vision, to acquire spectral features via NDVI, and to map the
the SSI system. Calibration techniques for NDVI computation spectral features back onto the 3D data via computer-graphics
using two spectral band images were developed by referencing a rendering.
diffuse reflectance panel. We also developed a texture mapping
technique for rendering NDVI values in 3D space. In this paper, II. MATERIALS AND METHODS
the performance of the SSI system was evaluated with an
artificial plant with spectral properties similar to real plant’s A. Overview of Stereo Spectral Imaging (SSI) System
green leaves. Stereo vision is a cost-effective and rapid imaging
technique for measuring and reconstructing object structures
I. INTRODUCTION in 3D. The fundamental idea of stereo vision is based on the
intersection of rays from two views in 3D, assuming that
Imaging systems exploiting both stereo vision and
camera intrinsic parameters and the baseline of cameras
spectroscopy have been developed for various agricultural and
(relative positions of the cameras) are known (or can be
remote sensing applications and space missions, e.g. rice
estimated). If an object feature in 3D is visible to both camera
seedling row detection [1], wheat plant geometric properties
views, the feature may appear at different locations on the
[2], autonomous vehicle navigation [3], digital elevation maps
images captured by the cameras. The amount of spatial
for land use/cover databases [4] and the Surface Stereo Imager
displacement, called disparity, is an important cue to obtain
for the Phoenix Mars Lander of NASA [5]. In this paper, a
the depth of the object feature. Then, depth information is
stereo spectral imaging (SSI) system collecting and processing
computed directly from disparity values [11]. In practice,
spatial and spectral information from two camera views and
disparity estimation is performed by a search algorithm called
two spectral bands was developed to study plant health.
stereo matching to find correspondence between two separate
3D measurements of whole plant canopies using stereo
images. Stereo matching methods to solve this correspondence
vision may provide detailed spatial structure information
problem can be categorized into two groups: region
about plant health, disease propagation path, or phenotypes of
correlation-based and feature-based stereo matching. The
plants used in biomass production such as cotton, poplar and
region correlation-based stereo matching algorithm finds
switch grass, and also can complement existing X-ray systems
correspondence for every pixel in the image by using pre-
for the below-ground part. In addition to the characterization
determined search windows to compute correlation whereas

181
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_31, © Springer Science+Business Media B.V. 2010
182 YOON AND THAI

the feature-based method finds correspondence only for a holders were designed and attached to the outside of the
sparse set of image features (e.g. edges, lines and corners). camera body such that the filters were placed in front of the
Two common criteria for correlation computation are the sum lenses to configure the 24 cm baseline setup (Fig. 1).
of absolute difference (SAD) and the sum of squared
difference (SSD). The SSI system was based on the region
correlation-based stereo matching algorithm and the SAD
criterion.
Live green plant canopies absorb solar radiation at the
blue and red spectral regions for photosynthesis whereas most
of the radiation at the spectral region longer than 700 nm is
scattered. Hence, images of green plants appear bright in the
near-infrared (NIR) spectral region and dark in the red spectral Fig. 1. Stereo Spectral Imaging setup.
region. The NDVI was developed to utilize the properties of The selection of the NIR filter at 750 nm was to avoid the
the different light energy absorption characteristics of plants at strong signal attenuation due to the oxygen absorption of
the NIR and red spectral regions. The NDVI at each image natural sunlight energy at 760 nm in the consideration of the
pixel is obtained by possibility of deploying the SSI system outdoors and the low
~ ~ quantum efficiency of the CCD sensor beyond 800 nm. Two
I NIR( x,y ) − I RED( x,y ) . (1)
NDVI ( x,y ) =
~ ~ tungsten halogen lamps (AC) were used as the lighting source
I NIR( x,y ) + I RED( x,y )
where ~ and ~ in a laboratory setting for this study.
I RED ( x,y ) I NIR( x,y) refer to the calibrated (or
The software was written in C++ and the C++ libraries for
normalized) relative reflectance values at the visible (red) and stereo image acquisition and processing were provided in the
near-infrared (NIR) spectral regions. The NDVI value itself Point Grey Research’s Software Development Kit (SDK). The
varies between -1.0 and +1.0. In practice, a value less than 0.2 graphical user interface of the software was written in the
usually means unhealthy or sparse vegetation and a value Microsoft Foundation Class 6.0 (i.e. Visual C++ 6.0). An open
close to +1 (more than 0.6) means very healthy green leaves source computer vision library OpenCV 1.0 was used to
or very dense vegetation. NDVI values for vegetation implement image processing algorithms mentioned in the
typically range between 0.2 and 0.7. A zero or less than zero paper. OpenGL [12] codes were written for implementing 3D
means no vegetation. However, actual wavelengths and rendering and texture mapping. Fig. 2 shows a screen capture
bandwidths for NDVI calculations varied with different of the software’s graphical user interface.
sensors and applications.
The SSI system consisted of a stereo vision camera
(Bumblebee XB3, Point Grey Research Inc., Richmond, BC,
Canada), a dual-wavelength filter system at 690 nm (red) and
750 nm (NIR) for NDVI computation, a PC and software. The
Bumblebee XB3 is a 3-CCD monochrome stereo camera with
the IEEE-1394b communication link (800 Mbits/s). The three
lenses were horizontally placed 12 cm apart, thus three Fig. 2. Software Interface Design.
different combinations of baselines were selectable by the user
via software: one wide-baseline (24 cm) and two narrow- B. Stereo Image Acquisition and Processing
baselines (12 cm). The lenses were sealed inside the camera For each experiment, 3 sets of stereo images were
unit. The focal length of each lens was set to 3.8 mm. The captured: raw (i.e. no filter was used), red filters on each lens,
horizontal field of view of each lens was 70°. The camera was and NIR filters on each lens (see Fig. 3). Each set was
factory pre-calibrated for lens distortions and camera acquired one at a time by manually removing or changing to
misalignments. Therefore, there was no need for in-field the proper filters at the filter holders as shown in Fig. 1.
camera calibration and images were aligned within 0.05 pixel
RMS error. The maximum image resolution per lens was
1280x960. Smaller resolutions were selectable by the user via (a)
software in order to speed up the stereo processing and to
increase the robustness of the stereo matching process. The
frame transfer rate was 15 frames per second at the full-
(b)
resolution.
Optical band-pass filters (50.8 mm x 50.8 mm) were used
for the NDVI computation. The center wavelengths of the
near-infrared and red band-pass filters were 750 nm and 690 (c)
nm (Part No. 20BPF10-690 and 20BPF10-750 respectively
from Newport Corporation, Irvine, CA) and two sets of these
filters were used. The bandwidths (full width at half Fig. 3. Raw (a), red (b) and NIR (c) stereo images.
maximum) were 13 nm (NIR) and 11 nm (red). Two filter (Note: the right camera view is located on the left of each stereo image)
 STEREO SPECTRAL IMAGING SYSTEM FOR PLANT HEALTH CHARACTERIZATION 183

The camera parameters and lighting conditions were not computation of the NDVI became possible with raw digital
changed during the data acquisition period (except for a flat numbers in the place of normalized values, as defined by
~ ~
field calibration to be mentioned later). The captured three I NIR ( x, y ) = I NIR ( x, y ), I RED ( x, y ) = IˆRED ( x, y ) ,. (4)
stereo images (raw, red and NIR) were used for the following
tasks. First, the raw stereo image was used for stereo matching where IˆRED( x,y) is the intensity value of the red-band image
leading to 3D reconstruction. Second, both NIR- and red- captured after the adjustment of the camera parameter.
filtered stereo images were used for NDVI computation. Stereo processing in the SSI system was implemented by
Third, the NIR-filtered stereo image was used for foreground using the library functions provided in the Point Grey
object segmentation because background clutters were better Research’s SDK and OpenCV. The following steps were
suppressed by the NIR filter. applied to all 3 stereo images:
The NDVI computation required the intensity calibration a) After two images (8-bit monochrome 1280x960)
(or called normalization) to obtain the relative reflectance at was captured, they were interleaved into a single image
each pixel. We implemented a relative reflectance (RR) according to a pre-determined image format (2560x960)
calibration method for the SSI system. This method was to in order to transfer the data according to the DCAM
calibrate every single pixel on the image by using a 99% specification. The acquired stereo images were then pre-
diffuse reflectance panel which completely covered the plant processed by a low-pass filter and spatially rectified to
object in the field of view. The relative reflectance calibration un-warp the lens distortions. Without low-pass filtering,
equation at each pixel required to collect a reference image the image rectification would produce aliasing effects.
and a dark current image, and was calculated by After the rectification, the radial image patterns were
~ Imeasurement ( x,y ) − Idark current ( x,y ) . (2) straightened. This process was applied to all stereo
I m easurement ( x,y ) =
I 99% reflectance( x,y ) − Idark current ( x,y ) images whether the band-pass filters were used or not.
The RR calibration method may be impractical for outdoor b) The rectified NIR-band stereo image was sent to
applications because it is not possible to put a reflectance an image segmentation function. For segmentation, an
panel large enough to cover the area in the entire field of view. intensity threshold was applied to the rectified stereo
Thus, we developed a flat field (FF) calibration technique image for suppressing out background clutters. Then, a
more suitable for outdoor imaging of plants. The original FF morphological open filter was applied to remove small
calibration technique [13] was based on an assumption that the and isolated background clutters. The remaining clutters,
relative reflectance of a calibration target remained unchanged if remained, were removed manually. Then, this
over wavelengths (i.e. red and NIR spectral regions). For the segmentation image was used as a binary mask image to
FF calibration, a reflectance target having very little variation be applied to all stereo images.
in reflectance over wavelengths was placed in the scene and c) The raw stereo image masked by the segmentation
the digital number at each pixel was normalized to the mean map was sent to the stereo matching algorithm to obtain
value of all pixels in the region of interest (ROI) of the 3D information of the object in the scene. The stereo
calibration target. The normalization was obtained by matching was based on the sum of absolute difference
~ I ( x, y ) ~ I ( x, y ) ,. (3) (SAD) correlation method for estimating disparity at
I ( x, y ) = NIR
NIR , I ( x, y ) = RED
RED every pixel in the reference view image. The right camera
I A, NIR I A,RED
view was used as the reference view image. The
where I A, NIR and I A, RED were the average intensity values of secondary (optional) stereo matching was based on the
the ROI (i.e., the target area A) at the NIR and red spectral edge detection that utilized the relative difference in
bands, respectively. If I A, NIR and I A, RED are equal (i.e. a intensity values. The output of the stereo matching
algorithm was a disparity image. When displaying it, each
constant I A ) and (3) is plugged into (1), then the constant pixel on the disparity image was mapped by a look-up
I A will be cancelled out and the un-calibrated digital table representing a close object feature with red and a
numbers, IRED( x,y) and INIR( x,y) will replace the normalized far-away object feature with blue. The disparity image
~ was represented with either an 8-bit (full-pixel
values I~RED( x,y) and I NIR( x,y) in (1). For making the average displacement) or 16-bit (sub-pixel displacement)
intensity values of the ROI equal (or approximately equal) precision.
between the two spectral bands, we implemented an algorithm d) The NIR and red images were input to the NDVI
to adjust either a shutter time or a camera gain of the red-band. computation engine. Again, they were also masked by the
More specifically, we captured the NIR-band image first and segmentation map. A pseudo-color image was generated
then captured the red-band image by changing a camera by a color look-up table in order to display the resulting
parameter (a shutter time or a gain) such that the average NDVI image. The color look-up table of the NDVI values
intensity value of the ROI in the red-band image becomes was made according to the following rule:
(approximately) equal to that in the NIR-band image. The • If NDVI is less than or equal to 0.2, then grey
algorithm found an optimal camera parameter for the red-band value. The black refers to -1.
capture by minimizing the sum of the absolute intensity
difference between the ROIs of two bands. Then, the
184 YOON AND THAI

• If NDVI is greater than 0.2 and less than or equal calibration (Eq. 2). Another 99% diffuse reflectance target
to 0.8, then mixture of pure red (0.2) and pure (SRT-99-050, Spectralon, 5”x5”, Labsphere) was used as the
green (0.8). reference material for the flat field (FF) calibration (Eq. 4). To
• If NDVI is greater than 0.8, then mixture of evaluate the RR calibration technique, we used seven target
green and pure blue (1.0). materials: a 5”x5” multi-step diffuse reflectance target (SRT-
The floating-point NDVI values were also saved for MS-050, Labsphere) having four sub-panels of 99%, 50%,
future numerical quantification. 25% and 12% nominal reflectance values, two vinyl sheets
used for military camouflages (uniform dark and light green)
C. Rendering of NDVI Image in 3D and their back sides (black). The nominal reflectance values of
After stereo processing, each pixel in the right camera the vinyl sheets were measured by a four channel fiber-optic
view (main view – 2D) was mapped into a point in the camera spectrometer (SQ2000, Ocean Optics, Dunedin, FL). For
coordinate system (3D) where its origin was at the focal point evaluating the accuracy of the computed NDVI values, we
of the right camera (please note that by convention this image used the three camouflage vinyl sheets.
was located on the left of each stereo image as shown in Fig. An artificial orchid plant (Fig. 1) was used for the SSI
3). The set of these points in 3D were called point clouds. system development because its leaves showed similar NIR
Obtaining point clouds from stereo matching was the first step spectral reflectance as compared to real leaves. Also it was
for 3D surface reconstruction. Simply, the point clouds were essential to have a reference plant during the system
nothing but a set of 3D coordinates representing surface points development cycles because the artificial plant was not
in 3D. Thus, it was important to spatially connect these perishable and provided good references over time.
surface points in order to create an approximate surface
representation. This representation was obtained by a III. RESULTS
geometric triangular mesh model. Rendering of point clouds A. Calibration and NDVI Computation
and triangular meshes in 3D was implemented by OpenGL The performance of the RR calibration technique was
functions. The quality of the 3D mesh representation was evaluated by using seven test materials placed in the camera
greatly affected by the quality of generated cloud points. field of view: the multi-step reflectance target (99%, 50%,
Typically, point clouds needed to be cleaned and edited to 25% and 12%) and the two vinyl sheets (dark and light green)
remove duplicated and unreferenced points. For this task, and their back side (black). The performance of the RR
when necessary, we removed isolated and tiny surface features calibration was compared to spectrometer measurements. The
by applying a median filter directly to a disparity image. The calibration of the spectrometer measurements was done on the
3D coordinates of the point clouds were also saved as a fly by the Ocean Optics’ software using its own dark-current
standard file format (xyz format) so as to be used in the free & reference (Spectralon) spectra. Calibration values at 689.92
open-source 3D data processing software (MeshLab) for off- nm and 749.99 nm of the spectrometer were compared against
line processing of these point clouds. the calibration values of the SSI system at the NIR and red
Image rendering based on OpenGL was implemented for bands. Fig. 5 shows the scatter plot of calibrated image and
mapping the computed NDVI values onto 3D point clouds as spectrometer measurements of the test materials. Each point in
well as a mesh model in the camera coordinate system such the figure represented an average calibration value of one
that a 3D virtual image was generated at any given viewpoint. particular material at one of the two spectral wavelengths, thus
Also, 3D anaglyph NDVI images were generated in real-time 14 points were obtained. A regression line was overlaid to
to provide a 3D stereoscopic perception when viewed with represent the relationship between the two instruments. The R2
two color (red and green) glasses. Fig. 4 is the data flowchart value (goodness-of-fit) of the linear regression was 0.9912.
of the 3D rendering algorithm of the SSI system. Overall, the measured values from the imaging system were
slightly lower than the spectrometer measurements. But, the
Raw stereo image Stereo matching Point clouds
difference was consistently small across the entire calibration
Background
value range from 0 to 1.
NIR stereo image matching
3D surface
reconstruction
Red stereo image NDVI
compute

3D rendering

Fig. 4. Data flow from stereo image acquisition to 3D rendering.

D. Other materials needed for experimentation

A 99% diffuse reflectance target (SRT-99-180, Fig. 5. Scatter plot of calibrated relative reflectance measurements from the
Spectralon, 18”x18”, Labsphere, North Sutton, NH) was used SSI system and the spectrometer. The coordinates represented normalized
relative reflectance values. The linear regression line was drawn.
as the reference material for the relative reflectance (RR)
 STEREO SPECTRAL IMAGING SYSTEM FOR PLANT HEALTH CHARACTERIZATION 185

The calculated NDVI values were evaluated in terms of Table II.

NDVI values of the test sample: spectrometer measurements
the RR calibration and the FF calibration by comparing them
Reflectance at Reflectance at NDVI
with NDVI values calculated by the spectrometer. The vinyl 689.92 nm 749.99 nm
sheets (dark green, light green, black) were used as test Leaf 0.099 ± 0.063 0.372 ± 0.110 0.619 ± 0.136
materials. Fig. 6 shows the scatter plot of the NDVI values Stem 0.135 ± 0.065 0.160 ± 0.048 0.100 ± 0.096
computed by the SSI system and the spectrometer. Numerical Flower 0.244 ± 0.091 0.261 ± 0.078 0.041 ± 0.059
numbers are presented in Table I. A linear regression model
was applied to find the linear relationship between two B. 3D Surface Reconstruction and Rendering
instruments in NDVI computation. The R2 values of the linear The parameters used for stereo processing of the sample
regression were 0.9974 (RR), 0.9921 (FF-Shutter), and 0.9919 plant were: stereo search window size, 11x11; maximum
(FF-Gain). The performance of the RR calibration was better disparity search range, 302; minimum disparity search limit,
than the FF calibration methods. However, the FF calibration 185; image resolution, 512x384; intensity threshold for object
technique also produced NDVI values close to what the segmentation, 8; morphological open filter size, 3. The 24 cm
spectrometer produced. In fact, as shown in Fig. 6, the baseline was used for stereo image acquisition and processing.
performance gap among three calibration methods decreased The synthetic diffuse white light was used for 3D rendering.
as the NDVI value increased. The overall performance of the In OpenGL, the orthographic projection transformation was
SSI system adopting either of three different calibration adopted for 3D rendering. The surface normal vector was
techniques was similar to that of the spectrometer. computed for providing OpenGL with the correct direction of
The NDVI values of the sample plant were also measured each mesh’s surface. The surface normal vector was positive
by the spectrometer. Table II summarizes the NDVI values when the vector direction was coming off the surface (not
measured at various points of the plant. Five different points getting into the surface). The 2D NDVI image (pseudo-color)
were measured for leaf, two different points were measured was input to the OpenGL function for texture mapping of the
for stem, and three different points were measured for flower. 3D meshes. The rotation, translation, and zoom in/out by a
As shown in Table II, the average and the standard deviation computer mouse was implemented in the software so that the
values were obtained. The average NDVI value of the leaf was user can freely move around the 3D object.
0.619 which was similar to normal healthy leaves of green Fig. 3 showed the acquired stereo images: (a) Raw, (b)
plants. The average NDVI value of the stem was 0.1 less than Red and (c) NIR, where we can see that the NIR image
0.2. The average NDVI value of the flower (0.041) was even suppressed the background clutters shown in the Raw image.
lower than the stem’s. The NIR image also showed good contrast for segmentation of
the plant on the foreground. Fig. 7(a) shows the segmentation
result for the plant. This segmentation mask was applied to the
subsequent disparity and NDVI images, as shown in Figs. 7(b)
and 7(c), respectively. The disparity map was a pseudo color
image similar to temperature. The hotter (red) pixels meant
closer to the camera (or viewer). The cooler (blue) pixels
meant away from the camera. The NDVI image shown in Fig.
7(c) was just a 2D image. As mentioned in Table II, the
average NDVI value for leaves was 0.619 which should be
close to green in the NDVI-mapped color image. The blue in
Fig. 7(c) meant NDVI values were greater than 0.8. The stems
Fig. 6. Scatter plot of NDVI values of the SSI system and the spectrometer. painted with the red color meant NDVI values close to 0.2.
The linear regression line was drawn. The “RR” and “FF” refer to the RR and The gray color shown in flower areas meant NDVI values
FF calibration-based NDVI calculations, respectively. The “Shutter” and below 0.2. The NDVI values at different parts of the plant
“Gain” refer to one of the FF calibration techniques: shutter time control and
gain control. were not quantified in this paper but it can be possible by an
image segmentation technique to partition the plant into
Table I. multiple segments like leaves, stems and flowers. The 3D
Average NDVI values of the vinyl sheets: SSI system and spectrometer rendering results are represented as point clouds in Fig. 7(d),
measurements as triangular meshes in Fig. 7(e) and as texture maps of NDVI
Spectrometer- RR-based FF-Shutter- FF-Gain-
based NDVI NDVI based NDVI based NDVI values in Fig. 7(f). The overall quality of the 3D rendering
Dark 0.656 0.623 0.614 0.612 was acceptable with this sample. The imaging system,
Green however, showed difficulties with highly textured plants,
Light 0.591 0.583 0.591 0.588 whether real plants or not, and still needs to make some
Green
Back-side 0.203 0.177 0.163 0.121
improvement.
The quality of the 3D rendering was greatly dependent on
the performance of the stereo matching algorithm. The current
stereo vision camera system showed a difficulty in capturing
small surface features from highly textured areas due to the
inherent limitation of the stereo matching algorithm. Thus, the
186 YOON AND THAI

use of the camera on small plants was challenging. The most REFERENCES
important parameter for depth estimation was the min-max
range of the disparity values to search for. In order to improve [1] Y. Kaizu and K. Imou, “A dual-spectral camera system for paddy rice
the quality of the 3D rendering, the accuracy of the depth seedling row detection,” Computers and Electronics in Agriculture, vol.
63, no. 1, pp. 49-58, 2008.
estimation needs to be improved by a better stereo matching
algorithm or the point clouds may be cleaned by multiple [2] H. J. Andersen, L. Reng and K. Kirk, “Geometric plant properties by
stereo images taken at different viewpoints. relaxed stereo vision using simulated annealing,” Computers and
Electronics in Agriculture, vol. 49, no. 2, pp. 219-232, 2005.

[3] M. Kise and Q. Zhang, “Creating a panoramic field image using multi-
spectral stereovision system,” Computers and Electronics in
Agriculture, vol. 60, no. 1, pp. 67-75, 2008.

(a) Segmentation mask (b) Disparity map (c) NDVI image [4] F. Paul, C. Huggel, and A. Kääb, “Combining satellite multispectral
image data and a digital elevation model for mapping debris-covered
glaciers”, Remote Sensing of Environment, vol. 89, no. 4, pp. 510-518,
2004.

[5] M.T. Lemmon, P.H. Smith, C. Shinohara, R. Tanner, P. Woida, A.

Shaw, J. Hughes, R. Reynolds, R. Woida, J. Penegor, C. Oquest, S.F.
(d) Point cloud rendering at different viewpoints Hviid, M.B. Madsen, M. Olsen, K. Leer, L. Drube, R.V. Morris, D.
Britt, “The Phoenix surface stereo imager (SSI) investigation,” in Proc.
39th Lunar and Planetary Science Conf. (Lunar and Planetary Science
XXXIX), vol.1391, pp. 2156-2157, 2008.

[6] C.L. Jones, N.O. Maness, M.L. Stone, and R. Jayasekara, “Chlorophyll
(e) Triangular mesh rendering at different viewpoints estimation using multispectral reflectance and height sensing”,
Transactions of the ASABE, vol. 50, no. 5, pp. 1867-1872, 2007.

[7] R. Drissi, J.-P. Goutouly, D. Forget, and J.-P. Gaudillere,

“Nondestructive measurement of grapevine leaf area by ground
normalized difference vegetation index,” Agronomy Journal, vol. 101,
pp. 226-231, 2009.
(f) NDVI texture rendering at different viewpoints
Fig. 7. Screen captures of the processed images and 3D rendering results. The [8] R.B. Boone, K.A. Galvin, N.M. Smith, and S.J. Lynn, “Generalizing El
rendering results at different viewpoints were captured. Nino effects upon Maasai livestock using hierarchical clusters of
vegetation patterns,” Photogrammetric Engineering and Remote
Sensing, vol. 66, no. 6, pp. 737-744, 2000.
IV. CONCLUSIONS
[9] J.P. Wang, P.M. Rich, K.P. Price, and W.D. Kettle, “Relations between
We developed a stereo spectral imaging (SSI) system for
NDVI and tree productivity in the central Great Plains,” International
quantifying plants’ health by rendering normalized difference Journal of Remote Sensing, vol. 25, no. 16, pp. 3127-3138, 2004.
vegetation index (NDVI) values in 3D. We used a stereo
vision camera system and custom designed a dual-wavelength [10] R.E. Plant, D.S. Munk, B.R. Roberts, R.L. Vargas, D.W. Rains, R.L.
Travis, and R.B. Hutmacher, “Relationships between remotely sensed
filter system at 690 nm and 750 nm to develop the SSI system.
reflectance data and cotton growth and yield,” Transactions of the
The bandpass stereo images of the SSI system were calibrated ASAE, vol. 43, no. 3, pp. 535-546, 2000.
for the NDVI computation. A new calibration technique based
on the flat field calibration was developed for outdoor [11] E. Trucco and A. Verri, Introductory Techniques for 3-D Computer
Vision, Upper Saddle River, N.J.: Prentice Hall, 1998.
imaging. The NDVI values measured by the SSI system were
compared with the NDVI values measured by a spectrometer. [12] D. Shreiner, M. Woo, J. Neider, and T. Davis, OpenGL(R)
Both performed very similarly (the best goodness-of-fit, R2 Programming Guide: The Official Guide to Learning OpenGL(R),
was 0.9974). We also developed a texture mapping technique Version 2. 5th ed., Reading, MA: Addison-Wesley Professional, 2005.
for rendering NDVI values in 3D. The performance of the SSI
[13] D.A. Roberts, Y. Yamaguchi, and R.J.P. Lyon, “Comparison of various
system was evaluated with an artificial plant showing spectral techniques for calibration of AIS data,” in Proc. of the Second Airborne
properties similar to green leaves of real plants. The test of the Imaging Spectrometer Data Analysis Workshop, JPL Publ. 86-35,
SSI system with real plants is planned. pp. 21-30, 1986.
 Transition Network for Automation
Modeling: A Case Study
Yuval Cohen*
Department of Industrial Engineering
The Open University of Israel
Raanana, Israel, 43107

Ming-En Wang, Bopaya Bidanda

Department of Industrial Engineering
The University of Pittsburgh
Pittsburgh, USA, 15261

Abstract –This paper illustrate through a case study the advantages validation. We also adopt the convention that there can be only
of the proposed Transition Network (TN) for modelling discrete one task (state) node following each transition, and that all the
automation on the shop floor. TN is associated with high-level state required resources for each task (state) are available by
diagram The proposed technique can easily simulate and emulate incorporating the inputs related to their availability into the
the automation logic and can be automatically translated to PLC
transition conditions.
(Programmable Logic Controller) code.
II. THE EXAMPLE CASE STUDY
Keywords: Transition Network, PLC, Automation, Ladder- For the case study a drill-press system is chosen to illustrate the
diagram
remaining steps of the proposed technique. The system is
adapted from [6]. Figure 1 depicts the main components and the
I. INTRODUCTION layout of the automated drill-press that will be used throughout
this paper. The drill press consists of an electric drill mounted on
The traditional modelling approach for state diagrams of a movable platform. The platform is attached to a piston of a
discrete automation approach typically characterize a state as hydraulic cylinder. The movable platform moves vertically
any possible combination of inputs and outputs [1,2,3]. The between the up position and the down position. The limit switch
emphasis is on the states. Each change in this combinatory design LS1 senses when the drill is in the up position and the limit
is a transition of one or more inputs and/or outputs [2, 3,4]. switch LS2 senses when it is in the down position. The hydraulic
Typically the logic of the generated models grows rapidly with cylinder is controlled by a hydraulic three-position, four-way
the number of inputs and outputs. Moreover, tracking the logic is valve with two solenoids. Solenoid A drives the hydraulic
a very difficult task [2,4,5]. In this paper we suggest a different cylinder down. Solenoid B drives it up. When both solenoids (A
approach with intuitive representation of cause and effect which and B) are deenergized the platform is stationary.
facilitate tracing of logic. In our model the states represent only A stationary table supports the workpiece and a pneumatic clamp
the output combinations. These are either machine actions or clamps the workpiece against a bracket. Limit switch LS4 is
some machine states of rest. The transitions are invoked by input actuated when the clamp is in the clamped position. Limit switch
changes and result in changes to some outputs. LS3 is actuated when the clamp is in the unclamped position. An
The technique presented in this paper not only relieves the ejector is also attached to the table (oriented 90 degrees from the
automation engineer from dealing with Programmable Logic clamping cylinder). The ejector ejects the workpiece when the
Controller (PLC) code when constructing the logic, but also in drilling is finished. Limit switch LS5 is actuated when the ejector
later stages, when changes are required. This approach offers is retracted and LS6 is actuated when the ejector is in the eject
intuitive action representation facilitating verification and position.

* Corresponding author: Full name: Yuval Cohen, Academic degree: PhD, Working place: The Open University of
Israel, Research field: Automation and production modeling, Post address: 108 Ravutski, P.O.Box 808, Raanana, Postcode

43107, Tel.: (972) 9-778-1883; Fax: (972) 9-778-0668; E-mail: [email protected]

187
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_32, © Springer Science+Business Media B.V. 2010
188 COHEN ET AL.

POWER Sol. A

Hydraulic
OFF cylinder Pressure line Hydraulic
pump
ON
Fluid
tank
Drill Sol. B
motor
up
Movable platform
(drill head)
LS1 Chuck down

Drill bit
LS3 LS4
LS2
A
Bracket
Workpiece

Stationary Table A
Pneumatic
cylinder (clamp)

Sol. C
View A-A
Exhaust

Air supply LS6 LS5 Pneumatic cylinder (ejector)

Sol. D

Air supply
Exhaust
Fig. 1. The case study system: an automated drilling machine (adapted from [6])

Next, the operation of the system described in figure 1 has to be

described by a state diagram that is constructed as explained in II. STATE-DIAGRAM CONSTRUCTION
section II. Note that the states in section II are related to the
status of outputs only. The aim of the state diagram is to draw a framework for the
transition network. This is very different from a regular state
diagram. The states represent actions or no action of the
 TRANSITION NETWORK FOR AUTOMATION MODELING: A CASE STUDY 189

machines and are therefore related only to combinations of the Next, essential data manipulation is described to enable the
machines outputs. Ignoring inputs dramatically simplifies the generation of the TN. The main input required for the proposed
state diagram. In a later stage only the necessary input scheme, is a specification of the output and timers status for
combinations are considered. The design of actions (legal each state. An example for such input (for the drill-press
states) is facilitated by analysing the steps. machine) is depicted in Table 1. This specification depends
For the case study of figure 1, the eight legal system states for heavily on the specific system at hand, and must be done
the drill press example are as follows: carefully and correctly.
1) OFF
2) Idle (only the hydraulic cylinder is ON) TABLE I
3) Ready - (similar to Idle but the drill motor is ON and the STATUS OF OUTPUTS FOR STATES IN FIG. 2.
clamp is in the clamped position) (Entries representing actions or active timers are shaded.)
4) Descend - (similar to Ready but Solenoid A is activated)
5) Ascend - (similar to Ready but Solenoid B is activated can OUTPUTS
be with or without clamping) Solenoids
6) Release clamp - (similar to Ascend but Solenoid C is State Pump Drill Coo Tim-
deactivated to release the clamp) A B C D motor motor -lant er 1
7) Eject workpiece - (similar to Release but Solenoid D (the 1 off off off off off off off off
ejector) is activated) 2 off off off off on off off On
8) Retract ejector - (similar to Release but Solenoid D is 3 off off on off on on off off
deactivated) 4 on off on off on on on off
At this point a state diagram can be constructed. Figure 2 5 off on on off on on off off
depicts the state diagram for the drill press specification. States 6 off off off off on off off off
are denoted by numbers and transitions by capital letters. 7 off off off on on off off Off
8 off off off off on off off Off
Start/End I
state Each transition in the state diagram has origin state and
A destination state. The status changes between the origin and
1.OFF 2. Idle
destination outputs and timers are the transition effects.
B Therefore, by comparing the lines in Table 1 that relate to the
J
origin and destination states, the effects could be generated
L 3. Ready automatically.
Transition effects are summarized in Table 2. For example, it is
C easy to see that the changes between lines 1 and 2 in table 1 are
in the pump motor and timer. These changes appear in line one
H 4. Descend of Table 2.

8. TABLE II
D THE EFFECTS OF TRANSITIONS OF FIG. 2.

Transition Its effect/s

5. Ascend A (1→2) Pump motor ↑ Timer ↑
G
B (2→3) Drill motor ↑ Solenoid C ↑
K E
C (3→4) Solenoid A ↑ Coolant ↑
D (4→5) Solenoid A ↓ Coolant ↓ Solenoid B ↑
7. Eject 6. Release E (5→6) Solenoid B ↓ Solenoid C ↓ Drill motor ↓
F F (6→7) Solenoid D ↑
G (7→8) Solenoid D ↓
H (8→2) Timer ↑ Reset ↓ Done ↑
Fig. 2. The case study State Diagram I (2→1) Pump motor ↓ PROCESS2↓ Done ↓
J (1→5) Pump motor ↑ Solenoid B ↑ Drill motor↑
Reset↑
The states do not include all the details regarding inputs, K (5→7) Solenoid B ↓ Solenoid D ↑ Drill motor ↓
outputs, and variables. These details are embedded in the TN. L (1→7) Pump motor ↑ Solenoid D ↑ Reset ↑
190 COHEN ET AL.

The generation of significant portions of Tables 1 and 2 can be the states of the system. For example, in table III transition C
automated. In addition to the elements discussed so far, there from state 3 to state 4 is triggered by the corresponding ST3 and
are external variables originating in higher control levels. additional input change. Its effects include activating ST4 while
Moreover, different history of past events may set different deactivating ST3.
modes of operation. Such modes are identified by internal
variables. The additional variables for the drill-press example Two arc types used to activate triggers are as follows:
are as follows: 1. An enable arc ( ) the triggers can fire only while the
source place is ON.
External Variables 2. A disable arc ( ) the triggers can fire only while the
1. PROCESS2 signifies that the resources for task 2 are source place is OFF.
occupied. It is turned ON when the workpiece is ready to be
drilled (the robot finished placing the workpiece at the drill- Enable and disable arcs are drawn with dashed lines to denote
press). It shuts OFF when the drill is ready to execute a new that they do not activate or deactivate elements. A tokens is
task. used to denote the current activated state. Two types of arcs
2. FINISH2 signifies that the drilling task is finished. It is used to identify the effects of a transition as follows:
turned ON when the drilling task is finished, and turned OFF 1. Activate arc ( ) turns ON the place when the
after invoking the next transition. corresponding E-Transition is activated.
2. Deactivate arc ( ) turns OFF the place when the
Internal Variables corresponding E-Transition is activated.
1. Timer1 is used to warm up the hydraulic engine and cool it
down (turned ON by transitions A and H). TABLE III
2. Reset variable (signifies that the system is recovering from CASE STUDY TRIGGER EVENTS
power failure). Reset is turned ON whenever power is turned Trigger Trans Source state: Detected Supplementary
ON (transitions J and L) and turned OFF after workpiece event ition Var. (name) Inputs Identifiers
ejection (transition H). Change
3. Done variable signifies that the next time the system will be 1. Start A ST1 (closed) PROCESS2
idle is a cool down period followed by system shut OFF. ↑
4. Eight state-variables: ST1,ST2, ST3, ST4, ST5, ST6,ST7, 2. Reset ascend J ST1 (closed) Reset↑ LS1/OFF
ST8, signifying that the system is at a certain state. 3. Reset L ST1 (closed) Reset↑ LS1/ON
ejection
4. Warm-up B ST2 (Idle) Timer ↓ Done/OFF*
completed
III. TRANSITIONS NETWORK GENERATION 6. Cool Down I ST2. (Idle) Timer ↓ Done/ON**
completed (Internal
The state diagram is further exploded into a new type of a variable
graphical scheme named Transition Network (TN). TNs turned ON by
describe the changes in low level elements such as inputs, Transition H)
outputs, and registers, required for executing the state diagram. 7. Clamp C ST3. (Ready) LS4↑
TNs arrange the elements in a meaningful way that enables 8. Down D ST4. LS2↑
immediate comprehension of a low level code. The TNs operate (Descend)
much like Petri networks. However, instead of tokens, the 9. Up E ST5. LS1↑ Reset/OFF
transition is triggered by change of inputs and its effects are (Ascend)
modeled as turning On and OFF certain outputs. The transitions
10. Reset K ST5. LS1↑ Reset/ON
are composed of the following elements: 1) places, 2) triggers,
(Ascend)
and 3) arcs. These elements are all depicted in figure 3. Each
transition is activated by one or more triggers. The triggers are 11. Released F ST6. LS3↑
denoted by triangles pointing at a thick vertical line that (Release)
symbolizes the transition. Places (denoted by circles) represent 12. Ejected G ST7. (Eject) LS6↑
the inputs, outputs, events, and variables. 13. Retract H ST8. LS5↑
Events are assigned places with additional symbol to denotes (Retract)
the type of event (turn ON, and shut OFF). Places that use non- *Internal variable turned OFF by Transition I
binary data (e.g., timers and counters) are denoted by ** Internal variable turned ON by Transition H
rectangles. Additionally, places are added for logically denoting
 TRANSITION NETWORK FOR AUTOMATION MODELING: A CASE STUDY 191

Each trigger is invoked by places linked to the trigger by enable trigger. Since the system starts at an initial state and the current
or disable arcs. Note the usage of the source state (STi) variable state is updated as integral part of each transition, tracking the
of the transition to facilitate trigger’s identification as one of the current state is done automatically. Moreover, the trigger
trigger’s conditions. After the trigger is activated, a transition identification is facilitated as a result. Trigger identification for
from the source state (i) to another state (j) occurs immediately. the case study is elaborated in Table III, and is depicted in
Each transition also resets the source state variable (STi) and figure 3.
sets the destination state variable (STj). Note that each trigger
has only one transition, but a transition may have more than one

ST3 C ST3
State 3. Ready Sol. A State 4. Descend
LS4 ↑ Coolant
ST4

D
ST4
ST4 D Sol. A
Coolant
LS2 ↑ Sol. B
ST5

State 5. Ascend

K E

ST5 ST5 ST5

Sol. B
ST5
K E Sol. B
LS1 ↑ Drill motor LS1 ↑ Sol. C
Sol. D Drill motor
Reset Reset
ST7 ST6

State 7. Eject State 6. Release

Fig. 3. A segment of Y net integrating EASD with the embedded E transitions

192 COHEN ET AL.

IV. Trnslating the TN to PLC Code 3. Activation arcs are translated into latched outputs, and Turn-
A Ladder Diagram (LD) is a popular PLC code [3,4,5,6] which off arcs are translated into Unlatched outputs.
was chosen to illustrate the translation of the model to PLC
code. Today, most of the control programs are developed using
LD [7]. The generated LD rungs are arranged in three main V. CONCLUSION
blocks as follows: 1) events identification 2) transition triggers,
and 3) transition effects. The construction of the above three This paper presents a new modeling technique for discrete
blocks is presented next. control modeling technique that is intuitive (based on causes
and effects) and has a potential to be used also for simulation,
A. Events Identification validation, code generation and maintenance. The technique is
illustrated using a case study and shows a great potential. Such
Inputs and outputs change their voltage level when turned ON simplification could reduce the cost of discrete automation and
or OFF. These changes are referred as rising or falling edges. could contribute to progress towards more flexible automation.
The international standard IEC 61131-3 [8] defines special LD The proposed technique could be easily translated to code of
contacts for detecting rising and falling edges. A rising edge existing control equipment (PLCs). For each processing task, a
corresponds to a place with “↓” and a falling edge to a place simple state diagram that describes only the elementary actions
with “↑”. G of the system has to be built. From that point the process of the
B. Transition Triggers model generation could be automated, provided that the
following input is specified: the status of outputs for each state;
Each trigger activates one E-Transition. Each transition is the related inputs of each state, and supplementary trigger
assigned an internal variable in the LD. When the E-Transition identifiers.
is enabled that variable will be turned ON. In order to Future research includes implementation of the proposed
implement this logic, a set of rules is described as follows: technique on an industrial shop floor. This would enable
I. Each trigger forms an LD rung. simulation and visualization of operation; it would assist in real-
II. Each place (in E-Transition) that is input to a trigger forms a time tracking and failure analysis of the control system.
contact: (enable arc forms a normally open (NO) contact, and Currently the construction of a translator (model to LD) is
disable arc a normally closed (NC) contact. under way, to enable such future implementation.
III. The LD rung output is a variable that corresponds to the
activated transition.
REFERENCES
Figure 3 depicts a ladder diagram segment corresponding to the
triggers of transition D. These variables are used in figure 4.
[1] Ferdinand Wagner (Author), Ruedi Schmuki (Author), Thomas Wagner
(Author), Peter Wolstenholme (Author) "Modeling Software with Finite State
ST4 D Machines: A Practical Approach", AUERBACH, 2006
↑LS2 Trigger for
Transition D [2] Peng S. and Zhou M. C., "Sensor-based Petri net modeling for PLC stage
programming of discrete-event control design", Proceedings of IEEE
International Conference on Robotics and Automation (ICRA), Volume 2, 1907
Note: each output is activated for one - 1912, 2002.
scan only - ↑ or ↓ are inputs and they last [3] Manesis S. and Akantziotis K., "Automated synthesis of Ladder automation
one scanning cycle circuits based on state-diagrams", Advances in Engineering Software, Volume
36 , Issue 4, 225 – 233, 2005.

Fig. 4. Ladder Diagram segment for triggering transition D [4] Collins K., "PLC Programming for Industrial Automation", Exposure
Publishing, 2007
[5] Bolton W., "Programmable Logic Controllers", Fourth Edition, Newnes,
C. Transition Effects 2006.
The rules for establishing the ladder diagram portion of [6] Bateson N. R., "Introduction to Control System Technology" (7th Edition),
transition’s effects is as follows: Prentice Hall; 2001.
1. Dedicate a rung for each output place of the transition and [7] Lee G. B., Zandong H., Lee J. S."Automatic generation of ladder diagram
add to it a corresponding LD output. with control Petri Net", Journal of Intelligent Manufacturing, 15, 2, 245-252,
2. In each rung add a contact that corresponds to the relevant 2004.
transition. [8] Karl-Heinz J. and Tiegelkamp, M., IEC 61131-3: Programming Industrial
Automation System, w. CD-ROM : Concepts and Programming Languages,
Requirements for Programming Systems,6th edition, Springer, 2001.
 Computational Modeling and Simulation
of a New Energy-Saving Sealing Unit
in Industrial Applications
Zheng (Jeremy) Li
University of Bridgeport, Bridgeport, CT 06604, USA

Abstract leakage problems in industrial machine applications.

This research is conducted based on theoretic
This new type of magnetic energy-saving sealing analysis, computational modeling and simulation, and
system can reduce the leakage of lubrication oil prototype testing. The recent development of rare
pollution and media gases in general industrial machine
earth permanent magnets have brought the renovation
applications including reciprocating machines, air
compressors and refrigerating units. This article
and provided the magnetic products with stronger
analyses the feasible and reliable function of this new magnetic properties than that of conventional ferrite
magnetic energy-saving sealing system through magnets. This leads to the development of high-
computer aided modeling / simulation and prototype intensity magnetic circuits that operated energy free
testing, and the results verified the reliable performance and surpasses the electromagnets in strength and
of this sealing system. The main features of this new effectiveness.
sealing system include: better sealing performance to
prevent the industrial machines from lubrication oil The results of prototype testing indicated that this
and gaseous leakage, energy-saving due to higher new magnetic sealing system can significantly reduce
sealing functioning efficiency with less frictional work the leakage problem in industrial machine application
loss and reduced internal media leakage, cheaper in
industrial production and manufacturing because of
and it can replace the regular sealing units including
lower precision requirement to the surfaces of rubber seal, diaphragm seal, corrugated pipe seal and
reciprocating components including pistons and shafts, magnetic fluid seal.
more durable sealing function and longer sealing
lifetime.

Magnetic Circuit Analysis in Sealing System

Introduction The rare-earth magnet has the higher density of

magnetic flux Br, strong magnetic field Hg, and larger
The lubrication oil and gaseous leakage in product of magnetism and energy (BH)max, and it can
industrial machines are common problems that were be used as the permanent magnet in the magnetic
not resolved efficiently ([1]), and the machinery circuit shown in Fig. 1. All the above good features
performance were therefore degraded ([2] and [3]). lead magnetic particles to be firmly bonded to the
The research and development of new effective inside wall of magnet. This magnetic circuit shows
sealing system are being conducted in these years to higher Br in circuit working gap, longer and durable
improve the sealing performance ([4] and [5]). lifetime in sealing capacity, compact in system
structure, light in unit weight, higher in performance
In this new magnetic energy-saving sealing efficiency, and stable in functioning.
system, a rare-earth magnet is used as a permanent
magnet to control the sealing capacity to solve the

193
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_33, © Springer Science+Business Media B.V. 2010
194 LI

H * L = - Lg * Φ (2)
U 0 * Ag

Here,
H = intensity of magnetic field of magnet steel
L = length of magnet steel
Hg = intensity of magnetic field in working gap
Lg = length of working gap
Φ = magnetic flux
U0 = magnetic conductivity of vacuum
Ag = cross section area of working gap

Fig. 1 Magnetic Sealing System Assume Fm( Φ ) = H * L, the intersection of

Fm( Φ ) and straight line – [Lg / (U0 * Ag)] * Φ at
This new magnetic sealing system can be briefly ordinate in Fig. 2 is the magnetic flux in working
described as follows. When piston/shaft reciprocates gap required to be found. This gap decreases from Lg
inside of the cylinder, lubricating oil is sealed by to Lg’ after magnetic particles being added into the
magnetic particles that firmly bonded to the inside space in magnetic circuit gap. As the thickness of
wall of magnet, as oil particles move to the seal. magnetic particles in the space between walls of
Finally the oil droplets will drop to the main shaft magnet and cylinder changes from 0 to b, the
chamber at the bottom of machine by its gravity and working point of magnet moves along the straight
this can prevent the oil in crank chamber from line QK and corresponding solution of magnetic flux
entering the gaseous cylinder. Also the gaseous in working gap can be found from line QK. The
leakage can be prevented because the gas could not computational modeling simulation verifies that the
pass through the strong bonded magnetic particle magnetic field is well maintained in this sealing
layers. Two critical factors in this new magnetic system.
sealing system design are density of magnetic flux
and magnetic stability of the magnet. The magnetic The coefficient of magnetic efficiency f is used to
flux of magnetic circuit in this sealing system should judge if the magnetic field in this sealing mechanism
be kept steady for a long period and magnetic field of is properly designed.
this magnet must be stable enough to resist the
fluctuation caused by external/disturbed magnetic Bg 2 * Vg
fields, temperature change, mechanical f= (3)
vibration/shock, and severe environment. The surplus ( B * H ) max * V
density of magnetic flux Br, surplus intensity of
Here,
magnetic field Hg, and maximum product of
magnetism and energy (BH)max are required to keep Bg = density of magnetic flux in working gap
at their peak values in this magnetic sealing system
design. V = volume of magnet steel
Vg = volume of working gap
The magnetic circuit of this sealing system is in
static condition which can be analyzed using ampere B = half length of working gap
enclosed circuit and H-B curve of this rare-earth
magnet. This magnetic circuit can be considered as a The higher f value indicates more feasible and
series magnetic circuits mainly composed of rare- reasonable design to the magnetic circuit. The f value
earth magnet and working gap. From Fig.1, we can is normally 40% in standard conditions and the
get: computational modeling solution shows that f value
in this sealing magnetic system is 49.3% which
H * L + Hg * Lg = 0 (1) verifies the proper magnetic circuit design in this
sealing system.
 COMPUTATIONAL MODELING AND SIMULATION OF A NEW ENERGY-SAVING SEALING UNIT 195

Computational Modeling and Simulations Conclusion

The results of computational modeling and Currently the lubrication oil and gaseous media
simulation of this prototype are shown in Table 1 and leakages are common and difficult engineering
2. problems that affect the industrial machinery function
and performance. This new magnetic energy-saving
Table 1. Air leakage at different piston linear speed. sealing system has been developed to reduce the oil
and gaseous media leakages in industrial machinery.
Piston Linear Estimated Air All the theoretical, mechanical and magnetic
Speed (Ft/Min) Leakage (SCFM) analysis, computational modeling and simulation, and
prototype tests verify that this sealing system can
5 0.0005 significantly reduce the oil and gaseous media
leakages in industrial machinery. Its sealing
20 0.007 performance is reliable due to the firmly bonded and
strong forces between the magnetic particles and
35 0.012
reciprocating pistons/shafts. This seal mechanism is
45 0.023 also durable if compared with regular seals including
rubber seal, diaphragm seal, corrugated pipe seal.
60 0.034 This new sealing technology brings energy-saving to
the industrial machines because of higher machine
75 0.046 efficiency with less frictional force between surfaces
of seal and pistons/shafts and decreased internal
gaseous media contamination. Furthermore this
magnetic sealing product will contribute to the future
Table 2. Air leakage at different air pressure. exploitation, popularization, and application of the
rich rare-earth elements/materials.
Air Pressure Estimated Air
(PSIG) Leakage (SCFM)

50 0.001 References
[1]. H. Quelle, G. González-Jorge, L. Domarco, Magnetic field
200 0.005
cancellation on the surface of superconducting rings: Influence
350 0.009
on the flux creep, Cryogenics 48 (2008), pp. 2-5.
450 0.014 [2]. H. González-Jorge, B. Liñares, I. Quelle, E. Carballo and G.

560 0.023 Domarco, Measurement of critical current in superconducting

rings, Measurement 40 (2007), pp. 378-380.

700 0.035
[3]. H. González-Jorge, J. Peleteiro, E. Carballo and G. Domarco,

Inducting technique and trapped field in ring-shaped

The above results indicated that the preliminary Superconductors, Physica 433 (2005), pp. 132-134.
results from prototype testing and computational
simulation are very close, and both results verify the [4]. U. Topal, L. Dorosiskii, H. Ozkan, and H. Yabuz, Effect of
creditability and feasibility of this new magnetic thermal neutron irradiation in boron-doped melt-textured
energy-saving sealing system.
YBCO superconductors, Physica 388 (2003), pp. 401-403.

[5]. L. Shlyk, G. Krabbes, G. Fuchs, K. Nenkov and B. Schupp,

Flux pinning and magnetic relaxation in melt-processed

YBa2Cu3O7-d dopped with Li, Journal of Applied Physica 96

(2004), pp. 371-374.

 Design of An Intelligent Solar Tracking System Using
PIC18F452 Micro Controller
Mostefa Ghassoul
[email protected]
Department of Chemical Engineering
College of Engineering
University of Bahrain

Abstract: mid way between the geographical east and west

This paper presents the design of a microchip PIC with approximately 30 degrees towards the south.
18F452 micro controller based tracking solar energy Studies have shown that this is not ideal positioning
system, where the extracted solar energy is optimized. in order to maximize energy extraction.
The main advantage of the system is that it is A better way is to continuously orient the panels
programmed, so that it could rotate the solar panels towards the sun. This necessitates the continuous
at a pre-programmed angle and at a pre-determined rotation of the later at a very low speed so that they
periods by using short pulses. This is done by lock towards the sun all the time. As it is known, at
delivering programmed pulses which drive the very low speed, the electric motors draw a very high
rotating motors through the same programmed angle. current, so an important part of the extracted solar
The micro controller detects two light dependent energy will be fed back to the motor and this not
sensors, one detecting the optimal inclination angle only results in the loss of most the energy but heats
towards the sun, and the second guiding the solar up of the motor as well. This renders the scheme
panels to align themselves with it when required. inefficient.
This is done by measuring the voltage difference An improved scheme(1) was implemented by
between the two sensors. If he difference is greater continuously running a timer, delivering constant
than the preset offset value, the panels are rotated so short bursts at predetermined periods distributed over
that they align themselves with the pilot. If the the length of the day. These pulses drive the motor
difference is less than the offset, then the panels through pre determined angles, at constant intervals
maintain their current position while the pilot keeps of time. Although this scheme has improved the
tracking the sun. The choice of the flash micro efficiency considerably, it still suffers from a major
controller is because it could be reprogrammed on- draw back. It is not intelligent enough so that it only
line using the technique known as In-Circuit Serial rotates the panels to the new position if they deliver
Programming as well as its very low price and size. higher energy. In fact, sometimes the driving motors
consume more energy than the cells produce.
This paper discusses a better improvement to this
Keywords: scheme. It presents an optimal positioning
Panels, pilot, micro controller, LDR, sunrise, sunset. mechanism of the panels so that maximum energy is
Introduction: extracted. The position is determined by using a
Solar energy has been used for the last few decades search technique, using a pilot scheme. The pilot is
and certainly, it will continue to be used extensively built around a small sensor mounted on a miniature
in the future, especially with the diminishing of electric motor which only consumes a very tiny
alternative resources particularly carbon based amount of energy. A comparison procedure
energy resources such as oil, gas and coal. This has compares between the energy of the panels and that
been encouraged by the improvement of the cells of the pilot. If the latter is bigger, then the panels are
efficiency where it was merely 17% in the early 90’s aligned with it, otherwise they stay in their current
to pass the 30% mark lately. But, unfortunately, it is position. The procedure repeats itself during the day.
still suffering from few draw backs. The first one is The tracking system is designed around a PIC
the relatively high cost of the cells. The second one is 18F452 micro controller. The choice of this type of
that the efficiency is still low even at the current controller is due to the fact that it is cost effective,
rate of 30%, and the third one is the positioning of easily programmable, and has many features such as
the solar cells, especially during the long summer the analogue to digital conversion. It could also be
days where the sun shines for more than 16 hours a erased and reprogrammed online using the In-Circuit
day and fixed cells do not extract maximum energy. Serial Programming technique (ICSP). It has also
Currently, people using solar energy fix the panels four timers ideal for this application.

197
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_34, © Springer Science+Business Media B.V. 2010
198 GHASSOUL

System design: used to trigger the controller accordingly. If the

The design depends on two light dependent sensors. panels voltage is less than that of the pilot by the
One mounted on the solar panel and the second on predetermined threshold, then the panels are moved
the miniature motor (pilot). The miniature motor to align themselves with the pilot. If this condition is
only draws a very small current, and is used to search not met, then the processor records the total number
for the position where maximum possible energy of pulses by which the panels lag the pilot so that
could be extracted. Before sunrise in the morning, when the condition is met, the panels are rotated
both panels and pilot are oriented towards the through an angle equivalent to the total number of
geographical east (known as start run). For missed pulses in order to align themselves with the
synchronization purpose, this position is detected by pilot (the duration is determined using the cascaded
two sensors. A light dependent resistor (LDR) detects timers mentioned earlier). This procedure repeats
the sun rise and a proximity switch detects the itself with each driving pulse until one of the
position. On sun rise, the pilot starts rotating through following conditions is met. Either the end proximity
a fixed angle determined by the micro controller switch is activated or the sun sets activating the pilot
internal timers. Two internal timers Timer0 and LDR. In the first case, the proximity switch is
Timer1 in cascade are used. Timer0 is programmed oriented towards the west. During longer days of the
as a pre- scalar dividing the system clock (500KHz year, it makes sure that the panels do not over run
obtained from the 4 MHz crystal) by 1:256, and their trajectory. If this is the case both panels and
Timer1 is used to count the clocks produced by the pilot stop and wait for the sun to set which is
pre-scalar. When the time elapses, the timer delivers detected by the LDR. On sun set, the system rotates
a programmed pulse proportional to the angle which back all the way to the starting position. When the
the pilot is driven through to the new position, so that start proximity switch is activated, the system comes
it keeps facing the sun. On the trailing edge of the to a halt ( facing geographical east) and waits for the
pulse, the panels and pilot voltages are compared sun rise of the next day to repeat the procedure.
through a hardware comparator, and the output is A block diagram of the tracking system is shown in
figure (1).

Motor drive: rotation whereas the panels are driven through pin
The drives for pilot and panels motors are identical RB5 for the forward rotation and RB6 for the return
except the amount of power delivered to each of rotation. When RB3 is activated, Q5 and Q6 are both
them. The only difference is that the pilot is driven closed, and the pilot motor is energized, and rotates
by the micro controller through its output port B pin in clockwise direction. When RB4 is high, Q7 and
RB3 for the forward rotation and RB4 for return Q8 are energized and the pilot motor rotates in
 DESIGN OF AN INTELLIGENT SOLAR TRACKING SYSTEM 199

counter clockwise direction. Unlike RB3, RB4 is Sun rise and sun set capture:
connected to a set software flip/flop (refer to figure2) The sun rise and sun set detection is straightforward.
which drives the system back to the origin by hitting It is done through a light dependent resistor as shown
a proximity switch which resets the flip flop and the in figure (4). When sun rises, the LDR resistance
system comes to a stand still. decreases and the current through it increases. The
The same procedure takes place with the panels. voltage at the transistor base builds up until the
When RB5 is activated, they rotate forward and transistor is forced into saturation. The collector
when RB6 is activated, they rotate backwards. As it current in turn increases and the relay closes
is well known, the flywheel diodes are used to connecting the supply to the micro controller input
protect the micro controller from any current surge PORTA pin RA0. When darkness falls, the LDR
due the coil effects (refer to figure (3). resistor increases and the transistor base current
decreases cutting the collector current, and in turn
disconnecting the relay. The flywheel diode is
connected to protect the transistor against the di/dt
effect. An identical circuit is used for the panels but
the input is through controller pin

RA3

From microcontroller

Fig(2) Pilot motor control

Voltage level detection:

The voltage difference between that of the panels
and that of the pilot is detected using the circuit
shown in figure (5). The MAIN CELL LDR is
connected to the panel and the PILOT CELL LDR is
connected to the pilot. The two variable resistors, at
the bottom of the figure, are used so that the voltage
difference could be controlled. If the panels voltage
is higher, the output of the comparator is high. This
in turn activates the micro controller to rotate the
panels so that they will be aligned with the pilot. If
the pilot voltage is higher, then the panels are left
From microcontroller standstill as explained earlier.

Figure (3): power drive of panel motor

200 GHASSOUL

Chronology of the system:

Figure (6) shows the itinerary of the system
trajectory during the course of the day. The first
pulse shows the correct orientation of the pilot and
panels towards the east waiting for sun rise. When
the sun rises (sun rise), the pilot sensor is activated
and it starts rotating at a pulsating rate(say 2seconds
on and half and an hour off. This is software
programmable) as explained earlier. On each falling
edge, the comparison process takes place. When the
pilot voltage is smaller than or equal to that of the
panels, the required pulses to drive the panels motor
are not generated (panels do not follow the pilot).
And when the pilot voltage is bigger than that of the
panels by the programmed offset, the solar panels
follow the pilot by the same number of missed pulses
(solar panel following the pilot). When the pilot hits
the end proximity switch, the system stop and waits
for sun set. When this happens, the system rotates
back to the start and waits for the following morning.

Software angle of approximately twenty degrees for each

The program was written using the microchip pulse.
assembly language. It was tested using the On the pulse trailing edge, the pilot and the panels
MPLAB utility and using the ICD2 emulator. voltages are compared. Because the two are still
close to each other, their voltage difference is still
Test and conclusions: less than the offset.(the offset is set using the two
The prototype is tested in the laboratory using a potentiometers as to 2V shown in figure(5)). As a
mobile lamp as a light source, a pilot and a single result, the panel maintains its current position. This
panel delivering 24 volts. At the beginning, the procedure repeats itself for the next three pulses.
prototype is oriented towards the east where the On the fourth one, the voltage difference is more
proximity switch (start proximity switch of Omron than the offset and the panel rotates for
M8 eight millimeter spacing type mounted on the approximately eighty degrees to align itself with
panel pivot refer to figure (1)) is closed forcing the pilot. This is equivalent to the four missed pulses.
pin RA3 of micro controller input PORTA to the After 180 degrees, the pilot activates the end
high state, forcing in its turn the motors (panel and proximity switch, which in turn interrupts the
pilot) to the stand still state. controller to drive the system to standstill. The lamp
When the lamp is turned on, the timer starts and is moved manually When the lamp is turned off, the
delivers pulses of one millisecond every five system rotates back to its initial state.
minutes interval. The pilot sensor rotates through an As a conclusion, a cost effective intelligent sun
tracking system to extract maximum solar energy
 DESIGN OF AN INTELLIGENT SOLAR TRACKING SYSTEM 201

possible was designed. Unlike what was reported in [2] "Design of PLC based tracking system to maximize solar
energy conversion" Ghassoul, M., proceedings of the 8th
literature, the main advantage of the system is that it IEEE TEM, KFUPM Dhahran,17-18 april 2001
is intelligent enough so that the panels track the sun [3] "28/40 pin 8 bit CMOS flash micro controllers", microchip
only if that contributes to extra energy extraction and datasheet 30292, 1998
at the same time, the energy consumed by panels [4] "RC model aircraft motor control", publisher J.,microchip
technology INC, 2002
driving motor is less than that extracted. The system [5] “Maximum power solar converter” Application AN1211,
can also align itself to perfection either on sunrise or Charais,J. Microchip Technology, 2008
sunset so no drift is allowed. Another main [6] “Low power capacitive sensing with the capacitive sensing
advantage is that it takes the days variation during module” Application AN1103, Floricica,A. Microchip
Technology, 2008
the year into consideration so to make sure that the [7] “Software handling for capacitive sensing” Ferme,T.
panels do not lead the sun during longer days nor lag Microchip Technology, 2007
it during shorter ones otherwise the system will [8] "Speed control of three phase induction
completely go out of synchronization. motor using PIC 18 micro controller"
Yedamale P., Microchip Technology INC, 2002
Finally It is hoped that the system could be further [9] "Brush DC servomotor implementation using PIC 17C756A",
looked at for commercial exploitation. Benling S., Microchip Technology INC,1999
[10] "Servo control of a DC brush motor", Bucella T., Teknic
INC,1997
References: [11] "Programmable logic controllers", Par E.A., 1996, published
[1] "A programmable logic system to control a solar panel to by Newnes
pump water from a well" Ghassou M. and Radwan F., 5th [12] "Programmable logic devices and logic controllers",
IEEE TEM, KFUPM Dhahran, pp 67-70 Manlado E. and Co, published by Prentice Hall, 1997
 Adaptive Two Layers Neural Network
Frequency Controller for Isolated Thermal
Power System
Ognjen Kuljaca, Krunoslav Horvat*, Jyotirmay Gadewadikar
Systems Research Institute, Department of Advanced Technologies, Alcorn State University
1000 ASU Drive #360, Alcorn State, MS 39096, USA
[email protected], [email protected]
*
Brodarski Institute
Av. V.Holjevca 20, Zagreb, HR 10020, Croatia
[email protected]

Abstract: - An adaptive neural network control scheme for environment are subject to frequent parameters changes
thermal power system is described. Neural network control which may diminish the quality of control when
scheme presented in this paper does not require off-line nonadaptive controllers are used.
training. The online tuning algorithm and neural network
architecture are described. The performance of the It is thus natural to examine adaptive control schemes.
controller is illustrated via simulation for different changes Conventional adaptive control schemes, like one
in process parameters and for different disturbances. described in [7] can perform well. However, they require
Performance of neural network controller is compared with parameter identification and with fast and constant
conventional proportional-integral control scheme for changes this could pose problems in controller
frequency control in thermal power systems. Neural realization. Thus neural network control may be
network control scheme described here is not linear-in-
parameter. Neural network has two layers and nodes employed with no need of constant parameters
weights in both layers are tuned. identification. NN load-frequency control is described in
[8], [9] and [10]. The results obtained by using NN
I. INTRODUCTION controllers are good. However, the described controllers
require training. We provide here a performance analysis
The paper deals with the neural network (NN) of adaptive NN controller that does not require training.
frequency controller for isolated thermo power system. In The neural network is capable of on-line learning. The
modern power systems stable frequency is one of the controller described here is an improved version NN
main measures of power quality. Frequency control control scheme given for the first time in [30]. Unlike
becomes more and more important as power systems neural network controller given, our scheme here is
enter the era of deregulation. Frequency of the system augmented in such way that both layers of neural network
changes if change in load is not met with the adequate are tunable, not only output layer like in [30].
change in generation. It is becoming very hard, if not The paper is organized as follows. In Section II. are
impossible to schedule loads precisely, thus the load given some mathematical preliminaries. Model of
fluctuations in the power system are becoming more isolated thermo power system is given in Section III.
explicit. Also, it is becoming very hard to have Neural network control scheme is described in Section
determined and constant primary controllers and turbines. IV. In Section V. the simulation results are given and in
Emerging markets of ancillary services mean that primary Section VI. is given the conclusion.
controllers and turbines that are used in secondary control
change constantly. These changes can cause serious II. MATHEMATICAL PRELIMINARIES
problems when conventional control schemes are used,
including problems with stability, unless the primary Let R denote the real numbers, R n the real n-vectors,
turbines and controllers are carefully selected. To avoid R mxn the real mxn matrices. Let S be a compact simply
possible instability with changes of parameters in the connected set of R n . With map f : S → R m , let us define
system, conventional secondary controllers are
implemented using smaller integral gains than optimal C m (S) the space such that f is continuous. Let • be
performance would require. any suitable vector norm. The supremum norm of f ( x )
The literature about frequency and load – frequency over S is defined as:
control is numerous ([1], [2], [3], [4], [5], [6], [7], and
many others). Many non-adaptive schemes are given in
([1], [2], [3], [4], [5] and [6]. However, as described sup f ( x ) , f : S → R m , ∀x ∈ S (1)
before, the modern power systems in deregulated

203
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_35, © Springer Science+Business Media B.V. 2010
204 KULJACA ET AL.

Given x ∈ R N1 , a two-layer NN (Fig. 1) has a net output Gt =

1 ,
1 + s ⋅ Tt
(5)
given by

Ks ,
Gs = (6)
y = W T σ( V T x ) (2) 1 + s ⋅ Ts

where Gg, Gt and Gs are representing turbine governors,

where [
x = 1 x1 ⋅ ⋅ ⋅ x N1 ]
T
, y = y1 [ y 2 ⋅ ⋅ ⋅ y N3 ]
T
and control turbines and the power system respectively.
σ(•) the activation function. If z = [z1 z 2 ⋅ ⋅ ⋅] , we Parameters are: Tg – turbine governor time constant, Tt –
T

turbine time constant, Ks – power system gain, Ts –power

define σ(z) = [σ(z1 ) σ(z 2 ) ⋅ ⋅ ⋅]T . Including "1" as a first
system time constant. Such models are described in more
term in σ(V T x) allows one to incorporate the thresholds details in [1], [2], [3], [4], [5], [6], [7], and many others.
as the first column of W T . Then any tuning of NN
weights includes tuning of thresholds as well. 1/R

ΔPL
1 1
ΔPr - ΔPm - Δf
θv,1 Gg Gt Gs
+ +
z1 σ1
¦ w1,1 θ w,1
v1,1
x1 ¦ y1
z2 σ2 Fig. 2: The model of isolated thermo power
¦
system
x2 ¦ y2
z3 σ3
¦

θ w,m It is also shown that the system in Fig. 2 is always

θv, N h
xn vN h ,n ¦ ym asymptotically stable if R is positive number. In real
zNh σ wN h , m
¦
Nh
systems that is always the case. The system is linear and
V W the need for adaptive control or use of the function
Fig. 1: Two layer neural network approximation property of the neural network is not
obvious since there are no nonlinearities in the controlled
plant.
The main property of NN we are concerned with for
However, all the parameters can and do change during
control and estimation purposes is the function
the operation. Thus, it is conceivable that adaptive control
approximation property ([16], [20]). Let f ( x ) be a
scheme would perform better than nonadaptive control.
smooth function from R n → R m . Then it can be shown The most usual way of control is to use linear PI
that if the activation functions are suitably selected, as controllers. The controller in that case takes the change of
long as x is restricted to a compact set S ∈ R n , then for power system frequency Δf as the input and produces the
some sufficiently large number of hidden-layer neurons control output ΔPr as output. That signal is fed to turbine
L, there exist weights and thresholds such that governor in order to counter the changes caused by the
change in the load ΔPL. The turbine output is the
mechanical power ΔPm. However, introduction of integral
f ( x ) = W T σ( V T x ) + ε ( x ) (3) action also means that the system can become unstable.
The system shown in Fig. 2 can be represented in state
space as
The value of ε(x ) is called the neural network functional
approximation error. In fact, for any choice of a positive ª 1 1 º
«− T 0 − ª1 º
number ε N , one can find a neural network such that RTg » « 0 »
« g »
ε( x ) ≤ ε N for all x ∈ S . « Tg » ª ΔP º
x = « »
1 1
− 0 x+« 0 0 »« r »
« T Tt » K » ¬ΔPL ¼ (7)
« t » «
Ks 1 » «0 − s»
III. ISOLATED THERMOPOWER SYSTEM « 0 − Ts
¬« ¼»
¬« Ts Ts ¼»
The model of an isolated thermo power system is
= Ax + Bu
shown in Fig. 2.
The transfer functions in the model are:
The state vector x is
1 ,
Gg =
1 + s ⋅ Tg (4) ª yg º
x = ««ΔPm »» (8)
«¬ Δf »¼
 ADAPTIVE TWO LAYERS NEURAL NETWORK FREQUENCY CONTROLLER 205

where yg is output of the turbine controllers. These states  = Fσ(V T z)Δf − k z FW.
W (13)
w
are physically available, so this representation will allow
for the NN control scheme design.
 = Gz(σ(V′T z) T WΔf ) T − k z GV.
V (14)
IV. ADAPTIVE NEURAL NETWORK CONTROL v

The neural network is build as shown in Section 2. The

first layer weights, sigmoid functions, are initialized These update laws where obtained from Lyapunov like
stability analysis of the system as described in [25] and
randomly and then fixed to form basis ϕ(x). The NN
[28].
output is
V. SIMULATION RESULTS
y = W T σ( V T x ) (9)
The simulations were performed for the following set
of parameters: Tq = 0.08 s, Tt = 0.3 s, Ks = 120 pu/s, R =
This architecture is an adapted form of the tracking NN 2.4 Hz/pu. The neural network had 20 nodes in the hidden
controller described in [23], [24], [25], [26], [27] and layer; initial network values were initialized to small
numerous other papers. However, there are some random numbers. The neural network parameters were
differences. Namely, here the problem is control, not kv=10-2, kw=10-3, G=diag(7) and F=diag(0.05). The
tracking. Also, there is no special robustifyng term or PD network used sigmoid activation function. The responses
controller parallel to neural network. Since there are of the system with NN control is compared with the usual
significant time constants present in the controlled plant, PI controlled system with proportional gain of the
derivative part will not have an effect. Proportional part is controller kp = 0.08 and integral gain ki = 0.6.
not needed to initially stabilize the system since The load disturbance was simulated as ΔPL = 0.1 p.u.
uncontrolled system is always stable. Proportional gain K and ΔPL = 0.1sin(2Πf L t ) p.u. Amplitude of 0.1 p.u for
was still used in scheme given and analyzed in [30], but
disturbance presents the disturbance of 10% of available
simulation analysis had shown that proportional gain does
generation and it is quite big in practical situations.
not improve the performance of the control scheme. At
The simulation results for different frequencies of
last, unlike in papers mentioned above, we don’t use
disturbance are shown in Fig. 3 – Fig. 10.
filtered error approach.
Fig. 3 depicts the responses of the system controlled
It is assumed that the load disturbance ΔPL is bounded
with PI controller and system controlled with neural
so that
network controller for step disturbance at 0 s and 500s.
The details for this situation are shown in Fig. 4 and Fig.
ΔPL ≤ ΔPM (10) 5. It can be seen that in case of step disturbance NN
controller settles faster and has smaller steady state error.
Since our neural network acts like tunable gain (no Fig. 6 – Fig. 9 give the response for different
dynamics), (10) means that state vector x is restricted to a disturbance frequencies. It can be seen that NN control
compact set. scheme outperforms conventional PI controller in all
This assumption is always true as long as we deal with cases. Frequency deviation is two or three times smaller
the power system in the normal mode of operation. If the when NN controller is used.
load disturbance is too big there cannot be any control
action since in that case the system just doesn’t have 0.3

enough power generation capability available. The 0.2

PI
NN

protection functions take over in that case and some loads 0.1

have to be disconnected. 0

-0.1
Δf

Let us define vector z as: -0.2

-0.3

ª1 º -0.4

z=« » (11) -0.5

¬x ¼ -0.6
0 100 200 300 400 500 600 700 800 900
t [s]

Let the control signal be given by Fig. 3: The step input PL=0.1 p.u.

ΔPr = W T σ(V T z) (12)

and the weight updates are provided by

206 KULJACA ET AL.
0.15
PI PI
0 NN NN
0.1

-0.1
0.05

-0.2 0

Δf
Δf

-0.3 -0.05

-0.1
-0.4

-0.15
-0.5

-0.2
440 460 480 500 520 540 560 0 100 200 300 400 500 600 700 800 900
t [s] t [s]

Fig. 4: The step input PL=0.1 p.u. - detail

Fig. 8: The response for fL=0.02
-4
x 10
0.015
1.5 PI PI
NN NN
1
0.01
0.5

0 0.005

-0.5
Δf

Δf
0
-1

-1.5
-0.005
-2

-2.5 -0.01

-3
-0.015
550 560 570 580 590 600 610 0 100 200 300 400 500 600 700 800 900
t [s] t [s]

Fig. 5: The step input PL=0.1 p.u. - detail

Fig. 9: The response for fL=0.002 and changed
governor and turbine parameters
0.015
PI
NN
0.01
VI. CONCLUSION
0.005
The results of an initial research in neural network
control of power systems with two layers nonlinear in
Δf

-0.005
parameter neural network controller are shown. The
simulation results show that controller performs well and
-0.01
adapts to changing parameters. The controller does not
-0.015
require off-line training phase. Analysis of responses
0 100 200 300 400 500 600 700 800 900
t [s] shows that neural network control scheme outperforms
conventional control scheme. By defining the neural
Fig. 6: The responses for fL=0.002
network differently, having output layer weights W
defined as a matrix this scheme can be adjusted to deal
The response of the system for changed turbine with the multivariable systems. The performance analysis
governor and turbine parameters (Tg=0.09s and Tt=0.8s) here shows that is would be worth to continue with the
is shown in Fig. 6 and Fig. 7 . Again, NN control research effort toward two tunable layer neural network
outperforms PI controller. controllers for interconnected systems as well as for the
0.04
systems with generation rate constraint.
PI
NN

REFERENCES
0.03

0.02

0.01 [1] Elgerd, O.I. "Control of Electric Power

Systems", IEEE Proceedings, pp. 4-15, 1981
Δf

-0.01 [2] Elgerd, O.I., Fosha, C.E., "Optimum

-0.02
Megawatt-Frequency Control of Multiarea electric
-0.03
Energy Systems" IEEE Transactions Power
-0.04
Apparatus and Systems, PAS-80, No. 4, pp. 556-562,
0 100 200 300 400
t [s]
500 600 700 800 900
1970
[3] Fellachi, A., "Optimum Decentralized Load
Fig. 7: The response for fL=0.005 Frequency Control" IEEE Transactions on Power
Systems, Vol. PWRS-2, No. 2, pp. 379-384, 1987
 ADAPTIVE TWO LAYERS NEURAL NETWORK FREQUENCY CONTROLLER 207

[4] Fosha, C.E., Elgerd, O.I., "The Megawatt- [17] C. A. Desoer and S. M. Shahruz, “Stability
Frequency Control Problem: A New Approach Via of dithered nonlinear systems with backlash or
Optimal control Theory" IEEE Transactions Power hysteresis,” Int. J. Contr., vol. 43, no. 4, pp. 1045-
Apparatus and Systems, PAS-80, No. 4, pp. 563-577, 1060, 1986.
1970 [18] B. Friedland, Advanced Control System
[5] Hamza, M.H., Agathoklis, P., Chan, W.C., Design, Prentice-Hall, New Jersey, 1996.
"A Combined Selftuning and Integral controller for [19] B. Igelnik and Y. H. Pao, “Stochastic
Load Frequency Control of Interconnected Power Choice of Basis Functions in Adaptive Function
systems "Regelungstechnik 30. Jahrgang Heft 7 pp. Approximation and the Functional-Link Net,” IEEE
226-232, 1982 Trans. Neural Networks, vol. 6, no. 6, pp. 1320-1329,
[6] Kundur, P., Power System Stability and Nov. 1995.
Control. McGraw-Hill, Inc.,1993 [20] K. Hornik, M. Stinchombe, and H. White,
[7] Liaw, C.M., "Design of a Reduced-Order “Multilayer feedforward networks are universal
Adaptive Load-Frequency Controller for an approximators,” Neural Networks, vol. 2, pp. 359-
Interconnected Hydrothermal Power System 366, 1989.
"International Journal of Control, Vol. 60, No. 6, pp. [21] B.S. Kim and A.J. Calise, "Nonlinear flight
1051-1063, 1994 control using neural networks," AIAA J. Guidance,
[8] Demiroren, A., Zeyneigly, H.I., Sengor, Control, Dynamics, vol. 20, no. 1, pp 26-33, 1997.
N.S., "The application of ANN technique to load- [22] J.-H. Kim, J.-H. Park, S.-W. Lee, and E. K.
frequency control for three-area power system”, 2001 P. Chong, “Fuzzy precompensation of PD controllers
IEEE Porto Power Tech Conference, Porto, Portugal for systems with deadzones,” J. Int. Fuzzy Syst., vol.
[9] Bevrani, H. "A novel approach for power 1, pp. 125-133, 1993.
system load frequency controller design", [23] S.-W. Lee and J.-H. Kim, “Control of
Transmission and Distribution Conference and systems with deadzones using neural-network based
Exhibition 2002: Asia Pacific. IEEE/PES , Volume: 1 learning control,” Proc. IEEE Int. Conf. Neural
, 6-10 Oct. 2002 Networks, June 1994, pp. 2535-2538.
[10] Birch, A.P., Sapeluk, A.T., Ozveren, C.T., [24] F. L. Lewis, C. T. Abdallah, and D. M.
"An enhanced neural network load frequency control Dawson, Control of Robot Manipulators, Macmillan,
technique", International Conference on Control, New York, 1993.
1994. Control '94. Volume 1, 21-24 Mar 1994 [25] F. L. Lewis, A. Yesildirek, and K. Liu,
[11] R. Barron, “Universal approximation “Multilayer neural-net robot controller with
bounds for superpositions of a sigmoidal function,” guaranteed tracking performance,” IEEE Trans.
IEEE Trans. Info. Theory, vol. 39, no. 3, pp. 930-945, Neural Networks, vol. 7, no. 2, pp. 1-11, Mar. 1996.
May 1993. [26] F. L. Lewis, K. Liu, and A. Yesilidrek,
[12] A.G. Barto, "Reinforcement learning and “Neural net robot controller with guaranteed tracking
adaptive critic methods," Handbook of Intelligent performance,” IEEE Trans. Neural Networks, vol. 6,
Control, pp. 469-492, Van Nostrand Reinhold, New no. 3, pp. 703-715, 1995.
York, 1992. [27] F. L. Lewis, K. Liu, R. R. Selmic, and Li-
[13] J. Campos and F.L. Lewis, "Adaptive critic Xin Wang, “Adaptive fuzzy logic compensation of
neural network for feedforward compensation," Proc. actuator deadzones,” J. Robot. Sys., vol. 14, no. 6, pp.
American Control Conf., vol. 4, pp. 2813 –2818, San 501-511, 1997.
Diego, 1999 [28] F.L. Lewis, S. Jagannathan, and A.
[14] F.-C. Chen, and H. K. Khalil, “Adaptive Yesildirek, Neural Network Control of Robot
control of nonlinear systems using neural networks,” Manipulators and Nonlinear Systems, Taylor and
Int. J. Contr., vol. 55, no. 6, pp. 1299-1317, 1992. Francis, Lindon, 1998.
[15] M. J. Corless and G. Leitmann, “Continuous [29] N. Sadegh, “A perceptron network for
state feedback guaranteeing uniform ultimate functional identification and control of nonlinear
boundedness for uncertain dynamic systems,” IEEE systems,” IEEE Trans. Neural Networks, vol. 4, no. 6,
Trans. Automat. Contr., vol. 26, no. 5, pp. 850-861, pp. 982-988, Nov. 1993.
1982. [30] Kuljaca, O., Lewis, F. Tesnjak, S., "Neural
[16] G. Cybenko, “Approximation by Network Frequency Control for Thermal Power
superpositions of a sigmoidal function,” Math. Contr. Systems", IEEE Control and Decision Conference,
Signals, Syst., vol. 2, no. 4, pp. 303-314, 1989. 2004
 Optimal Controller Comparison Using Pareto Fronts
David Moore
Department of Electrical Engineering
University of Cape Town
Rondebosch
South Africa

Abstract – New design methods in Control System are The framework proposed in this paper aims to make as
regularly proposed. These new methods are typically compared broad an analysis of controller applicability to a specific
to existing methods in a focussed manner than highlights certain problem as possible. Hence it uses a posteriori method,
criteria, but may neglect other criteria. This paper deals with the Differential Evolution (DE) [8], where as little subjective
use of Level Diagram visualisation for Pareto Fronts as a device
for comparing modern control schemes in a robust and objective
inference as possible is added before the decision maker is
manner. Specifically it uses a comparison of two methods: PI presented with information. DE is a stochastic population-
and Sliding Mode Control as a test case. based multi-objective optimizer. It uses real numbers and the
Keyword – Multiobjective Optimization, Differential evolution based ideas of crossbreeding and mutation to find
Evolution, Pareto Front, Level Diagrams, Sliding Mode Control. solutions in the form of sets that are mutually optimal. Such
solutions are typically represented as Pareto Fronts.
Pareto Fronts are traditionally visualised on a 2-D plot of
cost vs. cost showing the trade-offs that can be made between
I. INTRODUCTION
various costs 2]. When the front exists in more than 2
New methods are constantly being proposed for the design dimensions it can become difficult to analyse [1] and even
of Control Systems (e.g. VSC in 1970s, H-infinity in the more difficult to visualize.
1980s, etc). More often than not, the new methods are Level Diagrams 3] are a novel method for visualization of
compared to previous ones in a focused manner that high dimensional Pareto Fronts that overcomes this limitation.
highlights one or two criteria. This can be misleading in some Most MOO are coupled to a decision maker and attempt to
cases [l] and a more comprehensive comparison is required. find a single or directed set of solutions to a given problem
This paper presents a framework for comparing Control [2] rather than presenting a visualization of the trade-offs.
design methods using high dimensional Pareto fronts, that are
computed using Multi-Objective Optimization (MOO) 2] and II. METHOD OVERVIEW
investigated using the novel Level Diagram [3] visualisation. The success of a control method is typically dependent on the
Specifically it compares the Sliding Mode Control 4] and PI
specifics of its design and implementation. It makes sense to
Control methods in terms of many criteria. compare two methods only when both are very well designed.
To ensure that the comparisons are unbiased, the In order to ensure this, optimization is used to find the best
framework uses the best controllers that each method can controller that a method can produce, based on the costs that
produce. The performance of the controllers is quantified by have been defined for the problem.
cost functions (or costs) that are assumed to identify the best
controllers. These costs are frequently and typically mutually
A. Multiobjective Optimization
exclusive.
Multiobjective Optimization finds the best costs by
Multi-Objective Optimization methods are utilized to find
simultaneously optimizing two or more conflicting objectives
optimal (best) solutions to a given problem [2] where optimal
subject to certain constraints 2]. It can be formalized
is defined in terms of the trade-offs between computed cost
mathematically as follows
functions and the concept of Pareto Efficiency 2].
A Pareto Efficient or Optimal situation is one where any 1
change to make a member better off is impossible without
making some other member worse off [2]. Solutions that are
Pareto Efficient are described as non-dominant 5][6].
Many different types of MOO methods exist; they fall into
four groups depending on when the decision maker adds
preference information. These four groups are priori (before), where  is the input or decision vector,  is the decision
progressive (during), posteriori (after), and no preference 7]. space and is the cost or objective vector. This

209
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_36, © Springer Science+Business Media B.V. 2010
210 MOORE

optimization can then be explained as simultaneously

(1)
minimizing all objectives of for every .
Multi-Objective Optimization yields a set of mutually
optimal solutions . It is unique for a given set of costs. This (2)
optimal set is commonly referred to as the Pareto Front [5].
A norm is then applied to the normalised costs to evaluate
i. Differential Evolution the distance to the ideal point. The Euclidean norm (3) gives
Evolutionary Algorithms (EA) are a type of optimization the best conventional geometric representation of the shape of
method increasingly being used in engineering design and the Pareto Front [ HYPERLINK \l "XBl08" 3 ] and is used in
practice 5]. There are many different EA methods available, this paper.
each with different characteristics. One such method is
Differential Evolution, which is a popular method for multi-
objective optimization problems [5]. } (3)
The Pareto Differential Evolution Algorithm 6] is a
population-based optimizer, that finds a population of n Each cost ( ) and input ( ) is plotted on a separate set of
members making up a discrete approximation of the axes. They are plotted on the X axis against their
Pareto front . corresponding norm on the Y axis.
This effectively places all corresponding information for
ii. 2-D Pareto Front any or at the same position on the Y axis.
2-D Pareto Fronts are a representation method used to
compare trade-offs between optimal solutions to multi-
objective optimization problems, based on their costs.

Figure 2: Level Diagrams

Figure 2 gives the Level Diagrams for the same set given in
Figure 1.
An example of Level Diagrams with higher dimension, (in
Figure 1: 2-D Pareto Front this case 7 costs) is shown in Figure 7 and Figure 8.

An example of a 2-D Pareto Front is given in Figure 1, in C. Proportional Integral Control

this case showing a discrete approximation of the Pareto The two-term PI controller is an industry standard control
Front for two costs. method developed in the 1940s. A complete analysis and
development of design techniques is given in [9].
B. Level Diagrams
These visualisations [3] are used for demonstrating higher D. Sliding Mode Control
dimensional Pareto fronts (fronts of dimension greater than 3) Sliding Mode Control (SMC) is a robust non-linear control
such that useful information can still be visually interpreted technique [4][10]. It is based on the concept that it is easier to
from them. control 1st-order systems than nth-order systems. The
In Level Diagrams each cost is classified based on it’s technique works by defining a sliding surface (4), (sliding
proximity to an ideal point. Each cost is first normalised with mode, manifold or hyper-surface) in the state-space:
respect to its minimum and maximum values, such that each
has equal importance.
 OPTIMAL CONTROLLER COMPARISON USING PARETO FRONTS 211

(4)

This surface is defined such that the system trajectory has

desirable dynamics when on the surface.
And the scheme is designed such that the feedback forces
the system trajectory onto the sliding surface [4] in a reaching
motion. It then remains on the surface through input
switching in a sliding motion. Figure 3: General Control Setup
A reasonable modification [4] includes a term that Some definitions of the simulation are given below.
allows direct specification of both reaching and sliding rate
dynamics.
A. Plant Definition
The test plant is a small sized DC motor that represents a
(5)
typical first order dynamic system with transfer function:

(6) (7)

SMC is meant to be insensitive to parameter changes and The motor speed is regulated by the PI and (Integrated)
be able to completely reject matched disturbances [11] due to SMC controllers.
the non-linear term .
Complete SMC design methods are presented in both B. Cost Definitions
[10][4]. The Pareto Differential Evolution algorithm works by
One of the inherent difficulties of SMC is chatter that is optimizing the outputs (i.e. costs) for a given set of inputs (i.e.
induced by non-ideal switching. Minimization of this chatter decision variables). The choice of these costs affects the
is one of the primary focuses of research on SMC. A review outcome of the optimization.
of methods for reducing chatter is presented in [11]. In order to keep the comparison as general as possible, the
For this paper a pragmatic engineering approach, called costs were chosen to be as application agnostic as possible.
Integrated SMC, was used to deal with chatter. The method The first three costs are based on the Integral Square Error
works by augmenting the SMC controller with an integrator (9) of the plant error e (8), resulting from changes in the
on its output. The output of this integrator is the manipulated setpoint r, input v and output d disturbances introduced to the
variable for the plant. There are three main reasons for the system.
integrator’s inclusion: Firstly, SMC design techniques call for
systems with greater than first order dynamics, so first order (8)
systems need augmenting. Secondly, the integrator acts as a
filter, removing the need for any other chatter correction. And (9)
finally it eliminates the finite offset error that would occur
with first order plant models due to input disturbances. Similarly, the next three costs are based on the Integral
One expected result from this modification is that Square Error for the plant control u (ISU). It is modified such
integrating the SMC output reduces the controller’s that the resulting cost is a finite integral. This is done by
effectiveness at rejecting disturbances due to the delayed subtracting u’s final value from u at each time step.
response that the integrator introduces.
(10)
III. COMPARISON DEFINITIONS Both the and costs are abbreviated to
As stated above, PI and SMC control are compared to one and in the diagrams that follow.
another using Level Diagrams that are computed by the The last cost included is a rough measure of the robustness
Pareto Differential Evolution Algorithm. Furthermore, a set of the controlled system to model changes. It finds the largest
of PI controllers that cancel the plant poles and lead to poor positive percentage change for each of the 6 other costs when
internal performance [12] was included for extra comparative the plant parameters A and T are varied by 50%. In order for
purposes. Figure 3 shows the general plant setup with setpoint this measure to be useful on the Level Diagrams, the base ten
r, input v and output d disturbances. logarithm of this value is used.

C. Optimizer Setup
The optimizer was run for 20 generation with a crossover
rate and mutation rate of 0.15.
212 MOORE

The input parameters for the PI controller are the P and I as

shown in (11).

(11)

The parameters are bounded to the range .

Similarly for SMC, the input parameters are ,  and K.
They are bounded to the ranges , and
.
These limitations on  and  are introduced based on the
analysis of the eigen values of the resulting system dynamics.
They are primarily introduced as a result of limited sampling
speed.
The PI cancel set’s parameters are specified such that on
the range [0,10] pole – zero cancellation occurs for all
controllers.
Figure 4: PI (black) and PI pole- zero cancellation (grey) cost Level
D. ODE Solver Setup Diagrams.
The ODE solver uses an RK4 algorithm with a sampling
time of 0.05. When two more costs (Ev and Uv) are included in Figure 5
the situation changes. Some of the resulting optimal
E. Pole-Zero Cancel Set controllers are still on the pole-zero cancellation curve (as
The pole-zero cancelling PI controller set is included as a they are still optimal), but additional optimal controllers are
reference set which is known to be undesirable in practice; found not on this curve, thus avoiding the phenomenon.
since pole – zero cancellation can lead to poor internal
performance [12], and as such should be avoided.
Optimization does not take account of this and depends only
on the costs being used. Pole-Zero cancellation typically
results in very good output disturbance rejection behaviour
but poor input disturbance rejection behaviour. Due to the
good output disturbance rejection the controllers are often
classified as optimal. Common costs used in Control
Engineering optimization, such as the and costs
used in this paper, do not account for this issue of internal
performance. As such, optimizers will often optimize onto
values where pole-zero cancellation occurs.
The costs included in this paper are unable to avoid this
phenomenon, but they do show alternative solutions.
This is demonstrated below, where in Figure 4, two costs
(Er and Ur) are used initially. The optimized PI controllers
are shown in black, and the pole-zero cancellation set is
shown in grey. From the figure one can see that all the Figure 5: PI (black) and PI pole-zero cancellation (grey) cost Level
optimal PI controllers lie on the pole-zero cancellation curve. Diagrams
One thing to note is that the part of the curve where the
optimal PI controllers does not lie is non-dominant and thus IV. RESULTS
sub-optimal. It is included for visual completeness.
The results are split into two sections: General results from
comparing the SMC and PI methods and results from using
Level Diagrams.

A. General Results
During testing it was found that neither PI nor SMC
dominated the other on any significant portion of their fronts.
This indicates that both Pareto Fronts exist in a mutually
exclusive region of the n-dimensional space as shown in
Figure 6.
 OPTIMAL CONTROLLER COMPARISON USING PARETO FRONTS 213

Figure 6: 2-D visualisation of separation between PI and SMC

while a large portion of the PI solutions are very sensitive to

model changes.
A number of observations can be drawn from the Level
Diagrams for output disturbance and setpoint tracking (Ed,
Ud, Er, and Ur). The PI controllers respond in a very similar
fashion for both setpoint and disturbance; this is expected as
there are no modelled feedback dynamics in the simulation.
However, the SMC controller’s ISE for the two disturbances
changes: this is a function of this controller design.
Further, on the Ud and Ur Level Diagrams, SMC generally
requires less controller effort u, having a maximum value of
around 15 and 12 respectively. While a large number of the PI
controller solutions require far larger controller effort to
achieve their control.
That being said, the PI achieves better Output disturbance
rejection and setpoint tracking, shown on Ed and Er with their
higher controller effort.
This situation differs for the input disturbance (Ev and Uv).
Figure 7: Cost Level Diagrams for PI, PI Cancel and SMC PI manages equal or better input disturbance rejection as
shown in Ev, but generally requires less controller effort than
B. Level Diagram Results the SMC controller to a achieve it, shown in Uv.
Figure 7 and Figure 8 shows the resulting Level A couple of observations can be drawn from the Input
Diagrams for the Pareto front for the PI (black), PI cancel Level Diagrams shown in Figure 8. Firstly, lower values
(light grey) and SMC (dark grey) methods. In this case it is of lambda ( ) are undesirable as they limit the reaching
limited to a maximum Euclidean norm value of 0.75; this speed on the SMC and thus would increase the various ISE
is done on the assumption that only reasonably balanced measures. Higher values of lambda tend to have lower norm
cases are to be examined as it effectively removes the values.
extremes of the Pareto Fronts. The PI cancel set is The PI pole-zero cancellation set exists as a subset of the
retained as a known undesirable set. optimal PI controllers. This implies that further costs are
A number of results can be drawn from visual inspection of needed.
the Level Diagrams. While there are fewer results that can be drawn from the
Firstly, from the Robustness Delta Level Diagram, the input plots in this case, they can be useful for design decisions
SMC controller is much more robust to model changes than and trade-off analysis.
the PI controller; having a maximum value of less than 1,
214 MOORE

Finally, as the proportional gain Kp for the PI controller While Integrated SMC degrades the disturbance rejection of
increases above around 5, the controller norm worsens. SMC, it still retains the other robust characteristics of SMC.
As such, it is a useful method for chatter avoidance in
V. CONCLUSIONS situations where its inferior disturbance rejection is deemed
A number of general conclusions can be drawn with regard acceptable.
to both the comparison of PI and SMC design methods, and Level Diagrams proved to be a useful tool for visualisation
the use of Level Diagrams. and visual analysis of optimal controller sets. As with other
As expected, SMC proved far more robust in general that visualisation techniques, Level Diagrams would benefit from
PI, and was clearly shown using Level Diagrams. interactive features that cannot be easily shown in print.
Including more costs does not prevent the phenomenon of
pole-zero cancellation; including a set of pole-zero cancelling
controllers in the Level Diagrams does help with identifying
controllers that avoid this phenomenon.

Figure 8: Input Level Diagrams for PI, PI Cancel and SMC

REFERENCES
[1] M. Rapson, "Pareto Analysis of Controller Design Methodologies for [7] Johan Andersson, "A Survey of Multiobjective Optimization in
Intergrator plus Dead Time Processes," CISSE, 2007. Engineering Design," 2000.
[2] J.B. Yang, J.F. Whildborne G.P. Liu, Multiobjective Optimization and [8] Kenneth Price Rainer Storn, "Differential Evolution - A simple and
Control.: Research Studies Press LTD., 2002. efficient adaptive scheme for global optimization over continuous
[3] J.M. Herrero, J. Sanchis, M. Marínez X. Blasco, "A new graphical spaces," 1995.
visualization of n-dimensional Pareto front for decision-making in [9] Cecil L. Smith, Digital Computer Process Control. Scranton: Intext
multiobjective optimization," Information Sciences: an International Education Publishers, 1972.
Journal, vol. 178, no. 20, pp. 3908-3924, October 2008. [10] W. Li J.J.E. Slotine,.: Prentice Hall, 1991, ch. 7, pp. 276-309.
[4] S.K. Spurgeon C. Edwards, Sliding Mode Control: Thoery and [11] V.I. Utkin, Ü. Özgüner K.D. Young, "A Control Engineer's Guide to
Applications.: CRC Press, 1998. Sliding Mode Control," IEEE Trans on Control Systems Technology
vol. 7, no. 3, pp. 328-342, 1999. ,
[5] Carlos A. Coello, "Twenty years of Evolutionary Multiobjective
Optimization: A Historical View of the Field," IEEE Computational [12] P.A. La Plante, Comprehensive dictionary of electrical engineering
Intelligence Magazine, vol. 1, no. 2, pp. 28-36, 2006. CRC Press & IEEE Press, 1999. .:
[6] Ruhul A. Sarker Hussein A. Abbass, "The Pareto Differential Evolution [13] Keith J. Burnham Ali J. Koshkouei, "Control of DC Motors Using
Algorithm," International Journal on Artificial Intelligence Tools, vol. Proportional integral sliding mode,".
11, no. 4, pp. 531-552, 2002.
 Interaction with Objects in Virtual
Environments
Jesús A. Verduzco-Ramírez1, Enrique Araiza-Montes-de-Oca2, Daniel A. Gaytán-Valencia1, Nicandro Farías-Mendoza1
and Bruno Raffin3
1
Departamento de Sistemas y Computación, Instituto Tecnológico de Colima, Av. Tecnológico #1, C.P. 28976. Villa de
Álvarez, Colima, México
2
Facultad de Telemática, Universidad de Colima, Av. Universidad #333. Colonia Las Víboras. C.P. 28040. Colima,
Colima, México
3
INRIA, LIG, Avenue J. Kuntzmann, 38330 Montbonnot, France.

Abstract— The applications of virtual and augmented reality them within the environment in which the user can examine,
require high performance equipment normally associated with manipulate and interact with the objects shown [2].
high costs, inaccessible for small organizations and educational
In addition to the VR, there is the technology of augmented
institutions. By adding computing power and storage capacity of
several PCs interconnected by a network we have an alternative reality (AR), which consists of a collection of devices that
of high performance and low cost. The objective of this project is enhance the capabilities of human senses [3]. The AR is a
to implement an augmented reality platform integrated with technology that integrates elements of the environment
items available in the laboratories of an educational institution, typically captured in video format and audio with three-
such as PCs, webcams, Ethernet networking and free software. dimensional computer graphics objects resulting in new
This platform is an attractive and economical alternative for the
consistent, supplemented and enriched worlds [4].
development and implementation of virtual and augmented
reality applications to enable low-budget institutions simulate Both technologies are closely related but have differences.
processes, physical phenomena, weather, etc. In this first phase, VR allows immersion of the user in an artificial world that
we have a prototype that creates a scene consisting of a set of completely replaces the user's real world. The RA maintains
virtual objects, the video stream acquired from a Webcam and the user's real world enriched by the presence of virtual items.
processes the captured image into a graphic with a set of virtual The applications of this technology are found in the following
objects. A user facing the angle of acquisition of the webcam can areas: military, medical, recreational, archaeological and
interact with virtual objects in the scene with only the move of his
hand. Although our prototype is reduced to only two computers pedagogical.
the first in which images are acquired through the webcam and There are several platforms that use this approach
the second where the mixture, processing and visualization of the successfully; one of the most outstanding is Grimage [5],
scene is made, these early results demonstrate the feasibility of which performs 3D modeling of images acquired by multiple
our idea and are a strong motivation for continuing the project. cameras, physical simulation and parallel execution that offers
As future work we have planned to increase the number of
a new immersive experience. On this platform it is possible to
Webcams, to explore the interaction with complex virtual scenes
as well as distributing the processing of the scene among a group capture objects to be instantaneously modeled in 3D and
of computers to increase the size of the scene displayed. entered into a virtual world populated by rigid or flexible solid
objects with possible interaction with them. Grimage requires
Index Terms—Augmented reality, virtual reality, interactive for their operation high-performance equipment. The existing
applications, computer graphics. configuration at the INRIA Rhone Alpes has six high
definition cameras, a wall of images with 16 video projectors
and two interconnected clusters, one with 11 dual-Xeon 2.6
I. INTRODUCTION GHz for the application and another with 16 dual-Opteron 2

V irtual reality (VR) is a technology that superimposes

virtual information on the user's vision. Potential
applications have place in many fields such as medicine,
GHz for the display [5]. Figure 1 shows one of the
implementations of Grimage:

military and entertainment.

The goal of VR is to create an experience that makes the
user feel is immersed in a virtual world apparently real, for it,
VR uses 3D graphics and sound that surrounds the scenes
shown [1]. The VR uses the vision of the user, who moves
into the virtual world using appropriate devices, such as
goggles or electronic gloves. The VR exploits all the
techniques of image reproduction and extends them, using
Fig. 1. Interacting with virtual objects in Grimage.

215
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_37, © Springer Science+Business Media B.V. 2010
216 VERDUZCO-RAMÍREZ ET AL.

The software used in Grimage, FlowVR [6] is a VR API images and sends them in a video stream to the rendering
that offers support for heterogeneous codes executed in node.
parallel. FlowVR allows the development and implementation As a video server we selected Spook [9], a free software
of interactive applications in high performance computing that allows the delivery of video streams over IP networks.
clusters and grids. FlowVR reuses and extends the data flow Spook takes the images stored in the buffer, compresses the
paradigm commonly used for scientific visualization data in sequential order, and sends them as RTP packets
environments [7, 8]. through the network. In the rendering node, the
One of the limitations of the AR platforms such as Grimage is client starts the MPlayer immersed in FlowVR. Thus,
that they are made up of costly and inaccessible equipment for it is possible to display the images captured by the webcam in
educational institutions that have a small budget. the rendering the node.
Our approach aims to avoid such problems. Our proposal is to The mixture of images captured with the webcam and a 3D
develop an augmented reality platform that integrates low-cost scene component is provided by the FlowVR Render, which
equipment such as Ethernet, PC, Webcams and free software. allows the combination of the video stream to a virtual
One use expected to achieve with the implementation of this environment rendered by FlowVR.
platform is that teachers of an educational institution can
simulate physical models and manipulate virtual objects in III. IMMERSION OF OBJECTS
applications that are intended to support teaching.
A. Description of our Application
II. ARCHITECTURE OF OUR SOLUTION To test the functionality of our platform, we have developed
some examples that demonstrate the recognition of the
To implement a platform for low-cost RA, our solution is contours of the objects captured by the webcam and the
divided into four functional modules, as shown in Figure 2: interaction with virtual objects without using any device of
• One or more webcams to capture the video stream. interaction. For both operations we used the OpenCV library
• A module for the acquisition and distribution of images [10].
captured with a webcam.
• A rendering module to mix the flow of video and
B. Contours detection
interactive 3D scenes.
• A display wall for viewing the results. A contour is a list of points representing the curvature of an
image. This representation may be different depending on the
circumstances. There are many ways to represent them. In
OpenCV the contours are represented by sequences where
each entry in that sequence encodes information about the
location of the next point on the curve.
To get to show the contours of an image, we first need a
method for managing memory. OpenCV has an entity called
Memory Storage, which is a method for managing the location
of dynamic objects in memory. Any object of OpenCV that
requires the use of memory as part of its normal operation,
needs a Memory Storage. The Memory Storage is a linked list
of memory blocks for fast data storage and retrieval.
The type of object that can be stored in a memory storage is
Fig. 2. Interacting with virtual objects in Grimage.
a sequence. The sequences are linked lists of other data
structures.
In this first phase, to verify the feasibility of our idea we have
The first developed example is the recognition of a series of
implemented a prototype consisting of only two nodes. letters and figures drawn by hand. Examples are shown in
A. Acquisition Node Figure 3.
The acquisition node connects one or more web cameras,
which acquire images of the environment. A video server
stores and transmits these images to the rendering node.

B. Rendering Node
The rendering node is running an instance of FlowVR, it
receives the video frames, makes the rendering and subsequent
display through MPlayer immersed in FlowVR.
Fig. 3. Detection of contours.
C. Operation
As described in Section A, the node captures
 INTERACTION WITH OBJECTS IN VIRTUAL ENVIRONMENTS 217

C. Interacting with Virtual Objects IV. CONCLUSION

By using our prototype it is possible to capture objects In this paper we describe the first phase in the development
using the webcam and embed them into a 3D application, of an augmented reality platform of low cost.
which provides opportunities to interact with virtual objects Our solution uses low cost components, present in the
that make up the 3D scene. In an application of this type we laboratories of almost any educational institution, such as, data
identified three elements, shown in figure 4: networking, Webcams, PCs and free software, which makes
our proposal an attractive and relatively simple to implement.
This platform allows educational institutions; simulate
processes, physical phenomena, weather, etc.
A user in front of a webcam can interact with virtual objects in
an application without using any interaction device.

V. FUTURE WORK
The results obtained so far encourage us to continue this
project. We want to increase the number of nodes involved in
the rendering scenes, the number of cameras allowing the
acquisition of more than one video stream, to add textures to
Fig. 4. Architecture of the applications developed. objects displayed on the interaction scene and study the
system performance in these applications.
• The interaction scene: represented by a set of images from
the camera. REFERENCES
• The application scene: represented by a set of images in [1] C. Cruz-Neira, D. J. Sandin, T. A. DeFanti, R. V. Kenyon, and J. C.
which there are some features of the scene of interaction. For Hart. The Cave Audio Visual Experience Automatic Virtual
example, contours, corners and colors. Environment. Communication of the ACM, 35(6):64_72, 1992.
• Finally, the virtual objects embedded in scenes of interaction [2] J. Hilera, S. Otón, J. Martínez, “Aplicación de la Realidad Virtual en la
Enseñanza a Través de Internet”, Universidad de Alcalá, septiembre de
and application, which react to events within the application 1999. [Online]. Disponible en:
stage and are represented through images or textures in the http://www.ucm.es/info/multidoc/multidoc/revista/num8/hilera-oton.html
scene of interaction. [Accesado en diciembre 10, 2008].
[3] J. Ahrens, C. Law, W. Schroeder, K. Martin, and Michael Papka, A
Parallel Approach for Efficiently Visualizing Extremely Large, Time-
To test the interaction with virtual objects, we developed a Varying Datasets.
couple of examples intended to show how a user can interact http://www.acl.lanl.gov/Viz/papers/pvtk/pvtkpreprint/.
with them. [4] L. Heras, J. Villareal, “La Realidad Aumentada: Una Tecnología en
The first application shows a set of bubbles that the user can Espera de Usuarios”, Revista Digital Universitaria, vol. 5, no. 7, Agosto,
2008.
exploit to make moves on the interaction scene. The [5] J. Allard, C. Menier, B. Raffin, E. Boyer, F. Faure. 2007. Grimage:
application scene is the difference between a frame image and Markerless 3D Interactions. ACM SIGGRAPH 2007 emerging
its predecessor. Thus, a white shadow is formed as shown in technologies.
Figure 5. [6] J. Allard, V. Gouranton, L. Lecointre, S. Limet, E. Melin, B. Raffin and
S. Robert. 2004. FlowVR: a Middleware for Large Scale Virtual Reallity
The second application example shows a set of balls Applications. In Euro-Par 2004 Parallel Processing: 10th International
bouncing in the interaction scene. When the user touches one Euro-Par Conference, 497-505. http://flowvr.sf.net.
or when they collide with each other, they change their [7] Covise Programming Guide, 2001.
direction. http://www.hlrs.de/organization/vis/covise.
[8] S. Jordà. Realidad Aumentada. Taller de Sistemes Interactius 2003 –
The application scene consists of the contours of the figures 2004. Universitat Pompeu Fabra.
appearing in the interaction scene, which upon contact with [9] Spook Video Live Streamer http://www.litech.org/spook/ “OpenCV
virtual objects, they are running a routine to change its Wiki”, IBM, 2002. [Online]. Disponible en:
direction of motion. http://opencv.willowgarage.com/wiki/ [Accesado en mayo 11, 2009]
Both example applications are shown in figure 5.

Fig. 5. Example applications, to the left are the interaction

scenes and to the right the application scenes.
 One Approach for Training of Recurrent Neural
Network Model of IIR Digital Filter
Stefanova S.A.
Department of Electronic Engineering and Technologies, Technical University of Sofia,
8 Kliment Ohridski Str.,
Sofia, 1797, Bulgaria

Abstract - One approach for training of recurrent neural design of IIR digital filters with given amplitude and phase
network model of 1-D IIR digital filter is proposed. The responses is discussed in [11]. A method for 2-D recursive
sensitivity coefficients method has been applied in the training digital filters design is investigated in [12], where the design
process of the neural network. The set of time domain data is problem is reduced to a constrained minimization problem the
generated and used as a target function in the training
procedure. The modeling results have been obtained for two
solution of which is achieved by the convergence of an
different cases - for 4-th order bandpass IIR digital filter and for appropriate neural network.
partial response IIR digital filter. The frequency domain Some methods for the non-linear digital filters design
behavior of the neural network model and the target IIR filter
using neural networks are considered in [3]. A design method
has been investigated. The analysis of the frequency responses
shows good approximation results.
for nonlinear adaptive digital filters using parallel neural
networks is given in [4].

I. INTRODUCTION A various methods have been developed for the purposes

of the training process of the recurrent neural networks [13],
In the last years various approaches for modeling and [14], [15]. The back-propagation through time method is the
simulation of the analog and discrete dynamic systems based basic approach for recurrent neural network training,
on different type of neural networks have been developed. discussed in [13]. Some training algorithms based on the
Basic results related to the discrete dynamical systems sensitivity theory are considered in [14], [15]. The application
approximation using neural networks are discussed in [1]. of the Lagrange multipliers method as a training procedure of
Many approaches based on neural network have been the recurrent neural network model of IIR digital filter is
successfully applied to the problems of design, modeling and discussed in [14], [16]. Reference [9] describes the learning
simulation of analog filters [2], linear and non-linear [3], [4] algorithm for the neural network structure based on the state
digital filters,1-D and 2-D non-recursive (FIR) digital filters space representation of 1-D and 2-D IIR digital filters.
[5], [6], [7], [8], 1-D and 2-D recursive (IIR) digital filters [9],
[10], [11], [12] and adaptive digital filters [4]. In this paper one approach for training of IIR digital filter
One approach for the 1-D FIR digital filter design based model based on two layer recurrent neural network is
on the weighted mean square method and neural network to proposed. The structure and mathematical description of the
state the approximation problem is proposed in [5]. The recurrent neural network is given in section II. The training
optimal design approaches of high order FIR digital filters process of the neural network using sensitivity coefficients
based on the parallel algorithm of neural networks, which its method is discussed in Section III. The neural network model
activation matrix is produced by cosine and sine basis has been trained in such a way that with given predetermined
functions are given in [6],[7] . The main idea is to minimize input signal, the output variable approximates the target
the sum of the square errors between the amplitude response function in mean square sense. The neural network model of
of the desired FIR filter and that of the designed by training the IIR digital filter is considered in section IV. The
the weight vector of neural networks, then obtaining the effectiveness of the proposed neural network model of the IIR
impulse response of FIR digital filter. The reference [8] digital filter is demonstrated in Section V. The modeling
provides an approach for the design of 2-D FIR linear-phase results have been obtained for two different cases - for 4-th
digital filters based on a parallel back-propagation neural order bandpass IIR digital filter and for partial response IIR
networks algorithm. digital filter which is a special case of the Nyquist recursive
One method for time domain design of 1-D and 2-D digital filter [17]. The frequency domain behavior of the
recursive digital filter, based on recurrent neural network is neural network model and the target IIR filter has been
proposed in [9]. An approach for the training algorithm of a investigated and the analysis of the frequency responses
fully connected recurrent neural network where each neuron shows good approximation results. Some simulations are
is modeled by an IIR filter is proposed in [10]. The weights of realized using harmonic signals with different frequencies in
each layer in the network are updated by optimizing IIR filter the filter’s passband and stopband. The proposed approach
coefficients and the optimization is based on the recursive can be extended and successfully applied to the case of
least squares method. A Hopfield-type neural network for the modeling of nonlinear IIR digital filters.

219
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_38, © Springer Science+Business Media B.V. 2010
220 STEFANOVA

II. STRUCTURE AND DESCRIPTION OF THE NEURAL NETWORK The equations (2), (3) can be used to describe the recurrent
neural network shown in Fig. 1.
The two layer recurrent neural network is considered. The
The following additional vectors have been defined:
structure of neural network is shown in Fig. 1.
• a vector of the first layer inputs of neural network
The recurrent neural network is described with the
recurrent system of equations as follows: z (k ) = [ x1 (k ), x 2 (k ),...x n (k ), u1 (k ), u 2 (k ),..., u nu (k )]T ,

ª x1 (k ) º z ∈ R nz (4)
« » • a vector of the neural network outputs
ª w11 w12 ... w1n ...w1nz º « x 2 (k ) » y (k ) = [ y1 (k ), y 2 (k )... y m (k )]T
ª s1 (k ) º «...........»
(5)
« » « »
« w21 w22 ... w2 n ...w2 nz » « » •
« s 2 (k ) » a vector of the neural network weighting coefficients
«........ »
= «...................................... »
« »
« x n (k ) » ,
«u (k ) » [
p = w11 w12 ...w1nz , w21 w22 ...w2 nz ....wn nz , c1 , c2 ...cm , ]
« » « 1 »
¬« s n (k )¼» « wn1 wn 2 ... wn n ...wn n » p∈R
np
¬ z ¼ «...........» (6)
« » where n p = n.n z + m is the number of neural network
¬«u nu (k )¼»
(1) coefficients and the elements of matrix W ∈ R n × nz are
ª x1 (k + 1) º ª f ( s1 (k )) º introduced row by row.
« » « » Mean square error objective function is defined in
« x 2 (k + 1) » = « f ( s 2 (k )) » k = k 0 , k 0 + 1,..., k f − 1 , following form:
«........... » «........ »
« » « » 1 k f −1 m
«¬ x n (k + 1)»¼ «¬ f ( s n (k ))»¼ J (p) = ¦ ¦ [ yi (k ) − yˆ i (k )] ,
2
(7)
2 k =k0 i =1
where f(.) is the activation function for the first layer of the where { yˆ i (k )} is a set of samples of the target .
neural network
Let be introduced the vector function f(s) as follows: III. ALGORITHM OF SENSITIVITY COEFFICIENTS

f(s) = [ f ( s1 ), f ( s 2 ),..., f ( s n )]T , The main problem in the neural network training process is
the gradient calculation of the mean square objective function
then the neural network description in the matrix form is: (7) with respect to weights.
x (k + 1) = f ( W z (k )), x(k 0 ) = x 0 , (2) For that reason, it is necessary to determine the gradient
components of this objective function.
y (k ) = Cx(k ) , (3)
∂J ( p) k f −1 m ∂y
= ¦ ¦ [ y i (k ) − yˆ i (k )] i , l=1,…,np (8)
∂pl k = k0 i =1 ∂pl
x1 ( k )
For the first layer of the neural network the derivatives
∂y i
w11 s1 (k ) x 1 (k + 1) x 1 (k ) ŷ(k ) have been determined from (9) as follows:
¦ f (s1) z −1 ∂pl
x 2 (k ) x2(k +1)
¦
s 2 (k )
z −1
x 2 (k )
Cx (k )
y( k )
∂yi (k ) ∂x (k )
= ci i , i = 1,, m , l = 1, nw , nw = n.n z (9)
f (s2 )

∂pl ∂pl
x n (k )
  xn (k +1)
 x n (k ) ∂xi (k )
s n (k )
¦ f (sn ) z −1
The derivatives from (9) can be obtained from
wnn ∂pl
u 1 (k )
the recurrent system:
Fig. 1. Recurrent neural network structure for digital filter modeling n ∂x ( k + 1) f ∂x j (k )
∂xi (k + 1) ∂xi (k + 1) f
where = +¦ i ;
∂pl ∂pl j =1 ∂x j ( k ) ∂pl
x ∈ R n is a state space vector,
; i = 1,..., n ; l = 1,.., n w ; (10)
u ∈ R nu is a vector of neural network input excitations,
∂xi (k 0 )
y ∈ R m is a vector of neural network output, =0 ; k = k 0 , k 0 + 1,..., k f − 1
∂pl
W∈R n × nz
and C ∈ R m ×n
, m ≤ n , C = cij , [ ] where
­c i = j ∂xi (k +1) f ­ z j . f ′(si ) ; i = q = 1,, n ; j = 1,, nz
cij = ® i are matrixes of system coefficients, =® (11)
¯0 i ≠ j ∂wqj ¯0 ; i≠q; j = 1,, nz
n - is a number of neurons at the first layer, ∂xi (k + 1) f
nu - is a number of input excitations, = wij f ′( si ) ; i = 1,  , n ; j = 1,  , n (12)
∂x j (k )
n z = n + nu - is a number of neural network inputs.
 ONE APPROACH FOR TRAINING OF RECURRENT NEURAL NETWORK MODEL 221

and the derivatives (11), (12) are determined from the neural V. MODELING RESULTS
network shown in Fig. 1 in the case when network structure
Case 1. Modeling of 4 th order bandpass IIR digital filter
has been considered as a feedforward neural network.
For the second layer of the neural network the The following requirements to the magnitude response of
∂y the target bandpass digital filter are specified:
derivatives i have been determined as follows: passband [300–3400] Hz, passband loss – 3 dB; stopband [0–
∂pl
100] Hz and [3600–4000] Hz, stopband loss – 15 dB;
∂y1 (k ) ∂y (k ) sampling frequency – 8000 Hz.
= x1 (k ) , , m = xm (k ) (13)
∂p nw +1 ∂p n p The training sequences are generated using the impulse
Then the objective function (7) can be minimized response of target IIR digital filter shown in Fig. 2. This
target IIR digital filter has been designed by MATLAB’s
applying the standard optimization procedure.
There are two ways to realize the training algorithm of Filter Design Toolbox.
the neural network. The first one, described above, insists on The model of the bandpass IIR filter is realized as
recurrent neural network structure with 4 neurons.
determining the solution of the recurrent system (1) and
calculating the summary error from (7) for all time The effectiveness of the proposed algorithm is
samples k = k 0 , k 0 + 1,... . Simultaneously with this demonstrated by simulation with neural network model. The
calculation process, the sensitivity coefficients from (10) have simulation is realized applying input harmonic excitations
to be calculated and the gradient components of the objective with frequencies in the digital filter’s passband and stopband.
function (7) have to be obtained using equations (8). After These sinusoidal signals are different from the set of data
that the objective function (7) has to be minimized applying used in the training process of neural network model. The
standard gradient procedure. This training procedure supposes impulse responses of the target digital filter and the neural
the processing of all calculated data i.e. the training process is network model are shown in Fig. 2 and Fig. 3.
off-line. The second way to train the neural network requires
calculation of the error function (7) on the neural network
output by solving the system (1) for every time sample
k = k 0 , k 0 + 1,... . The gradient (8) of this error function has
to be determined using the sensitivity coefficients obtained
from system (10). The next operation is a correction of the
neural network weights using the components of the error
gradient (8). This training process is recurrent or on-line
training which can be realized in the real time. The possibility
of on-line realization of the training procedure is the main
advantage of the sensitivity coefficients algorithm in
comparison with the Lagrange multipliers training method,
described in [16], when the training process can be realized
only off-line.
IV. NEURAL NETWORK MODEL OF IIR DIGITAL FILTER
The IIR digital filter can be considered as a linear discrete Fig 2. Impulse response of the target bandpass IIR filter
dynamic system described in time domain with n – order
difference equation that has been stated using the delayed
samples of the input excitation and the response signal at the
output. Transforming the difference equation the state space
description of the IIR digital filter can be obtained as a linear
system of “n” number of first order recurrent equations. The
advantages of the state space representation are the simplicity
of time domain analysis, the matrix form description of digital
filter impulse and step responses and possibility of parallel
calculations.
The neural network structure shown in Fig. 1 must be
modified to linear structure. In this case the activation
function f(x) is linear. The corresponding system of equations
(1)-(3) is linear recurrent system. The sensitivity coefficients
algorithm modified for linear network is used as a training
procedure of the neural network model. Fig. 3. Impulse response of the neural network model
222 STEFANOVA

The magnitude responses in frequency domain of the The comparison between the input harmonic excitation
target IIR bandpass digital filter and neural network model with frequency 700Hz chosen from digital filter’s passband
are shown in Fig. 4 and Fig. 5, respectively. and the output reaction of neural network model is illustrated
in Fig. 7.

Fig 4. Magnitude response of the bandpass target filter

Fig 7. Input signal with added noise and frequency 700 Hz

and output response of neural network model

The time responses of the target bandpass IIR digital filter

and the neural network model in the case of input signals with
frequencies 50 Hz and 1kHz are given in Fig. 8 and Fig. 9.

Fig. 5. Magnitude response of the neural network model

Some simulation experiments have been implemented

when on the input of neural network model have been applied
sinusoidal signals with different frequencies and normal
distributed white noise N (0, 0.1) added to the harmonic
excitation. The input harmonic signal with frequency 60 Hz
chosen from the digital filter’s stopband and output response
of the neural network model are shown in Fig. 6. Fig. 8. Harmonic responses of target filter and neural network model – 50 Hz

Fig. 6. Input signal with frequency 60 Hz and output

Fig. 9. Harmonic responses of target filter and neural network model – 1 kHz
response of neural network model
 ONE APPROACH FOR TRAINING OF RECURRENT NEURAL NETWORK MODEL 223

Case 2. Modeling of partial response IIR digital filter MAGNITUDE RESPONSE OF THE TARGET FILTER

The class 1 partial response IIR digital filter is the special

case of the IIR Nyquist filter [17]. 1
Nyquist recursive digital filter plays an important role in
digital data transmission for its intersymbol interference (ISI)- 0.8
free property. Also it can be adopted in decimation or

MAGNITUDE
interpolation multirate systems. To achieve zero ISI, Nyquist
0.6
filter must satisfies some criteria in time domain that they
should have zeros equally spaced in the impulse response
coefficients except one specified. 0.4

Impulse response h( n ) of the (IIR) Nyquist filter with the

0.2
time domain constraints is defined in the form [17]:
­1
° ≠ 0 , if k = 0 0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
h( K + kN ) = ® N (14) NORMALIZED FREQUENCY

°̄0 , otherwise
Fig 11. Magnitude response of the target IIR partial response filter
where K and N are integers
The transfer function H ( z ) of the (IIR) Nyquist filter can
be expressed as:
Nn
¦ bi z − i
−K i = 0 , i ≠ kN = K 1
H ( z ) = bK z + Nd / N
, bK = (15)
−i N N
¦ ai N z
i =0
where N n , N d are integers, all the filter coefficients ai , bi are
real, a0 = 1 , N d is the multiple of N.
The specifications for the IIR partial response filter are the
following: N n = 10, N d = 2 , N = 2 , K = 6 [17].
The impulse response of the partial response filter, shown
in Fig. 10, is used as target function in the training procedure
of the neural network model. The magnitude response of the
partial response IIR digital filter has a cosine shape which is
illustrated in Fig. 11.
The impulse response and magnitude response of the
Fig 12. Impulse response of the neural network model of
neural network model with 6 neurons of IIR partial response IIR partial response filter
filter are shown in Fig.12 and Fig. 13 respectively.
MAGNITUDE RESPONSE OF THE NEURAL NETWORK MODEL

0.8
MAGNITUDE

0.6

0.4

0.2

0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
NORMALIZED FREQUENCY

Fig 10. Impulse response of the target IIR partial response filter
Fig.13. Magnitude response of the neural network model
of IIR partial response filter
224 STEFANOVA

VI. CONCLUSION [7] Zeng Zhe-zhao; Wen Hui, “Optimal design study of three-type FIR
high-order digital filters based on sine basis functions neural-network
algorithm”, IEEE International Symposium on Communications and
One approach for the IIR digital filter modeling based on Information Technology, Vol. 2, No 12-14 ,pp. 921- 924, October 2005.
recurrent neural network is proposed. The sensitivity [8] X. H. Wang, Y. G. He, H. Li, Y. L. Peng, M.J. Li, “Design of 2-D
coefficients method has been used to develop an effective digital filters using back propagation neural networks”, Proceedings of
International Conference on Machine Learning and Cybernetic, 18-21
training algorithm for the neural network model. The training Aug. 2005, Vol. 8, pp 4684 – 4689.
process also can be on-line and has a real time realization. [9] Wang D. and A. Zilouchian, Intelligent control systems using soft
The partial response IIR and bandpass IIR digital filters have computing methodologies, Editors: Ali Zilouchian, Mo Jamshidi,
been modeling using the recurrent neural network structures. Prentice Hall, March 2001, pp, 93-110.
[10] Chow, T.W.S. Siu-Yeung Cho “An accelerated recurrent network
To investigate the behavior of the digital filter models some training algorithm using IIR filter model and recursive least squares
simulations have been realized using input harmonic method” IEEE Transactions on Circuits and Systems I: Fundamental
excitations with different frequencies. The analysis of the Theory and Applications, Vol. 44, No 11, Nov, 1997, pp. 1082-1086.
obtained results shows a very good approximation of the [11] Bhattacharya, D.; Antoniou, A., “Design of IIR filters with arbitrary
amplitude and phase responses by feedback neural networks”, IEEE
impulse responses in time domain and the magnitude International Symposium on Circuits and Systems, ISCAS ’96, Vol. 3,
responses in the frequency domain. This approach can be 12-15 May 1996, pp. 457 – 460.
extended and successfully applied to the case of modeling of [12] Mladenov, V.M., Mastorakis, N.E., “Design of two-dimensional
nonlinear recursive digital filters. recursive filters by using neural networks”, IEEE Transactions on
Neural Networks, Vol. 12, No 3, May 2001, pp. 585 – 590.
REFERENCES [13] W. T. Miller, R. S. Sutton and P. J. Werbos, Neural Networks for
Control., Cambridge, MA:MIT Press, 1990.
[1] Jin L., P. Nikiforuk, and M. Gupta, “Approximation of discrete-time [14] R. Grino, G. Cembrano and C. Torras, “Nonlinear system identification
state-space trajectories using dynamical recurrent neural networks”, using additive dynamic neural networks – two on-line approaches”,
IEEE Transactions on Automatic Control, Vol.40, No 7, 1995. IEEE Trans. on Circuits and Systems I, vol. 47, no.2, 2000, pp. 150-
[2] H. Cruse , Neural Networks as Cybernetic Systems, Brains, Minds and 165.
Media, Bielefeld, Germany, October 2006. [15] Y. Fang, M. Yagoub, F. Wang and Qi-J. Zhang, “A new macromodeling
[3] M. Pedersen, “Optimization of recurrent neural networks for time approach for nonlinear microwave circuits based on recurrent neural
domain series modeling”, Ph.D Thesis, Department of Mathematical networks”, IEEE Trans. on Microwave Theory and Techniques, Vol.
Modeling, Technical University of Denmark, 1997. 48, no. 2, Dec. 2000, pp. 2335-2344
[4] J. Cao, Yahagi T., “Nonlinear adaptive digital filters using parallel [16] Stefanova S., “One Dimensional IIR digital filter modeling based on
neural networks”, Proceedings of IEEE International Conference on recurrent neural network”, International Joint Conferences on
Neural Networks, Vol. 2, Nov-Dec 1995 pp. 850 – 853. Computer, Information, and Systems Sciences, and Engineering
[5] D. Bhattacharya, and A. Antoniou, Real-time design of FIR filter by (CISSE 08), University of Bridgeport, Bridgeport, U.S.A, December 5 -
feedback neural network, IEEE Signal Processing Letters, Vol.3, No 4, 13, 2008.
May 1996, pp. 158 – 161. [17] S. Pei, Ch. Hsu, P. Wang, “Design of a class of IIR eigenfilters with
[6] Zhe-Zhao Zeng; Ye Chen; Yao-Nan Wang, ”Optimal Design Study of time and frequency domain constraint, IEEE Trans. on Circuits and
High-Order FIR Digital Filters Based on Neural-Network Algorithm”, Systems-II Analog and Digital Signal Processing, vol. 49, No 2,
International Conference on Machine Learning and Cybernetics, 13-16 February 2002, pp 145 - 151.
Aug. 2006, pp 3157 – 3161.
 Classification of the Power Transformers using
Dissolved Gas Analysis
T. C. B. N. Assunção, J. T. Assunção and A. D. Neto
Department of Electric Engineering, Federal University of São João del-Rei (UFSJ), Brazil
(e-mail: [email protected], [email protected], [email protected]).

To facilitate the diagnosis of incipient faults by DGA

Abstract— Least squares support vector machines (LS-SVM), several methods of interpretation have been developed, as
artificial neural networks (ANN) and the traditional methods based in Rogers, Doernenburg, Key Gas, Duval Triangle, IEEE
the dissolved gas analysis were employed for the detection of C57.104-1991, and IEC 60599 [1]. These methods are used
incipient faults in power transformers. The diagnosis criteria’s for to interpret the concentration of gases and/or the relation
commonly used for dissolved gas analysis (DGA) in transformer
between them, which are compared to thresholds to conclude
insulating oil are Doernenburg Ratio Method, Rogers Ratio Method,
about the state of transformer insulation. The gases analyzed
IEC 60599, IEEE C57.104-1991 and Duval Triangle. In this
paper, a comparative study using ANN, LS-SVM and DGA for the
are: Hydrogen (H2); Methane (CH4); Ethane (C2H6); Ethylene
incipient fault detection in power transformers is presented. The (C2H4); Acetylene (C2H2); Carbon Monoxide (CO) and
artificial neural network and least squares support vector machines Carbon Dioxide (CO2).
approaches results are compared with the DGA results obtained The interpretation of the DGA is made by ratio methods
from analysis of the traditional methods already consolidated in that are coding systems that assign a certain combination of
the technical literature. codes to a fault specific type. The codes are generated by
calculating gas ratios and comparing the ratios to pre-defined
I. INTRODUCTION ratio intervals. A fault condition is detected when a code
combination fits the code pattern of the fault.

P ower transformers are high cost important equipment

used in the transmission and distribution of the electric
energy. Its right performance is important for the electric
systems operation, because the loss of a critical unit can
The entire phenomenon related at gases and their
correlations with incipient faults have been characterized for
imprecision, in doubt measurement and unmodelled non-
linearity. So, there is necessity of an expertise to interpret the
generate great impact in safety, reliability and in the cost of difficult or inconclusive DGA tests results.
the electric energy supply, mainly under an ever-increasing In order to overcome this difficulty in the interpretation of
competitive environment. The power transformers due to tests results, there has been a substantial effort in the
natural ageing, loading regime or electrical and mechanic developing intelligent diagnostic in this area. This intelligent
efforts, operate under harder conditions than they were diagnostic is a combination of the artificial neural networks
installed. (ANN) and DGA samples for an incipient faults efficient
System power reliability improves when incipient faults in detection in power transformers.
power transformers are identified and eliminated before they For this purpose ANN can be used, due to its learning
deteriorate to a critical operation condition. capacity in the modeling complex and nonlinear relations [2].
Incipient transformer faults lead to electrical and thermal ANN is submitted to a training process from real cases, and
stresses on insulating materials. As a result of these stresses then handling appropriately new supplied data. Several ANN
the insulating materials can breakdown and several gases are structures have been proposed by researchers that can be
released on oil insulating. The analysis of these gases provides classified as static (SNN), dynamic temporal processing
useful information about the fault conditions and the type of (TPNN), recurrent (RNN) and radial basis function (RBFN).
materials involved. The most well-know ANN configuration is the multi-layer
There are several techniques used in the detection those feedforward network that have been applied successfully to
gases, but the dissolved gas analysis (DGA) was recognized solve some difficult and assorted problems including
as the most informative method. So, the DGA is the well- nonlinear system identification and control, financial market
known diagnostic technique in the world because it’s analysis, signal modeling, power load forecasting etc. The
sampling and analyzing procedures are simple, inexpensive fast learning procedure and its great generalization capability
and easy to be standardized. have promoted the use of the ANN in the areas of non-linear
The analysis of specific types of gases in the oil-filled identification, approximation and interpolation theory.
transformers is useful for early failure detection. The most Recently, the Support Vector Machine (SVM) has been
important aspect of fault analysis is taking data that has been proposed as a new and promising technique for classification
generated and correctly diagnosing the fault that is generating and regression of the linear and nonlinear systems [3].
of gases. The LS-SVM is a learning machine proposed in [3]
corresponding a modified version of the SVM. Like the SVM,
LS-SVM can be used in classification problems and

225
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_39, © Springer Science+Business Media B.V. 2010
226 ASSUNÇÃO ET AL.

approximation functions. The main characteristic of LS-SVM AR = Arcing; CT = Sparking; DPCO = Partial Discharge with
is smallest computational cost in relation to the ANN, without Tracking (note CO).
loss in the quality of the solutions. TABLE II
FAULTS DESCRIPTION TABLE – ROGERS METHOD.
In this paper, a comparative study of ANN, LS-SVM and
DGA for the calculation of efficiency for the incipient fault
CH4/ C2H2/ C2H4/ C2H2/
detection is presented. The results of this study are useful in Diagnostic
H2 CH4 C2H6 C2H4
the development of a reliable transformer diagnostic system
0 0 0 0 N
using intelligent systems and DGA.
5 0 0 0 DP
1 or 2 0 0 0 OH1
II. INCIPIENTS FAULTS DIAGNOSTIC
1 or 2 1 0 0 OH2
0 1 0 0 OH3
The power transformers in operation are subject to failures
0 1 0 0 OHC
due to electric and thermal stresses and the stresses due to
1 0 1 0 CCE
environmental influences. The effect of any stress results in
1 0 2 0 CCTC
the ageing of insulation, it leads up to decomposition of the
0 0 0 1 FL
insulating oil and paper, with the generation of dissolved
0 0 1 or 2 1 or 2 ARC
gases in the insulation oil.
0 0 2 2 CT
The oil dissolved gases can signify a possible failure in the
5 0 1 or 2 1 or 2 DPCO
power transformers [1- 2].
The gases more generated inside power transformers during
operation normal or during the degradation of isolation system
are: Hydrogen (H2); Methane (CH4); Ethane (C2H6); Ethylene Doernenburg Method is also based in ratio between gases,
(C2H4); Acetylene (C2H2); Carbon Monoxide (CO) and but, unlikely Roger’s Method, its ratio between gases is not
Carbon Dioxide (CO2); and they can indicate: codified. In this method, the relative value of gases is
associated with the fault conditions, in accordance with the
• Hydrogen (H2): Corona;
Table III. In the Doernenburg Method is not considered the
• Hydrogen (H2); Methane (CH4); Ethane (C2H6);
normal state, but there are restrictive conditions for
Ethylene (C2H4): Oil Thermal Decomposition;
application of this method, as showed in the Table IV. If at
• Carbon Monoxide (CO) and Carbon Dioxide (CO2: least one of the gases in each ratio between gases exceeds the
Paper Thermal Decomposition; limit L1, the ratio key gases procedure is valid, otherwise, if it
• Acetylene (C2H2): Electric Arc. is not significant d the unit should be resample and
investigated by alternative procedures.
In this paper were studied the methods more used for the
dissolved gas analysis, as follows [3-5]. TABLE III
RATIO KEY GASES – DOERNENBURG METHOD.
Rogers Method uses the following relations of gases:
CH4/H2 ; C2H2/CH4; C2H4/C2H6; C2H2/C2H4. In the Tables I Range Gases Ratio Range Gases Ratio
and II are presented the relations between gases and the Fault CH4/ C2H2/ C2H6/ C2H2/
codification related to fault type of the Rogers Method. H2 C2H4 C2H2 CH4
TABLE I
TD >1 < 0,75 > 0,4 < 0,3
RATIO KEY GASES – ROGERS METHOD DPL < 0,1 NS > 0,4 < 0,3
DPH < 1 to
Gases Ratio Range Gases Ratio Code > 0,75 < 0,4 > 0,3
> 0,1
< 0.1 5
0.1 to1.0 0 where: NS = Not Significant; TD = Thermal Decomposition;
CH4/H2
1.0 to 3.0 1
> 3.0 2
DPL = Partial Discharge (Low-Intensity); DPH = Partial
< 1.0 0 Discharge (High-Intensity).
C2H2/CH4
> 1.0 1
< 1.0 0 TABLE IV
C2H4/C2H6 1.0 to 3.0 1 LIMIT CONCENTRATIONS OF DISSOLVED GAS [PPM].
> 3.0 2
< 0.5 0 Gas H2 CH4 CO C2H2 C2H4 C2H6
C2H2/C2H4 0.5 to 3.0 1 L1 100 120 350 35 50 65
> 3.0 2

The concentration in “parts per million” (ppm) of CH4,

In the table II, N = Normal State; DP = Discharge Partial; C2H4 and C2H2 in the Duval Triangle Method are expressed as
OH1 = Overheating Temperature < 150º C; OH2 = Overheating percentages of the total (CH4 + C2H4 + C2H2) and plotted as a
Temperature between 150 and 200º C; OH3 = Overheating point (% CH4, % C2H4, % C2H2) in a triangular coordinate
Temperature between 200 and 300º C; OHC = Overheating system chart (Figure 1), which has been subdivided into fault
Conductor; CCW = Winding Circulating Current; CCTC = zones. The fault zone in which the point is located designates
Tank and Core Circulating Current; FL = Flashover;
 CLASSIFICATION OF THE POWER TRANSFORMERS USING DISSOLVED GAS ANALYSIS 227

the likely fault type produced by combination of gas prevention or after the occurrence of these faults, but they will
concentration. have a little sensibility in relation to incipient faults. In spite
In the Duval Triangle (Figure 1), PD = Discharge Partial; of, DGA conventional methods can be used as standards for
T1 = Thermal Fault, Temperature < 300º C; T2 = Thermal the intelligent systems based in the artificial intelligence [2, 8-
Fault, 300 º C < Temperature < 700º C; T3 = Temperature 12]. The intelligent systems applications for the incipient
>700º C; D1 = Discharges of Low Energy; D2 = Discharges of fault diagnostic in transformers are useful, because the
High Energy; DT = Mixture of Electrical and Thermal Faults. intelligent systems are adaptative, capable of handling highly
non-liner relationship and also generalize solutions for a new
solution.

III. 1. ARTIFICIAL NEURAL NETWORK

ANN has been established as a useful tool for classification

and regression problems, mainly for pattern recognitions and
function approximations. An important characteristic of the
ANN is that is not necessary to obtain a complete knowledge
about the relations among the variables involved in the
problem.

The main characteristics of the ANN are [6]:

• The basic information-processing unit of the ANN
is called neuron;
• Due to complexity of the problems (classification or
Figure 1. Duval Triangle Method.
regression) the neurons are organized in three or
more layers: the input layer, the output layer and
The fault diagnostic methods of Rogers, Doernenburg and
one or several hidden layers;
Duval Triangle as presented previously, are based in the ratio
gases. These methods are the basis of the IEEE, IEC e ABNT • Each neuron receives input signals, which are
Methods [9], as showed in the Tables V and VI. multiplied by synaptic weights;
• Input signals are transformed into signal output to
TABLE V the next neuron and so on, by an activation
RATIO KEY GASES - IEEE, IEC AND ABNT METHODS. function;
Rang Code • The synaptic weights in the network are adjusted
e CH4 C2H2/C2H4 C2H4/C2H6 according to external signals received by neural
< 0.1 0 1 0 network training process.
0.1-1 1 0 0
1-3 1 2 1 III.2. LEAST SQUARES SUPPORT VECTOR MACHINES
>3 2 2 2
Least Squares Support Vector Machines (LS-SVM) is a
TABLE VI method used for solving non-linear classification or modeling
FAULTS DESCRIPTION TABLE - IEEE, IEC AND ABNT METHODS. problems and has been applied to classification, function
Diagnosti Code estimation and nonlinear system optimal control problems.
c CH4 C2H2/C2H4 C2H4/C2H6 The basis of the method is the mapping of all available data
N 0 0 0 points to a feature space, thus transforming the problem into a
DPL 0 1 0 simple linear problem. LS-SVM is an extension of the
DPH 1 1 0 standard SVM and it expresses the training in terms of solving
D1 1 0 0 a linear set of equations instead of a quadratic programming
D2 1 0 0 problem, which allows LS-SVM to be faster than SVM and
T1 0 1 0
provides the reduced computation time. Efficient and scalable
T4 0 1 0
T2 0 0 1
algorithms, such as those based on conjugate gradient can be
T3 1 0 1 applied to solve LS-SVM. Extensive studies have shown that
LS-SVM is comparable to SVM in terms of generalization
where: DPL = Partial Discharge (Low Energy); DPH = Partial performance.
Discharge (High Energy); T4 = Thermal Fault, 150 º C <
Temperature < 300º C. IV. METHODOLOGY

III. INTELLIGENT SYSTEMS FOR FAULTS In these paper is proposed a tool for power transformers
DIAGNOSING classification in relation to their failure condition, using
dissolved gas analysis.
The conventional methods of interpretation of the DGA are DGA methods, ANN and LS-SVM approaches were
easy implemented and present good results for the hard faults studied and implemented for verification of operation
228 ASSUNÇÃO ET AL.

condition of power transformer. The data of this study were Normal State; OH = Overheating; FE = Electric Failure; LS =
obtained from electricity utility, and each method was tested LS-SVM; NI = Fault is Unidentifiable; R = Rogers Method;
with all the data set. DV = Duval Method; DB = Doernenburg; Method; NBR =
NBR 7274 Method; IEC = IEC Standard 6059 Method.
IV. 1. ANN IMPLEMENTATION
TABLE VII
ANN used is a multi-layer perceptron (MLP) with 5 CONCENTRATION OF GASES IN PPM.
neurons in the input layer, 12 neurons in the intermediate
layer and 3 neurons in the output layer. The activation Concentration of Gases [ppm]
TR
functions used in each of the layers were, respectively, tangent H2 CH6 C2H4 C2H2 C2H6
sigmoid function, tangent sigmoid function and logarithmic 1 40 62 11 1 51
sigmoid function. MLP ANN doesn’t presents binary output, 2 600 1800 3800 130 520
so it was used the winner neuron strategy to make the 3 6 2 7 0,4 5
classification. As the MLP used is composed of 3 output 4 1230 163 233 692 27
neurons, then it was determined that: 1. the output pattern of 5 630 670 1100 1700 81
the neuron that to present higher value is defined as 1; 2. the 6 20 3 5 0,4 2
output pattern of the remaining neurons is defined as 0. 7 24 3 8 0,4 4
Because designing an innovative ANN learning procedure is 8 30 520 1000 0,4 310
beyond the scope of this paper, the ANN used in this paper 9 645 86 110 317 13
was implemented using Neural Toolbox MATLAB routines 10 110 860 1500 1800 300
[13]. 11 19 9 7 0,4 20
12 16 1 1 0,4 1
IV. 2. LS-SVM IMPLEMENTATION 13 1600 55000 74000 9300 42000
14 95 10 11 39 0,4
Classification with the LS-SVM approach was made with 15 200 110 150 230 12
two networks. The first network was used to determine the 16 13 10 55 0,4 3
transformer actual operation condition (normality or fault) and 17 48 12 18 0,4 14
the second network was used to determine the fault type 18 44 95 13 0,4 33
(heating or electric failure). Because designing an innovative 19 595 80 89 244 9
LS-SVM learning procedure is beyond the scope of this paper, 20 94 5 25 20 4
the LS-SVM network used in this paper was implemented 21 290 16 12 0,4 24
using the LS-SVMlab Toolbox MATLAB routines [14]. 22 56 6 7 10 5
23 420 1400 1500 7 640
V. PRESENTATION AND ANALYSIS OF RESULTS 24 1790 580 336 919 321
25 0 1200 1900 4400 130
In order to training of the ANN and LS-SVM were used in 26 14 3 12 0,4 3
this study 246 samples of DGA and were selected and used 27 8800 64064 9565 0,4 72128
data of 30 transformers for ANN and LS-SVM testing. The 28 34 41 5 0,4 69
same data were used for the comparison between the results of 29 1330 10 66 182 20
the traditional methods, ANN and LS-SVM. In this study 30 230 120 220 740 14
were used five key gases, all of them combustible, as inputs
features of networks: H2, CH4, C2H2, C2H4 e C2H6.
Overheating, electric failure or normal are the output features. TABLE VIII
DIAGNOSIS ACCURACY RESULTS.
Then, input and output patterns for each diagnosis criterion
are defined according each method (Doernenburg Ratio
Method Diagnosis Accuracy [%]
Method, Rogers Ratio Method, IEC 60599, IEEE C57.104-
Rogers 33.33
1991 and Duval Triangle. The concentration of gases data in
Duval 40.00
ppm are shown in the Table VII.
Dornenbur 50.00
Table VIII shows the accuracy results for the incipient fault
g
detection of the DGA conventional methods, ANN and LS-
NBR 7274 50.00
SVM. It is demonstrated in the Table VIII the superior
IEC 60599 50.00
efficiency of the ANN methods over the other methods.
LS-SVM 70.00
Rogers’s method presented 33.33% efficiency, Dorneburg’s
MLP 83.33
method presented 50%, NBR7274 presented 50%, ANN
presented 83.33% and LS-SVM presented 70%. These results
VI. CONCLUSION
demonstrate that intelligent systems (ANN and LS-SVM) can
be a promising tool in the detection of incipient faults in
power transformers. A comparative study was made employing intelligent
The resulting tests employing the DGA conventional systems (ANN and LS-SVM) and DGA conventional methods
methods, ANN and LS-SVM approaches, are shown in the (Doernenburg Ratio Method, Rogers Ratio Method, IEC
Table IX, where: TR = Transformer; D = Diagnostic; N = 60599, IEEE C57.104-1991 and Duval Triangle) for the
 CLASSIFICATION OF THE POWER TRANSFORMERS USING DISSOLVED GAS ANALYSIS 229

efficiency calculation for incipient fault detection on power ANN due to its generalization characteristic is an adequate
transformers. The MLP and LS-SVM were trained with the tool for to solve the conflicts between methods. The results
gases relation of the DGA conventional methods cited presented show the actual capacity of the ANN for the
previously. identification of incipient fault in power transformers and also
to solve the standardized inconsistencies of the methods. In
TABLE IX spite of the superior results obtained with MPL, it was verified
RESULTS SET DISPOSED BY DGA CONVENTIONAL METHODS,
that with the default parameters of the respective algorithms
ANN AND LS-SVM.
the implementation of the LS-VM model is easier than the
MLP model. MLP network involves more experience for its
METHODS modeling and training, mainly for the definition of the number
TR D
R DV DB NBR IEC MLP LS of hidden layers. Therefore, LS-SVM would be studied as an
1 N SA NI N OH OH OH OH important alternative to ANN and DGA conventional methods
2 N NI NI OH OH OH OH FE for the identification of incipient fault in power transformers.
3 OH NI FE N OH OH N N We intend to continue the studies on the application of LS-
4 FE FE FE NI FE FE FE FE SVM in modeling and simulation for the identification of
5 FE NI FE NI NI NI FE FE incipient fault in power transformers, because it showed to be
6 N OH FE N OH OH N N a useful tool in this study.
7 OH NI FE N OH OH N N
8 FE NI FE OH OH OH OH FE ACKNOWLEDGMENT
9 FE FE FE NI FE FE FE FE
10 N NI FE NI NI NI FE FE The authors acknowledge CNPQ – Brazil, for the financial
11 N OH FE N N N N N support.
12 OH NI FE N NI NI N N
13 FE NI FE OH NI NI OH FE REFERENCES
14 FE FE FE N FE FE FE FE [1] Assunção, T. C. B. N. Contribuição à Modelagem e Análise do
15 N FE FE NI FE FE FE FE Envelhecimento de Transformadores de Potência, Tese de Doutorado,
Universidade Federal de Minas Gerais, UFMG, 2007.
16 N NI FE N NI NI N N [2] Singh, J., Sood, Y. R., Jarial, R. K., Verma, P., Condition Monitoring of
17 OH NI FE N OH OH N N Power Transformers - Bibliography Survey, IEEE Electrical Insulation
18 FE OH FE N OH OH N OH Magazine, vol. 24, n° 3, May/June 2008, pag. 11-25.
19 FE FE FE NI FE FE FE FE [3] IEEE Guide for the Interpretation of Gases Generated in Oil-Immersed
Transformers, C57.104-1991, 1992.
20 N NI FE N NI NI FE FE [4] Mineral Oil-Impregnated Electrical Equipment in Service, Guide to the
21 N NI FE N FE FE N FE Interpretation of Dissolved and Free Gases Analysis, IEC Standard, 1999.
22 OH FE FE N FE FE FE FE [5] Interpretação da Análise dos Gases de Transformadores em Serviço, NBR
23 FE NI FE OH OH OH OH FE 7274, ABNT, 1982.
[6] Haykin, S. ; Redes Neurais Princípios e prática, 2ª edição, Prentice Hall
24 FE FE FE NI FE FE FE FE INC., ARTMED Editora S. A., Bookman, 1999.
25 N FE FE NI FE FE FE FE [7] Suykens, J. A. K., L., L., Doren, P. V. e Moorand, B., V. J., Least squares
26 OH NI FE N NI NI N N support vector machines classifiers: A large scale algorithm·, European
27 OH NI FE OH OH OH OH FE Conference on Circuits Theory and Design, pp. 839-842.
[8] Moreira, M. P., Santos, L. T. B., Vellasco, M. M. B. R., Power
28 OH OH FE N OH OH OH OH Transformers Diagnosis Using Neural Networks, Proceedings of International
29 FE NI FE FE NI NI FE FE Joint Conference on Neural Networks, Orlando, Florida, August 2007..
30 FE FE FE NI FE FE FE FE [9] Guardado, J. L., Naredo, J. L., Moreno, P., Fuerte, C. R., A Comparative
Study of Neural Network Effciency in Power Transformers Diagnosis Using
Dissolved Gas Analysis, IEEE Transactions on Power Delivery, vol. 16, nº 4,
The main problem of the DGA conventional methods is that October 2001, pag. 643-647.
different methods applied to the same sample result in [10] Yanming, T., Zheng, Q., DGA Based Insulation Diagnosis of Power
different and often in contrary diagnostic decisions. An Transformers Via ANN, Proceedings of The 6 th International Conference on
adequate utilization of the DGA conventional methods is Properties and Applications of Dielectric Materials, June 2000, Xi'an Jiatong
University, China.
important; because most of expertise’s in incipient fault [11] Farag, A. S., Mohandes, M., Al-Shaikh, A., Diagnosing Failed
diagnostics use these methods for elaboration of expertness of Distribution Transformers Using Neural Networks, IEEE Transactions on
power transformers operation condition. The diagnostic made Power Delivery, vol. 16, n° 4, October 2001, pag. 631-636.
using DGA conventional methods often are based in the [12] Salami, A., Pahlevani, P., Neural Network Approach for Fault Diagnosis
Transformers, 2008 International Conference on Condition Monitoring and
empiric knowledge of the expertises, and then it is necessary Diagnosis, Beijing, China, April 2008.
to mind mainly with the inconsistency (non-decision [13] Demuth, M. B. H. e Hagan, M., Neural network toolbox, for use with
problems) and the conflicts between the methods. MATLAB, The Math Works;
[14] Pelckmans, K.; Suykens, J. A. K, LS-SVMlab toolbox, 2003, Version 1.5.
 Modelling Control of Pore Number and Radii
Distribution in Single-Cell Electroporation
Sadhana Talele Department of Engineering
University of Waikato
Hamilton, New Zealand
Email: [email protected]
Paul Gaynor Department of Electrical and Computer Engineering
University of Canterbury
Christchurch, New Zealand

Abstract—Electroporation EP, in which external electric field on the type of application of EP it would be good to know the
pulses create transient pores in a cell membrane, is an important starting parameters of electric field to be applied in order to
technique for delivery of genes and drugs into the cell. To enable achieve a good number of pores in a particular range of size,
a useful level of entry of genes into cells, the pores should have
sufficiently large radii, and remain open long enough without causing as pores that are too small are not of any therapeutic use and
membrane rupture. A numerical model for a single spherical cell reduce the strength of the cell membrane, while pores that are
electroporated by application of direct and/or alternating external too big create a possibility of them becoming permanent leading
electric field pulses has been developed. The model is used to indicate to cell lysis. To date, such controllability of pore number and size
the actual number of pores and their radii distribution developed has not been reported in the literature.
in response to various electric field pulses, as function of time and
position on the cell surface. This study briefly describes the model Previous reported studies have modelled EP to calculate Vm
briefly which is then used to investigate the ability to control the and the pore density N [10], [11]. These models consider the non-
number and distribution of pore radii by choice of electric field linear behavior of EP, which is due to the dynamics of membrane
parameters. We believe this would be one of the first papers to pore formation and its effect on the electrical properties of the
investigate the ability to CONTROL the number and distribution membrane in turn. However, these models do not include the
(range) of pore radii (as opposed to other papers that merely report
the pore number and range with varying pulse parameters. evolution of pore sizes. They assume a constant pore radius.
Recent models [12], [13] include spatial and temporal aspects of
Index Terms—electroporation, numerical model, fractional pore
pore radius evolution. However, the form of electric fields used
area, pore radii
in these models were single unipolar pulses. The study presented
here develops a model of single spherical cell EP and provides
I. I NTRODUCTION simulation of spatial and temporal aspects of pore radius as an
Exposure of biological cells to electric fields can lead to effect of multiple pulses of varied magnitudes. As such it is
a variety of biophysical and biochemical responses [1]–[3]. possible to investigate potential control of both the number of
Electroporation (EP) is a feature common to all lipid bilayers pores created and their radii distribution.
in which electrically conductive pores are induced to form in
the membrane, when the transmembrane potential Vm exceeds a II. M ODEL OF A SINGLE CELL
semi-critical value creating a state of high permeability [4], [5]. Consider a spherical cell of radius a with intracellular conduc-
These pores shunt excess stimulus current across the membrane, tivity σin . The cell is immersed in a medium with conductivity
limiting the growth of Vm . The pores are long lived, sometimes σex . This system is exposed to a time-varying electric field E(t).
surviving in the membrane for up to several minutes, and provid- Azimuthal symmetry about the axis of applied electric field is
ing pathways for the movement of large ions and high molecular assumed. The model discussed in [14] is described below briefly
weight molecules such as drugs and DN A into the cell [6]–[8]. for completeness.
These properties have resulted the application called EP which is
now a common tool in biotechnology and is used in a number of
medical treatments [1]–[3], [6], [9]. However, the process of EP is A. Transmembrane potential
still not completely understood. There is a need for supplementing It is assumed that the intracellular and extracellular regions are
experimental knowledge with theoretical models [10]. Depending charge free. The homogeneous external electric field of strength

231
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_40, © Springer Science+Business Media B.V. 2010
232 TALELE AND GAYNOR

E(t) is used as a boundary condition, the potential being fixed pore, n is the relative entrance length of the pore and vm is the
at nondimensional transmembrane potential [10] given by
Φ(3a, θ) = −3aE(t) cos θ, (1)  
F
on a sphere of radius 3a surrounding the cell, where θ is the polar vm = Vm . (10)
RT
angle. Since there are no sources inside the region, the potential
obeys Laplace’s equation This equation for pore current isml (equation 8) accounts for the
∇2 Φ = 0, (2) electrical interactions between the ions and the pore wall [10].
Now, assume Q larger pores exist, and ilge is the current through
except at the cell membrane at a radial distance of (r = a) where the electropores of radius larger than 1 nm. Then, Jlge is the total
the potential is discontinuous because of the abrupt change in current density through Q larger pores; rq being the radius of the
conductivity and the approximated infinitely thin membrane. Thus q th pore. Hence,
internal and external potentials can be defined as,
1 
Q

Φin (r, θ) r < a, Jlge = ilge (rq ) (11)
Φ(r, θ) = (3) A q=1
Φex (r, θ) a < r < 3a.
The current across the membrane is used to relate the internal where A is the corresponding cell surface area. For these larger
and external potentials [10], that is, pores, the current–voltage relationship assumes that the trans-
∂Vm membrane potential Vm occurs across the sum of pore resistance
−r̂ · (σin ∇Φin (a, θ)) = −r̂ · (σex ∇Φex (a, θ)) = Cm + Jm
∂t Rp and the series input resistance Rin [13], [15], as follows:
(4)
where Vm
Vm (θ) = Φin (a, θ) − Φex (a, θ). (5) ilge (r) = , (12)
Rp + Rin
Here r̂ is the unit outward radial vector, Cm is the specific
membrane capacitance, Vm is the transmembrane potential and where
Jm is the current density at the cell membrane due to existing h
Rp = , (13)
pores. The current density is made up of three terms as given in πσps rq2
Equation 6
Jm = Jion + Jsml + Jlge , (6) and
1
where Jion = gm (Vm − Vrest ) is the ionic current density [10] (gm Rin = . (14)
2σps rq
is the specific membrane conductance, and Vrest is the membrane
rest potential). The remaining two terms are explained in the
following paragraphs. B. Formation of pores:
Jsml is the current density through small pores and is given by
Initially pores are assumed to be formed with the minimum-
Jsml = N isml (r), (7)
energy radius rm = 0.76 nm, at a rate given by [10]
where N is the pore density of the initial small pores formed and  
isml (r) is the diffusion current through a single pore of radius r dN 2 N
= ψe(Vm /Vep ) 1 − , (15)
(true for small pores only). A previously derived expression for dt Neq (Vm )
isml [10], based upon the Nernst-Planck equation models, is used
for pore radius below 1 nm, and is, where N is the pore density of the initial small pores formed,
2
ψ is the creation rate coefficient and Neq is the equilibrium pore
πr σps νm RT (e − 1) νm
isml = · (8) density for a voltage Vm given by
Fh (G− eνm − G+ )
2
with Neq (Vm ) = N0 eb(Vm /Vep ) . (16)
w0 ew0 ±nνm ± nνm
G± = . (9)
w0 ± nνm Here, N0 is the initial pore density with no applied electric field,
Here σps is the conductivity of the aqueous solution that fills the Vep is the characteristic voltage of electroporation and b is the pore
√
pore (approximated by σin σex ), F is Faraday’s constant, R is creation constant equal to (rm /r∗ )2 , where r∗ is the minimum
the universal gas constant, T is the absolute temperature, h is the radius of hydrophilic pores and rm the minimum-energy radius
thickness of the membrane, w0 is the energy barrier inside the [16]. All parameter values are as given in Table 1.
 MODELLING CONTROL OF PORE NUMBER AND RADII DISTRIBUTION 233

a 15.0 (μm) cell radius

C. Evolution of pore radii: Cm 10−2 (Fm−2 ) specific membrane capacitance [12]
h 5.0 (nm) membrane thickness [?], [5], [10]
The pores that are initially created with minimum-energy
gm 1.9 (Sm−2 ) specific membrane conductance [10]
radius rm change in size. Evolution of the radius of a pore is Vrest –80 (mV) membrane rest potential [10]
governed by the following equations 17 - 21 [13]. The lipid σin 0.3 (S m−1 ) intracellular conductivity [5]
σex 1.2 (Sm−1 ) extracellular conductivity [?]
bilayer energy wm depends on a number of parameters and is r∗ 0.51 (nm) minimum radius of hydrophilic pores [13]
given by rm 0.8 (nm) minimum energy radius at Vm =0 [13]
T 295 (K) absolute room temperature [10]
Q »
X „ «4 Z rq
– n 0.15 relative entrance length of pores [10]
r∗ 2 b 2.46 pore creation constant [10]
wm = wst + 2πwed rq − πξeff (Ap )rq + F (rq , Vm )dr . (17)
q=1
rq 0 Vep 258 (mV) characteristic voltage of electroporation [10]
N0 1.5 × 109 (m−2 ) initial pore density [10]
The terms in this equation are explained in the following w0 2.65 energy barrier within pore [10]
paragraphs. ψ 1 × 109 (m−2 s−1 ) creation rate coefficient [10]
wst 1.4 × 10−19 (J) steric repulsion energy [13]
In the first term, wst is the steric repulsion energy [16]. The wed 1.8 × 10−11 (J m−1 ) edge energy [13]
lipid head groups which line the pore interior, tend to repel each ξ0 1 × 10−6 (J m−2 ) tension of the bilayer without pores [12]
other due to steric and/or electrostatic interactions [17]–[19] and ξ 2 × 10−2 (J m−2 ) tension of hydrocarbon-water interface [10]
Fmax 0.70 × 10−9 (N V−2 ) max electric force for Vm = 1 V [10]
are taken into account by this term in equation 17. rh 0.97 × 10−9 (m) constant [13]
The appearance of a circular pore in a membrane is balanced rt 0.31 × 10−9 (m) constant [13]
by the presence of two energy terms: reduction in energy barrier D 5 × 10−14 (m2 s−1 ) diffusion coefficient for pore radius [12]

proportional to removal of pore area πrq 2 (third term in equation TABLE I

17) and increase in energy barrier by a linear edge component G EOMETRIC , ELECTRIC AND ELECTROPORATION PARAMETERS USED IN
SIMULATION .
proportional to pore edge of length 2πrq [4], [20], [21] (second
term in equation 17). Here, wed is the pore edge energy, and ξeff
the effective tension of the membrane is given by
2ξ  − ξ0 The rate of change of pore radii is given by [13]
ξeff (Ap ) = 2ξ  −  2 (18)
A drq D ∂wm
1 − Ap =− , q = 1, 2, ..., Q (21)
dt kT ∂rq

where ξ is the energy per area of the hydrocarbon-water interface where D is the diffusion coefficient of the pore radius, k is
[16], [18], ξ0 is the tension of a membrane without pores and A Boltzman’s constant and T is the absolute temperature.
is the total area of the lipid bilayer. A varying value of the total
area Ap occupied by the pores at any given time is given by
III. N UMERICAL I MPLEMENTATION

Q
The model described above is implemented in MATLAB and
Ap = πrq2 (19) simulations carried out for a spherical cell with 15 μm radius. The
q=1
following description of model geometry is illustrated in Figure 1.
and contributes to a dynamic effect of pore formation both Azimuthal symmetry about the axis of the applied electric field is
temporally and spatially. assumed. The membrane is divided into a number of individually
The last term in Equation 17 is the contribution of the mem- modelled slices (25 including the two poles). Each section of
brane potential to the bilayer energy [16]. Assuming the inner cell membrane is assigned a pore density N , of ‘small pores’,
surface of a pore as toroidal [16], [22], the electric force F acting idealised as having equal radius rm . Each section also has a
on the pore is given by transmembrane voltage (Vm ) calculated. All results are simulated
Fmax for a radial discretization step of 3a/51.
F (r, Vm ) =   Vm2 . (20)
1+ rh
r+rt IV. S IMULATION AND R ESULTS
This equation is a heuristic approximation [16] of the numerical The number of pores and their radius is significant when
solution which has been computed for the electrical force acting administering an application. It is important to select appropriate
on a pore derived from first principles; Here, rh and rt are values of experimental electric field parameters in order that
constants taken from reference [16]. This equation is thought to expected results occur. For example, a sufficient number of larger
be appropriate for larger pores as it predicts that F approaches pores may be required if a particular gene transfer is expected.
a constant value Fmax as the pore radius increases, rather than The model discussed above can simulate results of EP, given
increase linearly [21], or decrease to zero [23]–[25] as radius any form of applied electric field and most other EP system
increases. All parameter values are as given in Table 1. parameters. Of importance in this study is the effect of applied
234 TALELE AND GAYNOR

electric field
As seen in Table 2, the first pulse was chosen to be of amplitude
θ12 = 90˚
r51
135 kV/m with a time duration of 3 μs, as this gave a reasonable
∆r FPA and a maximum pore radius of 10 nm. The aim is to increase
r50
the total FPA, as well as the contribution of FPA by pores in the
range greater than 5 nm, preferably larger than 10 nm. Several
amplitudes for the second pulse were tried (135kV, 175kV, and
rα + 1
195kV) with a break of 3 μs between two pulses.

···
rα θ2 = 15˚
rα – 1 As seen in Table 3, the second pulse of 195 kV/m, increases
θ1 = 7.5˚
the total FPA substantially to 0.05%. The total number of pores

···
r1
∆θ
θ24 = 180˚ θ0 = 0˚ remain nearly the same (a few new pores form as an effect of the
a second pulse), while increasing the number of pores in the range
5-10 nm from 660 to 862, pores in the range 10-20nm from 275
cell membrane
to 376 and pores in the range 20-100 nm from 0 to 14. As seen
2a
in Table 3, the FPA contribution due to large pores in the pore
radius range 5-100 nm has increased from about 18% of total to
about 27 % of total FPA.
In an attempt to increase this ratio, a higher amplitude second
pulse was tried (results not shown). This, instead of increasing
the size of pores, formed new smaller pores. Using a higher
magnitude first pulse hasve been seen to give a higher total FPA
Fig. 1. Spherical cell model geometry showing the angle and radius discretization but with more small pores (under 5 nm).
used in the numerical model
Another way of reaching the target is to use the pulses for a
longer time. The time duration of the first pulse (135kV/m) and
the second pulse (195kV/m) was increased to 4 μs, with different
electric field on pore size distribution and the fraction of cell
break times of 3 μs, 8 μs and 26 μs. The results are shown in
surface area occupied by pores, defined as the fractional pore
Table 4 to see the effects on FPA and pore sizes. The total FPA
area (FPA). Several simulations were carried out and important
increases in both cases, so does the contribution of FPA by pores
results tabulated as shown in Tables 2-5. These tables give
larger than 5 nm. A shorter break ensures that the initial pores
information about the final pore numbers, pore density, %FPA,
do not shrink too much and thus a few pores are able to expand
maximum mean radius, maximum overall radius, and the pore
to a size larger than 20 nm. It is not possible for these pores to
size distribution. The numbers on right hand side of the pore
exist with a longer break. However, they could be possible with
numbers is the %FPA contribution due to corresponding pores
longer first and/or second pulse duration.
in that row. Electric field applied is represented as [135 0 135 ]
To determine if use of a much lower amplitude first pulse or
for a time duration [0 3 6 9] . This represents an electric field
reversing the sequence of the higher and the lower amplitude
pulse of 135 kV/m applied from 0 to 3 μs, a break from 3 to
pulses would have any favorable result, more simulations were
6 μs, followed by an electric field of 135kV/m from 6 to 9 μs
done and results are shown in Table 5. As seen in the first column
and so on. The magnitude of first pulse was chosen such that it
of Table 5, the number of large pores and their FPA contribution
is just enough to create pores resulting in an FPA of more than
(radius larger than 5 nm) drastically reduces although the total
0.03%, as the appropriate level of electroporation is in the range
FPA is similar. Column 4 of Table 5 indicates that while a very
of 0.01% to 0.1% [26]. The magnitude and the time duration of
small (85 kV/m) first pulse followed by a high amplitude second
second pulse were varied to investigate its effect on the number of
pulse (195 kV/m) gives a reasonable value of total FPA, and
pores in the range 10-20 nm. The model was used to investigate
a good number of large pores (5-10 nm), it does not support
the mechanism /usefulness of the two-pulse protocol for control
formation of many pores in the range (greater than 10 nm).
of pore size and number distribution. Different electric field
amplitudes for different time durations were used as the second
pulse for simulations, keeping other parameters like cell radius, V. C ONCLUSION
internal and external conductivity, etc., constant. The final extent There seems to be an interdependent relationship between the
of electroporation (measured by FPA) was targeted to be about pulse amplitude, pulse duration, number of pulses and the time-
0.05 % , with at least over one tenth of the FPA contributed by gap between pulses in order to control the total FPA and the
pores with radii larger than 5 nm, making sure that the number desired pore radii distribution. Generally, the first pulse seems
of these pores larger than 5 nm in radius is also at least 2% of to control the number of pores, and the second pulse helps to
the total number of pores on a spherical cell of radius 15 μm. grow these pores to a larger size. This is in agreement with the
 MODELLING CONTROL OF PORE NUMBER AND RADII DISTRIBUTION 235

Table 2

Time in microseconds 2us 2us 3us

Electric field Amplitude in Volts/m 125K 135k 135k

(total)Final pore number 56246 78720 78720

(total)Final FPA (%) 0.02767 0.03345 0.03405

(total)Maximum mean radius (nm) 2.0101 1.8779 1.8779

(total)Maximum overall radius (nm) 8.3189 7.456 10.9606

Table 3

Time in microseconds [0 3 6 9]us [0 3 6 9]us [0 3 6 9]us

Electric field Amplitude in Volts/m [135k 0 135k] [135k 0 175k] [135k 0 195k]

(total)Final pore number 78720 78739 79011

(total)Final FPA (%) 0.03539 0.04599 0.05257

(total)Maximum mean radius (nm) 1.8779 2.0671 2.1638

(total)Maximum overall radius (nm) 18.6556 25.1665 25.9748

pores less than 1 nm 1465 0.00009 1468 0.00009 1639 0.00011

pores in the range 1-5 nm 76320 0.029 76121 0.03586 76120 0.03873
pores in the range 5-10 nm 660 0.0023 861 0.00421 862 0.00487
pores in the range 10-20 nm 275 0.0035 282 0.00544 376 0.0081
pores in the range 20-100 nm 0 0 7 0.00039 14 0.00076
pores larger than 100 nm 0 0 0 0 0

Table 4

Time in microseconds [0 4 7 11] [0 4 12 16] [0 4 30 34]

Electric field Amplitude in Volts/m [135k 0 195k] [135k 0 195k] [135k 0 195k]

(total)Final pore number 78850 83903 84191

(total)Final FPA (%) 0.05573 0.05394 0.05444

(total)Maximum mean radius (nm) 2.1705 2.1838 2.1903

(total)Maximum overall radius (nm) 33.6247 22.7617 16.6944

pores less than 1 nm 1494 0.0001 1926 0.00012 1816 0.00012

pores in the range 1-5 nm 76120 0.03826 80120 0.0424 80320 0.04193
pores in the range 5-10 nm 861 0.00557 1779 0.00889 1781 0.00799
pores in the range 10-20 nm 359 0.01035 65 0.00222 274 0.00441
pores in the range 20-100 nm 16 0.00145 13 0.00066 0 0
pores larger than 100 nm 0 0 0 0 0 0

Table 5

Time in microseconds [0 4 7 11] [0 3 6 9]

Electric field Amplitude in Volts/m [195000 0 135000] [85000 0 195000]

(total)Final pore number 211812 (pores form at second pulse)126748

(total)Final FPA (%) 0.05313 0.05778

(total)Maximum mean radius (nm) 1.6206 2.7365

(total)Maximum overall radius (nm) 24.829 11.6053

pores less than 1 nm 6500 0.00062 1580 0.0001

pores in the range 1-5 nm 20500 0.04774 12310 0.0495
pores in the range 5-10 nm 40 0.00013 2060 0.00807
pores in the range 10-20 nm 270 0.00442 10 0.0001
pores in the range 20-100 nm 1 0.00023 0 0
pores larger than 100 nm 0 0 0
236 TALELE AND GAYNOR

theoretical research by citeneu2003 and experimental findings by [13] K. C. Smith, J. C. Neu, and W. Krassowska, “Model of creation and
[27], who reported better uptake of DNA for a two pulse protocol evolution of stable electropores for DNA delivery,” Biophysical Journal,
vol. 86, pp. 2813–2826, 2004.
compared with a single pulse protocol. The break between the two [14] Talele, S., P. Gaynor, J. van Ekeran, and M. J. Cree. Modelling single cell
pulses will let the initial pores reduce in size. Thus if the break electroporation with bipolar pulse: Simulating dependence of electroporated
is too long, the second pulse tends to form new pores and has to fractional pore area on the bipolar field frequency. In: Proceedings
International Joint Conferences on Computer, Information, and Systems
expand the initial pores from a smaller size, thus ending up with Sciences, and Engineering. USA (2008).
a comparatively smaller size of pores than if a shorter break was [15] J. Newman, “Resistance for flow of current to a disk,” Journal of The
used. Electrochemical Society, vol. 113, pp. 501–502, 1966.
[16] J. C. Neu and W. Krassowska, “Modeling postshock evolution of large
electropores,” Physical Review E, vol. 67, pp. 1–12, 2003.
R EFERENCES [17] J. C. Weaver and Y. Chizmadzhev, “Theory of electroporation: a review,”
Bioelectrochemistry and Bioenergetics, vol. 41, pp. 135–160, 1996.
[1] M. Cemaẑar, D. Miklavĉiĉ, and G. Serŝa, “Intrinsic sensitivity of tumor cells [18] J. Israelachvili, Intermolecular and Surface Forces. London, UK: Academic
to bleomycin as an indicator of tumor response to electrochemotherapy,” Press, 1992.
Japanese Journal of Cancer Research, vol. 89, pp. 328–333, 1998. [19] J. C. Neu and W. Krassowska, “Asymptotic model of electroporation,”
[2] G. Serŝa, “Electrochemotherapy,” in Electrochemotherapy, Electrogenether- Physical Review E, vol. 59, pp. 3471–3482, 1999.
apy, and Transdermal Drug Delivery : Electrically Mediated Delivery of [20] J. C. Weaver and K. T. Powell, “Theory of electroporation,” in Electropo-
Molecules to Cells (Methods in Molecular medicine), M. J. Jaroszeski, ration and Electrofusion in Cell Biology, E. Neumann, A. E. Sowers, and
R. Heller, and R. Gilbert, Eds. Totowa, New Jersey: Humana Press, 2000, C. A. Jordan, Eds. New York: Plenum Press, 1989, ch. 7, pp. 111–126.
ch. 6, pp. 119–133. [21] I. G. Abidor, V. B. Arakelyan, L. V. Chernomordik, Y. A. Chizmadzhev,
[3] G. Serŝa, S. Novaković, and D. Miklavĉiĉ, “Potentiation of bleomycin V. F. Pastushenko, and M. R. Tarasevich, “Electric breakdown of bilayer
antitumor effectiveness by electrotherapy,” Cancer Letters, vol. 69, pp. 81– membranes: I. The main experimental facts and their qualitative discussion,”
84, 1993. Bioelectrochemistry and Bioenergetics, vol. 6, pp. 37–52, 1979.
[4] R. W. Glaser, S. L. Leikin, L. V. Chernomordik, V. F. Pastushenko, , and [22] M. Kanduŝer, M. Foŝnaria, M. Ŝentjurc, V. Kralj-Igliĉ, H. Hägerstrand,
A. I. Sokirko, “Reversible electrical breakdown of lipid bilayers: Formation A. Igliĉ, and D. Miklavĉiĉ, “Effect of surfactant polyoxyethylene glycol
and evolution of pores,” Biochim Biophys Acta, vol. 940, p. 275287, 1988. C12 E8 on electroporation of cell line DC3F,” Colloids and Surfaces A, vol.
[5] T. Kotnik, D. Miklavĉiĉ, and T. Slivnik, “Time course of transmembrane 214, pp. 205–217, 2003.
voltage induced by time-varying electric fields: a method for theoretical [23] A. Barnett and J. C. Weaver, “Electroporation: A unified, quantitative theory
analysis and its application,” Bioelectrochemistry and Bioenergetics, vol. 45, of reversible electrical breakdown and mechanical rupture in artificial planar
pp. 3–16, 1998. bilayer membranes,” Bioelectrochemistry and Bioenergetics, vol. 25, pp.
[6] L. M. Mir, “Therapeutic perspectives of in vivo cell electropermeabilization,” 163–182, 1991.
Bioelectrochemistry, vol. 53, pp. 1–10, 2000. [24] R. P. Joshi, Q. Hu, K. H. Schoenbach, and H. P. Hjalmarson, “Improved
[7] T. Budak-Alpdogan, D. Banerjee, and J. R. Bertino, “Hematopoietic stem energy model for membrane electroporation in biological cells subjected to
cell gene therapy with drug resistance genes: An update,” Cancer Gene electrical pulses,” Physical Review E, vol. 65, no. 4, pp. 041 920–041 928,
Therapy, vol. 12, pp. 849–863, 2005. Apr 2002.
[8] J. R. Bertino, “Transfer of drug resistance genes into hematopoietic stem [25] V. F. Pastushenko and Y. A. Chizmadzhev, “Stabilization of conducting pores
cells for marrow protection,” The Oncologist, vol. 13, pp. 1036–1042, 2008. in BLM by electric current,” General Physiology and Biophysics, vol. 1, pp.
[9] S. B. Dev, D. P. Widera, and G. A. Hofmann, “Medical applications of 43–52, 1982.
electroporation,” IEEE Transactions on Plasma Science, vol. 28, pp. 206– [26] M. Hibino, M. Shigemori, H. Itoh, K. Nagayama, and K. Kinosita, Jr.,
223, 2000. “Membrane conductance of an electroporated cell analyzed by submicrosec-
[10] K. A. DeBruin and W. Krassowska, “Modeling electroporation in a single ond imaging of transmembrane potential,” Biophysical Journal, vol. 59, pp.
cell. I. Effects of field strength and rest potential.” Biophysical Journal, 209–220, 1991.
vol. 77, pp. 1213–1224, 1999. [27] S. Ŝatkauskas, M. F. Bureau, M. Puc, A. Mahfoudi, D. Scherman,
[11] T. R. Gowrishankar and J. C. Weaver, “An approach to electrical modeling of D. Miklavĉiĉ, and L. M. Mir, “Mechanisms of in vivo DNA electrotransfer:
single and multiple cells,” Proceedings of the National Academy of Sciences, Respective contributions of cell electropermeabilization and DNA elec-
vol. 100, pp. 3203–3208, 2003. trophoresis,” Molecular Therapy, vol. 5, pp. 133–140, 2002.
[12] W. Krassowska and P. D. Filev, “Modeling electroporation in a single cell,”
Biophysical Journal, vol. 92, pp. 404–417, 2007.
 Single Allocation P-Hub Median Problem to
Monitor Land Borders by Using Unmanned
Aircraft
Yasemin Akku, nci Sarçiçek
Eski ehir Osmangazi University,
Department of Industrial Engineering

Abstract - During the last two decades, hub location problems (single allocation) or to more hubs (multiple allocations).
have become popular with successful applications in Capacitated versions of hub problems may involve different
transportation, telecommunication, networks, retailing and kinds of capacity constraints [3].
other logistic systems. In this paper we consider Turkey’s land In general, problems are handled in 4 types: p-Hub
borders security. Because security is the one of the most
important deal nowadays, countries are spending so much to
Median Problem; p-Hub Center Problem; Hub Covering
prevent threats that may come from neighbor countries’ Problem; Hub Location Problem with Fixed Costs. The
borders. Also several geographical restrictions at our land objective of the p-hub median problem is to minimize the
borders cause insufficiency to monitor and to take the requiring total transportation cost (time, distance, etc.) needed to serve
data. We focus on the problem of the single allocation p-hub the given set of flows, given n demand nodes, flow between
median that is addressed to monitor the movement at Turkey’s origin–destination pairs and the number of hubs to locate (p)
land borders by using unmanned aircraft. In addition, aiming [4]. p-hub center problem is a minimax type problem. The p-
assignment of demand points that are defined as cities at land hub center problem is to locate p hubs and to allocate non-hub
borders to hubs mathematical model is built and the results are nodes to hub nodes such that the maximum travel time (or
evaluated. Model is encoded by using GAMS software and
solved by CPLEX solver.
distance) between any origin–destination pair is minimized
[5]. Hub covering problem’s aim is to maximize the covered
area by the hubs obeying the maximum time bound on travel
Key Words- P-Hub Median Problem, Unmanned Aircraft time [4]. In the p-hub location problem the fixed cost of
opening facilities is disregarded. On the other hand, the
I. INTRODUCTION simple plant location problem includes fixed facility costs. In
1992, O’Kelly introduces the fixed facility costs into a hub
Hubs form critical elements in many airline, location problem and thereby making the number of hubs a
transportation, postal and telecommunications networks. They decision variable [6].
are centralized facilities in these networks whose functions In this study, we addressed to monitor the movement at
are to consolidate, switch and sort flows. Typical applications Turkey’s land borders by using unmanned aircraft. Within the
of hub location include airline passenger travel, scope of the problem, we consider single allocation p-hub
telecommunication systems and postal networks [1]. The hub median problem.
location problem is concerned with locating hub facilities and The research on the p-hub median problem with single
allocating demand nodes to hubs in order to route the traffic assignment started with the works of O’Kelly (1987). O’Kelly
between origin–destination pairs. (1987) presented the first mathematical formulation for the
Studies on the hub location problem often assume three single allocation p-hub median problem as a quadratic integer
things [2]: program which minimizes the total network cost. This
1. The hub network is complete with a link between every quadratic integer program is considered as the basic model for
hub pair; hub location problem. Campbell (1994) produced the linear
2. There are economies of scale incorporated by a discount integer programming formulation for the single allocation p-
factor for using the inter-hub connections; hub median problem. His formulation has (n4+n2+n) variables
3. No direct service (between two non-hub nodes) is allowed of which (n2+n) are binary and it has (n4+2n2+n+1) linear
There are several kinds of hub location problems based on constraints [2].
characteristics of a particular hub network. If the number of Ernst and Krishnamoorthy (1996) proposed a different
hub nodes is fixed to p, we are dealing with p-hub location linear integer programming formulation which requires fewer
problems. Each non-hub node can be allocated either to one variables and constraints in an attempt to solve larger

237
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_41, © Springer Science+Business Media B.V. 2010
238 AKKU AND SARIÇIÇEK

problems. They treated the inter-hub transfers as a multi

commodity flow problem where each commodity represents
the traffic flow originating from a particular node [2].
Sohn and Park (1997) studied the single allocation two-
hub median problems. They transform the quadratic 0-1
integer program for single allocation problem in the fixed two
hub system into a linear program and solve in polynomial
time when the hub locations are fixed [7].
Ebery (2001) presented a formulation for the single
allocation p-hub median problem with two or three hubs [8]. Fig. 1. (a)Single allocation (b) and multiple allocation
Yaman (2008) considered a version of this problem where
service quality considerations are incorporated through p-hub median problems can be classified into two groups
delivery time restrictions. She proposed mixed integer according to the connection type of demand points to hubs as
programming models for these two problems and report the single and multi allocation. If each demand point is assigned
outcomes of a computational study using the CAB data and to exactly one hub, the problem is called the single
the Turkey data [9]. assignment (allocation) problem. If a demand point can be
Various heuristic algorithms have been developed by assigned to more than one hub then the problem is called the
O’Kelly (1987), Aykin (1990), Klincewicz (1991, 1992), multi assignment problem. Our problem is the single
Skorin-Kapov and Skorin-Kapov (1994), Smith et al. (1996), assignment (allocation) p-hub median problem.
Campbell (1996), Ernst and Krishnamoorthy (1996), O’Kelly
et al. (1996), Skorin-Kapov et al. (1996), Ernst and III. SINGLE ALLOCATION P-HUB MEDIAN PROBLEM
Krishnamoorthy (1998), Pirkul and Schilling (1998),
Abdinnour- Helm (2001), Elhedhli and Hu (2005) and Kratica A. The Definition of The Problem
(2007) .
In this study, we addressed to monitor the movement at The land borders of the republic of Turkey are 2,573 km
Turkey’s land borders by using unmanned aircraft. Within the in total, and coastlines (including islands) are another 8,333
scope of the problem, weight factors of the possible hubs are km, Turkey has two European and six Asian countries for
determined; a mathematical model was built and solved to neighbors along its land borders.
decide hub locations. The land border to the northeast with the commonwealth
The remainder of this paper is organized as follows. In the of Independent States is 610 km long; that with Iran, 454 km
next section, we introduce problem. The problem is solved by long, and that with Iraq 331 kilometers long. In the south is
in Section 3. Finally, our results are summarized in Section 4. the 877 km-long border with Syria, which took its present
form in 1939, when the Republic of Hatay joined Turkey.
II. p-HUB MEDIAN PROBLEM Turkey's borders on the European continent consist of a 212-
km frontier with Greece and a 269-km border with Bulgaria.
Hubs are centralized facilities in these networks whose The border of Turkey with Syria, Iran, and Iraq are
functions are to consolidate, switch and sort flows. Flow especially important when Turkey's geo-strategic location, its
concentration and consolidation on the arcs that connect hub difficult neighborhood, the closeness to the energy sources in
nodes (hub arcs) allow us to exploit transportation flow the Middle East, international transportation, and terrorist
economies. It is also possible to eliminate many expensive incidents in the east and southeast are considered. There is no
direct connection arcs between origin destination pairs [1]. physical security system at these borders except with Syria
There are two basic types of hub networks: single [11].
allocation and multiple allocations (Fig. 1) [10]. They differ Nowadays, developed countries use state-of-the-art
in how non-hub nodes are allocated to hubs. In single technologies to protect their borders such as Unmanned
allocation, all the incoming and outgoing traffic of every Aerial Vehicle (UAV), satellite-based navigation system, and
demand center is routed through a single hub; in multiple sensors. With this study, the handiness of the UAV will be
allocation, each demand center can receive and send flow investigated for Turkey’s borders.
through more than one hub [2]. Unmanned Aerial Vehicles (UAVs) are remotely piloted
The p-hub median problem is to locate p hubs in a network or self-piloted aircraft that can carry cameras, sensors,
and allocate demand points to hubs such that the sum of the communications equipment or other payloads. They have
costs of transporting flow between all origin destination pairs been used in a reconnaissance and intelligence-gathering role
in the network is minimized [4]. So p-hub median problem is since the 1950s, and more challenging roles are envisioned,
a minimum type problem. including combat missions [12].
UAVs will make significant contributions to the war
fighting capability of operational forces. They greatly
improve the timeliness of battlefield information while
reducing the risk of capture or loss of manned RECCE assets.
 SINGLE ALLOCATION P-HUB MEDIAN PROBLEM 239

When compared to manned RECCE aircraft they are cost Table 2. Potential hubs
effective and versatile systems. While reconnaissance,
intelligence, surveillance, and target acquisition (RISTA) are NUMBER OF HUBS POTENTIAL HUBS

the premier missions of UAVs, they can also provide 1 ADANA

substantial support to intelligence preparation of the 2 ADIYAMAN

battlefield (IPB), situation development, battle management 3 ARI

(BM), battle damage assessment (BDA) [13]. 4 ANKARA
Each Region has its own unique regional features. The 5 ANTALYA
features of all the possible hubs and demand nodes are not 6 BALIKESR
assumed the same in order to have a realistic model. So, 7 ÇANAKKALE
parameters that present the regional features are defined and
8 DENZL
used to reflect the differences between the regions. We used
9 DYARBAKIR
these features to find β j ( the weight of the j th possible hub) 10 ELAZI
by using multi criteria decision making model. 11 ERZNCAN
Cities with land border were taken as demand nodes 12 ERZURUM
(Table 1). Existing airports in Turkey were chosen (Fig. 2) 13 GAZANTEP
while possible hubs were being determined (Table 2). 14 ISPARTA
Therefore, the installation cost of the hubs was not 15 STANBUL
considered.
16 ZMR

17 KARS

18 KAYSER

19 KONYA

20 MALATYA

21 MARDN

22 KAHRAMANMARA

23 MULA

24 MU

25 NEVEHR

26 SAMSUN

27 SRT

28 SNOP

29 SVAS
Fig. 2. Airports
30 TEKRDA
Table 1. Demand nodes 31 TRABZON

32 ANLIURFA
NUMBER OF
DEMAND NODES 33 UAK
DEMAND NODES
34 VAN
1 HATAY

2 KLS

3 G.ANTEP Proposed mathematical model is built by assumptions

4 URFA which determine which possible hub(s) would be appropriate
5 MARDN to be used.
6 IRNAK

7 HAKKAR – IRAN
B. Assumptions
8 HAKKAR - IRAQ
• The aircraft altitude is constant.
9 VAN

10 ARI
• There is always a sight available.
11 IDIR • The model of all the aircrafts is same.
12 KARS • The speed of the aircrafts is constant (220km/h).
13 ARDAHAN • The speed of the aircrafts during the observation
14 ARTVN operations is 110 km/h.
15 EDRNE • Observation diameter is 15 km.
16 KIRIKKALE
• The communication range of the aircrafts is 300 km.
240 AKKU AND SARIÇIÇEK

• Aircrafts turn back to their hubs. Equation (1) defines the aim function. It decides about
• Duties are performed in good weather conditions. which hub will be activated for maximum benefit using
weighting parameter. Equation (2) assigns a demand node for
• There is no threat for the aircrafts.
each hub using (3). The maximum number of the active hubs
• Considered hubs are controled by General Directorate is limited by p with (4). Equation (5) does the hub selection
of State Airports Authority of Turkey. and assignment using the maximum range of the aircraft.

C. Parameters F. Results

H ; {1,..., m} is the sequence of the main possible hubs The GAMS/CPLEX solver is used to select the hubs to be
N ; {1,..., n} is the sequence of the demand nodes. activated among the possible hubs.
Model is solved for each p value individually
i ; demand node i = 1,2,...,16 while p = 1,2,...,16 . Current airports are defined as possible
j ; possible hub j = 1,2,...,34 hubs. Therefore, the cost of adding a new hub is not
considered. According to the results of our model, Tekirdag,
β j ; the weight of the j th possible hub Trabzon, Gaziantep, and Van airports are decided to be used
d ij ; distance between i th demand node and j th hub for UAV system (Table 3).

S ; the maximum range of the aircraft Table 3. Results

p ; total number of the hubs (p<=the number of the demand
nodes)
­1, d ij < S
a ij = ®
¯0, otherwise
gi; the number of the illegal transportation at the i th node in
(Jan-June 2009)
oi ; the number of the illegal incidents at the i th node in (Jan-
The reasons of recommending Tekirda as a hub instead
June 2009)
of Istanbul are to be close to the land border, to have a current
g i + oi ; the demand of the i th node airport, and to have low traffic. The weight factors of the
other hubs were effective on the decision. The weight factors
are calculated using the following criteria:
D. Decision Variables i) whether the airport belongs to the military or not,
ii) whether there is threat in the area or not,
iii) whether the city has a land border or not,
iv) whether the airport has low air traffic or not, and
v) the duration that the airport is in service.
Our results show that, the weight factors that are calculated
using many criteria are very useful.

E. Model IV. CONCLUSION

n m
max Z = ¦ β j * Y j + ¦¦ ( g i + oi ) * X ij (1) The problem of the single allocation p-hub median is
i j addressed to monitor the movement at Turkey’s land borders
m by using UAV. The range of the UAV is used to determine
¦X
j
ij =1 ∀i ∈ N (2) the best location to install the UAV system. In future work,
the cost and the altitude of the UAV may be added to the
X ij ≤ Y j ∀i ∈ N ∀j ∈ H (3) model.
m

¦Y
j
j =p (4)

X ij ≤ a ij ∀i ∈ N ∀j ∈ H (5)

X ij ∈ {0,1} ∀i ∈ N ∀j ∈ H (6)
 SINGLE ALLOCATION P-HUB MEDIAN PROBLEM 241

REFERENCES [7] J. Sohn and S. Park, “A linear program for the two-hub
location problem”, European Journal of Operational
[1] A.T. Ernst, H. Hamacher, H. Jiang, M. Krishnamoorthy, Research, 100, pp.617-622, 1997.
and G. Woeginger, “Uncapacitated single and multiple [8] J. Ebery, “Solving large single allocation p-hub problems
allocation p-hub center problems”, Computers and with two or three hubs”, European Journal of Operational
Operations Research, 36, pp. 2230-2241, 2009. Research”, 128, pp.447-458, 2001.
[2] S. Alamur and B.Y. Kara, “Network hub location [9] H. Yaman, B.Y. Kara and T. Barbaros, “The latest arrival
problems”, European Journal of Operational Research, 190, hub location problem for cargo delivery systems with
pp.1-21, 2007. stopovers”, Transportation Research, Part B(43), pp. 643–
[3] J. Kratica, Z. Stanimirovic, D. Tosic, and V. Filipovic, 658, 2009.
“Two genetic algorithms for solving the uncapacitated single
[10] A. Özger and H. Oktal, “Integer model approach for
allocation p-hub median problem”, European Journal of
uncapacitated multiple allocation p-hub median problem in
Operational Research, 182, pp.15-28, 2007.
air cargo transportation”, Havac l k Ve Uzay Teknolojileri
[4] J.F. Campbell, “Integer programming formulations of
Dergisi, 4, pp.47-60, 2009.
discrete hub location problem”, European Journal of
[11] www.stratejikboyut.com.article_detail.php (Cafer
Operational Research, 72, pp. 387-405, 1994.
Serinkan)
[5] J.F. Campbell, “Hub location for time definite
[12] http://www.fas.org/irp/program/collect/uav.htm
transportation”, Computers & Operations Research, vol.36,
[13] http://www.globalsecurity.org/intell/ systems/ uav-
pp.3107-3116, 2009.
intro.htm
[6] S. Elmasta, “Hub location problem for air ground
transportation systems with time restrictions”, PhD Thesis,
Bilkent University, 2006.
 LPT to RS232 communication converter
Grofu Florin “Constantin Brancusi” University – Tg-Jiu
Cercel Constantin “Constantin Brancusi” University – Tg-Jiu

Abstract: The base idea of this paper came as a result of using

of some data acquisition systems controlled by an application In case of using of a 8051 family microcontroller there is
made in LabWindows CVI. The communication between PC and the possibility to use of multi processor communication.
the acquisition systems is made through RS232 interface. The In these modes, 9 data bits are received, followed by a
main problem appeared is the impossibility to use the acquisition
system for fast bidirectional data transfer using a RS232
stop bit. The ninth bit goes into RB8. Then comes a stop bit.
protocol. As solution for this problem we propose a LPT-RS232 The port can be programmed such that when the stop bit is
communication converter. received, the serial port interrupt is activated only if RB8 = 1.
Keywords: Acquisition System; Communication; RS232 This feature is enabled by setting bit SM2 in SCON. When
interface; Converter the master processor must transmit a block of data to one of
several slaves, it first sends out an address byte that identifies
I. INTRODUCTION the target slave. An address byte differs from a data byte in
that the 9th bit is 1 in an address byte and 0 in a data byte.
The achievement of data acquisition systems by an With SM2 = 1, no slave is interrupted by a data byte. An
automation engineer is an easy thing to do. This can be made address byte, however, interrupts all slaves, so that each slave
by connecting a microcontroller to an AD or DA converter, can examine the received byte and see if it is being addressed.
depends on application. For communication of this system The addressed slave clears its SM2 bit and prepares to receive
with a software application the easiest way is to use the the data bytes that follow. The slaves that are not addressed
RS232 interface. The schematic diagram is presented in set their SM2 bits and ignore the data bytes.[4]
figure 1. In this data acquisition system data we use the follow
For hardware specialist is easy to implement for the structure of data package, presented in table 1.
acquisition system a communication protocol which can This structure allows to connect more acquisition systems
ensure the data transfer in optimal condition without blocking to the same software application. If the application sends one
the system or misunderstanding the data. data package all the systems receive on the interrupt first
byte, which mean the address of destination module using
multiprocessor mode. Only the module with this address goes
in data receive mode, the rest of system ignore the data that
follows. On the complete receive of package it check the
CRC as an additional correction.[1]

Table 1
Octet 1 Address byte address byte
Octet 2 Nr. bytes data bytes
Octet 3 Actions data bytes
Octet 4 Data data bytes
............ .............. ..............

Octet n Data data bytes

Octet n+1 CRC data bytes

RS232 Interface II. SOFTWARE APLICATION

The software application for connecting with the system

it is realized in LabWindows CVI.[7] LabWindows/CVI is an
Microcontroller system integrated American National Standards Institute (ANSI) C
environment (C is a computer programming language)
developed by National Instruments Corporation and designed
primarily for engineers and scientists creating virtual
instrumentation applications. LabWindows/CVI helps you
ADC/DAC
leverage the power of the computer to create flexible,
........ ........ reusable, and inexpensive measurement applications that
outperform traditional test and measurement methods.
Analog I/O Digital I/O
Fig. 1 Data acquisition system

243
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_42, © Springer Science+Business Media B.V. 2010
244 FLORIN AND CONSTANTIN

LabWindows/CVI is a programming environment that has An alternative solution to serial communication is the
been widely adopted throughout industry, academia, and parallel communication that provides high speed
research labs as the standard for data acquisition and communication but on short distance. If there is already exist
instrument control software. It is a powerful and flexible serial communication equipment then must be implemented a
instrumentation software system that embraces the traditional parallel to serial communication converter. This converter
programming methodologies and enables you to create simply takes parallel data package and then sends it serial to
structured code, and it features an easy-to-use programming acquisition system. It also takes serial data package from the
environment.[8] acquisition system and forwards the data parallel to the
LabWindows/CVI includes all the tools necessary for data application.[9]
analysis and presentation of the results on the Graphical User Thus, all changes of state of serial communication port is
Interface (GUI) on your computer screen. It comes with a transferred to the parallel to serial conversion module that
complete set of integrated input/output (I/O) libraries, carried out in real time. The structure of such a
analysis routines, and user interface tools that aid you in rapid communication system is shown in figure 2.
application development, reducing the time that would be The converter is made bidirectional, meaning it can
required by other conventional programming transmit and also receive data, so the information will flow
environments.[6] not only from computer to peripheral, but also from the
LabWindows/CVIs forte lies in building virtual computer peripheral.
instrumentation systems with General Purpose Interface Bus
(GPIB), VME (Versa-Modular Eurocard) extensions for
Instrumentation (VXI), PCI (Peripheral Component
Interconnect) extensions for instrumentation (PXI), serial
interface communication (RS-232), Transmission Control
Protocol/Internet Protocol (TCP/IP) based devices with plug-
in data acquisition (DAQ) boards without spending too much
effort to create the applications. It combines an interactive,
easy-to-use development approach with the programming
power and flexibility of compiled ANSI C code.
The full potential of LabWindows/CVI is used to automate
test systems, bench top experiments, DAQ monitoring
projects, verification tests and measurements, process
monitoring, and controlling systems.
Parallel Interface
III. LPT-RS232 CONVERTER
Microcontroller system for
In case we use the RS232 communication between PC and LPT/RS232 converter
data acquisition system up to 10 times per second the RS232 Interface
application is easy to use.
If we need to increase the speed of communication the
problems begin to appear and only a very good LabWindows
programmer can resolve them. This problem appears as effect
of serial port accessing mode through the existing functions.
The communication is delayed because the ninth bit (parity
bit) must be different set for a command byte or a data byte. DAS
From experiments we observed that delays occur when
accessing the change configuration functions of the serial port
or on change direction of transfer. Thus if the complete Fig. 2. Using LPT/RS232 converter
communication means transmission and reception of a Knowing the communication protocol for serial transfer,
package of 6 bytes of data at a transfer rate of 9600 bits / sec, how to transfer data using PC parallel port and the structure
theoretically should last up to 15 ms, practically takes about and functioning of a 89S52 Intel microcontroller we designed
100 to 150 ms depending on the architecture and performance and achieve a parallel to RS232 communication converter.
of computer system used. Functional diagram of the converter is shown in figure 3:
This leads to a decrease of almost 10 times of the effective
speed of communication resulting the possibility of a
bidirectional data transfer up to 10 times per second.
 LPT TO RS232 COMMUNICATION CONVERTER 245

R/W

P1
1 Data Strobe VCC_+5V VCC_+5V
14 3 1
2 D0 S1
15 RESET 4 2 R12
3 D1 360 R12 U7
Sens 16 R1 6K8 360
4 D2 5 4

+
VCC_+5V SDA Vss
17 C1 10uF
5 D3 6 3
18 SCL A2
6 D4 7 2
19 WP A1

20
9
7 D5 U1 8 1
20 D0 1 39 Vcc A0

GND
RESET
8 D6 D1 2 P1.0/T2 P0.0 38 VCC_+5V 24LC256
21 D2 3 P1.1/T2X P0.1 37
9 D7 D3 4 P1.2 P0.2 36
22 D4 5 P1.3 P0.3 35
10 D5 6 P1.4 P0.4 34
23 D6 7 P1.5 P0.5 33 VCC_+5V
11 ACK D7 8 P1.6 P0.6 32
24 P1.7 P0.7
12 21 Led_1 Led_1 R2 360 D1 LED
25 R/W 12 80C52 P2.0 22 Led_2
13 Busy Data Strobe 13 INT0 P2.1 23 Led_3
DB25 ACK 14 INT1 P2.2 24 Led_4 Led_2 R3 360 D2 LED
Sens 15 T0 P2.3 25 Led_5
Busy 16 T1 P2.4 26 Led_6
WR P2.5 27 Led_7 Led_3 R4 360 D3 LED
U2 RS485 17 P2.6 28 Led_8
2 RS485 RD P2.7
6 RE 3 30 Led_4 R5 360 D4 LED
ARK350/2

2 A DE RX_date 10 ALE/P 29
7 1
<----
RX_date_2 RXD PSEN
1 B R 4 TX_date TX_date 11 Led_5 R6 360 D5 LED
D TXD

EA/VP
---->

VCC
X1

X2
SN75176 Led_6 R7 360 D6 LED
RX_date

19

18

31

40
2 RX_date_2 Led_7 R8 360 D7 LED
3 74HCT08
P2 RX_date_1 Y1 VCC_+5V
1
5 Led_8 R9 360 D8 LED
24MHz Fund
4
COMUNICATIE C2 C3
7
U5 RS232 33pF 33pF
2 8 9 RX_date_1
<---- <----
R2in R2out
3 ----> 7 10 TX_date ---->
T2out T2in
13 12
DB9_T R1in R1out
14 11
T1out T1in
5 15
C4 C2- GND C5
100n 4 6 100n
C2+ Vs-
3 2
C9 C1- Vs+ C7
100n 1 16 100n
C1+ VCC VCC_+5V
Fig. 3. Structure of LPT-RS232 converter
MAX232

In structure of converter we see an EEPROM memory CONCLUSION

which is necessary to use the converter in more complex
application were we transfer more then 128 bytes. The From performed analyze result that in case of acquisition
EEPROM memory is used also for back-up some setting of system with serial communication and where is necessary to
module. increase the effective communication speed we need to
To explain better the advantage of using this type of contact a programmer which may realize a software module
converter we achieve a software application that we connect which can be executed faster (in LabWindows), module
three identical acquisition systems to the PC in this way: which depend on the hardware structure of computer.
- one directly to PC serial port (COM1); A simply way is to use the proposal parallel to serial
- one through an USB to serial converter; converter which ensure the increase of effective
- one through the proposed parallel to serial converter. communication speed indifferently of computer architecture
The application, with the graphical interface presented in considering that to send and receive an 6 byte data package
figure 4, send sequential by the three acquisition systems a 6 takes on parallel port approximately 1ms.
bytes data package. In case of correct communication the In figure 4 it is presented the interface of the software
acquisition systems replay also a 6 bytes data package.[1] application which can demonstrate the advantage of proposed
As result of our experiments we obtain the follows communication system. The software application has the
communication times as in table 2. possibility to communicate with three identical acquisition
devices connected with PC by different ways: one with direct
Table 2 connection to PC serial port (RS232), one using an USB to
Nr. DAS Time RS232 converter, and one using the proposed LPT to RS232
Crt. communication communication
1. RS232 (COM1) 402,7 ms
converter.
2. USB-RS232 converter 372,3 ms As we can see in the figure 4, if we consider the values
3. LPT-RS232 converter 14,6 ms displayed in the Time Value boxes, the time for
communication corresponding to the proposed converter is at
least ten times smaller than others. So, the real speed of
246 FLORIN AND CONSTANTIN

communication is increased with at least ten times. We The LPT to RS232 converter is very useful in situations
consider this time the time necessary to send and receive a that the data packages sent and received is not so big (under
data package. 128 bytes), case occurred in monitoring and control
applications for industrial process.

Fig. 4. User interface

REFERENCES

[1]. Cercel Constantin, Grofu Florin, USB to Parallel communication [6]. www.ni.com/pdf/manuals/373551c.pdf
interface, Annals of „Constantin Brancusi” University of Tirgu Jiu, [7]. Vektrex Application Note: 11082002, Using IVI-COM Drivers From
The XIII-th edition of the international scientific conference of the LabWindows/CVI
Engineering Faculty, November 07-08, 2008 , ISSN 1844-4856 [8]. Advanced Topics in LabWindows/CVI by Khalid, Shahid F
[2]. Porat B., A Course in Digital Signal Processings, John Wiley & Sons, [9]. Grofu Florin, Popescu Luminita Georgeta, Cercel Constantin
Inc., New York, 1997. Acquisition system for monitoring vibrations, CISSE2007 – Novel
[3]. Thomas B. Cho, Circuit Techniques for Low Power, High Speed Algorithms and Techniques in Telecommunications, Automation and
Pipelined A/D chapter 2, University of California, Berkeley 2000 Industrial Electronics, Springer, 2008, ISBN: 978-1-4020-8736-3
[4]. Dallas Semiconductor, Ultra-High-Speed Flash Microcontroller [10]. Runceanu Adrian, Fragmentation in distributed databases,
User’s Guide , http://www.maxim-ic.com/ 2004 CISSE2007 - Innovations and Advanced Techniques in Systems,
[5]. Maxim-IC, App Note 604: Fast Memory Transfers with the Ultra Computing Sciences and Software Engineering, Springer, 2008,
High-Speed Microcontroller - DS89C420, ISBN: 978-1-4020-8734-9
 Simulation Model of Coremaking Plant
Stefan Beleva, Miroslav Penchevb,
a,b
University of Rousse, 8 Studentska str., 7017 Rousse, Bulgaria

Abstract
The paper presents simulation model of coremaking plant which gives a possibility for optimal utilization of working machines. The
aim of modeling is investigation of coremaking equipment idle time in dependence of a section production plan within a working day.
The model is realized like simulation program which is built in the system for operative management (SCADA) at one of coremaking
plants in Vidima Ltd. Sevlievo, Bulgaria.
Keywords: coremaking plant, simulation model

container”. In the railway hopper cars technological scheme the

variant “mobile mixer” is also possible.
1. Introduction Both technological schemes (mobile container and mobile
mixer) are based on servicing a group of concurrently operating
machines by one mixer and one transport device. The main
advantage of these schemes is the economy, because one mixer
The main method applied for production of casting cores for and one transport device serve all working machines. The
needs of non-ferrous metallurgy and mostly for producers of potential of delaying the service, and appearing of the idle time
sanitary fittings is the hotbox method. In this coremaking could be pointed as a disadvantage, because the queue is formed
technique, alkali free dry silica sand is mixed with latent (heat when several machines must be served at the same time. It should
activated) hardener and liquid binder (Phenolic resole or Urea- be noted that in most cases the separate core-blowing machines
modified furan resin). The ready made moist mixture for casting produce different products which increases the risk of idle time
cores contains about 1.4 – 2.0 % resin and can be used within 4-6 caused by lack of servicing because of the different technological
hours, depending on the type of the used resin. The mixture is times and the different sand core mixture consumption.
transferred to an automatic core-blowing machine where the core The task of the developed model is to show the idle times
is moulded and cured at 180-250°C in a previously heated caused by lack of servicing in quantitative (as idle times for the
corebox, Fig. 1. separate machines) and qualitative (as moments of occurrence of
Phenolic resole or Urea- Latent hardener the idle times) aspect depending on the daily job of the plant.
modified furan resin 0.3 – 0.5 %
1.4 – 2,0%
Silica sand
2. Conceptual model of coremaking plant
Sand cores

Moist sand mix Dry

Mixing
The investigating plant consists of 10 linearly situated core-
(4÷ 6 h) 200÷250 °
blowing machines, applying the hotbox method. The mixing
Distribution 4÷120 sec equipment, connected to the coremaking plant is of the mobile
Batch mixer system
Concurrently worked container type, Fig. 2.
Automatic core sand mix installation Core blowing machines
Fig. 1. Coremaking plant - hotbox processes. 2.1. Model hierarchical structure
Modern coremaking plants are equipped with fully
automated sand mixing equipments including a batch mixer and a A hierarchical model of the separate components, derived
core sand mixture distribution system. The core sand mixture through functional decomposition, may represent the coremaking
distribution system is designed and adjusted to the particular plant.
coremaking plant depending on the specific plant architecture and The plant produces about 300 types of casting cores to
the customer’s requirements. Usually the distribution systems satisfy the needs of a foundry for sanitary fittings. For the
which transfer the ready core sand mixture from the mixer to the purposes of this research, the following may be pointed as
core-blowing machines is based on railway hopper cars or special characteristics of the products: the time for the technological
suspended monorail cars, technological scheme – “mobile cycle TCJ and the characteristics for the consumption of sand

247
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_43, © Springer Science+Business Media B.V. 2010
248 BELEV AND PENCHEV

core mixture (the mass of the respective product m J and the (FIFO) discipline of servicing. The equipment parameters are
determined and can be presented by the set Mix:
number of products, made within one technological cycle n J ).
The index “J” designates the type of the produced item: Mix = {d, M,  , TtrI , Tul} , (6)

Z = {z 1 ,z 2 ,  , z J ); J = 1,2,  ,300 where:

z j = z (Tc j , m j , n j ), ∀ z J ∈ Z d is the number of the currently requested doses of sand
. (1) core mixture;
Automatic Mixing Plant  - the mass of one dose of sand core mixture;
Batch Mixer  - the time necessary for the preparation of one dose of
sand core mixture;
TtrI - the time necessary for conveying the mixture to the
Core Machine Hopper

hopper of the core shooting machine, which has sent the

request;
Tul - the time necessary for discharging the mixture into the
Hopper car machine’s hopper.
Some special features of the mixing equipment follow from
railway the plant’s architectural scheme, shown in Fig. 2. They are:
• the service device has a constant component – the
working time of the mixer I = dTMX defined by
10 parameter d ;
1 2 3 8 9
• the service device has a different time for conveyance
Core blowing machines - “Hot box”method ( TTR , I = 1 ÷ 10 ), depending on the route, defined by
I

Fig. 2. Coremaking plant – architectural diagram. parameter I .

In the conceptual model of coremaking plant (see Fig. 3)
The daily job may be represented as a section of the these special features are described by denoting the events:
multitude of the machines (2) and the multitude of the produced • Elimination of the request after loading the core
items (3). The index I designate the position of the core-blowing shooting machine - Departure (I,t).
machine in the technological scheme of the plant. • Readiness of the transporting device - Car Ready (t).

h = {mh 1 , mh 2 ,  , mh I }; I = 1 ÷ 10 (2)

Count of requests -
Car Backward (t)

Active requests:
Car Forward (t)

Mixer Start (t)

Car Unload (t)

Active (t)
Car Ready (t)

MPE = Z  Mh (3)

NR(t)
Each of the core-blowing machines in the plant is equipped
with a hopper for sand core mixture with adjustable minimum
Core sand mix installation

level rI , (in % of the hopper’s content). When the mixture level Flow of requests
M,
falls below the preset minimum, the machine transmits request TUL, TTR TMX 10

for serving to the mixing equipment1: Batch

R(t ) =  R (I, d, t)
I=1
I

Shuttle Car Mixer Queue

R I = R(I, d, t ) , (4) Car Ready (t)

Supply Cain:
where: Request departure: Departure (I, t)
I is the location of the core-blowing machine which has Machine load: Load (I, dI*M, t)
transmitted a request;
Requests chain:
d - the number of the requested (by the respective machine)
Load (dI*M,t)

Load (dI*M,t)

Load (dI*M,t)

doses of sand core mixture (d = 1 or d = 2);

RI = R (t, d)

RI = R (t, d)

RI = R (t, d)

t - the moment of originating of the request, as a moment

of occurrence of a characteristic event, changing the
Core Machines
condition of the observed system.
I=1 I=2 I=1
zJ zJ zJ
In this case d and I are the request parameters. Machine Machine Machine
dI No 1 dI No 2 dI No 10
As for the aims of this research, it could be considered that a rI
λJ1 TIT(t)
rI
λJ2 TIT(t) rI
λJI TIT(t)
united flow of requests is received by the mixing equipment:
10
(5) Σ
R =  R (I, d, t)
I=1
I
.

10

The mixing equipment as an algorithm for servicing the

TIT = ¦T
I=1
ITI = min

requests may be considered as a queuing system with standard Fig. 3. Conceptual model of coremaking plant.

1
The mixture consumption model in the hopper is reviewed in following In our case the subject of investigation is a casual flow of
section. events: moments of turning off & on and the duration of intervals
 SIMULATION MODEL OF COREMAKING PLANT 249

TIT during which the separate machines do not work because of where: t is the current time of the simulation experiment, and kJ –
lack of sand core mixture (Fig. 3 and Fig. 4). the current number of the machine cycle from the moment of
The aim of the investigation (in order to enhance the loading the machine.
effectivity of loading of the core-blowing machines) may be The individual technological cycle may be described by
defined as: minimizing the idle time caused by lack of servicing of equation (11), Fig. 5a:
the machines in the plant (7), by changing the location of the (11)
TC = TRW + TFIL + TFIX + TREM + TSERV =  hC + TSERV
product of the current daily job in the technological scheme of the
plant. where: TRW is the time for moving the operating organs from
10
(7) initial to operating position; FIL – the time for filling and
TIT = ¦ TIT = min .
I congesting the mould; TFIX – the time for curing the core into the
I =1
mould; TREM – the time for opening the mould, removing the
2.2. Core Blowing Machine model core and returning the operating organs to initial position; TSERV
– time for cleaning and lubricating the corebox.
A model of consumption of the mixture may represent the TFIX + TREM
moment of sending the request and an idle time interval that is
caused by lack of sand mixture, Fig. 4. TFIL
TSERV
The following denotations are used in the Figure:
L – current hopper level, %;
r – request level, %: TRW
0 L >r (8)
Re quest =
1 0 ≤L ≤r TC
C – core blowing machine cycle, sec;
TAT – arrival time interval, sec; ) Cycle Time b) Service Time Model
WT – waiting time interval, sec; Fig. 5. Core blowing machine – technological cycle.
IT, – idle time interval, machine does not work, sec.
The time intervals TRW, FIL, TFIX and TREM are functions
of the machine settings for the particular product J. And the
Machine Run Machine Run Machine stopped. operation of cleaning and lubricating the corebox is performed
No Request Request Request manually by the operator and is of probability nature.
Observation of the work of machine 6 with product: faucet “bath
L, % L ≥ r 0≤L≤r L=0
body SL – A” is shown in Fig. 5b. The realized observations and
100% statistic investigations show that the time for service may be
described by an exponential distribution which parameter is
defined by the average time.
TC

r Request
3. Model implementation
Level
0% t
TAT TWT TIT On a system level the simulation model assumes description
of the functioning of the simulated system using the concepts
activities, events and processes by means of a definite
Work cycle of Core Blowing Machine programming language. The methods known in practice for
organizing the computing process in the simulation programs
Fig. 4. Core blowing machine – hopper level.
(activity scanning, event schedule and processes interaction) are
Taking into consideration the discrete character of the striving to formalize and solve the problem with the concurrent
machine cycle and the casual character of the interval TWT for the processes, common to all computer simulators. In this particular
mixture level in the hopper of machine I, producing item J at the case due to the elementary organization of the simulation
moment of supplying, it could be written down: machine and the relatively easy realization by means of a
universal programming language, for realization of the simulator
LI, J (t, k JTC J ) = LI, J (t, k JTC J ) + 100 * dI ,% , (9) the “activity scanning” method is chosen.
kJ = 0
3.1. Model formalization
and for the level after each cycle:
mJnJ
LI, J (t, k JTC J ) = LI, J (t, (k J − 1)TC J − 100 ,% Definition 1: The dynamic model of a simple object may be
M , (10) represented as an arranged five, [1]:
LI, J (t, k JTC J ) ≥ 0
DM = (S; EINP; EOUT; SC; g), (12)
kJ = kJ + 1
where S is the endmost set of states of the object;
EINP – the set of the input events of the object;
EOUT – the set of the output events of the object;
250 BELEV AND PENCHEV

g : S × E INP → S - transition function. preset and/or calculated time of life in state SI. (Computation
SC – the set of conditions, which may be defined as a of the times is performed beforehand).
subset of the events, admissible for the separate states Q(1)
MhLoad

SC : ∀s ∈ S; EINP
S
⊂ EINP .
c : (Q(1) = I) ∧ (MhLoad = 1)
This model of the states and transitions is interpreted in the a : Re set Ÿ MhLoad
following manner. If the object is in state SI and event  ∈  INP Machine
Load S0I
occurs, which is admissible for that state, then the object transits
in a new state defined by the function g : S × E INP → S . c : TRUE
L I = L I + 100d Q(1) ,%
The visual formalism of that interpretation is “State
Re set Ÿ R(I, d) Ÿ Queue Re duce
Transition Diagram”, well known from the theory for the finite a:
Set Ÿ TS1I = t + Tc I
state machine and widely used in automation. Store Ÿ Depart(I, t )
In this paper the diagram specification defined by IEC 61131
[2] is used. It is the base of many programming languages for Machine
controlling devices. The diagram is a directed graph whose Cycle S1I
points (denoted by a circle) represent the object’s states and the
c : (t ≥ TS1I )
directed line denotes the transition to the next state (Fig. 6). In a m In I
a : L I = L I − 100
rectangle to the right of the state the actions performed by the M
,%
State
object in that state are denoted. In a random moment of time the
Evaluate S2I
object may be only in one state – active state.

SK -1
c : L I ≥ rI ,% c : L I ≤ 0,%
a : Set Ÿ TS 3I c : 0 < L I ≤ rI ,% a : Store Ÿ Empty(I, t )
TS 3I = t + RND(λ I ) Set Ÿ TS 3I = t + RND(λ I )
Logical Segment
a : Set Ÿ R(I, d)
Store Ÿ Arive(I, t )
State
SK : - State Actions Corebox Stop
service S3I

c: - Transition Logical Condition c : (t ≥ TS3I )

Transition : - Transition Actions
Fig. 7. State transition diagram of core machine model.
Figure 7 represents the atomic model of the core-blowing
SK+1 machine which is developed according to specification IEC
61131 and the rules considered above. The overall model of the
Fig. 6. State transition diagram – definitions. core plant reflects the synchronization between the associated
elementary models of the objects participating in the modeled
Definition 2. The transition is instantaneous and may be production system. They are the group of 10 concurrently
activated if the preceding state is active and corresponding logical operating core-blowing machines; mixing equipment including
condition is executed. The activation of the transition leads to batch mixer and hopper car; a queue with a standard FIFO
activation of the next state and deactivation of the preceding discipline and a database for the simulation experiment.
state. One or several single actions may be associated to the
transition.
Tow special states are foreseen: initial state (initialization),
in which all the rest of the states are brought to zero, and
“blocking” state (stop) which blocks the work of the object. The
object can leave that state only at initialization.
Since the specification does not provide for synchronization
between the actions of the separate objects and measuring the
lifetime of a separate state according to the model time, the
following additional rules are accepted for the purposes of the
simulation:
1. Each initial for the object event is registered by
means of setting of a synchronizing signal.
2. After occurrence of an incoming event the object
nullifies the signal, which has initiated the event.
3. The object settles in initial state only through
external signals, by means of transition to initial state.
4. The lifetime of the separate state is associated to the
model time as an action in the transition preceding the state:
t SI = t + TSP _ SI . Here t is the current moment of the model Fig. 8. Coupled model of coremaking plant.
time in which the transition is activated, and TSP _ SI is the
 SIMULATION MODEL OF COREMAKING PLANT 251

Event No 173: Request from Core Machine 6

Modeling Time: 9500 sec
Request in Queue:1

Event No 181: Core Machine 5 - Full

Modeling Time: 9669 sec
Request in Queue:2

Fig. 9. Result of modeling experiment.

The chosen model for organizing the computation process of 4. Conclusion

the simulator (organization of the simulation machine) does not
reflect the lifetime of the separate states of the objects
participating in the simulation system. This problem is solved by The SCADA systems usually do not include a simulator of
means of registering the events which lead to activation and the controlled system, which could be used for supporting short-
deactivation of the states concerning the user, into a database. For term management resolutions. The developed simulator affords
this purpose, the respective events are declared outgoing and the opportunity for measuring the time, during which the
synchronizing signals are sent to the database, which are simulated plant remains in the separate states, as well as a
processed according to the above reviewed rules. At the detailed tracing of the moments of occurrence of the events and
consequent processing of the database the lifetime of the may be used as an instrument by the plant management staff.
respective state is calculated as the difference between the model It could be considered that the method applied for the
times of deactivating and activating events, Fig. 8. simulator development is applicable to the practical tasks for
The simulator was realized using C++ and was built in the measuring the time of the separate activities of relatively simple
operating control system of the core plant (SCADA package industrial systems with random events, for which a detailed
WinCC 6.0) in VIDIMA JSC, Sevlievo, Bulgaria. The widely tracing of the moments of occurrence of the individual events is
available data base software MS Access was used, which is also necessary.
supported by the SCADA package. The results of modeling
experiment are shown in Fig. 9. References

[1] . .Law, W. D. Kelton: Simulation Modeling and

Analysis, Mc. Graw-Hill, 2000.
[2] W. Lewis. Programming Industrial Control Systems Using
IEC 1131-3, volume 50 of IEE Control Engineering Series.
The Institution of Electrical Engineers, London, UK, 1996.
 Depth Calculation using Computer
Vision and Sift
Edson CAVALCANTI Neto∗ , Haroldo Cabral MAYA ∗ , Antonio Themoteo VARELA, ∗

Mayara Machado CARVALHO∗ , Auzuir Ripardo de ALEXANDRIA ∗

∗ IFCE- Instituto Federal de Educação, Ciência e Tecnologia do Ceará
Av. Treze de Maio, 2081 - Benfica
Fortaleza, Ceará, Brazil, +55 (85) 3307-3666
Email: [email protected]; [email protected]; [email protected];
[email protected]; [email protected]

Abstract—Robotics has been developed in a high overpass obstacles and others can be made. One of the
level in the past few years, especially in mobile ways to perform distance measurement is through digital
Robotics. Its main contribution is shown in human images. Image stereo systems are made with two images
entertainment and performance in dangerous works.
However, those are not just the only applications they taken simultaneous, or not, of the same scene, similar to
perform. There is a wide field to be developed. Distance animals that have two eyes. With this system you can
calculation is something really important for localiza- recover the depth information in the images acquired [1].
tion and navigation of mobile robots. The classic way This work aims to develop a system of distance mea-
to calculate distances are using sonar and laser radar.
surement using computer vision to use in modules of
But, the use of computer vision in stereo, based on
two cameras, is widespread in Robotics. Therefore, perception in mobile robots.
algorithms, to detect new points, allows the creation The development of a computer vision system is of
of new methods to execute stereo vision. In those great importance to complete tasks by robots, with special
algorithms there is one that has been used, the SIFT regard to navigation of autonomous robots in unfamiliar
(scale-invariant feature transform). It may be used
with classic algorithms of stereo vision in the Com-
environments. This statement is true, since vision is the
puter Vision 3D system development. This project’s primary sense used by humans and that guides their
objective is to develop a distance acquisition system daily lives, in their work and their leisure time. There
using computer vision in stereo with the help of SIFT is a tendency of current mobile robots incorporate vision
method. The results are very close with what we were systems. As the use of stereo systems allows the mea-
expecting. The next steps to be done are the execution
of quantitative tests to determinate the focal distance
surement of depth, it is going to be useful in the task of
of the utilized cameras and to calibrate the system. locating and overpass obstacles. It should be mentioned
that several other algorithms can worked together with
I. Introduction the vision system in stereo, using the same cameras and
A typical mobile robot is constituted by subsystems images acquisition system [4]. This work is organized as
such as: perception, localization, navigation and movement bibliographic review from Transformed SIFT and Depth
control [1]. The perception system is responsible to acquire calculation using Digital Image Processing in the second
data from the space where the robot is placed. This data, part, there is the methodology used to achieve the results
after being stored, is processed to be used to generate in the third part and in the forth there are the results of
information on the location where the robot is placed, the research.
through its location system. In possession of its location,
the robot is able to generate the environment map where II. Bibliographic Review
it is situated. After it may compare its local position with
a more comprehensive map generated or delivered by its The main methods to execute the implementation and
developer or user [2]. The ability to navigate using the to understand this system are the transformed SIFT and
map and the mission that the robot should perform are the use of stereo images to calculate the coordinate z
the main tasks that should be performed by its navigation distance.
system. This last task informs which path the system of
A. SIFT Method
motion control should perform.
One of the most important information for a mobile SIFT is a robust method to extract and describe key-
robot is the distance measurement to barriers. This infor- points. The features extracted by this technique are in-
mation can be obtained through various sensors, among variant to translation, rotation, and scale and partially
them images from the environment might be obtained. invariant to changes in lighting and perspective [2] [3] [5].
In possession of this distance, decisions such as location, Basically, the algorithm is performed in four steps:

253
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_44, © Springer Science+Business Media B.V. 2010
254 NETO ET AL.

• Detect extremes: it is a search on all scales and

locations of images with different Gaussian filters
(DOG - difference of Gaussian) to identify interest
points with invariance in scale and rotation;
• Locate keypoints: the places where there is an ex-
treme, sets a detailed model to determine the exact
location and scale. Keypoints are selected based on
stability measures;
• Define orientation: set up an orientation for each
keypoint using the local image gradient and;
• Describe keypoints: measure the local gradient of each
keypoint, based on their neighborhood.
These measurements are transformed to a representation
that is tolerant to significant levels of distortion and
changes in lighting. In possession of the keypoints captured
from two images, they are able to be compared with
Euclidean distance calculation and find its correspondent.
Fig. 1. User Interface.
Therefore, it is possible to trace points in images, even in
real time and make the mosaics’ composition.
B. Depth calculation using Digital Image Processing C/C++, using the program C++ Builder from Borland,
A stereo vision system is scheme in which two images running on Windows XP, as well as simulation platform
of the same scene are acquired under different conditions of MATLAB as we see on Figure 1.
in order to be able to make measurements in three dimen-
sions. Animal’s vision system accomplish this task easily,
that is why animals generally have two eyes. Therefore,
the system used is shown in Figure 3 and Figure 4. It
consists of two webcams with focal length f equidistant
from one source at a distance b/2. Cameras from the
left and right are called l and r, respectively [1]. When
viewing an object, every point of its visible contour has
a coordinate (x, y, z) in physical space in relation to the
original localization between the left and right lenses. To
perform the coordinates x, y and z measurement are used
the following equations (1).

(xl + xr )/2 (yl + yr )/2 f

x=b ;y = b ;z = b (1)
(xl − xr ) (yl − yr ) xl − xr

Therefore, beyond the distance between the cameras and

the focal length f of its optical assemblies, it requires
the coordinates xl and xr of a point to measure its z
depth. The challenge is how to find the given point in
the previewed object’s contour. It needs to locate the
point in the image obtained with the left camera and
the same point in the image from the right camera. A
classic way to make this relationship between xl and xr
is using correlation. Another useful way today is the use
of SIFT method, and this is more robust because of its
characteristics. Invariance to translation, rotation, scale,
and luminous intensity of the pixels are the characteristics Fig. 2. Process fluxogram.
[1].
III. Methodology A stereo system images acquisition must be developed
with two cameras and specific routines for images captur-
After bibliographic review, SIFT method’s algorithms ing. The depth calculation process can be seen in Figure 2.
are chosen and implemented. The language chosen is
 DEPTH CALCULATION USING COMPUTER VISION AND SIFT 255

Routines for distance measurement should be imple-

mented from the SIFT method and the acquired images.
The system calibration is made by using pattern images
such as checkerboards with distances, inclinations and
rotations known. Tests using standard images, controlled
environments and complex should be also made. From the
results obtained, the system is evaluated and the results
and evaluations must be published in scientific events.
IV. Results
To perform the preliminary tests with the implemented
program, two cameras are used with a distance of 120 mm
from center to center. Two images are captured, Figure
3 and 4, one from the left camera and one from the
right respectively. It is observed in these figures that there Fig. 4. Picture taken from the right camera.
are some main objects: the computer monitor and the
paper holder, the first is nearer to the camera than the
second. When applied the SIFT method to these images
and made a comparison between the keypoints found in
each image, we obtain the results shown in Figure 5. The
Figure 6 shows the 5 subregions. The result is the average
of each keypoints’ depth distance in the 5 subregions.
They are: left = 650.08mm; right = 550.24mm, top =
331.95mm, bottom = 542.09mm, center = 550.24mm. The
focal length f of the cameras used is unknown, but it might
be calculated experimentally. The distance b between the
cameras used is 120mm. It is able to be seen with the result
that the subregion where the monitor is, the subregion in
the center, has an average depth lower than the subregion Fig. 5. Result of transformed SIFT in the first application.
where the paper holder is, the right subregion. This entire
process occurs in 5.8 seconds and can be observed through
a graphical user interface in Figure 3. obtained in the first test. There are some times which the
program does not find keypoints in certain parts of the
images. This implies a zero value in one or more subregions
displayed in the interface.

Fig. 3. Picture taken from the left camera.

It was realized a second test, two more pictures from the

same scene were captured and were processed with by the Fig. 6. Defined subregions.
system. The obtained results from the second test are: left
= 535.10mm; right = 730.59mm, top = 321.28mm, bottom V. Conclusion
= 481.55mm, center = 554.77mm. We obtain the numbers This work aims to develop a system based in computer
from the distances and this make possible to conclude that vision to obtain the depth distance. It is developed in
the system it is working because we moved the cameras MATLAB with two cameras of same focal length and
away from the scene and the numbers are higher than the spaced with a known distance.
256 NETO ET AL.

Using the system we captured two pictures from the Acknowledgment

cameras and prosecuted the images and the SIFT to get We thank to Elder Teixeira.
the distances. The obtained results are very promising and We thank to the laboratories LEM and ITTI from IFCE-
consistent with the expected. The simulation approaches Instituto Federal de Educação, Ciência e Tecnologia do
to the real situation. It occurs because the acquired dis- Ceará and the laboratory LESC from UFC-Universidade
tances are very close to the original. There were realized Federal do Ceará.
a few tests with the system. In these tests we changed
the main objects in the scene and changed the distances References
between the cameras and the objects. These changes was [1] SIEGWART, R. and NOURBAKHSH, I. R. Introduction to
shown in the interface, as we saw in Figure 3. The numbers Autonomous Mobile Robots, Bradford Book, 2004.
[2] ANGELI, A., DONCIEUX, S., MEYER, J.-A. and FILLIAT, D,
were coherent with the reality. If we moved back the Incremental vision-based topological slam, Intelligent Robots and
cameras, we obtained higher numbers in the interface. Systems. IEEE/RSJ International Conference, 2008.
If we moved just one side of the cameras, the numbers [3] LOWE, D. G. Distinctive image features from scale-invariant
keypoints. Int. J. Computer Vision, 2004.
changed just in the area we modified. [4] TOON GOEDEMÉ, MARNIX NUTTIN, T. T. e GOOL, L. V.
The next steps to be performed are to execute a quanti- Omnidirectional vision based topological navigation. Interna-
tative test to calibrate the measurement system and embed tional Journal of Computer Vision, 2007.
[5] VALGREN, C., LILIENTHAL, A. e DUCKETT, T. Incremental
the system in a mobile robot. topological mapping using omnidirectional vision. Intelligent
Robots and Systems, 2006 IEEE/RSJ International Conference,
2006.
 Dynamic of New Technologies in Switching Power
Devices IGBTs
Christian Ayala Esquivel
Jorge Ochoa Somuano
Isidro Moctezuma Cantorán
Universidad del Mar Campus Puerto Escondido (UMAR)
Carretera San Pedro Mixtepec-Sola de Vega, Cd. Universitaria, Puerto Escondido, Oaxaca,71980 México
{christiancae,ochoa,moctezuma}@zicatela.umar.mx

Abstract: This paper presents the results from considered a hybrid between a bipolar transistor and
a study on the dynamic new technologies that a field effect transistor (MOSFET) obtaining the best
present under various conditions when the features of both. The IGBT copied the best features
current varies in hard switching, these data were of MOSFET: simple control circuit, safe working
obtained in response to use as a case study of the area and high peak currents; and the bipolar
isolation gate bipolar transistor, which is a power transistor [1].
semiconductor device.
II. MANUFACTURING TECHNOLOGIES
I. INTRODUCTION
Currently, in the market there are three IGBTs
The proper selection of a power electronic device, from manufacturing technologies: Punch-Through,
one must consider aspects such as cost, the parasitic Non-Punch-Through and Field-Stop [9] - [11].
elements and the power dissipated. This latter aspect
directly influences the efficiency of the converter, A. Technology “Punch-Through (PT)”
especially topologies that used hard switching on the This technology IGBT has a non-homogeneous
switches. Therefore, correct evaluation of the losses structure. In this device fabrication technology is
generated in the type of converter and its operating built on a (p+) type substrate thickness (≈300m).
conditions are very important to optimize the circuit The (n-) layer is obtained by epitaxial growth and is
design. The information collected for the assessment relatively thin. “Fig. 1” shows a very thin (n+) layer
of the manufacturer's data sheet, is often very limited between the base and emitter of the internal pnp
or incomplete and represents approximate values, named buffer, which limits the expansion of the
taking into account the tolerances of the devices space charge region in a state of Punch-Through
commonly present. Another drawback stems from blocking effect and reduces the emitter injection in
the fact that these data do not lend themselves to conducting.
comparison between devices from different Gate Gate Gate
manufacturers, since they were obtained under Emitter
Metal
Emitter
Metal
Emitter
Metal

different operating conditions, temperature, current

Oxide Oxide Oxide
Poly Si Poly Si Poly Si

level, gate resistances and so on. The comparison of n+

various technologies is in terms of conduction losses p+

n-
and switching. In this paper, we analyzed only the n+ FS

case of hard switching, since it generates the most

losses during transients. Among the power electronic Collector
Collector

devices, the IGBT of insulated gate is currently the Collector

DSEP power semiconductor device with the widest
range of applications and possibilities for
Fig.1: Internal structure of different IGBT technologies (PT, NPT and FS).
optimization in the future. The IGBT can be

257
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_45, © Springer Science+Business Media B.V. 2010
258 ESQUIVEL ET AL.

The PT (IGBT) reaches low conduction losses C. Technology “Field-Stop (FS)”:

(low resistance) through a high emitter coefficient of The FS (IGBT) technology was developed for
internal BJT, low-switching losses through a applications that require a saturation voltage
reduced lifetime [1]. This produces a carrier “VCE(sat)”. This technology was recently launched by
distribution with high slope (dp/dx) in the (n-) and a the Infineon Technologies Company (formerly
transient off current with large amplitude that Siemens Company).With this structure, conduction
decreases rapidly and depends largely on the “VCE(sat)” losses will by reduced, making the (n-)
lifetime. A device with these characteristics is the even thinner (≈120 μm) through the integration of an
International Rectifier IRG4PH50S [11]. additional layer (n+) layer called Field-Stop (similar
to a buffer in a PT structure) as shown in "Fig. 1".
B. Technology “Non Punch-Through (NPT)” The initial problem for the manufacture of the
The NPT (IGBT) has a structure technology device was handling very thin wafers. A device with
originally developed by the company Siemens. This these characteristics is the IKW25T120 from
device is built in a homogeneous (n-) type substrate Infineon Technologies [9].
of approximately (220μm) wide. The issuer is
developed through the implementation of a very thin III. NEW GATE STRUCTURES
layer (p+) doped and low (transparent emitter) on
the back of the substrate, as shown in "Fig 1". Lately it has been applied to all technologies, a
Therefore, the IGBT is made homogeneous by the new gate structure "trench-gate” instead of the
modulation of the base resistance through a low conventional planar gate, as shown in "Fig. 2".
emitter coefficient in combination with a very long
carrier lifetime. These characteristics lead to a
homogeneous carrier distribution in the (n-) zone.
Therefore, with the switch in the off position the
space-charge displaces a significant portion of the
charges stored on the side of the collector of internal
(pnp) blocking voltages still small by checking one
of the principal differences with the PT technology.
Moreover, the stored charges of the emitter side of
the internal (pnp) become dislodged through
recombination at the transparent emitter surface.
Both effects mentioned ensure low switching losses
of NPT type IGBT without the use of lifetime
control. The structure was electrically characterized
by a transient current in the tail-shaped switch in the
off position of reduced amplitude that decays slowly
Fig. 2: Planar gate structure and trench.
and is almost invariable with the temperature and
does not show the effect of lifetime [1]. This is the This structure helps to reduce conduction losses,
most appropriate technology for high voltage, so as the MOSFET channel is shaped vertically to the
most manufacturers now produce the NPT with chip surface and requires a less active area. This will
breakdown voltages of 1200V. Furthermore, increase the density of cells. In short, the benefits
homogeneous IGBT is very robust, since it has a obtained from the trench-gate structure including a
good performance under hard conditions and under decrease in the MOS channel resistance and the
short circuit conditions (safe operating square area), elimination of the JFET region [2]-[8]. The gate
a device of this type is the IGTG11N120CN from structure can be applied to any manufacturing
Fairchild [10], [2] - [5]. technology of MOS devices. The combination of
Field-Stop technology with trench gate structure led
to a new Trench Stop IGBT (TS).
 DYNAMIC OF NEW TECHNOLOGIES IN SWITCHING POWER DEVICES IGBTS 259

IV. SWITCHING MODES Depending on the time in which it occurs, there are
two types of short circuits:
The following describes the different switching Short circuit type I: the short circuit occurs during
modes that can be put a switch-controlled power the locking phase, before the device turn on "Fig. 3".
converter in an application [1]. Short circuit type II: the short circuit occurs
during the driving phase, before the device off in the
"Fig. 3".

V. TEST CIRCUIT
The test circuit used for the characterization in
hard switching converter is a “DC/DC” type gearbox
or also called chopper. The circuit diagram is shown
in "Fig. 4", including the device under test “DUT”
and the auxiliaries.

LC Rind

AUX1
Fig 3: Different switching converter. Dfw
Lcarga
1) Hard Switching: The switching of a circuit
breaker is defined as hard when there is VO
simultaneous presence of current and voltage in the
time period that lasts the switching phase. It can
DUT
occur in the ignition and off the device and is AUX2
D1
characterized by high losses "Fig. 3". IMP1 IMP2
2) Soft Switching: The switching of a circuit
breaker is defined as mild when the voltage or
current is zero at the start of phase switching. It can
occur in the ignition and shut off and is Fig 4: Test circuit for hard switching.
characterized by low losses "Fig. 3". The soft
The “AUX1” auxiliary device works on one side a
switching is classified into:
free movement diode in the normal operation tests
Zero voltage switching or ZVS: This type of
and on the other side as for short-circuit tests. The
switching occurs only during power when the
“AUX2” auxiliary device is placed parallel with the
voltage is zero before it starts to flow a current
device being tested and it circulates the initial
through the device.
current of the inductive load to the level of current
Zero current switching or ZCS: This type of
necessary for testing. Thus, the “DUT” leads the
switching occurs only during shutdown, when the
minimum time required and avoids the effect of
current is zero before the device starts to block a
internal heat losses by conduction (conduction
voltage. [3], [4].
current by saturation voltage), obtaining a better
3). Short circuit: The short is a type of switching
control junction temperature of the device. “LC” is
is not desirable in a converter, which in principle
the parasitic inductance wiring. “Rind” is inductance
corresponds to a hard switching on the ignition
resistance load. “Lcarga” is load inductance where the
under certain conditions the external circuit, being
energy is stored for the current test. “IMP1” and
the most frequent short-circuiting the load. This
“IMP2” are drivers that perform the action to control
switch is characterized by very high losses that lead
the activation of auxiliary devices under test.
to the device to the limits of its safe operation.
260 ESQUIVEL ET AL.

The nominal test circuit: collector current caused by the same reverse recovery diode containg
IC=25A, gate resistance RG=22Ω, voltage VO=600V, in the testbank causing the same effect. This displays
wiring inductance LC=116nH, gate voltage an increase in power and the energy dissipated as it
VGG=15V and junction temperature Ti=30°C. switches to the on position. Infineon Technologies in
hard switching consumed to the maximum power in
VI. EXPERIMENTAL RESULTS testing (6mJ) as well as International Rectifier.
While Fairchild only consumed (5mJ).
In this section we present the results in three 800 80

devices with hard switching characterized by 600 60

varying the parameters of the IC current. Energies

Vce [V]
400 40

Ic [A]
are shown only in the ignition and shutdown, which 200 20

were calculated by integrating the instantaneous 0 0

power dissipated during the switching transient. -200

0 0.5 1
(1) t:0.5us/div
1.5 2
-20
0 0.5 1
(2) t:0.5us/div
1.5 2

“Fig. 5” shows tensions and currents of VCE, IC 40 6

respectively as well as the instant potency calculate 30 5

as a product of this and the instantaneous power is 4

Pon [KW]

Eon [mJ]
20

calculated as a result of these and the energy

3
10
2

obtained by integration of the power, see “Fig. 5”. 0

1

800 80 -10 0
0 0.5 1 1.5 2 0 0.5 1 1.5 2
(3) t:0.5us/div (4) t:0.5us/div
600 60

Fig. 6: Behavior of the voltage VCE, current IC, power dissipated energy PON
Vce [V]

400 40
Ic [A]

200 20 and EON in the ignition hard switching, “IRG4PH50”.

0 0

-200 -20
Lastly "Fig. 7" shows the switching on the
0 0.5 1
(1) t:0.5us/div
1.5 2 0 0.5 1
(2) t:0.5us/div
1.5 2
International Rectifier were, where the collector-
40 8 emitter voltage VCE generating the same step
30
6
decreases due to parasitic inductance. This displays
an increased potency similar to that of Infineon and
Pon [KW]

Eon [mJ]

20
4
10
dissipated energy at the time of power up the device
from company International Rectifier. Note that the
2
0

-10
0 0.5 1 1.5 2
0
0 0.5 1 1.5 2 time scales are similar and can see very well the best
(3) t:0.5us/div (4) t:0.5us/div
benefits in switching the ignition.
Fig. 5: Behavior of the voltage VCE, current IC, power dissipated energy PON 800 80
and EON in the ignition hard switching, “IKW25T120”.
600 60

In "Fig. 5" the ignition switch is displayed, where

Vce [V]

400 40
Ic [A]

the collector-emitter voltage VCE generating step

200 20

decreases due to the parasitic inductance. The

0 0

-200 -20

inductance affects the slope of the current “di/dt”. 0 0.5 1

(1) t:0.5us/div
1.5 2 0 0.5 1
(2) t:0.5us/div
1.5 2

While the current spike is caused by the diode 40 8

reverse recovery is the test bench and on a current 30

6

plotted peak is noticed as a result. This displays an

Pon [KW]

Eon [mJ]

20
4

increase in power and the energy dissipated during 10

the on phase, on the device from Infienon

2
0

Technologies.
-10 0
0 0.5 1 1.5 2 0 0.5 1 1.5 2
(3) t:0.5us/div (4) t:0.5us/div
"Fig. 6" also demonstrates the Fairchild Company
Fig. 7: Behavior of the voltage VCE, current IC, power dissipated energy PON
power switching device, where the collector-emitter and EON in the ignition hard switching, “IGTG11N120CN”.
voltage VCE generating the same step decreases due
to the parasitic capacitance that is similar to above,
with a much greater slope. The current spike is
 DYNAMIC OF NEW TECHNOLOGIES IN SWITCHING POWER DEVICES IGBTS 261

The case of commutation when you want to turn a good behavior by having a relatively small tail off
off a device is different, see “Fig. 8, Fig. 9 and Fig. and has an energy of (2.3mJ) which is about half the
10”. energy the company of Infineon Technologies.
"Fig. 8" again this switching device Infineon, is 800 60

shown off the device showing the effect of parasitic 600

40

capacitance when the device off. Subsequent voltage

Vce [V]
400

Ic [A]
20

slope (dv/dt). This also leads to increased energy in

200

the off causing losses at high frequency. About the

0

-200 -20

maximum is (4mJ). (1) t:0.5us/div (2) t:0.5us/div

0 0.5 1 1.5 2 0 0.5 1 1.5 2

800 60
25 2.5

600 20
40 2

15
Vce [V]

Poff [kW]

Eoff [mJ]
400
Ic [A]

1.5
20
10
200
1
5
0
0 0.5
0

-200 -20 -5 0
0 0.5 1 1.5 2 0 0.5 1 1.5 2 0 0.5 1 1.5 2 0 0.5 1 1.5 2
(1) t:0.5us/div (2) t:0.5us/div (3) t:0.5us/div (4) t:0.5us/div

30 5
Fig. 10: Behavior of the voltage VCE, current IC, power dissipated energy PON
20
4
and EON in the ignition hard switching shutdown, “IGTG11N120CN”.
Eoff [mJ]
Poff [kW]

3
10

0
2
VII. CONCLUSION
1

-10
0 0.5 1 1.5 2
0
0 0.5 1 1.5 2 The experimental results on tests (hard-
(3) t:0.5us/div (4) t:0.5us/div
switching), the three technologies analyzed show
Fig. 8: Behavior of the voltage VCE, current IC, power dissipated energy PON that the ignition energies during on and off
and EON in the ignition hard switching shutdown, “IKW25T120”.
switching is different as the parameters vary. There
In "Fig. 9" shows the switching behavior in are differences in the level of impact it can have a
shutdown where the device is the most deficient, parameter variation in each of the devices, those
approximately consumes the energy (23mJ) of belonging to different technologies. Considering the
Fairchild Company, showing that this device can’t results, is necessary extend the experimental
be used at high frequencies have a very slow characterization from hard switching, changes in
shutdown. current, voltage, temperatures of union, gate
resistance and parasitic inductance. The optimization
800 60 of devices, increases efficiency and therefore energy
600
40
saving.
Vce [V]

400
Ic [A]

20
200
VIII. FUTURE WORK
0
0

-200
0 1 2 3 4
-20
0 1 2 3 4 Obtain the complete mathematical model of hard
(1) t:1us/div (2) t:1us/div
switching behavior and maximize the efficiency of
40 25
the tested power devices.
30 20
Poff [kW]

Eoff [mJ]

20 15

10 10
REFERENCES
[1] J.Aguayo., “Análisis Comparativo de Transistores IGBT
0 5

-10
0 1 2 3 4
0
0 1 2 3 4 tipo PT y NPT en Diferentes Modos de Conmutación”,
(3) t:1us/div (4) t:1us/div
Tesis de Maestría, CENIDET, Cuernavaca, México, 2000.
Fig. 9: Behavior of the voltage VCE, current IC, power dissipated energy PON [2] H. Ruething, F. Umbach, O. Hellmund, P. Kanschat,
and EON in the ignition hard switching shutdown, “IRG4PH50S”. G.Schmidt, “600V-IGBT3: Trench Field Stop Technology
in 70μm. Ultra Thin Wafer Technology”, Memorias de la
Unlike earlier in the "Fig. 10", is shown device conferencia ISPSD, 2003, p.
off from International Rectifier, this device presents
262 ESQUIVEL ET AL.

[3] T. Laska, G. Miller, M. Pfaffenlehner, P. Türkes, D. Stop IGBTs”, IEEE Trans. Industry Applications, vol.
Berger, B. Gutsmann, P. Kanschat, M. Münzer, “Short IA-39/4, p. 922-928, 2003.
Circuit Properties of Trench-/Field-Stop-IGBTs – Design [7] R. Kraus, M. Redding, K. Hoffmann, "The Short-Circuit
Aspects for a Superior Robustness” Memorias de la Behaviour of IGBTs Based on Different Technologies",
conferencia ISPSD, 2003, p. 1-4 Memorias de la conferencia, EPE, 1995, p. 1_157-1_160.
[4] H. Iwamoto, H. Kondo,S. Mori, J. Donlon, A. Kawakami, [8] S.Azzopardi,C. Jamet, J.M. Vinassa, C. Zardini,
“An Investigation of Turn-off Performance of Planar and "Switching Performances Comparison of 1200V Punch-
Trench-Gate IGBTs under Soft and Hard Switching”, Through and Punch-Through IGBTs under Hard-
Memorias de la conferencia IAS, 2000, p. 1-6. Switching at High Temperature", Memorias de la
[5] R. Mallwitz,R. Tschirbs,M. Pfaffenlehner, A. Mauder, C. conferencia IEEE PESC, 1998, p. 1201-1207.
Schaeffer, “1700V Trench IGBT Modules”, Memorias de [9] www.infineon.com
la conferencia PCIM, 2001, p. 1-6. [10] www.fairchildsemi.com
[6] X.Kang, A. Caiafa, E. Santi, J. Hudgins, R. Palmer, [11] www.irf.com
“Characterization and Modeling of High-Voltage Field-
 Anisotropic Percolating Pathways in the Thin Films
of Polymeric PEDT/PSS Complex and their Relation
to the Electrical Conductivity as Revealed by
the Mesoscale Simulation
Toomas Kaevanda, Jaan Kaldab, Andres Öpika, Ülo Lillea1
a
Department of Material Sciences, Tallinn University of Technology, Ehitajate tee 5, 19086 Tallinn, Estonia
b
Institute of Cybernetics, Tallinn University of Technology, Akadeemia tee 21, 12618 Tallinn, Estonia

chains of beads) in a grid. The percolation phenomenon has

Abstract. The MesoDyn simulation code supplemented been treated on a discrete lattice and also in the continuous
with the percolation and cluster analysis concepts has been space, and the relevant parameters of the correlated
used to generate the subcritical percolating networks of conducting network of the PEDT particles have been
PEDT beads of the coarse-grained PEDT/PSS complex in a calculated [34-35].
range of PEDT/PSS mass ratio up to 20. Based on the The aim of this research is 1) to establish the degree of
comparison of the percolation parameters in three anisotropy of the percolation pathways and its dependence on
perpendicular directions, the anisotropy of the networks the composition of the complex; 2) to determine the
has been determined and its correlation with the relationship between the percolating current-carrying
macroscopic electrical conductivity has been evaluated. backbone density (strength) and the electrical conductivity;
3) to clarify the limits of the used simulation approach.
I. INTRODUCTION This is not an easy task, keeping in mind that the percolating
The term “percolation pathway” has been often used in the pathways are wispy, tenuous, insubstantial fractal objects [36]
interpretation and generalization of the experimental data in that have no characteristic length scale and furthermore, little
the research of the transport properties of electrically is known about anisotropic percolation [37].
conducting conjugated polymers, e.g. [1-3]. However, to our In this report, we demonstrate the anisotropic nature of
best knowledge, there are no reports devoted to the analysis of percolating pathways in a wide range of the composition of the
this topic (except for our efforts as shown below). At the same complex, and discuss their relation to the macroscopic
time, the respective studies of the carbon nanotubes are electrical conductivity.
progressing [4-5].
Poly(3,4-ethylenedioxythiophene) (PEDT) blended with
poly(4-styrenesulphonate (PSS) is one of the most intensively II. METHODOLOGY
studied inherently electrically conducting conjugated polymer, The methodology of the mesoscale simulation in a 323 cubic
and its thin films are widely used in several industrial applica- lattice using the MesoDyn simulation code (from Accelrys
tions [6-11]. The properties of these films depend on their Inc. Material Studio Library Bundle, MS 4.4) and the
morphology, which, in its turn, is a function of the PEDT/PSS treatment of the generated morphologies in the framework of
mass ratio and of the processing conditions. The morphology the isotropic percolation concept is described in detail in our
and the factors affecting it have been studied by a wide range previous reports [34-35]. The topology of the PSS
of physical methods, including electrical measurements, and mesomolecules corresponds to that of random copolymers (if
supported by atomic scale modeling [12-28]. not stated otherwise). Due to the phase separation in the
Current study is largely motivated by a recently found PEDT/PSS systems, the lamellar anisotropic structures are
convincing evidence on the highly anisotropic nature of the formed. However, we are forced to use isotropic percolation
electrical conductivity of the PEDT/PSS complex (1:6 by model, due to the lack of well-developed anisotropic models.
weight) [2,29 and citations therein]. The main conductivity To facilitate the further reading, some details on the
mechanism in the thin PEDT/PSS film at 300K is the variable generation of the percolation pathways are given below.
range hopping (VRH) of the charge carriers from one Our self-developed cluster analysis algorithm is used to
conducting center to another center [29-30]. generate the subcritical percolating networks. The lattice
In this report, the authors continue their earlier efforts to site occupation probability p (not distinct from the volume
simulate the behavior and the transport properties of the fractions of the PEDT beds) is taken to be very close to the
PEDT/PSS complex using dynamic density field theory (DDF) effective percolation threshold p*c, , at the distance Δp =
[31-32] and principles of classical percolation theory [33]. The p*c - p. The procedure is as follows. First, MesoDyn is used
simulated morphologies have been given in the form of the to generate a density field d(X,Y,Z) on a 323 lattice (i.e. each
density fields as ordered systems of the coarse-grained cell with coordinates X, Y and Z is characterized with a
polymer molecules (mesomolecules represented as Gaussian certain density d). Further, a cite (cell) is postulated to be

1
E-mail addres [email protected].
Ü. Lille thanks Estonian Science Foundation for financial support (Grant
No 6633).

263
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_46, © Springer Science+Business Media B.V. 2010
264 KAEVAND ET AL.

occupied, if its density is higher than (1 - y)dmax, and vacant TABLE I

otherwise; here, y is a control parameter, and dmax is the PSS/PEDT mass ratios (S/T), volume fractions of the PEDT beads (θT)
and macroscopic electrical conductivity (σ) S cm-1.
maximal density. The parameter y is increased slowly, until 1)from the data [3,19] using correlation function ln σ = 4.1738 lnθ + 2.9248, R2 = 0.961.2) PSS is
the global conductance along the surviving sites appears; at modeled as an alternating copolymer; 3) as in 2, but unassociated PSS chain is ca 3-fold longer; 4)
under shear 0.001 ns-1, repulsive interaction 2.6 RT; 5) doping level 0.03.
this moment, the critical density dc = (1 – y)ddmax is It S θT 10-3
recorded. The density interval |dmax − dc| ÷ dmax will be em /T xσ
referred to as the conducting window. The accuracy of the 1 1 0. 600
parameter y (4 significant digits) defines the value of the .4 495
2 2 0. 400
sub-criticality parameter Δp = 10-4 . Finally, the obtained .5 333
subcritical network of the PEDT beads is used to determine 3 4 0. 201)
the number of cells in the conducting window (Ncon) and in .2 193
the percolating pathway (Np). Removing the “dead ends” 4 6 0. 5
143
one gets the number of cells in the charge carrying 5 8 0. 21)
backbone Nb. The “dead” ends are the parts of the infinite 110
cluster connected to it only by one site (they lead nowhere). 6 1 0. 11)
Ratio of these three values to the total number of cells gives 0 091
the (effective) critical probability p*c and the densities of 7 2 0. 0.06
0 05
the percolation path P(p), and of the backbone P(p)b. 8 2 0. 400
Similarly, one can calculate the fraction of the singly 2)
.5 33
connected (“red”) sites fr. In principle, all current passes 9 2 0. 400
3)
through these “red” sites (however, see below section III. .5 33
1 2 0. 400
C).
04) .5 33
The percolation path is a cluster (effectively infinite) 1 2 0. ?
connecting the opposite sides of the 3-D lattice. Due to the 15) .5 33
system proximity to the percolation threshold, the correlation
length is certainly larger than the lattice side, i.e. our systems
are scale invariant (or self-similar) [37-38]. We generate the Item 11 is introduced to give an idea of how the change of the
percolation path for all three directions X ,Y, Z independently. doping level affects the results. This item is not included into
In what follows, the respective pathways are termed as the the correlation analysis shown below. In general, its behavior
percolation paths for the first, second and third direction; the is rather similar to that of others.
sub-indices of the parameters designate the used direction. Note that in all experiments, the macrophase separation of
Note that the first-appearing path does not necessarily always the neutral host mesomolecule (unassociated PSS) is observed
correspond to the same Cartesian coordinate axis. (as evidenced by the order parameter values over 0.1, except
for the item 7). The respective values of the guest PEDT beads
III. RESULTS remain in the range of 0.005-0.01, characterizing the
III. A. INITIAL SYSTEMS AND THEIR MACROSCOPIC microphase separation. The behavior of the polyanionic PSS
ELECTRICAL CONDUCTIVITY beads remains between these two extremes.
For the studied systems, the PSS to PEDT mass ratio
(termed as S/T) ranges from 1.4 to 20; the corresponding III. B. CHARGE CARRIERS HOPPING AND THE DISTANCE
PEDT volume fraction θT varies from 0.495 to 0.05. In this set BETWEEN THE CONDUCTING SITES
of experiments, the doping level is kept constant (0.25, i.e. one The hopping mechanism implies that the charge carriers
PSS anion per four PEDT cycles if not stated otherwise). The tunnel between the conducting sites. Neglecting the
macroscopic electrical conductivity (σ) of the studied systems fluctuations of the barrier height, the tunneling probability
ranges from ca 1 to 6×10-5 S cm-1 (Tab. 1) [3,19]. These data (and the local conductivity) decreases exponentially with the
are correlated with the θT values by the linear equation distance between the conducting sites. Therefore, there are
expressing the exponential function σi = A exp(bθT): very strong fluctuations of the local conductivity, and the
ln σ = 19.27 θT – 8.651; (items 1-7) R2 = 0.86 (1); global (macroscopic) conductivity along some axes (e.g. X)
pay attention that b>0. Naturally, ln σ versus S/T ratio can be estimated as the “bottleneck” conductivity σc [33], i.e.
correlates with b < 0. Such an empirical law reflects the fact such a largest conductivity σc that the region σ(x,y,z) > σc
that the macroscopic conductivity values range in a very wide percolates in the direction of the given axes (X).
(exponentially wide) range of values. It should be also stressed As argued above, the local conductivity is proportional to
that the morphologically induced conductivity variations for the exponent of the distance between the conducting sites,
systems with a fixed S/T value are ignored by this law. which in its turn is proportional to d-1/3, and at the bottleneck
site, to dc-1/3.
This “bottleneck model” reflects the physical nature of the
problem with a reasonable accuracy. However, it is still an
approximation, for which the resistance of the sites, other than
 ANISOTROPIC PERCOLATING PATHWAYS IN THE THIN FILMS 265

the bottleneck site, is completely neglected. In reality, Pay attention to the fact that these correlation analysis
however, this is not correct. Another drawback of the results are relatively insensitive with respect to the change of
bottleneck model is that for small samples (like our 323), the the fitting parameter (dc vs 〈Δd〉). Further, the values of the
statistical fluctuations of the bottleneck value are rather large. exponential factors for different directions are rather close and
Therefore, it can be argued that the average hopping length, as therefore, the expected high anisotropy, i.e. high ratio of the
reflected by the average field density within the conduction conductivity in parallel to the substrate surface plane (║, in-
window 〈Δd〉, can serve in our case as an alternative measure plane) and perpendicular to it (┴, out-of-plane) cannot be
with a lesser statistical variance. To summarize, dc is the lower identified. An exceptional case is the item 4 (S/T = 6), for
bound of the conducting window, and 〈Δd〉 is the median value which this ratio achieves a value of about 25. Thus, these field
of it. density values (dc and 〈Δd〉) are unable to reflect the electrical
Hence, it is expected that the global conductivity is anisotropy in the set of studied systems. In what follows we
proportional to exp(bdc-1/3), or to exp(b〈Δd〉-1/3), respectively, argue that this fact can be explained by the imperfections of
where b is an unknown constant; recall that dc and Δd are the the bottleneck model and that the strong anisotropy of the
bottleneck field density and the average density of the global conductivity can be ascribed to the strong anisotropy of
conducting window of the PEDT-beads, respectively. the topological structure of the percolation clusters at
criticality (which is, indeed, observed in our simulations).
TABLE II .
Maximal (dmax) and critical (dc) field densities in the perpendicular III. C. ANISOTROPY OF THE PERCOLATION PARAMETERS AND
directions 1, 2, 3. THE MACROSCOPIC ELECTRICAL CONDUCTIVITY.
It dmax dc Below are compared the percolation parameters
em 1 2 3 characterizing the subcritical networks generated for the
1 0.9 0.8 0.7 0.7 samples with different S/T ratios (see Tab III).
542 09 960 91 The relatively high p*c values give an evidence that a lot of
2 0.8 0.6 0.6 0.6
452 38 012 01 cells are located in the finite clusters. A rather modest
3 0.6 0.4 0.3 0.3 anisotropy is observed.
845 07 971 97 One can see that there is a considerable anisotropy for P(p),
4 0.5 0.3 0.1 0.0 as evidenced by the fact that the values for the first two
347 97 930 812
directions are noticeably lower than in the case of the third
5 0.4 0.0 0.0 0.0
301 794 778 763 direction. This means that more sites are needed to circumvent
6 0.3 0.0 0.0 0.0 the insulating PSS beads, i.e. in order to percolate in the third
536 694 694 691 direction.
7 0.1 0.0 0.0 0.0 Remarkable is that P(p) values are rather close to the P(p)b
191 495 495 494
8 0.5 0.4 0.4 0.4
values. Hence, the subcritical network has very few dead ends.
73 67 66 43 Note also that according to our previous report [34], P(p)
9 0.5 0.4 0.4 0.4 behaves according to a power law, P(p) ∝ |p−pc|β, where the
55 56 54 33 critical exponent for the infinite cluster density β >1. We also
1 0.5 0.4 0.4 0.4
0 41 47 21 21
showed there that our systems correspond to the correlated
percolation, and the power law of P(p) is observed up to p = 1.
The simulation data (see Tab. II), indeed, confirm the This is important, because in the real systems p − pc. can take
expected correlation, for all the three directions (items 1-10): non-small values.
ln σ = -4.8767 dc1-1/3 +4.3383, R2 = 0.8215 (3), The most anisotropic quantity is P(p)b : over the full range
ln σ = -5.1979 dc2-1/3 + 5.0938, R2 = 0.90 (4) of the mass ratios and simulation conditions one can observe
ln σ = -4.138 dc3-1/3 + 4.0129, R2 = 0.65 (5) the dependence of P(p)b on the direction. The average value of
Note that excluding the special cases 8,9, and 10 improves the the charge-carrying backbone strength in the third direction
correlation for the 3rd direction, resulting in σ ∝ exp(-4.5045 exceeds that of in the first direction ca four-fold.
dc3-1/3) and R2 = 0.89. Being guided by this finding, we also studied the correlation
Pay attention that b < 0, i.e. ln σ decreases when dc-1/3 between P(p)b1 and the macroscopic conductivity; a
increases (which corresponds to decreasing dc). moderately correlated linear relationship was found (items 1-
Similarly good (even slightly better) correlation is observed 10):
when the average conducting window density 〈Δd〉 is used ln σ = -84.79 P(p)b1 - 0.707, R2 = 0.51 (9).
instead of dc (items 1-7): This correlation improves somewhat, if the discordant item
ln σ = -7.107 〈 Δd 〉c1-1/3 +5.5708, R2 = 0.87 (6) 5 is discarded, resulting in
ln σ = -132.4 P(p)b1 +0.3641 R2 = 0.62. (10)
ln σ = -7.358 〈 Δd 〉c2-1/3 +6.071, R2 = 0.90 (7)
Pay attention that b < 0, thus, the conductivity decreases very
ln σ = -7.4394 〈 Δd 〉c3-1/3 +6.2799, R2 = 0.91 (8)
rapidly with the increase of P(p)b1 values (e.g. 3-6 orders of
Qualitatively analogous relationship (however, just for a
magnitude at 4-fold increase of the backbone density).
single direction) was observed in our recent report [39].
266 KAEVAND ET AL.

Although Eqs. (9) and (10) reflect just a modest correlation, respectively. From the point of view of the experimental data,
let us have a look on the consequences of them. Particularly these are quite reasonable results. A special case is the item
remarkable in Tab. III is the item 4 (S/T = 6, as was used in 10: a shear applied in the X direction, resulting in the equality
[29]), where the ratio 〈P(p)b1,2〉/P(p)b3 = 7.5 corresponds to of backbone densities in 2nd and 3rd directions, clearly
〈σ1,2〉/σ3 = 1018, i.e. the out-of-plane conductivity is differing from that of the first one, similarly to the item 9 (in
vanishingly small. Items 2 and 9 also contrast with the the latter, the unassociated PSS chain is ca 3-fold longer as
remaining items: the backbone strength in the directions 1 and compared to the conventional one). Note that shear (the
2 is very similar and differ strongly from that of the third processing tool to introduce an anisotropy) and subtle changes
direction, leading to the 〈σ1,2〉/σ3 values of 102 and 107, in topology increase the anisotropy.

TABLE III
Percolation functions in perpendicular directions (1, 2, 3) at various PSS/PEDT mass ratios (S/T) at the doping level 0.25 (if not shown
otherwise) and temperature 298K.
1) PSS is modeled as an alternating copolymer; 2) as in 1), but unassociated PSS chain is ca 3-fold longer; 3)under shear 0.001 ns-1, repulsive interaction 2.6 RT,; 4) doping level 0.03; 5) average
from 3 parallel experiments.
Item, 10-2 x p*c 10-2 x P(p) 10-2 x P(p)b
(S/T) 1 2 3 1 2 3 1 2 3
1,(1.4) 6.26 8.24 9.19 2.47 3.69 5.17 2.23 3.45 4.87
2, (2.5) 10.2 12.5 14.8 2.25 2.95 13.4 2.15 2.76 12.8
3, (4.2) 12.6 14.0 14.0 2.76 10.5 10.5 2.56 9.95 9.98
4, (6.0) 6.67 23.8 40.4 1.63 8.12 37.9 1.44 10.4 37.75)
5, (8.0) 35.5 37.4 39.5 9.27 10.7 20.7 9.24 10.7 20.3
6, (10) 36.8 35.8 37.7 6.15 6.53 15.9 5.97 6.14 15.4
7, (20) 25.5 25.5 27.0 7.36 7.36 11.6 6.10 6.13 9.95
81),(2.5) 6.97 7.14 16.6 3.69 4.72 12.3 3.05 4.02 11.4
92),(2.5) 6.10 6.76 12.6 0.82 1.71 10.5 0.65 1.44 9.86
103),(2. 7.86 19.0 19.0 1.22 10.2 10.2 1.0 9.42 9.48
5)
114),(2. 12.9 15.2 16.7 5.1 6.0 10.5 5.02 5.92 10.4
5)
Aver. 15.4 19.0 23.1 3.76 6.65 14.8 3.44 6.45 14.2
Stdev. 10.3 9.29 10.5 2.30 2.73 6.01 2.63 3.31 8.74

The previous paragraph 1) demonstrates an approximate densities (Tab. IV). Hence, not all red sites act as connectors
character of the correlations ln σ vs P(p)b [given by Eqs. of finite clusters into the infinite one, i.e. part of them do not
(9),(10)], and 2) clearly reflect the conductivity anisotropy. take part directly in the formation of the subcritical network.
Let us now study the structure of the charge carrying Therefore the role of the red sites in ambiguous.
backbone (some of them are shown in Fig.). The P(p)b values
are ca one order of magnitude higher than those of fr. (see Tab. TABLE IV
III and IV). Hence, the backbone contains a number of Fractions of the singly connected sites (fr) in the perpendicular directons
-3
multiconnected pieces which provide several routes in parallel Item 10 x fr
1 2 3
(loops). This is because the red sites connect the subsets of the 1 5.19 7.51 10.3
finite clusters into the infinite one. Note that the ratio of the 2 3.75 3.84 18.1
masses of the percolation path to the overall mass of all the 3 4.61 15.6 15.4
cluster pieces is ca ¼ for the direction 1 and 2/3 for the 4 2.01 4.70 7.48
direction 3 (compare the average values of P(p) and p*c in 5 1.95 1.37 7.96
6 4.58 6.32 8.88
Tab. III). These structural features may result in reducing the 7 17.8 17.7 24.4
correlation strength ln σ vs P(p)b. 8 5.19 7.51 10.3
Note that the average value of the fractions of the red bonds 9 2.78 4.94 17.9
in directions 1 and 2 is lower in comparison to that of in the 10 4.00 14.8 14.8
third direction, i.e. the behavior is similar to the backbone Average 5.19 8.43 19.4
 ANISOTROPIC PERCOLATING PATHWAYS IN THE THIN FILMS 267

A first two directions are related to the ║ conductivity and the

third one to the ┴ conductivity. Note that the ║ conductivity
constitutes the main part of the macroscopic (global)
conductivity.
In summary, we have treated in the framework of the
isotropic percolation concept the density fields of the PEDT
B
beads evolved as driven by the chemical potential and
Langevin diffusion in the coarse-grained PEDT/PSS system.
The obtained results have given evidence on the presence of
the anisotropic percolation pathways in the wide range of the
composition in the thin film of the complex. It is quite
C remarkable that such a simulation reflects the anisotropy
which in fact originates from the centrosymmetric forces in
the spin coating process used for the preparation of the thin
film [29]. Furthermore, we have used the classical isotropic
percolation concept and multiple occupancies characteristic
for the correlated systems have been neglected [43]. Therefore
Figure. Charge carrying backbones in 3-D lattice in the the DFT supplemented with the classical percolation concept
directions 1,2,3 (from the left to the right). is a valuable tool for the studies of the transport properties of
A – item 2 (see Tab 3); B – item 4; C – item 10. the conjugated polymers electrostatically bound to a
polyanion. As the further tasks one can consider a more
detailed study of the model (11) (including the analysis using
The paragraphs above are based on the random resistor the Kirhoff’s laws for the network of hopping-model
model [40] (an equivalent of the hopping conductivity model resistances, see above) and the research into the morphological
[41-42]). However, in subcritical neworks used in this report changes induced by the annealing at different temperatures
the variations of the conductivity by orders of magnitude taking into account the temperature dependence of the
cannot be explained within such a model alone. The most interaction and diffusion parameters. From the other side the
accurate model would be to ascribe to each cell the resistivity proper treatment of the systems on different doping levels this
based on the hopping model and to solve the system of methodology should be complemented with quantum-
Kirhoff’s laws. This will be the subject of our further studies. mechanical calculations.
Meanwhile, the reality can be approached reasonably well by The very recent application of the most effective physical
combining the bottleneck model with the random resistor methods, e.g. magnetoresistance measurements at low
model: the large variations induced by the S/T values are temperatures had shed light to intimate details of the electrical
mostly explained by the bottleneck conductivity and the conduction process in conjugated polymers, e.g. distinguishing
random resistor model leads to a prefactor in the form of the role of the isolating PSS layer and intrinsic disorder in the
P(p)b, explaining the anisotropy. Such an interpretation PEDT particles. Such details are out of grasp of the mesoscale
explains also the modestness of the correlations given in the simulation consisting itself other models and approximation
equations (9) and (10).Therefore, the results presented above (coarse- graining and Rouse dynamics of polymer chain, the
[the presence of a good correlation of the global conductivity use of the mean field etc).
versus exp(dc-1/3), together with the strong anisotropy of the
percolation backbones, as evidenced by P(p)b] suggest that the
overall dependence takes the form of
σ = AP(p)b exp(bdc-1/3), (11); CONCLUSION
hence ln σ – ln P(p)b1 = bdc1-1/3 + ln A (12), Anisotropic percolation pathways have been shown to be
and a good correlation is observed (items 1-7): present in the thin films of intrinsically electrically conducting
ln σ – ln P(p)b1 = -5.220 bdc1-1/3 + 7.8522, R2 = 0.91 (13). conjugated polymer PEDT/PSS complex. For a wide range of
This is a logical result because (12) normalizes the the PEDT/PSS mass ratio, evidence has been given that the
macroscopic conductivity to the density of the charge-carrying electrical conductivity correlates with the characterizing
backbone in this way smoothing the fluctuations of the measures of these pathways. A model which explains the large
calculated data. variations of the macroscopic conductivity induced by the
In all experiments, the percolation pathway densities for the PEDT/ PSS mass ratio and its directional dependence on the
first and second directions are lower than in the case of the substrate surface has been proposed. The applicability limits
third one. Naturally, this morphological inhomogeneity causes of the DFT theory, complemented with the percolation
the same feature in the electrical conductivity. Therefore, the concepts, have been evaluated.
268 KAEVAND ET AL.

REFERENCSES [21] J. Ouyang; Q. Xu; C. Chu; Y. Yang; G. Li; J. Shinar, "On the
mechanism of conductivity enhancement in poly(3,4-
[1] M. Kemerink; S. Timpanaro; M. M. deKok; E. A. Meulemkamp; ethylenedioxythiophene):poly(styrene sulfonate) film through solvent
M. J. Touwslager, "Three-dimensional inhomogeneities in PEDOT: PSS treament," Polymer vol. 45, pp. 8443-8450, 2004.
films," J.Phys.Chem.B vol. 108, pp. 18820-18825, 2004. [22] M. M. De Kok e. al, "Modification of PEDOT:PSS as Hole
[2] A. M. Nardes; M. Kemerink; R. A. J. Janssen, "Anisotropic Injection Layer in Polymer LEDs," in Physics of Organic Semiconductors,
hopping conduction in spin-coated PEDOT : PSS thin films," Physical Review W.Brütting, Ed.; WILEY-VCH: 2005, p 451-473.
B vol. 76, pp. 1-7, 2007. [23] C. Ionescu-Zanetti, Mechler, A., Carter, S.A., Lal,R.,
[3] D. M. DeLongchamp; B. D. Vogt; C. M. Brooks; K. Kano; J. "Semiconductive Polymer Blends: Correlating Structure with Transport
Obrzut; C. A. Richter; O. A. Kirillov; E. K. Lin, "Influence of a water rinse on Properties at the Nanoscale," Adv. Mater. vol. 16, pp. 385-389, 2004.
the structure and properties of poly(3,4-ethylene dioxythiophene):poly(styrene [24] P. Vacca; M. Petrosino; R. Miscioscia; G. Nenna; C. Minarini; D.
solfonate) films," Langmuir vol. 21, pp. 11480-11483, 2005. Della Sala; A. Rubino, "Poly(3,4-ethylenedioxythiophene): poly(4-
[4] B. E. e. a. Kilbride, "Experimental observation of the scaling laws styrenesulfonate) ratio: Structural, physical and hole injection properties in
for alternating current and direct current conductivity in polymer-carbon organic light emitting diodes," Thin Solid Films vol. 516, pp. 4232-4237,
nanotube composite thin flms.," Journal of Applied Physics vol. 92, pp. 4024- 2008.
4030, 2002. [25] A. M. Nardes; R. A. J. Janssen; M. Kemerink, "A morphological
[5] S. S. Rahatekar; M. S. P. Shaffer; J. A. Elliott, "Electrical. model for the solvent-enhanced conductivity of PEDOT : PSS thin films,"
percolation in carbon nanotube dispersions: A mesoscale modeling and Advanced Functional Materials vol. 18, pp. 865-871, 2008.
experimental study," Smart Nanotextiles vol. 920, pp. 63-69 [26] M. Deepa; S. Bhandari; R. Kant, "A comparison of charge
151, 2006. transport behavior in functionalized and non-functionalized poly 3,4-
[6] S. Ghosh; O. Inganas, "Self-assembly of a conducting polymer (ethylenedioxythiophene) films," Electrochimica Acta vol. 54, pp. 1292-1303,
nanostructure by physical crosslinking: applications to conducting blends and 2009.
modified electrodes," Synthetic Metals vol. 101, pp. 413-416, 1999/5. [27] H. Yan; S. Arima; Y. Mori; T. Kagata; H. Sato; H. Okuzaki,
[7] S. Kirchmeyer; K. Reuter, "Scientific importance, properties and "Poly(3,4-ethylenedioxythiophene)/poly(4-styrenesulfonate): Correlation
growing applications of poly(3,4-ethylenedioxythiophene)," Journal of between colloidal particles and thin films," Thin Solid Films vol. 517,
Materials Chemistry vol. 15, pp. 2077-2088, 2005. pp. 3299-3303, 2009.
[8] L. B. Groenendaal; F. Jonas; D. Freitag; H. Pielartzik; J. R. [28] A. Dkhissi; D. Beljonne; R. Lazzaroni, "Atomic scale modeling of
Reynolds, "Poly(3,4-ethylenedioxythiophene) and its derivatives: past, present interfacial structure of PEDOT/PSS," Synthetic Metals vol. 159, pp. 546-549,
and future " Adv.Mater. vol. 12, pp. 481-494, 2000. 2009.
[9] X. Crispin; e. al, "Conductivity, morphology, interfacial chemistry, [29] A. M. Nardes; M. Kemerink; R. A. J. Janssen; J. A. M.
and stability of poly(3,4-ethylene dioxythiophene)-poly(styrene sulfonate): A Bastiaansen; N. M. M. Kiggen; B. M. W. Langeveld; A. J. J. M. van Breemen;
photoelectron spectroscopy study," J.Polym.Sci.B vol. 41, pp. 2561-2583, M. M. de Kok, "Microscopic understanding of the anisotropic conductivity of
2003. PEDOT : PSS thin films," Advanced Materials vol. 19, pp. 1196-1200, 2007.
[10] A. B. Kaiser, "Systematic conductivity behavior in conducting [30] C. S. S. Sangeeth; M. Jaiswal; R. Menon, "Correlation of
polymers: effects of heterogeneous disorder," Adv.Mater. vol. 13, pp. 927- morphology and charge transport in poly(3,4-ethylenedioxythiophene)-
941, 2001. polystyrenesulfonic acid (PEDOT-PSS) films," Journal of Physics-Condensed
[11] F. Louwet; L. Groenendaal; J. Dhaen; J. Manca; J. Van Luppen; E. Matter vol. 21, 2009.
Verdonck; L. Leenders, "PEDOT/PSS: synthesis, characterization, properties [31] J. G. E. M. Fraaije; B. A. C. van Vlimmeren; N. M. Maurits; M.
and applications," Synthetic Metals vol. 135-136, pp. 115-117, 2003. Postma; O. A. Evers ; C. Hoffmann ; P. Altevogt; G. Goldbeck-Wood, "The
[12] K. Z. Xing; M. Fahlman; X. W. Chen; O. Inganas; W. R. Salaneck, dynamic mean-field density functional method and its application to the
"The electronic structure of poly(3,4-ethylene-dioxythiophene): studied by mesoscopic dynamics of quenched block copolymer melts," Journal of
XPS and UPS," Synthetic Metals vol. 89, pp. 161-165, 1997. Chemical Physics vol 106, pp. 4260 - 4269, 1997.
[13] K. E. Aasmundtveit; E. J. Samuelsen; L. A. A. Pettersson; O. [32] G. H. Fredrickson In International series of monographs on
Inganas; T. Johansson; R. Feidenhans'l, "Structure of thin films of poly(3,4- physics; Clarendon Press ;Oxford University Press: Oxford, New York, 2006,
ethylenedioxythiophene)," Synthetic Metals vol. 101, pp. 561-564, 1999. pp. 19-22, 196-199.
[14] G.Greczynski; Th.Kugler; M.Keil; W.Osikowicz; M.Fahlman; [33] D. Stauffer; A. Aharony, "Introduction to Percolation Theory,"
W.R.Salaneck, "Photoelectron spectroscopy of thin films of PEDOT-PSS Taylor&Francis: 1994; Second Edition., pp. 89-114.
conjugated polymer blend: a mini-review and some new results," J.of Electron [34] T. Kaevand; J. Kalda; A. Öpik; Ü. Lille; Springer: 2009; Vol.
Spectroscopy and Related Phenomena vol. 121, pp. 1-17, 2001. Advances in Computer and Information Science and Engineering.
[15] L. A. A. Pettersson; S. Ghosh; O. Inganas, "Optical anisotropy in [35] T. Kaevand; J. Kalda; A. Öpik; Ü. Lille, "Correlated percolating
thin films of poly(3,4-ethylenedioxythiophene)-poly(4-styrenesulfonate)," networks in the thin film of polymeric PEDT/PSS complex as revealed by the
Organic Electronics vol. 3, pp. 143-148, 2002. mesoscale simulation," Macromolecules vol. 42, pp. 1407-1409, 2009.
[16] J. Y. Kim; J. H. Jung; D. E. Lee; J. Joo, "Enhancement of electrical [36] R. Zallen; John Wiley Sons, Inc.: New York, Chicester, Weinheim,
conductivity of poly(3,4-ethylenedioxythiophene)/poly(4-styrenesulfonate) by Brisbane, Singapore, Toronto, 2004, pp. 135-251.
a change of solvents," Synthetic Metals vol. 126, pp. 311-316, 2002. [37] D. ben-Avraham, Diffusion and Reactions in Fractals and
[17] S. K. M. Jonsson; J. Birgerson; X. Crispin; G. Greczynski; W. Disordered Systems, vol. Cambridge: Cambridge University Press, 2004,
Osikowicz; A. W. Denier van der Gon; W. R. Salaneck; M. Fahlman, "The pp. 19-31..
effects of solvents on the morphology and sheet resistance in poly(3,4- [38] V. G. Gantmakher; Electrons and Disorder in Solids Clarendon
ethylenedioxythiophene)-polystyrenesulfonic acid (PEDOT-PSS) films," Press Oxford, 2005, p 206-211.
Synthetic Metals vol. 139, pp. 1-10, 2003. [39] T. Kaevand; A. Öpik; Ü. Lille; T., S., Ed.; Springer: 2008; Vol.
[18] A. Zykwinska; W. Domagala; A. Czardybon; B. Pilawa; M. Advances in Computer and Information Science and Engineering, p 540-546.
Lapkowski, "In situ EPR spectroelectrochemical studies of paramagnetic [40] S. Kirkpatrick, "Percolation and conduction," Reviews of Modern
centres in poly(3,4-ethylenedioxythiophene) (PEDOT) and poly (3,4- Physics vol. 45, pp. 574-588, 1973.
butylenedioxythiophene) (PBuDOT) films," Chemical Physics vol. 292, [41] V. Ambegaokar; P. I. Halperin; J. S. Langer, "Hopping
pp. 31-45, 2003. conductivity in disordered systems," Physical Rev.B vol. 4, pp. 2612-2620,
[19] R. R. Smith; A. P. Smith; J. T. Stricker; B. E. Taylor; M. F. 1971.
Durstock, "Layer-by-layer assembly of poly(ethylenedioxythiophene): [42] B. I. Shklovskii; A. L. Efros, Electronic properties of doped
poly(styrenesulfonate)," Macromolecules vol. 39, pp. 6071-6074, 2006. semiconductors vol. Berlin Heidelberg New York Tokyo: Springer-Verlag,
[20] S. Timpanaro; M. Kemerink; F. J. Touwslager; M. M. De Kok; S. 1984, pp. 95-136.
Schrader, "Morphology and conductivity of PEDOT/PSS films studied by [43] R. K. P. Zia; Y. Wu; B. Schmittmann, "Percolation of a collection
scanning-tunneling microscopy," Chemical Physics Letters vol. 394, pp. 339- of finite random walks: a model for gas permeation through thin polymeric
343, 2004. membranes," Journal of Mathematical Chemistry vol. 45, pp. 58-64, 2009.
 Predicting Trading Signals of Sri Lankan Stock
Market Using Genetic Algorithms and
Neural Networks
M.M.K Dassanayake
Dr. Chandima Tilakarathne
Department of Statistics, University of Colombo,Sri Lanka
[email protected]

Abstract-This study predict the trading signals of Sri Lankan which corresponds to approximately 24% of the Gross
stock market using two sophisticated machine learning techniques Domestic Product of the country. The All Share Price Index is
called Genetic Algorithm (GA) and Neural Networks. These two the principal stock index of the CSE in Sri Lanka. ASPI
techniques in combination predict the direction (going up or not) measures the movement of share prices of all listed companies.
of the close price of tomorrow’s (day t+1) ‘All Share Price Index’ Hence, this study predicts the direction of ASPI on behalf of
(ASPI) of the Colombo Stock Exchange (CSE). The study period the Sri Lankan stock market.
considered was from 1st November 2002 to 31st December 2008. However, published researches related to the predictions of
The influential factors considered in this study represent the the CSE are very limited. Among the published work, the aim
intermarket influence, political and environmental factors, was to predict the value (price level), but not the direction. Past
economic stability and microeconomic factors: such as interest studies [1] showed the importance of directional prediction as
rate and exchange rate. A software called ‘genetic model’ was the predictability and the profitability of such predictions are
developed to find the optimum input variable combination that higher. Also there are no past studies related to predicting
will affect the direction of tomorrow’s ASPI value. Two identical ASPI using these techniques (Genetic algorithm and neural
neural network models called A and B were created for two network in combination) and this study may be the first of that
different time periods, to predict the direction of ASPI of day kind. Section II of this paper describes the methodology and
(t+1). the techniques applied. Section III includes data description
and data preprocessing. Section IV discuses how optimum
I. INTRODUCTION input variable combination will be selected using genetic
algorithm. Section V discusses how the directional prediction
Trading on the stock markets has long been considered as an can be performed using neural networks. At last, section VI
alternative investment which can make fast profits. Stock presents the conclusion, the limitations of the study and the
markets are very dynamic, where the functionalities and the improvements suggested.
value of transactions highly depend on the other local and
global factors as well as financial and non-financial factors. II. METHODOLOGY AND THE TECHNIQUES APPLIED
Hence, to obtain a profit through the stock market, it is
necessary to make intelligent trading decisions. In an attempt There are several variables that may affect the close price of
to gain advancement in decision making, there have been tomorrow’s ASPI [2]. From them several potential input
innovations in stock market trading technologies and tools. variables were selected based on inter-market influence,
The aim of the present study is to employ such technologies macroeconomic factors and news.
(Genetic algorithm and neural network in combination) to A preliminary analysis was carried out in order to identify
predict the direction of Sri Lankan stock market. Initially, the the association between these potential input variables and the
genetic algorithm will be employed to find the best set of input response variable. Graphical description, line charts and other
variable combination that will affect the direction of Sri statistical tests were incorporated, in order to identify their
Lankan stock market. Then neural networks will be used to trends. Then the variables with trends were preprocessed again
check the prediction ability of the best input variable in order to clarify the associations. The variables that have high
combination. impact on the tomorrow’s ASPI were selected as the ‘final
The Colombo Stock Exchange (CSE) is the main stock input variables’. Past data of these variables (six years) was
exchange in Sri Lanka. It is one of the most modern exchanges transformed in to binary status using relative return function
in South Asia, providing a fully automated trading platform. (1).Genetic Model was used to select the optimum input
As of 31 May 2008, 234 companies are listed on the CSE, variable combination out of theses final input variables that
representing twenty business sectors with a market will affect the direction of tomorrow’s ASPI. ‘Gens’ of the
capitalization of 850 billion rupees (over US$7.2 billion), chromosome [3] should be final input variables and the

269
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_47, © Springer Science+Business Media B.V. 2010
270 DASSANAYAKE AND TILAKARATHNE

‘individuals’ should be binary past data. A fitness function was Where,

used to assign fitness values to these individuals. Genetic ሾ௏಴ೄಶ ሺ௧ሻି௏಴ೄಶ ሺ௧ିଵሻሿ
model should undergo several generations until the stopping
ܴܴ஼ௌா ሺ‫ݐ‬ሻ ൌ (1)
௏಴ೄಶ ሺ௧ିଵሻ
criteria is met and the generation at the stopping criteria has the
highest fittest individual and that also will be the combination
of optimum input variables that will affect direction of Here ܴܴ஼ௌா ሺ‫ ݐ‬ሻ and ܸ஼ௌா ሺ‫ ݐ‬ሻ denotes the Relative Return
tomorrow’s ASPI. (RR) and the Close Price of the ASPI on day t, respectively.
Then these optimum variables were inputted in to the
Neural Network model to create a system that will predict the Relative Return of the all variables was considered in the
direction of tomorrow’s ASPI. Here the past data of the study when predicting the close price index of the tomorrow’s
optimum variables was divided in to three sets. The training set ASPI.
consists of the observations for six years except the last six
months. Data of the last six months was divided equality in to The qualitative variable NEWS, will be classified the overall
two sets for testing and validation. Finally, unseen data of three situation in Sri Lanka as negative or non-negative news in the
months were given to the trained NN model and Stimulate the sense that the news item of day t negatively affect on ASPI or
Output for Evaluation. not, as in the Criteria B,

III. DATA DESCRIPTION AND DATA PREPROCESSING Criteria B

The variables were used in the study represent the inter NEWS = 1 if there was non-negative news on the day t
market influence, microeconomic factors and news. The US NEWS = 0 if there was negative news on the day t
S&P 500 index, Indian BSE Senex index, the Amex oil index,
Exchange rate, Interest rate and news were considered as the The classification of the NEWS was done by considering the
potential influential variables that affecting the direction of the factors such as political situation, civil war situation,
close prices of the ASPI of day (t+1). The Stock market data environment factors, effective price changes of commodities
was downloaded from www.finance .yahoo.com website and and oil price change in day t.
the corresponding symbols for these stock markets are as
follows IV. GENETIC MODEL

• All Share price Index - ^CSE In this study GA is employed and created a Genetic Model
• US S&P 500 - ^GSPC with the intention of finding the optimum variable combination
• BSE Sensex -^BSE that will affect the direction of the close price of ASPI of day
• AMEX Oil Index -^XOI (t+1). Fig. 1, illustrates the structure of the Genetic model.

The macroeconomic data were obtained from the Central

Bank of Sri Lanka. Daily news papers were observed to gather
‘news’.
The study period was from 1st November 2002 to 31st
December 2008. This period covers 1573 daily observations of
the respective variable. Therefore, the data set is large enough
for modeling and short enough to exclude unnecessary impact
from the very past observations. Since different stock markets
are close on different holidays, the regular time series data sets
considered have missing values. If no trading took place on a
particular day, the rate of change of price should be zero.
Therefore the missing values were replaced by the
corresponding value of the last day. Since the interest was to
predict the direction of the close price of the ASPI, the
response variable (Y) was introduced as in the Criteria A.

Criteria A

Y=1 if ܴܴ஼ௌா ሺ ‫ݐ‬ሻ>0

Y=0 Otherwise Fig. 1. Structure of the genetic model.
 PREDICTING TRADING SIGNALS OF SRI LANKAN STOCK MARKET 271

The programming language java was used to implement the

model. This Genetic Model was divided into twenty main
coding modules before starting the implementation. Final
implementation of this Model was carried out by
“optimumstockvariables.java” module by integrating the entire
main and sub coding modules.
Seven potential input variables were selected for this study.
Therefore the number of genes in the chromosome was seven.
Table I illustrates the structure of the chromosome. Since the
relative return of the each input variable was taken, values of
the genes should be binary. So the upper bound should be 1
Fig. 2. Genetic Model Starting.
and the lower bound should be 0.
However, chromosome has no knowledge about the specific
problem given by the study, and hence has no intrinsic way of
deciding if one potential solution is any better than another
potential solution. That's where the fitness function comes in:
it's a single method that must implement, which accepts a
potential problem solution and returns an integer value that
indicates how good (or "fit") that solution is relative to other
possible solutions. The higher the number, the better the
solution. The lower the number (one being the lowest legal
fitness value), the poorer the solution. In this study GA model
was used these fitness measurements to evolve the population
of solutions toward a more optimal set of solutions. Equation Fig. 3. Result of the Genetic Model
(2), illustrates the fitness function of the model.

௣ ௣ ଵ
݂ሺ‫ܥ‬ሻ ൌ ቂ ቃቀ ቁሺ ሻ (2) It emphasis that the optimum input variable combination that
௣ା௡ ௉ ௡
affects the direction (up or down) of the close price of the
p is the number of positive samples the rule covers, n is the tomorrow’s (day (t+1)) ASPI is ‘NEWS’ and ‘ASPI Lag1’.
number of negative sample the rule covers, P is the total ‘ASPI Lag 1’ stands for the close price index of ASPI of day t
number of positive samples and C is the rule or the condition. and ‘NEWS’ Represents the environment, political and war
The rule C of the study was, whether the close price of the related information of the day t.
direction of tomorrow’s ASPI will go up for a particular input
combination. V. NEURAL NETWORK MODEL
The termination criterion chosen for the Genetic Model is
related with the number of generations formed. Thus when the Neural network (NN) model was employed to predict the
algorithm reaches a specific number of generations it stops and direction of the close price index of the tomorrow’s ASPI value
returns the chromosomes structure of the fittest individuals. using the optimum input variable combination given by
Max allowed evaluations for the GA model [3] can be Genetic Model.
changed, by changing the value of the The data set considered for the study contains data over six
“MAX_ALLOWED_EVOLUTIONS“ variable in the years. But sometimes the earliest time series data did not
“OptimumStockVariables.java” module. After several changes affects the current time series data. Therefore the data set was
of the value of the “MAX_ALLOWED_EVOLUTIONS“ divided in to two time windows. So the first window was, from
variable, it has been decided that the two hundred evaluations 4th November 2002 to 24th February 2005 and the second
are high enough for successful convergence. So by adding window was, from 25th February 2005 to 31st December 2008.
further high value for number of evaluations will only add Hence two identical neural network models were employed for
more computational time. Fig. 3, illustrates the result of the the predictions; namely NN Model A and NN model B. Past
Genetic model. data related to optimum variable combination contains 1572
samples. Those data was divided in to two for the two NN
models. Since there should be training and validation data sets
TABLE I to NN to perform well, two samples from those data sets each
CHROMOSOME OF THE GENETIC MODEL have hundred observations) were separated for validation and
testing.
CSE 1st Lag GSPC BSE EX_R NEWS IN_R XOI There were two layers in the neural network model and the
weight adjustments did in batch mode [4]. The training
272 DASSANAYAKE AND TILAKARATHNE

algorithm was back propagation algorithm and the error By observing Fig. 5 and 6, it is clear that the NN Model B
function was mean square error function [5]. Both neural gives a high performance than NN Model A (since same
network models (A&B) had these same properties. The Fig. 4, criteria was used for the both NN Models, normalization of the
will illustrate NN model structure in a graphical way. performance values was not necessary for model comparison).
Neural Network tool box of MATLAB R2007a software was After Models were created successfully, another unseen data
used to create the NN models. Data related to training, testing of three months were given to the network models to evaluate
and validation were fed into the models and the outputs of the the model. Output of this data were recorded using the
models were shown in the Fig. 5 and 6. stimulate option of the NN tool box of MATLAB.
For evaluation, unseen data of three months will be given to
the NN model and stimulate the output. Since the transfer
function called tansigmoid [4] was used output values will be
within the interval [-1 1]. So following criteria was employed
to normalize those values.

Criteria C

Y=1 if output > 0

Y=0 Otherwise

Fig. 4. Neural network model structure Where,

Y will represent the direction of the close price value

of ASPI of day (t+1).
Y=1 represent going up
Y=0 represent going down.
output will be the output given by NN Models.

Then those values were compared with the real ASPI

relative return values of the day (t+1) and calculated the
percentage of success. The Table II will illustrate the results.

According to the Table II values, it is clear that NN Model B

was more successful in predicting than NN Model A. But there
is only 2.5% difference between two models so one can argue
Fig. 5. Training output of the NN Model A that the successfulness is nearly same. Also the fact that both
models predict the negative results well rather than the positive
results cannot be neglected. Positive tab of Table II indicates
the percentage number of going ups (Relative Return of ASPI
=1) that the NN Model predicts successfully and Negative tab
indicates the percentage number of going downs (Relative
Return of ASPI =0) that the NN Model predicts successfully.

TABLE II
HIT RATES FOR THE TWO MODULES

Positive Negative Total

NN Model A 58% 63% 60.5%
NN Model B 61% 64% 62.5%
Both Models 59.5% 63.5% 61.5%
Fig. 6. Training output of the NN Model B
 PREDICTING TRADING SIGNALS OF SRI LANKAN STOCK MARKET 273

VI. CONCLUSIONS This study incorporated news which is a qualitative

representation of the Sri Lankan political situation, civil war
Results of the Genetic model reveal that the best set of input situation, environment factors, effective price change of
variable combination for predicting the direction of the close commodities and oil price changes. The information was
price of the ASPI of day (t+1) is news and close price of the gathered from the past daily news papers. Hence the direct
ASPI of day t. This study also highlighted that the considered impact of the news on the Sri Lankan stock market was
foreign stock markets, US S&P 500 and Indian BSE did not untraceable. This was a disadvantage to the study.
have significant impact on Sri Lankan Stock Market. Amex oil
index and exchange rate are also not significant according to Furthermore the news published on the news papers is not
the Genetic model. But interest rates are somewhat significant the only news which has an impact on the stock markets. The
and affects negatively on the close price of the ASPI of day behavior of the brokers’ manipulation, insider trading,
(t+1). management and policy changes of the companies and many
other reasons create news, which affect the stock market
Additionally the results of the study provide evidence to trading. Such information is unavailable to the general public.
believe the fact that time series data can be somewhat period
oriented. By considering all the evidence it is clear that the The genetic model or the system that was developed in this
news and the news related to civil war has a high negative study to identify the optimum input variable combination is a
impact on the close price of the Sri Lankan stock market. console application. But if this system may goes as a
commercial system user may be liked to have a graphical user
Finally, the combination of genetic algorithm and neural interface rather than a console interface. So it is good that one
network can be employed to create a system that can predict can improve this model by adding a graphical user interface.
the direction of tomorrow’s (day (t+1)) close price of ASPI to a
satisfactory extent and this system can be used for taking This conclusion is based on the considered study period and
investment decisions in Sri Lankan Stock Exchange. the input variables. However these results may differ according
to the nature of the market, the study period and also the input
A. Comparison with Other Similar Studies variables.

According to Reference [6] overall hit rate of the modified REFERENCES

Neural Network approach for predicting the close price index
[1] M. L. Leung, H. Daouk, and A. S. Chen, “Forecasting Stock
of ASPI is 59.2%. According to [1] the best prediction Indices: A Comparison of Classification and Level Estimation
accuracy for predicting Japan Nikkkei225 index is 68%, US Models,” International Journal of Forecasting, 16, pp. 173-190,
S&P 500 is 63% and UK FTSE100 is 61%. In this study the hit 2000.
rate is 61.5%. Also it is widely accepted that the ASPI is highly [2] A. Wijayanayake, and T. Rupasinghe, Identifying the Factors That
volatile and hence prediction is little bit difficult. Therefore the Affect the Sri Lankan Stock Market, Proceedings of the
outcome of this study is good enough when comparing with the International Sri Lankan Statistical Conference: Visions of
Futuristic Methodologies, 2004, pp. 381-388.
other similar studies.
[3] E. D. Goldberg, Genetic Algorithm in Search, Optimization, and
B. Limitations and Improvements Suggested Machine Learning, New York: Addison-Wesley, 1989.

[4] M. C. Bishop, Neural Networks for Pattern Recognition, New

This study was only limited to two microeconomic York: Oxford University Press, 1996.
variables: the exchange rate and the interest rate. There are
other macroeconomic variables which may have an impact on [5] K. Gurney, An Introduction to Neural Networks, London: UCL
Press, 1997.
the CSE. And there may be other factors which may directly or
indirectly influence the direction of the close price index of the [6] C. D. Tilakaratne, J. H. D. S. P. Tissera and M. A. Mammadov,
CSE. Due to the time limitations and the difficulties faced in Predicting Trading Signals of All Share Price Index Using a
Modified Neural Network Algorithm, Proceedings of 9th
acquiring data those factors were not considered. But if one International Information Technology Conference, Colombo, Sri
can include these factors the outcome might be more accurate. Lanka, 2008.
 An Automatic Measurement Algorithm for the
Diameters of Carbon Nanotubes by Using
Image Processing
Yoichiro Iwasaki1, Toshiyuki Nakamiya1, Ryosuke Kozai1, Fumiaki Mitsugi2, and Tomoaki Ikegami2
1
Department of Electronics and Intelligent Systems Engineering, Tokai University
9-1-1, Toroku, Kumamoto 862-8652, Japan
2
Graduate School of Science and Technology, Kumamoto University
2-39-1, Kurokami, Kumamoto 860-8555, Japan

Abstract- We propose an automatic measurement algorithm for irradiation. Therefore, this method cannot provide the
the diameters of carbon nanotubes (CNTs) electron microscopy variations of numerous diameters, and also cannot estimate
images by using image processing. In the algorithm, Otsu’s the uniformity of diameters in a wide measurement area.
method (discriminant analysis method) was employed to Recently, a few methods to measure automatically the
determine automatically the threshold for image binarization. characteristics of nanomaterials using image processing were
Proposed algorithm provides the numerous values of diameters proposed.
detected by horizontal and vertical scans for the binary images. A measurement method for the characteristics of
The diameter measurements using developed program were MWCNTs was proposed [3]. This method provides outer
carried out for the multi-walled CNTs (MWNTs) taken by a radius, inner radius, and a physical characteristic of
scanning electron microscopy (SEM). MWCNTs. However, the selected areas to measure MWCNTs
We have confirmed that the proposed scan algorithm detects are restricted by the image segmentation procedures.
well the parts of diameters in the wide measurement areas. We Therefore, it is not effective to evaluate the uniformity of the
have also confirmed that the measured diameters ware close to diameters of CNTs in a wide area. Other image analysis
the values estimated manually. So, we can say that the developed
method for nanocomposites contained CNTs was proposed
program is able to detect the diameters for the clearly observed
CNTs.
[4]. However, this method cannot be used to measure CNTs
We suppose that this method is useful to evaluate the
diameters.
diameters of the CNTs not only for the management of the CNTs
In this paper, we propose a diameter measurement
products but also for the optimization of CNTs preparation
algorithm for CNTs electron microscopy images by using
processes.
image processing. In the algorithm, Otsu’s method
(discriminant analysis method) [5] was employed to
I. INTRODUCTION determine the threshold for image binarization. Proposed
algorithm provides the numerous values of diameters detected
Since the discovery of CNTs [1], numerous researchers by horizontal and vertical scans for the binary images. The
have investigated on single-walled CNTs (SWNTs), multi- diameter measurement was carried out for the MWCNTs
walled CNTs (MWNTs), their growth methods, their SEM images by the developed program.
diameters, their lengths, alignment mechanism, electron
emission property, gas sensing property, and device II. PROPOSED ALGORITHM TO MEASURE THE
applications. DIAMETERS OF CNTS
It is very important to evaluate uniformity of diameters of
CNTs in order to increase the uniformity of the characteristics Fig. 1 shows a MWCNTs image taken by a SEM. The
of CNTs for the above mentioned applications. The diameters proposed program measures the diameters of CNTs in the
of CNTs are generally estimate manually using the image rectangle surrounded by four points P0-P3. The mixture area
taken by a scanning electron microscopy or a transmission of image and characters is not measured because each pixel
electron microscopy (TEM). These human visual works are cannot be distinguished between image and character. Before
extremely inefficient. the image processing is started, the white color frame is added
There is a method using radial breathing mode (RBM) of for the circumference of CNTs image. The preparations are
resonance Raman spectroscopy to estimate automatically the necessary to begin the detection for CNTs regions.
diameters of CNTs [2]. However, this method is effective
only for SWNTs and is no use to MWNTs, and estimates only
the average diameter in the region of the Raman laser

275
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_48, © Springer Science+Business Media B.V. 2010
276 IWASAKI ET AL.

P0(0, 0) P3(xmax, 0) Start

Input a CNTs 24-bit

color image


P1(0, ymax) Copy the image

Convert the image into 8-

bit grayscale image
Mixture area of Added white
image and color frame
characters
P2(xmax, ymax) Histogram equalization

Median filtering
Fig 1. A sample of CNTs image.

Start Binarization by Otsu’s

method

Input the scale unit

Closing operation

Image Processing
Opening operation

Measure the diameters of CNTs

Return

End
Fig 3. Program of the image processing.

Fig 2. Main program of the diameter measurement.

and between class variances for black and white classes are
used to determine the threshold of the pixel value as shown in
The program is mainly composed from the image Fig 4. We have confirmed that the Otsu's method is effective
processing and the measurement of the diameters of the CNTs for the binarization of CNTs images. The closing and opening
as shown in Fig. 2. operations are followed to remove the binarization noise.
At first, the captured CNTs image is treated the noise After the binarization of the CNTs image, the
reduction and binarization by the image processing. Fig.3 measurement of diameters of the CNTs is performed. The
exhibits the flow chart and the explanation of the image flow chart of the measurement program is shown in Fig. 5. In
processing. The captured image is converted to 8-bit this program, x(horizontal) and y(vertical) direction scans are
grayscale image, and the histogram equalization is carried out executed to search CNTs, and measure the diameters. Before
to emphasize the contrast. Then, the noise is reduced by the the scan, the lower and upper limits for the diameter should be
median filter. After that, the automatic binarization is set. The procedure how to measure the diameters is explained
performed using Otsu’s method. In the Otsu's method, within in Fig.6 using x-direction scan.
 AN AUTOMATIC MEASUREMENT ALGORITHM 277

Total pixel number: Nt

Mean value: mt
Start
Variance: t2

Black class White class

Loop 1
Pixel number: Nb Pixel number: Nw
Frequency

y=0; y<=y_max; y++

Mean value: mb Mean value: mw
Variance: b2 Variance: w2
Loop 2
x=0; x<=x_max; x++

0 Pixel value 255

Histogram
Pixel value No
Within-class variance: V within
2 N bV b2  N wV w2
f(x,y)=0(black)
Nb  N w
Between-class variance: Yes
N b mb  mt  N w mw  mt
2 2
V between
2

Nb  N w
Pixel value No
Separation metrics: V between
2
f(x-1,y)=255(white)
V 2
within
Threshold is given at the maximum of separation metrics.
Yes
Fig. 4 The algorithm of Otsu's method.
Measure CNT lengths in
five directions 1-5
Start

The minimum value of the five lengths is

Input the binary image regarded as the diameter


Set the parameters Change color of the diameter line in the

CNTs color image

Search CNTs regions in x-direction

Loop 2

Search CNTs regions in y-direction

Loop 1

Show the output image Return

Fig.6 Flow chart of searching CNTs regions in x-direction.

Return

Fig.5 Flow chart of measuring diameters of CNTs.

278 IWASAKI ET AL.

a CNT region
x x
Direction 3 

Direction 2

Direction 1
y y Direction 10
Direction 6
Direction 5

Direction 4 Direction 7 Direction 9

End counting
pixels and convert Direction 8 
the number of
Start counting pixels into the
pixels if new length unit.
CNT region is
detected.
(a) Horizontal scan (b) Vertical scan

Fig. 7 Explanation of the procedure for searching CNTs regions in x-direction (horizontal scan) and y-direction (vertical scan).

At first, the y axis is fixed, and x-direction scan starts. If

the pixel value at x is 0(black) and the pixel value at the x-1 is
255(white), the program recognizes that the x position is the
edge of a CNT, and starts counting black pixels in 5
directions of 0, 25, 45, 315, 335 degrees shown in Fig. 7(a).
The minimum length in the 5 measured lengths is regarded as
the diameter. The color of the pixels regarded as the diameter
are changed, and finally the convert from the number of
scale bar unit value
pixels to its real length is carried out. This process is repeated
by increasing y positions.
After x-direction scan, the scan direction is changed, and Fig. 8 Scale area in a CNTs SEM image.
y-direction scan is executed like the horizontal scan. In
vertical scan, 5 directions of 225, 245, 270, 295, 315 are used should be inputted as shown in Fig. 2. The both ends of the
as shown in Fig. 7(b). scale bar are pointed by a mouse, and the unit value is
Finally, the output image which indicates the detected inputted as shown in Fig.8. The input data are used to convert
diameters is shown. the numbers of pixels into the real lengths.
In this experiments, we decided that the scanned vertical
III. EXPERIMENTAL RESULTS length between P0 and P1 in Fig. 1 was set to 70 % of the
height of the image.
We have developed our algorithm with Visual C++ 2008 Table I shows the summary of results of diameter
and Computer Vision Library OpenCV [6]. measurements which include information such as number of
The program was utilized to images of two samples. succeeded diameter detection, the mean values of diameters,
After the start of the image processing, the scale unit and the standard deviations of diameters.
 AN AUTOMATIC MEASUREMENT ALGORITHM 279

TABLE I the parts of diameters in the wide areas. We

SUMMARY OF RESULTS OF DIAMETER MEASUREMENTS have confirmed that the measured values of

Sample 1 Sample 2 diameters ware close to the values estimated

manually. So, we can say that the developed
program is able to detect the diameters for
Number of x-direction 31,287 12,671 the clearly observed CNTs.
detected y-direction 47,651 18,628 In the experiments, we used a personal
diameters total 78,938 31,299 computer with an Intel Core i7 2.93GHz
Mean values of x-direction 20.8nm 23.6nm CPU and 6.0 GB RAM under Windows
diameters y-direction 19.5nm 23.7nm Vista 64-bit Operating System, and
total 20.0nm 23.7nm confirmed that the diameter measurement
Standard x-direction 11.6 12.9 process was finished within 7 seconds for
deviations of y-direction the sample 1.
11.9 12.6
diameters total 11.8 12.7
IV. CONCLUSIONS
width x height of image (pixels) 4,440 x 3,200 4,400 x 3,200
Threshold of Otsu's binarization 125 123 The developed program could detect the
diameters of CNTs, and calculate precisely the statistics such
as the mean values of diameters, the standard deviations of
As shown in Table I, the statistical values of detected diameters.
diameters in case of x-direction and y-direction scans are We suppose that this method is useful to evaluate the
almost same because the CNTs are randomly aligned. Fig.9 diameter of the CNTs not only for the management of the
shows two samples of CNTs images, and diameter detection CNTs products but also for the optimization of CNTs
results in which detected diameters are shown in gray. preparation processes.
We can see that the proposed scan algorithm detects well We can obtain more effective information for the depth

(a) Original image of sample 1 (b) Original image of sample 2

(c) Detected diameters shown in gray of sample 1 (d) Detected diameters shown in gray of sample 2

Fig. 9 Detection results for CNTs diameters.

280 IWASAKI ET AL.

profile if the depth information is given by other microscopy [3] C. Gommes, S. Blacher, K. Masenelli-Varlot, Ch. Bossuot, E. McRae,
A. Fonseca, J. -B. Nagy, and J. -P. Pirard, “Image Analysis
like 3-dimensional SEM.
Characterization of Multi-Walled Carbon Nanotubes,” Carbon, Vol. 41,
pp. 2561-2572, 2003.
REFERENCES [4] J. Ryszkowska, "Quantitative Image Analysis of Polyurethane/Carbon
Nanotube Composite Microstructures," Materials Characterization,
[1] S. Iijima, “Helical Microtubules of Graphitic Carbon," Nature, Vol. Vol. 60, pp. 1127-1132, 2009.
354, pp. 56-58, November 1991. [5] N. Otsu, "A Threshold Selection Method from Gray-Level
[2] A. Jorio, R. Saito, J. H. Hafner, C. M. Lieber, M. Hunter, T. McClure, Histograms,” IEEE Trans. Systems, Man, and Cybernetics, Vol. 9, No.
G. Dresselhaus, and M. S. Dresselhaus, “Structural (n, m) 1, pp. 62-66, 1979.
Determination of Isolated Single-Wall Carbon Nanotubes by Resonant [6] G. Bradski and A. Kaehler, Learning OpenCV, O'Relly Media,
Raman Scattering," Physical Review Letters, Vol. 86, No. 6, pp. 1118- September 2008.
1121, February 2001.


 Object Detection in Foveated Images
1 2
Abulfazl Yavari , H.R. Pourreza
1
Payam Nour university of Kashmar, Kashmar, Khorasan Razavi, IRAN
2
Ferdowsi University of Mashad, Mashad, Khorasan Razavi, IRAN
E-mails: Abulfazl_yavari @yahoo.com, [email protected]

Abstract – In this paper, we present a new visual attention sys-

tem which is able to detect objects in the images with non-uniform
resolution. Since, one of the goals of the visual attention systems is
simulating of human perception, and in human visual system the
foveated images processed, therefore, visual attention systems
should be able to identify the object to these images. We test the
system by two types of the images: real world and artificial im-
ages. Real world images include the cloth images with some defect,
and the system should detect these defects. In artificial images,
one element is different from the others only in one feature and
the system should detect it.

Index Terms – Visual Attention, foveated Images, object

detection

Fig.1 a lamp attract our attention (bottom-up attention)

I. INTRODUCTION
A. Visual Attention B. Foveated Images
The visual attention is a selective process that enables a per- Human vision system (HVS) is a space variant system [4]. It
son to act effectively in his complicated environment [1]. We means that by receding from the gazing point, the resolution
all frequently use the word "attention" in our everyday conver- gradually decreases and only the totality of the scene will sur-
sations. A definition for attention is given in [2]: "attention vive. Images that have this feature are called foveated images.
defines the mental ability to select stimuli, responses, and The area with the highest resolution is called the fovea [5]
memories". These stimuli can be anything, for instance, think (fig.3).
of conditions in which you are studying and concentrated on We can find the source of this behaviour, by studying the
the text you are reading. Suddenly you are interrupted with a eye's structure. There are two kinds of vision cells on the ret-
loud sound or you smell something burning and attract your ina, cone cells and rod cells. Cone cells are too much less than
attention. Similarly, there are stimuli in the environment that rod cells in number, but they are sensitive to color, and each
affect our vision, for example, a moving object, picture on the cell is individually connected to a nerve. Cone cells are gath-
wall, or a bright area in a dark room. These are examples of ered in the fovea area in the retina. Rod cells are too much
cases where without any predetermined goal automatically more than cone cells and they are in the area around the fovea.
attract our attention. This type of attention is called bottom-up Multiple rod cells are connected to a single shared nerve and
attention [3](fig. 1). they are sensitive to light [6].
There are other cases in which we are looking for a special Because humans have non-uniform vision and in fact their
object in the environment, and all the things that have similar brains perform special processing on foveated images, in this
features to that object, will attract our attention. Assume, for paper we concentrate on visual attention on foveated images.
instance, that we are looking for a red pen on a table. Anything Up to now, visual attention is studied only on normal images
with red color or with a shape like a pen will attract us, and so (uniform resolution).
we may find the desired object in the first or the next focuses.
This type of attention is called top-down attention [3] (Fig. 2).

281
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_49, © Springer Science+Business Media B.V. 2010
282 YAVARI AND POURREZA

which is finally routed to a central representation. This ap-

proach is strongly biological motivated. It is based on the Fea-
ture Integration Theory (FIT) [8]. This architecture is merely
bottom-up.
One of the earliest implementation of a visual attention sys-
tem was introduced by Milanese [9], [10]. It is based on the
Koch and Ullman model, and uses filter operations for the
computation of the feature maps. These model are especially
well-suited to be applied to real world scenes since the filter
operation provide useful tools for the efficient detection of
scene properties contrast or oriented edge. In this model as
feature Milanese considers two color opponencies – red-green
and blue-yellow --, 16 different orientations, local curvature,
and if no color information is available, intensity. To compute
Fig.2 when we looking for a red pen on the table, regions that have red the feature specific saliency, Milanese proposes a conspicuity
color, attract our attention (top-down attention)
operator which compares the local values of the feature maps
to their surround. This operator is motivated from the On-Off
and Off-On cells in the cortex and is also a common technique
for detecting contrast in images; it is usually referred to as Cen-
ter-Surround mechanism or Center-Surround difference. The
resulting contrasts were collected in so called conspicuity
maps, a term that was since then frequently used to denote fea-
ture dependent saliency. The conspicuity maps are integrated
into the saliency map by a relaxation process that identifies a
small number of convex regions of interest. The output of the
system is the saliency map that shows a few regions of interest.
A drawback of the system is its high computational complexity
that results from the many filter operations on different scales.
Another system is Neuromorphic Vision Toolkit (NVT) [10],
[11]. The Ideas of the feature maps, the saliency map, the
WTA, and the IOR in this system were adopted from the Koch
and Ullman model, the approaches of using linear filters for the
Fig.3 foveated image
computation of the features, and conspicuity maps were
adopted from Milanese. In this system features are color, orien-
Section 2 is dedicated to related works and we introduce tation, and intensity and all computations are performed on
what currently is performed in the field of visual attention. In image pyramids. Additionally, in this system use a weighting
section 3 we present our work with enough details and then the function for the weighted combination of different feature
tests and evaluation results are discussed in section 4. Finally maps by promoting maps with few peaks and suppressing those
we present the conclusions and our future works in section 5. with many ones. This technique is computationally much faster
II. RELATED WORK than the relaxation process of Milanese and yields good results.
The system contain several details that were chosen for effi-
The increased interest on research on visual attention to- ciency reasons or because they represent a straight forward
gether with the increased power of computers and the resulting solution to complex requirement. This approach may lead to
ability to realize complex computer vision system has led to a some problems and inaccurate results in several cases. For ex-
wide variety of computational systems on visual attention. In ample, the Center-Surround mechanism is realized by the sub-
this section we introduce several of the most important compu- traction of different scales of the image pyramid, a method that
tational systems on visual attention. is fast but very precise. In this system, the color space is RGB
The first computational architecture of visual attention was that represents colors differently to human perception, which
introduced by Koch and Ullman [7]. This model provided use- seem not appropriate for a system simulating human behaviour
ful algorithms for later implementations and for many current and leads to implausible results, too.
computational models of visual attention. The idea is that sev- The NVT in its basic version does concentrate on computing
eral features are computed in parallel and their conspicuities bottom-up attention. The need for top-down influences is men-
are collected in a saliency map. A Winner-Take-All network1 tioned but not realised. Navalpakkam and Itti introduce a deri-
(WTA) determines the most salient region (MSR) in this map, vation of their bottom-up model which is able to deal with top-
down cues [11]. The idea is to learn feature values of a target
1
A neural network that determines the most salient in a topog- from a training image in which the target is indicated by a bi-
raphical map. nary mask. The system learns the feature values from the dif-
 OBJECT DETECTION IN FOVEATED IMAGES 283

ferent feature maps on different scales. One difficulty with this For each feature we first compute an image pyramid from
approach is that it is not clear how bottom-up and top-down which we compute scale maps. These present saliencies on
cues compete. Since there is evidence that two distinct brain different scales for different feature types. The scale maps are
areas are associated with bottom-up and top-down mechanisms fused into feature maps representing different feature type and
in human perception [12], it might be useful to separate the these again are combined to conspicuity maps, one for each
processing also in a computational system. feature. Finally, the conspicuity maps are fused to a single sali-
The next attention system is Hamker. It aims mainly a mod- ency map, with the degree of brightness proportional to the
eling the visual attention mechanism of the human brain [14]. degree of saliency. From the saliency map, the most salient
Hamker's model, shares several aspects with the architecture of region (MSR) is extracted and the focus of attention (FOA) is
NVT: he computes contrasts for several features (intensity, directed there.
orientation, red-green, blue-yellow) and combines them in fea-
ture conspicuity maps. The conspicuity of these maps is com-
bined in a perceptual map that corresponds to the common sali- A. Feature computation
ency map.
The channel for the feature intensity extracts regions with
Another currently considerable system is VOCUS, intro- strong intensity contrasts from the input image. First, the color
duced by Frintrop [1]. It has many similarities to NVT system image is converted into gray-scale. From the gray-scale image,
and it uses feature maps and saliency and IOR, which have a Gaussian image pyramid with five different scales s0 to s4 is
been used also in many currently introduced systems. Our pro- computed. The intensity feature maps is computed by center-
posed system is very similar to this one from the implementa- surround mechanism extracting intensity differences between
tion point of view and utilizes many of its ideas. But as one of image region and their surround, Similar to ganglion cells in
the goals of vision attention systems is to simulate the human's the human visual system [12]. The cells are divided into two
perception and his vision system, and foveated pictures are types: On-center cells respond excitatorily to light at the center
used in his vision system, it’s important to note that this system and inhibitorily to light at the surround, whereas Off-center
is unable to handle foveated images, but our system is capable cells respond inhibitorily to light at the center and excitatory to
of locating the goal in such images. light at the surround [12].
III. SYSTEM DESCRIPTION According to the human system, we determine two feature
In this section we introduce details of our system that try to types for intensity: the On-center difference responding
find region of interest in foveated images. The architecture of strongly to bright region on a dark background, and the Off-
center difference responding strongly to dark regions on a
system shares the main concepts with the standard models of
visual attention, especially based on the model of Koch and bright background. Then the maps for each center-surround
Ullman. The implementation is roughly based on the VOCUS variation are summed up by across scale addition: first, all
maps are resized to scale s2. The maps are added up pixel by
(Visual Object detection with a Computational attention Sys-
tem) [1], one of the best known attention systems currently. pixel. After summing up the scales this yields 2 intensity maps.
However, there are several differences in implementation de- Similar, orientation maps are computed from oriented pyra-
tails and type of the input images (Foveated images). mids [10]. The oriented pyramid in fact consists of four pyra-
The structure of our visual attention system is shown in fig.4. mids, one for each of the orientation 0, 45, 90,135. The orienta-
Before we go into the details of the system, we present a rough tions are computed by Gabor filters detecting bar like feature,
according to a specified orientation. The orientation scale maps
overview of the structure, as illustrated in figure; the input of
the system is a foveated image, which may be generated by are summed up by across scale addition for each orientation,
software or hardware. We use software tools to generate im- yielding four orientation feature maps of scales s2, one for each
orientation.
ages with the fovea in the centre of the image. First, a pre-
processing is performed on the input image, which can be per- To compute the color feature map, the color image is firstly
formed in two ways: whether by using Gaussian filters with converted into an LAB-image; an LAB image pyramid is gen-
different sizes, so that the larger filters are used in the center of erated. From this pyramid, four color pyramids are generated
the image and as we move from the center to the periphery, the for the distinct colors red, green, blue, and yellow. On these
size of the filter decreases. Another way is to use a 3*3 Gaus- pyramids, the color contrast is computed by center surround
sian filter and use it multiple times in the center of the image mechanism. The maps of each color are rescaling to the scale
and reducing the number of its use as we go from the center to s2 and summed up into four color feature maps.
the periphery.
Next three different feature dimensions are computed: inten-
sity, orientation, and color. For each dimension, the saliencies B. Fusing saliency
are computed on different scales and for different feature types,
e.g. red, green, blue, and yellow for the feature color. Thereaf-
To combine the feature maps, we should decrease the effect
ter the maps are fused step by step, thereby strengthening im-
of less important maps. To do so, we use the formula
portant aspects and ignoring others.
284 YAVARI AND POURREZA

A. Experiment 1
X = X / m (1 )
In which X indicates the map, and m is calculated like be- This image is a tile image that in the top of it there is a black
low: for instance, for the feature of color, we first calculate the dot (fig.5).
number of pixels (Np) with a value more than the threshold t, As illustrated in figure, our system detected the defect, but
and then determine the minimum value (min), and m for every VOCUS not able to detect it. Since the resolution in the center
map is equal to Np/min. Finally, formula 1 is used for saliency of image is better than periphery, the VOSUS worked wrongly.
maps and by adding them pixel to pixel, the saliency map is
obtained. Now we should detect the salient areas in the saliency
map. This is possible through finding pixels with largest val-
ues. These regions are the region that attracts the attention.
We use a green circle to show this region on input images.

(a) (b)
Fig.5. (a) the defect is detected by our system (b) but not by the VOCUS

B. Experiment 2
This image is a cloth image that there is a vertical red defect
on it: the defect is detected by our system, but not by the
VOCUS (fig.6)

(a) (b)
Fig.6. (a) the defect is detected by our system (b) but not by the VOCUS

C. Experiment 3
This image is an artificial image (pop-out). There is a red
blob among the blue one. In this image only one feature is dif-
Fig.4 structure of the system ference: color (fig.7).

IV. EXPERIMENTS AND RESULTS

In this section, we present experimental results on real world
images and artificial images. We implement our system with
MATLAB programming language.
Our sample pictures include real images of cloth and tile
with some defects, and the system should detect these defects. (a) (b)
In artificial images, one element is different from the others Fig.7. (a) red bolo is detected by our system (b) but not by the VOCUS
only in one feature and the system should detect it. In each
experiment, we first show that VOCUS is unable to find the
defects and then show that our system is able to detect the de-
fects correctly. It's worth noting that the small green circle in-
dicates the approximate location of the defect.
 OBJECT DETECTION IN FOVEATED IMAGES 285

D. Experiment 4 ACKNOWLEDGEMENT
This image is also a cloth image, that there is a yellow defect The authors would like to thank Simone Frintrop for useful
on it (fig.8) comments on this paper.
REFERENCES
[1] Frintrop, S: VOVUS: a Visual Attention System for Object Detection
and Goal Directed Search. PhD thesis University of Bonn Germany.
January 9, 2006
[2] Corbetta, M.: Frontoparietal cortical networks for directing attention
and the eye to visual locations: Identical, independent, or overlapping
neural system? Proc. Of the National Academy of sciences of the
United States of America, 95:831-838. 1990
[3] Desimone, R. and Duncan, J.: Neural mechanism of selective Visual at-
tention. Annual reviews of Neuroscience, 18:193-222. 1995
[4] Coltekin, A.: Foveation for 3D visualization and Stereo Imaging. PhD
(a) (b) thesis Helsinki University of Technology Finland. February 3, 2006.
[5] Chang, E.: Foveation Techniques and Scheduling Issues in Thinwire Vi-
Fig.8. (a) horizontal line is detected by our system (b) but not by the VOCUS sualization. PhD thesis New York University. 1998.
[6] Gonzales, R. C. and Woods, R. E.: Digital image processing. Addison-
Wesley Publishing Company, 1992
[7] Koch, C. and Ullman, S.:Shifys in selective visual attention: Towards
V. CONCLUSION AND FUTURE WORKS the underlying neural circuitry. Human Neurobiology 4 (4, 1985) 219-
227.
We presented a new visual attention system which is ena- [8] Treisman, A. M. and Gelade, G.: A feature integration theory of atten-
tion. Cognitive Psychology 12 (1980) 97–136.
bling of detecting attentive areas in the images with Non- [9] Milanese, R.: Detecting Salient regions in an Image: From Bilogical
uniform resolution, something that no other currently existing Evidence to Computer Implementation. PhD thesis, University of Ge-
system is capable of. According to the standard attention model neva, Switzerland. 1993.
of Koch and Ullman, and VOCUS, we concentrate on the three [10] Milanese, R., Wechsler, H., Gil, S., Bost, J., and Pun, T.: Integration of
bottom-up and top-down cues for visual attention using non-linear re-
features (color, intensity, orientation), since they belong to the laxation. In proc. Of the IEEE Conference on Computer Vision and
basic feature of human perception and are rather easy to com- Pattern Recognition, pages 781-785. 1994.
pute. Of course, this can only approximate human behaviour [11] Itti, L., Koch, C. and Niebur, E.: A model of saliency-based Visual At-
where many other features attract attention, e.g., size, curva- tention for Rapid Scene Analysis. IEEE Trans. On PAMI 20, pages
1254-1259, 1998.
ture, and motion [1]. [12] Palmer, S. E.: Vision Science, Photons to Phenomenology the MIT
In future works, we intend to complete the top-down part of Press 1999.
the system. In addition, we should improve the system's speed [13] Greenspan, H., Belongie, S., Goodman, R., Perona, P., Rakshit, S., and
to make it possible for being used in real time cases. We will Anderson, C.: Overcomplete steerable pyramid filters and rotation in-
variance. IEEE Computer Vision and Pattern Recognition (CVPR),
also study addition of the motion feature, which has an impor- pages 222-228, 1994.
tant role in attracting the attention. [14] Hamker, F. Modeling Attention: From computational neuroscience to
computer vision. In:proc. Of WAPCV'04 (2004) 59-66.
 Digital Semi-Controlled Rectifier using a
Microcontroller
Emmanuel E. Rached, Adel I. Chit, Michel J. Owayjan
American University of Science and Technology, Beirut, Lebanon
Department of Computer and Communications Engineering
[email protected], [email protected], [email protected]

Abstract- Controlled rectifiers are an important aspect in power angle, separating the control circuit from the rectifier and
electronics; they are used to modify the average DC voltage at the accurately triggering the thyristors. The proposed solution has
output. Their use is widespread and they hold an important role three main parts. The first part provides the monitoring of the
in speed control of motors. Traditional analog control for these AC sine wave and zero crossing detection. Fig. 1 shows the
rectifiers was not always accurate and some required complex
designs. This paper presents a new low cost, high accuracy digital
circuit diagram of this part. The transformer steps down the
control method using a microcontroller. The microcontroller 220 VAC source to a more manageable 6 VAC voltage. Filter
keeps track of the current AC half-wave and triggers the capacitors are used on the both sides of the transformer. The 6
corresponding thyristors based on the desired time delay through VAC is further reduced to 3 VAC by the use of a voltage
the use of a potentiometer. Experimental tests showed a high divider. The 3 VAC is required so as not to damage the
accuracy and low response time in controlling the thyristor’s MCP601 operational amplifiers [6]. The sine wave is then fed
firing angle. to two MPC601 to provide a negative and positive reference
for the microcontroller.
I. INTRODUCTION
Unlike the more commonly used 741 Operational Amplifier
Most industrial applications use thyristor-based controlled which requires a 15 V source to operate reliably, the MCP601
rectifiers, especially where high voltages and currents are can perfectly operate with the same 5 V supply as the
involved [1, 2]. Rectifiers are used industrially to provide a microcontroller allowing the use of a single power supply.
variable DC voltage from AC sources to control the speed of The lower MCP601 takes as input the sine wave and, in the
motors [3]. They are also used in high voltage converter configuration shown here with the +5 V to Vdd, ground to Vss
stations to provide long distance transmissions. Other uses and comparing to ground, outputs a constant 5 V whenever the
include light dimming, power switching, and certain circuit sine wave is in the positive half-wave. This provides a logical
breaker applications [1, 2]. Traditional analog control of “1” (ON) to the microcontroller so it can reliably know that
thyristor-based rectifiers not only requires complex designs, the signal is currently in the positive half-wave.
but also lacks accuracy [4]. Digital control using a
microprocessor can improve the accuracy of such rectifiers
that can be used to drive a motor as a load [5].

In this paper we present a new low-cost, high accuracy

digital control method using a microcontroller. The proposed
solution that was implemented and tested is composed of three
sections. The first section monitors the AC supply voltage in
order to calculate the zero-crossings of the sine wave. The
second part consists of a PIC16F877A microcontroller which
calculates and controls the firing angle of the thyristors. And
the last stage provides isolation of the microcontroller from
the load. The details of the proposed solution are discussed in
the next section. Then, we present the tests and the results,
and end with some important conclusions and future work.

II. PROPOSED SOLUTION

Controlling the rectifier in a reliable manner requires

monitoring of the AC sine wave, the ability to detect the zero
crossing, monitoring the potentiometer for the desired firing Fig. 1. Sine wave monitoring

287
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_50, © Springer Science+Business Media B.V. 2010
288 RACHED ET AL.

Fig. 2. Operational Amplifiers output

When the sine signal is in the negative half-wave, the

Operational Amplifier cannot output a negative voltage since
its negative power supply is connected to ground thus making
its output a logical “0” (OFF) allowing the microcontroller to
know that the signal is no longer in the positive half-wave.
The upper MCP601 operates in the same manner except it
outputs a signal when the sine wave is in the negative half-
wave portion. The output of the MCP601s is a square wave
signal with a 5 V amplitude level and 50% duty cycle. Fig. 2
shows the output of both the MCP601s versus the 50 Hz AC
input.

The second stage controls the firing angle through the use a Fig. 3. Microcontroller configuration
microcontroller. A potentiometer connected between +5 V
and ground provides the microcontroller with the desired
thyristors’ firing angle. This connects to the first Analog to As seen from the graph, the firing angle increases from 0º to
Digital converter of the microcontroller. A full +5 V as input 180º as the voltage decreases from 5 V to 0 V. This was a
instructs the microcontroller to provide a trigger pulse at 0º design choice. The microcontroller’s Analog to Digital
and 180º for the positive and negative half-waves respectively, converter allows for 10 bits of precision translating into 210 =
effectively turning the controlled rectifier into a normal bridge 1024 possibilities, dividing the 180º into 1024 steps of
rectifier. On the other hand, a 0 V input results in the approximately 0.1758º each. This provides a very high
microcontroller outputting a trigger pulse at 180º and 360º for accuracy in controlling the firing angle. The 4 MHz crystal
the positive and negative half-waves respectively, effectively oscillator provides fast execution time for the microcontroller
negating any output and making it 0 V. The remainder of the and the software optimizations were performed based on it.
range allows precise control of the firing angle between 0º and
180º. The signals coming from the first stage are directed
towards pin number 17 for the negative portion and pin
number 18 for the positive part allowing the microcontroller to
correctly know in what half-wave the sine signal is currently
in. The trigger pulse is then outputted on pin 24 for the
negative half-wave and on pin 23 for the positive half-wave.
Fig. 3 shows the configuration of the microcontroller. The
other elements shown in the illustration above are standard
configuration for a PIC16F877A microcontroller [7]. The
crystal oscillator used is operating at 4 MHz. Fig. 4 shows the
potentiometer voltage versus the firing angle.
Fig. 4. Firing angle versus Voltage
 DIGITAL SEMI-CONTROLLED RECTIFIER USING A MICROCONTROLLER 289

The following simplified pseudo code summarizes the

microcontroller’s main BASIC code:

Input: negative_input(boolean)
positive_input(boolean)
delay = A/D potentiometer voltage

Output: negative_output(boolean)
positive_output(boolean)

While(positive_input = 1)
{
delay = GetADValue()
wait(delay)
positive_output = ON Fig. 6. Semi-controlled rectifier example
wait(duration)
positive_output = OFF In the configuration shown, the transistors’ output remains
} off until a pulse is sent by the microcontroller on the
corresponding pin, at which point the transistors switches ON
The negative half-wave’s pseudo code is similar to what is and remains in this state for the duration determined by the
shown above. microcontroller.

The third and final part of the solution provides separation An added simplicity for this solution is that all the active
between its various components. 1 k resistors placed elements in the circuit can be powered from a single +5 V
between the output pins of the microcontroller and the base of power supply, simplifying power requirements by allowing
the BJTs limit the current draw from the microcontroller to a the use a single 5 V power supply. A standard 5 V, Zener
maximum of 5 mA to prevent damaging it. Pulse transformers regulated power supply was also implemented to accompany
provide the necessary separation between the transistors and this solution. Fig. 6 shows an example of a semi-controlled
the thyristors as those will most likely be used in high current bridge rectifier with a motor as the load implementing the
circuit. Fig. 5 shows the diagram of this stage. The upper solution presented in this paper with the corresponding
part handles the negative half-wave while the lower one negative pulse and positive pulse indicated.
handles the positive half-wave. Freewheeling diodes are used
between the components to eliminate flyback. III. TESTS & RESULTS

The solution presented in this paper was tested on a 50 Hz,

220 VAC source. A PIC16F887A microcontroller was used
as the main controller. The only reason that this particular
model was chosen is that it was readily available. A cheaper
microcontroller could be used with a slight code modification
to achieve cost reductions. A standard 5 V Zener regulated
supply was also designed to power all active components of
the circuit. Initial testing required some refinement for the
code due to delays by the microcontroller in executing the
given instructions as different instructions require a different
number of cycles to complete. Table 1 shows the results prior
to the modifications while Table 2 shows them after the
modifications and optimizations were made.

TABLE 1
RESULTS PRIOR TO MODIFICATIONS
Desired Angle Actual Angle Difference
0º 3º ~300%
2º 4º 100%
25º 27º 8%
75º 76º 2%
135º 137º 1.5%
Fig. 5. Final Stage 180º 188º 4.5%
290 RACHED ET AL.

TABLE 2 REFERENCES
RESULTS AFTER MODIFICATIONS
Desired Angle Actual Angle Difference [1] J. Millman and A. Grabel, Microelectronics, 2nd ed.,
0º 0.036º ~0.3% Mc-Graw-Hill. 1987.
2º 2º 0% [2] M. Rashid, Power Electronics: Circuits, Devices, and
25º 25º 0% Applications , 3rd ed., Prentice Hall, 2003.
75º 75º 0% [3] Wikimedia Foundation Inc., “Rectifier.” [Online].
135º 135º 0% 2009. Available: http://en.wikipedia.org/wiki/Rectifier.
180º 181º 0.5% [4] D. Maksimovic, R. Zane, and R. Erickson, “Impact of
digital control in power electronics,” Proceedings of
2004 International Symposium on Power
Semiconductor Devices & ICs, Kitakyushu, 2003,
IV. CONCLUSIONS pp. 13-22.
[5] K. Rahman, M. Choudhury, and T. Zulfikar, “An
This microcontroller powered digital control method is AT89C51 microcontroller based control circuit for dual
presented in this paper to provide a low cost, high efficiency three phase controlled rectifier,”Proceedings of the 3rd
alternative to currently available methods. Programmed in International Conference on Electrical & Computer
BASIC, the implementation shows both its high efficiency and Engineering ICECE 2004, Dhaka, Bangladesh,
low response time compared to other solutions. This presents December 2004, pp. 347-350.
a potential for cost and time saving for many industries. Some [6] Microchip Technology Inc., Datasheet of
limitations do exist for this particular solution. The code was MCP601/2/3/4, [Online]. Last accessed October 2009.
written for 50 Hz AC electrical sources. Also, the Available: http://www.microchip.com. 2004.
optimizations performed were based on the use of a 4 MHz [7] Microchip Technology Inc., PIC16F87XA Datasheet,
crystal oscillator. On the other hand, these drawbacks are [Online]. Last accessed October 2009. Available:
software based and thus can be tailored to other requirements http://www.microchip.com. 2003.
by simply modifying the microcontroller’s BASIC code.
Future work could include the use of a more general code for
the ability to work with any source’s AC frequency and
further optimizations. Other improvements would be the use
of a cheaper microcontroller as the PIC16F877A was only
used since it was immediately available. These improvements
are expected to further reduce the cost of this solution and
make it more widely available.
 Generalized Minimum Variance Controller with
Dynamic Pole Assignment to Improve Performance
in Industrial Applications
M.A. Paz Ramos1, S.C. Garibo Esquivel2, O.J. Santos Sánchez3, R. Fernández del Busto4,
J.F. Briones de la Torre1, J.L. Gallegos Ramirez1, J. I. Rodríguez Martínez1
1
Universidad Politécnica de Aguascalientes, Aguascalientes, Aguascalientes, México.
2
Instituto Tecnológico de Aguascalientes, Aguascalientes, Aguascalientes, México.
3
Universidad Autónoma del Estado de Hidalgo, Pachuca Hidalgo, México.
4
Tecnológico de Monterrey Campus Ciudad de México, México Distrito Federal, México.

Abstract – A comparison is made between the Generalized

Minimum Variance Controller (GMVC) and a new version of it, φ(t+k)=Py(t+k)+Qu(t)-Rr(t) (1)
the Generalized Minimum Variance Controller with a Dynamic
Pole Assignment (GMVCDPA) regarding stability and where P, Q, R (called cost polynomials) are used to modify
performance under process parameter variations. Simulation
the control performance [2] while y is the process output, u is
examples show that the proposed structure has better
characteristics than the original GMVC. To further illustrate
the control input, r is the reference signal and k is the system
the advances of the GMVCDPA, three laboratory experiments delay. The cost function to be minimized is the mathematical
are carried out: for DC motor, speed and position control and expectation of φ (t + k ) ,
for chamber prototype, temperature control. Results obtained J=Ex[φ2(t+k)] (2)
in temperature control experiment are compared with those
obtained with an industrial PID controller. The control law proposed for the GMV controller is
I. INTRODUCTION Hu (t ) + Gy (t ) + Er (t ) = 0 (3)
A good deal of control problems found in industry, are where the control polynomials H, G and E are directly
properly solved with conventional control strategies; estimated from (1), because in an implicit way, the following
however, there are some areas in which these techniques are relationship prevails:
not adequate, and where adaptive controllers have showed PC = FA + z −k G (4)
better results. Particularly, self-tuning regulators (STR) are where:
being used and have become the most popular adaptive
applications in the industry. H = BF − QC y E = −CR (5)
A particular STR structure is studied in this work: the
Generalized Minimum Variance Controller (GMVC), which In Figure 1, we can see a synthesis of the structure of the
is described in section II. The design of the cost polynomials GMVC where A, B and C are the process polynomials. The
for GMVC, according to a pole assignment is discussed in control algorithm is clearly explained in [2].
section III. In order to achieve better results in practice, we
propose an algorithmic modification of the original GMVC
in section III. In section IV, we discuss stability issues and in
section V performance issues are revisited. Finally in section
VI physical implementations are presented.

II. GENERALIZED MINIMUM VARIANCE

CONTROLLER

Clarke and Gawthrop proposed the generalized minimum

variance controller (GMVC) [1], defining a pseudo-output of
the system φ(t) which includes not only the system output,
but the reference and control signals as well. Among other
properties, this algorithm allows the user to tailor the
performance of the controller through a specific choice of the
optimality criterion, and also deal with non-minimum phase Figure 1. Functional diagram of a GMVC
systems. In the GMVC a generalized output, or pseudo
output is defined, by the following equation:

291
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_51, © Springer Science+Business Media B.V. 2010
292 RAMOS ET AL.

III. GENERALIZED MINIMUM VARIANCE IV. IMPROVING STABILITY

CONTROLLER USING DYNAMIC POLE ASSIGNMENT
Second order systems can be considered as
representative case in industrial process control. A common
The GMVC is a very useful controller when the process is
recommendation during implementation of self tuning
time invariant. Another algorithm related to GMVC is the
regulators is to include filters to achieve a relevant
Generalized Minimum Variance Controller with Pole identification [4]. In order to illustrate how GMVCDPA
Assignment (GMVCPA) [2], where the main purpose is to improves stability and performance, we choose a second order
tailor the transient response by pole assignment. For the system. Although the structure of the GMVCDPA is discrete
GMVCPA, the cost polynomials are designed for the when it is applied, in this paper we use a continuous analysis,
controller to achieve a pole assignment, thus solving the because it makes it easy to visualize the stability zones and,
identity: consequently it is easier to appreciate the improvement of the
original structure. Consider the next process:
PB + QA = T (6) B(s) K (8)
=
A( s ) s 2 + a 1 s + a 2
where T is the pole assignment polynomial. Equation (7) is a then nA=2 and nB=0, where n is the system order and the
conceptual representation of the closed loop using a GMVPA subscript refers to the polynomial, we design a GMVCPA
control, when the cost polynomials are designed from (6). controller where nT=n, if nP= 1 and nQ=1, then P and Q could
be expressed as:
BQ (7) P ( s ) = p 0 s + p1
G CL = (9)
PB + QA Q ( s ) = q 0 s + q1
The cost polynomials P and Q are fixed during control, and For the next analysis, we assume that the process parameters
then, if the process parameters are perturbed, the relationship are the true ones. Considering (6) and (9), when nT=2, q0=0
(6) could be invalid. Even when the GMVCPA preserve the then we have
ability to minimize the output variance, the performance T ( s ) = s 2 q 1 + s ( p 0 K + a1 q 1 ) + p 1 K + a 2 q1 (10)
could be affected due to changes in the denominator of (7).
The Generalized Minimum Variance Controller with If T(s) is monic, q1=1.
Dynamic Pole Assignment (GMVCDPA) was proposed in
[3], and allows a better control when parameter perturbations Applying the Routh criterion to the polynomial (10), we
are present. While in the GMVCPA the cost functions are obtain the array
proposed off line, in the GMVCDPA the cost polynomials
are calculated and updated on line. If we wanted to achieve a s2 1 p1 K + a 2
dynamic pole assignment in the traditional GMVCPA, we s1 p 0 K + a1 (11)
would need two parametric estimations: process estimation s 0
p1 K + a 2
and control estimation. In [3] the control estimation is
avoided and substituted by a control calculation using only There are necessary and sufficient conditions for closed loop
process estimation. Figure 2 illustrates the structure of the stability if
GMVCDPA. In this paper, we present a stability analysis and
performance test in order to compare the original formulation p o K + a1 > 0
(12)
of GMVCPA against its modification, the GMVCDPA. p 1 K + a2 > 0

Consequently, the stability boundaries for possible variations

in the process parameters are defined by

∂a 1 > − p o K
(13)
∂a 2 > − p 1 K
Thanks to all the evidence that has been presented through of
this section, we confidently postulate the following Theorem:

Theorem 1 Given the open loop process

K (14)
G (s) = 2
s + a1 s + a 2
and a GMVPA controller whose cost polynomials are
defined by
Figure 2. Functional diagram of the GMVCDPA
 GENERALIZED MINIMUM VARIANCE CONTROLLER WITH DYNAMIC POLE ASSIGNMENT 293

P ( s ) = p 0 s + p1 (15)
In Figure 4(a) the output of the process controlled is showed
while in the Figure 4(b) we can see the control signal.
Q( s) = 1
Suppose now, that there is a perturbation in one of the
assuming a K fixed in (14), the stability zone before parameters, particularly in the resistance one. The nominal
perturbations in the process parameters a1 and a2 when the value is drastically perturbed to R=37.5 in the 25th second.
loop is closed considering (15), is the S zone whose Incorporating the change to (17), the new process
boundaries are defined by coefficients are a1*=4.434 and a2*=26.14, which causes for
∂a 1 > − p o K the coordinate (a1*, a2*) to stay outside of the stability zone
(16)
∂a 2 > − p 1 K S. This situation can be appreciated thanks to the simulation
results showed in Figure 5.
In the parameters space a1-a2 (see Figure 3).
GMVPA Control
3

r,y
a) 1

-1
0 5 10 15 20 25 30 35 40 45 50
time (sec.)
3

1
b)

u
0

Figure 3. Stability zone, defined by the cost polynomial P -1

-2
0 5 10 15 20 25 30 35 40 45 50
In order to exemplify the Theorem 1, we choose to control time (sec.)
the capacitor voltage of an RLC circuit which is the low pass
Figure 4. GMVCPA for the RLC circuit example
filter whose transfer function is
GMVPA Control
20

Eo( s ) 1 / LC
= (17) 10
Ei ( s ) s 2 + R / Ls + 1 / LC
r,y

a) 0

-10
The values of the electric components are L=8.4572H,
C=0.004523F and R=69.02119Ω. Then the transfer function -20
0 5 10 15 20 25 30 35 40 45 50
can be rewritten as time (sec.)
20

Eo ( s ) 26 .14 (18)
= 10
Ei ( s ) s 2 + 8.161s + 26 .14
b) 0
u

-10
If the desired T polynomial is defined as
-20
0 5 10 15 20 25 30 35 40 45 50
T ( s ) = s 2 + 3.68s + 9.688 (19) time (sec.)

Figure 5. GMVCPA when there is a parameter perturbation

to achieve a ts=2.5 sec. with a Mp=10%, then, when
GMVCPA is considered, the cost polynomials are Considering now the parametric uncertainties treated in
Example 1, we present the polynomial
P ( s ) = −0.17141579 s − 0.62941652 (20)
Q( s) = 1 A * ( s ) = s 2 + ( a 1 + Δa 1 ) s + ( a 2 + Δ a 2 ) (23)
then the stability boundaries are
from (23) we define
∂a1 = 4.48 (21) a 1 * = a 1 + Δa1
∂a 2 = 16.45 (24)
a 2 * = a 2 + Δa 2
For a simulation we calculate the discrete solution for (6) as then the closed loop polynomial is

P ( z ) = −2.597 z + 1.7988 T ( s ) = s 2 + s ( p 0 K + a1 *) + p1 K + a 2 * (25)

(22)
Q ( z ) = 1.25784
294 RAMOS ET AL.

In the GMVCPA the cost polynomials are designed off line, whenever (t1, t2 ) > 0, being t1 and t2 the coefficients of the T
from the solution of a Sylvester matrix [2]. polynomial.

ª 1 º In Figure 6, we present the case shown in Figure 5, but now

ª q1 º « » we use the GMVCDPA. As we can observe in Figure 6, the
«q » « 0 »
1 (26) control system overcomes the parametric perturbation
« » = « (t − a )»
0

« p1 » « K 2 2 applied in the 25th second satisfactorily. The parameter

»
« » «1 » perturbation is overcome appropriately due to the capacity of
¬ p 0 ¼ « (t 1 − a1 ) » the GMVCDPA to update the cost polynomials on line, when
¬K ¼ it happens, also, can be observed a softer starting of the
where t1 and t2 are the parameters of the T polynomial. control signal. In Figure 7(a) the estimated parameters can be
observed and in Figure 7(b) the cost polynomials can also be
If we consider that the cost polynomial are designed with the observed.
nominal parameters a1 and a2, coordinate (a1*, a2*) could be GMVDPA Control
out of the stability zone S. On the other hand, when we apply 3

the GMVCDPA, the calculation of the cost polynomials was 2

performed on line using the estimation. Assuming the

r,y
asymptotic convergence of the estimation we have a) 1

aˆ 1 ≈ a1 + Δa1 (27) 0

aˆ 2 ≈ a 2 + Δa 2 -1
0 5 10 15 20 25 30 35 40 45 50
time (sec.)

then the calculus of the cost polynomials is defined by 3

2
ª 1 º
ª q1 º « » b) 1

u
«q » « 0 »
1
« » = « (t − aˆ )»
0 (28) 0

« p1 » « K 2 2
» -1
« » «1 »
0 5 10 15 20 25 30 35 40 45 50

¬ 0 ¼ « (t 1 − a1 ) »
p ˆ time (sec.)

¬K ¼ Figure 6. GMVCDPA when there is a parameter perturbation

if we substitute (28) in (16) 1.5

∂a1 = a1 * −t1
1
(29) a)
θ

∂a 2 = a 2 * −t 2 0.5

0
using (29), we can see that if T(s) fulfils the Stodola 0 5 10 15 20 25 30 35 40 45 50
condition, the boundaries of the stability wrap the perturbed time (sec.)

coordinate (a1*,a2 *). 4

cost polynomials

2 p1

From the previous results outlined, we state the following q1

b) 0
theorem: q0

-2 p0

Theorem 2 Given the open loop process -4

0 5 10 15 20 25 30 35 40 45 50
time(sec.)
K (30)
G (s) = 2 Figure 7. Parameters update in the GMVCDPA
s + a1 * s + a 2 *
with parameter perturbation V. IMPROVING PERFORMANCE

a1* = a1 − Δa1 Another attribute of the GMVCDPA is its ability to preserve

(31)
the dynamic performance originally contemplated, even in
a2 * = a2 + Δ a2
presence of perturbations. Suppose now that there is a
perturbation in the parameters in closed loop,
If we apply the GMVCDPA, then the coordinate (a1*,a2*) PB − QA * (33)
is inside of the stability zone defined by the margins
∂a1 = a1 * −t1 (32) where A* it is the polynomial of the perturbed parameters.
∂a 2 = a 2 * −t 2 Considering the original design, the closed loop when the
perturbation exists is
 GENERALIZED MINIMUM VARIANCE CONTROLLER WITH DYNAMIC POLE ASSIGNMENT 295

s 2 + t 1 s + ( sa 1 * − sa 1 ) + t 2 + ( a 2 * − a 2 ) (34) position control is implemented, and second, for a chamber

prototype, the results of a temperature control are compared
when GMVCDPA is used, according to (22), the results of with those obtained with the auto-tune mode of an industrial
differences in (29) are 0, consequently the pole assignment is PID controller
completed Speed control: in this experiment, we use a Minertia motor
PB − QA* = T (35) T03 manufactured by Yaskawa, which is usually used in
computer peripherals as:
In order to illustrate the characteristics of the GMVCDPA in
performance, we implemented practically the example 1
building the RLC circuit, and for the resistive element, we ƒ capstan and reel for magnetic tape
ƒ paper and ribbon feeding for line printer
use a potentiometer to cause a change from 57Ω to 251Ω, in
ƒ X-Y plotter
the 32th second. Is possible to see in Figure 8(a) the
GMVCPA achieve the control in presence of the ƒ and small general purpose machinery
perturbation, because it tries to reduce the process output
variance, but the performance is seriously affected. On the The aim of the experiment is to control the speed of the
other hand, we can see in Figure 9(a), that the performance is motor. The process output is the speed variable, measured in
regular, in spite of the perturbation when the GMVCDPA is revolutions per minute (rpm). The reference was developed
used, because the action of the controller is modified to in such a way that it involves ramps to be able to observe the
achieve the desired output [see Fig. 9(b)]. tracking and the regulation in the same experiment. In Figure
10, we can observe that the speed of the motor follows the
GMVPA Control
5 reference tightly.
4 Speed Control
700
r,y

a) 3
600
2

r,y(rpm)
500
1
400
0
0 10 20 30 40 50 60 300
time (sec.)
200
6 0 20 40 60 80 100 120 140 160 180 200
time(sec.)
4

b) 8
u

2
u(volts)

0 6

-2 4
0 10 20 30 40 50 60
time (sec.)
2
0 20 40 60 80 100 120 140 160 180 200
Figure 8. GMVCPA performance when there is a parameter time(sec.)
perturbation
GMVDPA Control
Figure 10. Motor speed control using GMVCDPA
5

4 Position control: for this experiment, we choose a small

3 conveyor belt moved by a DC PM Gearmotor 1.61.046
manufactured by Buehler. The process input is the electric
r,y

a)
2

1
feeding of the motor. The process output is the belt
displacement of the belt in centimeters (cm.). We use the
0
0 10 20 30 40 50 60 reference shape used in the speed example. We can see in
time (sec.)
Figure 11(a), that we have remarkable tracking. Usually for
6 position control we use a P+D Control; when the system has
attained the reference value, the control signal becomes null.
4
b)
In this practical case, the motor has a dead zone between the
u

2
-3 and 3 volts in the input process. The GMVCDPA
0 perceives this situation, and allocates the control signal under
-2 the frontier of the dead zone. We can see in Figure 11(b) that
0 10 20 30 40 50 60
time (sec.)
the control signal is barely lower than the dead zone frontier,
which allows a faster response before a sudden reference
Figure 9. GMVCDPA performance when there is a parameter
change.
perturbation
Temperature control: In this experiment, we use a prototype
which consists of in two chambers. In the first chamber there
VI. INDUSTRIAL APPLICATIONS
is a heating source and in the second one, there is a
In this section, three typical industrial applications are temperature sensor (K thermocouple). Between the chambers
presented, using GMVCDPA. First, for DC motor, speed and there is a fan that forces the air to flow between chambers
(Figure 12). The process input voltage is applied to the fan.
296 RAMOS ET AL.

Fan speed increase causes the temperature to increase in the PID Auto-tuning DC 1020 Honeywell
second chamber. The process output is the temperature in the 45

r,y(temperature °C)
second chamber (°C). The process input has a dead zone
40
between the 0 volts and 3 volts.
Position Control
100 35

80
30
60 0 100 200 300 400 500 600 700 800 900 1000
r,y(cm)

time(sec.)
40
10
20

0 5

u(volts)
0 20 40 60 80 100 120 140 160 180 200
time(sec.)
0
10

5 -5
0 100 200 300 400 500 600 700 800 900 1000
u(volts)

0 time(sec.)

-5 Figure 13. Temperature control using a PID Auto-tuning DC 1020

Honeywell
-10
0 20 40 60 80 100 120 140 160 180 200 GMVDPA Control
time(sec.) 45

r,y (temperature °C)

Figure 11. Conveyor belt position control using GMVCDPA
40
a)

35

30
0 100 200 300 400 500 600 700 800 900 1000
time (sec.)

10

8
u (volts.)

6
b)
4
Figure 12. Temperature process
2

To control the process showed in Figure 12, we first use an 0

0 100 200 300 400 500 600 700 800 900 1000
industrial field controller, the DC1020 PID auto-tuning time (sec.)

manufactured by Honeywell. When we connect the Figure 14. Temperature control using GMVCDPA
controller to the process, we select the auto-tune mode
(ATV). The control results are showed in Figure 13(a), the VII. CONCLUSIONS
reference deviation is approximately +/- 1.3°C, which for a
general industrial application is considered a good The behavior of the GMVC and GMVCDPA which is a
performance. We can see that the control signal goes near to modified version of GMVC, was analyzed under stability and
saturation periodically; one of the causes of such behaviour is system parameter variation viewpoints. GMVCDPA provides
the dead zone of the actuator and the control structure of the better results than GMVC in these two areas. The laboratory
commercial controller. experiments that were carried out with GMVCDPA, showed
the outstanding performance of this controller.
In Figure 14 (a), we show the ability of the system to control
the temperature control using the GMVCDPA. Using this VIII. REFERENCES
self tuning controller, we improve the performance in the
application. The reference deviation for this case is +/-0.3°C. [l] Clarke, D.W., Gawthrop B.A. “Self-tuning controller”, IEE Proc.
Control Theory Appl., 122, 1975, pp. 929-934.
In Figure 14 (b) we can see that the signal control is barely [2] Wellstead, P.E., Zarrop M.B. Self-tuning systems. John Wiley & Sons,
over the dead zone, allowing a better regulation. England 1991.
[3] Paz Ramos M.A., Quintero-Mármol M.E., Fernández del Busto R.,
“Generalized Minimum Variance with Pole Assignment Controller
Modified for Practical Applications”, Proceedings of the IEEE
International Conference on Control Applications, Taipei, 2004,
pp. 1347-1352.
[4] Poulin E., Pomerleau A., Desbiens A., Hodouin D., “Development and
Evaluation of an Auto-tuning and Adaptive PID Controller”,
Automatica, Vol. 32, No. 1, 1986, pp. 71-82.
 Automated Detection of Sunspots and Sunspot
Groups in Full-Disk Solar Images
Sarab F. Majed
Department Of Mathematic, University Of Western Ontario, ON, Canada
Mehmmood A. Abd
Department Of Electrical And Computer Engineering, University Of Windsor, ON, Canada
V. Zharkova
Department Of Computing And Mathematics, University Of Bradford, BD7 1DP,UK

The precise detection of sunspot groups is a fundamental task

Abstract___In this paper, a new adopted unsupervised for understanding the solar activities and their impact on The
segmentation technique is presented. This technique extracts Earth’s geomagnetic field and ionosphere parameters [5].
both sunspots and sunspot groups from the solar disk as whole Segmentation aims to divide the images into meaningful
group that eases the automated sunspot group classification. regions that corresponding to structural units in the scene of
MacIntosh Classification is used as standard to determine which
group belongs to which class. Sunspot data are extracted from
distinguished objects of interest. Therefore, it is very
daily white light (WL) images of a solar disk captured by the important to isolate the basic features of sunspots and sunspot
SOHO/MDI satellite. New unsupervised segmentation algorithm groups, in order to study and understand the results of the
is applied to extract sunspot groups from the full disk image with solar activity numbers. The paper is organized in the following
use of mathematical morphology in order to improve the quality sections. Preprocessing is presented in section 2 that includes
and segregate the sunspot groups. The results were compared Solar Disk acquisition and edge detection. Section 3 presents
with the standard data from EGSO, SFC catalogues and with Binarization and section 4 presents sunspot groups extraction.
Mt.Wilson Report. The obtained results reveal a good accuracy Section 5 presents an adopted segmentation algorithm for
of group classification, which is very promising in comparison sunspots and sunspot groups and their numbers. Finally,
with the manual classification, catalogues.
Section 6 presents the obtained results compared with SIDC,
Index Terms___Solar Disk, sunspots, edge detection, segmentation, EGSO SFC and Mt.Wilson [3][4][6].
Mathematical morphology
2. PREPROCESSING
1. INTRODUCTION
2.1 Solar Disk Acquisition

S unspots observations and classification are the very

essential tasks for solar astronomers. Sunspots and
sunspot groups give clues to the timing of solar flares and
The white and black images of a solar disk taken in Flexible
Image Transport System (FITS) file format are considered in
this study. These images are full solar disk calibrated synoptic
solar activities. Sunspots are dark area appear on the daily continuum and Line of Sight (LOS) magnetogram
photosphere and features of active regions of the sun. In observations. The images are taken from the Solar Oscillations
addition, sunspots are born as tiny pores and some of them Investigation Michelson Doppler Imager (SOI/MDI) aboard of
become developed into large spots while the other disappear the Solar Heliospheric Observatory (SOHO)[14]. The image
[1]. Generally, sunspots have different characteristics such as (Figure 1) has dimensions of 1024 x 1024 pixels with 16
size, longitudinal and shape and they are unified into groups different intensities. The resolution is 2.6 arcsec/pixels. As in
with a large and long-lived leader spot, which has a leading the first step, the perimeter and the solar center are detected.
polarity for the given hemisphere. The leader spots are Then the whole solar disk is extracted from the background.
matched with one or more trailing, smaller, spots of the Moreover, the coordinates of the centroid of a solar disk is
opposite magnetic polarity. Such groups can be easily read from the header of fits image (http://fits.gsfc.nasa.gov/)
observed with solar magnetograph and can be realized that the and checked by placing the solar disk diameters in
ends of sunspots groups have opposite magnetic polarity [2] perpendicular directions.
[3]. However, the irregularities of a sunspots shape and
variability of these shapes make automated detection very 2.2 Edge Detection
difficult. The impact of the solar activity on the terrestrial Accurate solar and sunspots edge detection is important and
atmosphere and human life is associated with a number of initial task to determine to sunspots position. The position of
sunspots and the size or shape of sunspot groups. The number sunspots and sunspot groups are expressed in spherical
of sunspots and sunspot groups, Wolf number, is an indictor of coordinates. However, diameter of the sun rely on the correct
the sun’s magnetic field eruption. The most important indices detection /calculation of solar limb. For achieving this task the
are those called international sunspot number which are threshold intensity level is found using trial and error.
summarized by the Sunspots Index Data Center (SIDC)[4].

297
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_52, © Springer Science+Business Media B.V. 2010
298 MAJED ET AL.

Fig. 1. The full disk image obtained from SOHO/MDI on 1 January 2000.

Fig. 2. Solar Edge Detection using Canny edge Detector.

Many researchers have been applied different techniques
Zharkova et al (2003)[10] employed canny edge detection that is expressed in terms of the class probabilities ‚i,
followed by limb fitting. Curto et al (2008) [9] were used
erosion, gradient transformation and threshold. In this paper separated by threshold ,t, and the variance σ i2 . Where
adopted Canny edge detection is used to detect the edges of classes’ means, i, in turn, can be updated iteratively [7]. After
solar disk. A pixel’s location is declared as the edge if the choosing the right threshold value, the following adopted
value of the threshold, T, is less than the value of a gradient, g, algorithm is applied to strengthen the intensities of sunspot’s
or when g ≤ TH ,H-high, and less than TL, L-low. Thus, penumbra and umbra and isolate them from background:
gradient components of X and Y, g x and, g y are computed. 1. Read the image f (x, y)
2. While(i and j ≤ image’s dimensions(max_x ,max_y))do
The edge will have the higher pixels intensity than other of If ( f (x, y) >= T )then f (x, y)= 0;
surrounding regions [3]. However, after defining solar disk If ( f(x, y) <=T )then f(x,y)=1 ;
perimeter’s pixels, the centroid needs to be found out by End if.
comparing the information from the image header with those 3.End
detected from the two perpendicular maximal diameters. However, the output image extracted with the above algorithm
Consequently, all the pixels that are located outside of the is shown in Figure 3. Then after, the resulted image is
solar disk are neglected. converted into a binary image to find out the regions of
interest, which encompasses each sunspot groups. The length
3. NOISE REMOVING AND FEATURE of a longitudinal location of the group is used to determine
ENHANCING which sunspot belonged to which group.
After defining perimeter and extracting the solar disk form 3. BINARIZATION
the background the noise and peppers are removed. All pixels,
which are outside the solar disk, are converted into black and Binarization is the conversion from gray-scale to a binary
those, which are inside solar disk, are remaining as they are. image .One obvious way to extract the objects from
In this approach intensity filtering is employed to reduce the background is by thresholding. Thus, in this phase, the
noise level in solar disk. Global threshold value is computed resulted cleaned image is converted into binary image based
and then multiplied by N, where N is a constant varying on the obtained threshold. The threshold value is calculated
depending on the image’s contrast. Otsu’s method is used to using trial and error due to the variation of intensity in each
automatically perform histogram shape-based image threshold sunspot. After that, the complement of the binary image is
[7]. It finds the threshold that minimizes the intra-class taken. In that image all the background pixels are masked to
variance, is defined as a weighted sum of variances of two be black and all the objects keep their pixels intensities. In this
classes: case the white sunspot regions remain have wide range of
pixels intensities .In other words not completely white but
σ w2 (t ) = ω1 (t )σ 12 (t ) + ω 2 (t )σ 22 (t ) (1) differ from the background. This technique is used to reserve
the relations among the leading and trailers spots in each
Otsu proves that minimizing the intra-class variance is as group. However, every object outside a solar disk
similar as maximizing inter-class variance: circumference is disregarded. Figure 4 illustrates the produced
resulting mask image. Consequently, all the objects attributes
can be computed by using the mathematical morphology.
σ b2 (t ) = σ 2 − σ w2 (t ) = ω1 (t )ω 2 (t )[μ1 (t ) − μ 2 (t )]2 (2)
 AUTOMATED DETECTION OF SUNSPOTS AND SUNSPOT GROUPS 299

Morphological processing facilitates the sunspots region Morphological image processing are dilation and erosion.
attributes extracting. Dilation is an operation that utilized to thicken or grow the
objects in binary images. On the other hand, the erosion
shrinks the objects. The amount and the way that the objects
grow or shrink is controlled by chosen selection of the
structure element, which can be a set of 4 or 8 connected
pixels. The structuring element that chosen in this research is
an 8- connected set. However, the dilation is mathematically
expressed as [3].
­
A ⊕ B = z | B
§ ∧ ·
∩ A ≠ φ
½ (3)
® ¨ ¸ ¾
¯ © ¹ z ¿
Where A is the empty and B is the structuring element. The
mathematically definition of erosion is.
{
AΘB = z (B ) z  A c ≠ φ } (4)
The combination of dilation and erosion is yielding closing
and opening that are higher order morphological operations.
The opening is erosion of A by B, followed by dilation.
Opening removes small regions and thins the filaments of
Fig. 3. Smoothed image. object pixels. It can be expressed as.
Meaningful shape information of geometric objects,
measuring typical characteristics area, bounding box, centroid
A $ B = ( AΘB ) ⊕ B (5)
coordinate, and the length of a perimeter. Hence, finding the While the closing morphological is dilation followed by
sunspots coordinates (x, y), area in pixels, eases a erosion. Closing removes regions and thins filaments of
determination of the sunspots location in relation to the solar background pixels. The mathematical expression of closing is
center. expressed as.
(
A • B = A ⊕ B ΘB ) (6)
Applying the mathematical morphology operations is useful in
representation and description of a region shape such as
boundaries, skeletons, and convex hull. Closing fill the small
gaps in a penumbra shadow to expose the umbra borders.
Opening operation is utilized to opens small gaps between the
touching objects. Thus, morphological operations are being
useful in breaking narrow isthmuses and eliminating small
(spurious) objects. The sunspots separation makes counting
their number much easy [3].
Moreover, binary mathematical morphology is carried out in
this study. The area, centroid, and boundaries are extracted.
Then, bounding box coordinates are calculated for each spot
Fig. 4.The masked solar image with 23 potential sunspots. and later used to segregate each group as independent object.
Moreover, area of any sunspots is utilized to estimate the area Moreover, the number of objects is utilized to find out the
in heliographic coordinates. number of sunspots in the solar disk. This number is
originated from the labeling-connected components of the
4. SUNSPOT GROUPS DETECTION binary image. All region area more than five pixels consider
as a spot otherwise not. The experimental executed on data set
Extracting sunspots and organizing them into groups phase taken by SOHO/MDI. In the image which is taken on 24
is consists of two phases: The first phase employs January 2000 at 20: 00: UT, 59 sunspots are detected. This
Morphological operations to enhance the image, and the number is very close to maximum number that detected by
second phase gathers all related sunspots into group based on Mt.Wilson [1]. However, Figure 1 shows five sunspots groups
neighborhood features of sunspots .The longitudinal of the clearly detected in an image taken by SOHO/MDI on 1
group is used to determine which sunspot belonged to which January 2000. Table 1 shows a comparison between detected
group. The MacIntosh classification catalogue is used as sunspots number in our study and EGSO (Zharkova et al.,
standard classification in this paper [11]. The Morphological 2005)[6],[13].The false detection rate (FAR) represents the
operations are used to compensate the poor quality of the sunspots that are detected in our study but not by EGSO. The
images and reduce the noise similar to Zharkov et al., False rejection rate represents the sunspots that detected by
2005a,b[8],[12],[13]. Besides that they are used to enhance EGSO by not detected in our study. However, the variation
the objects /sunspots attributes. The fundamental of between the results are because image quality. However, the
300 MAJED ET AL.

detection of sunspots number in this study around 80 percent, D–Bipolar group with penumbra on spots at both ends of the
in comparison with EGSO, which could be promising results. group, and with length ≤ 10 .
0
Finally, all the adjacent sunspots that are governed by a E-Bipolar group with penumbra on spots at both ends of
certain number, which is varying between 3 to 15 heliographic 0 0
degrees. The length of longitudinal, as that mentioned in group, and with length defined as: 10 ≤ Lenght ≤ 15 .
MacIntosh Classification, is used to group sunspots together F-Bipolar group with penumbra on spots at both ends of the
into a one group. Before computing the length of longitudinal 0
group, end length ≥ 15 . H-Unipolar group with penumbra.
the sunspots group must be determined as Unipolar or bipolar. The principal spot is usually the leader spot remaining from a
Based on the conditions: “ Unipolar group: A single spot, or a pre-existing bipolar group“[11]. However, the above-
compact cluster of spots with the largest greatest separation mentioned MacIntosh classification conditions are used as
0
between spots ≤ 3 . In the case of a group with penumbra standard in this segmentation approach.
(class H ) ,the nearest is measured between the center of the
attendant umbra and the nearest border of penumbra 5. SUNSPOT GROUPS EXTRACTION ALGORITHM
surrounding the main spot “[11]. Bipolar Groups: Two or
more spots forming an elongated cluster of length ≥ 3 .
0 Applying morphological operations enhances the sunspots
Unusually there will be space near the middle of the cluster characteristics and simplifies the way for determining
dividing it into two distinct parts. Groups with a large sunspots boundaries [3]. In this paper, 8-connected
0 0 0 components are utilized to separate each sunspot and its
principal spot must be ≥ 5 in length (i.e 2.5 plus 3 ) to be splinters in the single area, or group. After that, a leading
classified as bipolar " [11]. MacIntosh classification is used as sunspot is taken as the central spot from which the boundary
standard to determine the greatest separation between sunspots of sunspots groups is defined. The leading spot is defined by
in each group: “ A Unipolar group with no penumbra. B - it’s area and location. The ending spot is usually larger in size
Bipolar group without penumbra on any spots. and has a stronger magnetic field and located slightly closer to
the equator than the trailing spots. Moreover, it is usually
TABLE 1
THE COMPARISON OF DETECTED NUMBERS OF SUNSPOTS IN located at the western direction and the trailers /followers in
THIS STUDY AND EGSO [13]. the eastern direction. These attributes are marked with respect
Date
No# Spots No#Spots
FAR FRR to the solar center. Consequently, the sub-region of sunspots is
Our study EGSO grouped in a single larger region, or as a sunspot group.
1-July-2002 23 23 0 0
2-July-2002 22 22 0 0
Sunspots are grouped by using the length of longitudinal that
3-July-2002 39 41 0 2 confines the group as that described in MacIntosh
4-July-2002 47 53 0 6 classification. A longitude length variation for a sunspot group
5-July-2002 38 39 0 1 0
6-July-2002 37 37 0 0
is ranging in 10 ≥ legth ≥ 15 heliographic degrees. In this
7-July-2002 33 32 1 0 paper, the algorithm starts by computing the bounding box for
8-July-2002 25 26 0 1 each sunspot. The largest bounding box for a leading sunspot
9-July-2002 17 17 0 0 will be adjusted to encompass all the neighboring ones. The
10-July-2002 27 29 0 2
11-July-2002 28 28 0 0
algorithm is iterative and terminates when no more expansion
12-July-2002 30 31 0 1 occurs or the length exceeds the maximum longitude length.
13-July-2002 53 53 0 0 If the initial bounding box has vector
14-July-2002
15-July-2002
66
43
66
43
0
0
0
0
( )
V0 = x0 , y 0, xw0 , yw0 and the neighbor box has vector
16-July-2002 44 45 0 1 Vr = ( x r , y r , xwr , ywr ) then the expanded bounding box
17-July-2002 58 61 0 3
18-July-2002 57 56 1 0 has a vector V = ( x, y, xw, yw) .
19-July-2002 41 40 1 0
20-July-2002 41 40 1 0 Where x = min ( x0 , x r ) and y = min ( y 0 , y r )
xw = max ( x r + xwr − x, x r + xwr − x )
21-July-2002 41 41 0 0
22-July-2002 37 37 0 0
23-July-2002
24-July-2002
42
70
42
71
0
0
0
1
yw = ( y r + ywr − y, y r + ywr − y )
25-July-2002 86 86 0 0 This algorithm has to be repeated until no more bounding box
26-July-2002 114 114 0 0 expansion occurred. The output of this algorithm will be an
27-July-2002 133 138 0 5 independent image of each group. However, each group image
28-July-2002 141 142 0 1
is extracted with the detail of the original image, such as a
29-July-2002 118 118 0 0
30-July-2002 100 101 0 1 date and time. Figure 4 shows four sunspot groups are
31-July-2002 83 83 0 0 detected, in image that taken by SOHO/MDI on 12 July
2002,in the smoothed image in Figure 3 corresponding to five
C-Bipolar group with penumbra on one end of the group, in of Mt Wilson. However, this difference is occurred due to the
most cases surrounding the largest of the leader umbrae. image quality and the size and location of some sunspots
 AUTOMATED DETECTION OF SUNSPOTS AND SUNSPOT GROUPS 301

groups, which are very small and close to the edge of the sun. Mt Wilson did not detect them) and the false rejection rate
Besides, that the automated systems is based on extracting all (FRR) (Mt. Wilson detected the sunspots and we did not
available information from single image each day, whereas detect them). However, this variance of detection results due
sunspots number changes by the time. Table 2 shows the to the some external factors such image quality, clouds or poor
accuracy of our automated sunspot groups detection and visibility. From table 2 it is clear the result’s accuracy of
extraction in comparison with Mt Wilson for July 2002. detected sunspots groups in this study around 84 percent.

6. CONCLUSION

This unsupervised segmentation technique has shown

promising results that produce reliable sunspots and sunspot
groups detection and extraction.
The used data set is fits format images obtained from
SOHO/MDI instrument as a white light full disc image. The
proposed technique shows promising segmentation and
recognition results. Automotive recognition is minimizing
labors needs and time consuming in comparison with manual
methods.

Fig.4. Four sunspot group detected in an image that taken by SOHO/MDI on This algorithm can be applied on any type of image
12 July 2000. format with minor modification and can be applied in
biological applications. Further more, the rustles can be
TABLE 2
THE NUMBER Of SUNSPOT GROUPS THAT ARE DETECTED FOR
facilitating the automotive sunspot group detection and
JULY 2002 IN COMPARISON WITH THOSE DETECTED BY MT. classification.
WILSON [10].
#Spot
Group
#Spot
group,
REFERENCES
Date Mt Wilson Detected FAR FRR
01-July-02 5 5 0 0 [1] David H. Hathaway, Robert M. Wilson And Edwin J. Reichmann, ”
02- July -02 7 7 0 0 Group Sunspot Numbers: Sunspot Cycle Characteristics,” Solar Physics
Vol., 211,2002, pp. 357–370.
03- July -02 10 9 0 1
[2] Guseinov, M. D,” The Polarity Of The Magnetic Field In Bright Umbral
04- July -02 8 10 2 0
Points In Sunspots,” Krymskaia Astrofizicheskaia Observatories,
05- July -02 11 11 0 0
Izvestiia, Vol. 68, 1983, pp. 36-39.
06- July -02 11 10 0 1 [3] Gonzalez, R. C, Wood, R. E,”Digital Image Processing,” Prentice Hall,
07- July -02 6 6 0 0 2004 , pp. 338-446.
08- July -02 7 7 0 0 [4] Observatories Royal de Belgium (2009-9-1) Sunspots Index Data Center
09- July -02 9 9 0 0 (SIDC) [Online]. Available: http://www.icsu-fags.org/ps11sidc.htm
10- July -02 7 7 0 0 [5] Mitic, M. Djordjevic ,A. Mihajlovic,S. J. “Changes In atmosphere and
11- July -02 - - Ionosphere Caused By Solar-geomagnetic Activity ,Within Belgrade
12- July -02 5 4 0 1 Area ,” ,” EGS XXVII General Assembly, Nice,(21-4-2009) [Online].
13- July -02 7 7 0 Available: http://adsabs.harvard.edu/abs/2002EGSGA..27..631M
14- July -02 6 7 1 0 [6] Zharkova et al., Solar Physics.(1-5-2009). European Grid of Solar
15- July -02 9 9 0 0 Observation, Solar Feature Catalogues [Online]. Available:
16- July -02 9 8 0 1 http://solar.inf.brad.ac.uk/pages/ss_start.php
17- July -02 7 6 0 1 [7] M. Sezgin and B. Sankur, "Survey over image thresholding techniques
18- July -02 4 5 1 0 and quantitative performance evaluation," Journal of Electronic Imaging.
19- July -02 5 7 2 0 Vol.13 (1) ,2004, pp. 146–165
20- July -02 7 7 0 0 [8]. S. Zharkov, V. Zharkova, S. Ipson, A. Benkhalil,” The Technique For
21- July -02 8 8 0 0 Automated Recognition On Full Disk Solar ,” EURASIP Journal .Vol.
15, 2005, pp. 2573-2584
22- July -02 9 8 0 1
[9] J. J. Curto, M. Blanca and E. Martínez, ”Automatic Sunspots Detection
23- July -02 10 9 1 0
on Full-Disk Solar Images using Mathematical Morphology,” Solar
24- July -02 11 10 1 0 Physic. Vol. 250, 2008, pp. 411-429.
25- July -02 9 9 0 0 [10] Mount Wilson Observatory, (13-12-2008) [Online]. Available:
26- July -02 12 10 2 0 http://www.mtwilson.edu/hk/ARGD/Sunspot_Cycle/
27- July -02 15 11 4 0 [11]. Patrick S. MacIntosh, (1990),” The Classification Of Sunspot Groups “,
28- July -02 13 11 2 0 Solar Physics Vol. 125,1990, pp. 251-267.
29- July -02 10 8 2 0 [12]. Zharkov, S., Zharkova and V.,Ipson, S., 2005b,” Statistical properties
30- July -02 11 9 2 0 of sunspots in 1996-2004 : I. Detection ,North-South asymmetry and
31-July-02 8 8 0 0 area distribution ,” Sol. Phys. ,228, PP. 377-397.
[13]. Zharkova, V., Aboudarham, J., Zharkov, S., Ipson, S., Benkhalil, A.,
Fuller, N., “ Solar feature Catalogue in EGSO,” Sol. Phys., Vol. 228,
Table 1 shows a comparison between our method and Mt 2005, PP 361-375.
Wilson reports results. Moreover, it is worthy to compute the [14]. Solar Heliospheric Observatory (SOHO), (10-9-
false acceptance rate (FAR) (where we detect the sunspots and 2009, [Online]. Available:http://sohowww.nascom.nasa.gov/
 Stochastic Model Based Approach for Biometric
Identification
Mofakharul Islam1 Sitalakshmi Venkataraman2 Mamoun Alazab3
Internet Commerce and Security Laboratory
Graduate School of Information Technology and Mathematical Sciences
University of Ballarat, Australia
[email protected]
[email protected]
[email protected]

Abstract- In this paper, we present a new stochastic model based In biometric identification, where dental characteristics
approach for enhanced image segmentation in biometric such as spacing between teeth, teeth shapes, sinus patterns and
identification systems. Biometric features such as fingerprint, root / crown morphologies are used to identify human
face, iris, hand geometry and more recently dental features are individuals, image segmentation and classification is a
being used for human identification. Image analysis of each of
these biometric features has various challenges to overcome. To
complex task due to the non-standard regions of interest (ROI)
address such contemporary problems of image segmentation, we and information present in dental X-ray images. The content
provide a novel approach based on maximum a posteriori (MAP) variations are also attributed to the quality of contrast, non-
fitting Gaussian mixture model using Expectation-Minimization uniform intensities and digitisation / high frequency noise that
(EM) algorithm within the Bayesian framework. Our new are inherent in X-ray images. Hence, traditional image
algorithm captures the pixel intensity by the likelihood term in segmentation algorithms such as edge-based, region-growing
Bayesian Networks, and a priori biasing term of the spatial and threshold techniques that have been successful in certain
location information with the help of Markov Random Fields applications are not suitable for X-ray images [7], [8].
(MRF) model. We have employed a novel approach of using Similarly, contemporary approaches, which combine region
Daubechies wavelet transform for texture feature extraction that
uses MRF model and a robust technique of determining the
growing methods with edge-detection techniques or even
number of pixel classes based on Cluster Ensembles for a reliable statistical model based approaches, are either computationally
segmentation of dental X-ray images. We present how our expensive or more suitable for colour models [9], [10].
approach could be applied in dental biometrics to achieve very In order to address the above mentioned problems, this
fast and reliable human identification. Experiments show that paper proposes a novel stochastic model based approach to
our new unsupervised image segmentation technique provides perform image segmentation effectively. The application of
accurate feature extraction and teeth segmentation for effective our proposed approach in biometric identification using dental
biometric identification. images has been experimentally tested to produce effective
and accurate results for feature extraction and image
I. INTRODUCTION
segmentation. The rest of the paper is organised as follows. In
Human physical characteristics such as fingerprint, face, Section II , we present a modified classical finite mixture
retina scans, iris patterns, hand geometry, and others have model that employs Markov Random Fields (MRF) model
been used in various personal identification systems [1], [2]. with Expectation-Minimization (EM) algorithm under
However, in order to meet the specific requirements of Bayesian framework. Section III describes our novel approach
different biometric applications, more recently, researchers of image segmentation of dental X-ray images where the gray-
have been probing into other internal features such as teeth, level intensity and texture features play crucial role in pixel
spine and chest contours [3], [4]. Since internal features such clustering. Texture feature extraction are performed using
as iris and retina consist of soft tissues that could get damaged Daubechies wavelet transform and ensemble of clustering to
or deformed, recent research studies have established that arrive at reliable number of components/segments in an image
dental features form a more reliable candidate for biometric under unsupervised framework. In Section IV, we discuss the
identification as they are more robust and could be preserved experimental results of our proposed approach applied on
for a sufficiently long period of time, even after death [5], [6]. teeth X-ray images and in Section V we present our
Another motivation is that similar to medical diagnosis, X-ray conclusions and directions for future work.
image processing approaches could be adopted as low-cost
solutions for personal identification systems using human II. MAP FITTING GAUSSIAN MIXTURE MODEL
internal features such as teeth. Hence, in this research, we Among the stochastic model based methods, Gaussian
concentrate on the image segmentation approaches suitable mixture model or the finite mixture model has received
for dental biometric identification systems that are useful not significant attention from researchers in the last decade due to
only for ante mortem human identification but also for post- its improved segmentation results when compared to other
mortem identification, due to their survivability characteristic image segmentation techniques [11], [12]. A common
for a long period of time.

303
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_53, © Springer Science+Business Media B.V. 2010
304 ISLAM ET AL.

drawback of such statistical model based methods is their high efficiently handle X-ray images towards achieving increased
computational intensity. However, adapting a finite mixture accuracy in image segmentation.
model for image segmentation is recently becoming popular We present here a mixture model with an unknown
[13]. We propose a stochastic model based on classical finite number of components K, each of which has its own vector of
mixture model that employs Markov Random Fields (MRF) to density parameter θ j for each feature. Let x i denote the
capture the neighbourhood relationships among the
neighbouring pixels. The proposed model has some observation at the ith pixel of an image (i=1, 2,…..N) modeled
advantages, especially in the case of contextually dependent as independent and identically distributed (iid). So, the
physical phenomena like images, as it incorporates the probabilities of the ith pixel belonging to the jth class label are
outstanding features of the MRF model without additional given as follows:
computational burden. In the proposed method, MRF η ij = P ( j | x i ) (1)
introduces a priori distribution that takes into account the
neighbourhood dependency or relationship among the
neighbouring pixels. We elaborate a stochastic model based We assume that the density function f ( xi | Φ, Γ) at
segmentation framework based on Gaussian mixture models any observation, xi is given by the mathematical expression
below.
as follows:
The proposed model is an introduction of MRF model K
into the classical mixture model using Expectation-
f ( xi | Φ, Γ) = ¦ p ijψ ( xi | θ j ) (2)
Minimization (EM) algorithm under Bayesian framework.
j =1
Recently, approaches using MRF models and Bayesian
methods have provided answers to various contemporary where, ψ ( xi | θ j ) is a Gaussian distribution with
problems in image segmentation. The theoretical framework
parameters θ j = ( μ j , σ j ) , and K is the number of
relies on Bayesian estimation via combinatorial optimization
(Simulated Annealing). The application of a clustering method components in the mixture.
to image segmentation has the particular characteristics that
spatial information should be taken into account. Here, in We introduce Maximum a posteriori (MAP), a prior
addition to intensity-values of pixels, the pixel location must distribution for the parameter set Φ that takes into account
also be used to determine the cluster to which each pixel is spatial information based on the Gibbs function. According to
assigned. Conceptually, in most cases, it is desired that the the Hammersley-Clifford theorem [14], [15], Gibbs
same cluster label be assigned to spatially adjacent pixels. In distribution takes the following form:
implementing these ideas, the Bayesian framework provides a P (Φ ) = Z1 exp(− β ¦ V N i (Φ )) (3)
natural approach to address the problem of image
segmentation. Pixel intensity information is captured by the where, Φ is a vector of features, β is called the
likelihood term in Bayesian Networks, while a priori biasing regularization parameter and the normalizing constant Z
term captures the spatial location information with the help of is called a partition function. V N i (Φ ) denotes clique
MRF model.
m
In dental biometrics, where dental features are required to potential of the label configuration p within the
be extracted from X-ray teeth images, the existing image
neighbourhood N i of the ith pixel which can be
segmentation algorithms are ineffective if they are unable to
handle the inherent lighting and imaging conditions of X-ray calculated as;
images. We propose an adaptive Gaussian mixture model that
employs MRF to handle these changing conditions in X-rays. V N i (Φ ) = ¦ g (u i ,m ) (4)
The conceptual idea behind the MRF model is that the m∈N i
neighbouring pixels must have influences on a pixel. What we where, u i , m denotes the distance between the two label
capture through the MRF model is a pixel's interaction with its
i m
neighbouring pixels. In other words, we adapt the MRF model vectors p and p . The function g(u) must be gradually
to capture the spatial relationship among the neighbouring increasing and non-negative. The neighbourhood
pixels. As a result, it determines the probability of a pixel to
remain in the same cluster with its adjacent pixels. In this way,
N i consists of horizontal and vertical neighbours of i.
the spatial information captured addresses the various X-ray
imaging artefacts such as noise and non-homogeneity in To calculate the posterior probability distribution of one
intensity, as the neighbourhood information gathered is used random variable while the value of another is given, we make
to achieve a higher degree of discriminative ability in use of Bayes theorem. According to Bayes theorem [14], the
determining different pixel clusters or segments. We provide posterior probability density function (PDF) is defined as
below our mathematical model in adapting the MRF approach follows:
through the introduction of neighbourhood relationship among
the pixels in addition to the pixel features in order to Posterior PDF = Likelihood Term x Prior Probability
 STOCHASTIC MODEL BASED APPROACH FOR BIOMETRIC IDENTIFICATION 305

Normalising Constant intensity and texture could play a significant role in achieving
this goal. Hence, in this research work, we have effectively
Then, Posterior PDF for a random variable X, given that the utilized the pixel’s gray level intensities and texture features
data Y = y takes the form as follows: for clustering purpose. Here, Daubechies wavelet transform
has been employed as texture descriptor. We combine i) the
gray level intensity, ii) Daubechies transform wavelet
descriptor, iii) Cluster Ensembles, a robust number of
component / segment finding tool, where the SNOB [17], [10]
where, fX(x) is the prior density of X and the likelihood is used as a starting point, and iv) the proposed MAP fitting
function as a function of x is given by Gaussian mixture model, into an integrated segmentation
model. The main purpose of such a novel image segmentation
approach is to achieve greater accuracy in meaningfully
,
segmented components that is required for a wide variety of
industrial images.
is the normalizing constant, We adopt the following four steps for the experimental
and is the posterior density of X, given investigation of our proposed dental X-ray image
the data Y = y. segmentation model:
Step 1: Extract the texture features,
Now, we can derive a posterior log density function as follows: Step 2: Find the number of components,
N Step 3: Determine the cluster ensembles, and
¦
P (Φ, Γ | X ) = log( x i | Φ, Γ) + log P (Φ )
i =1
(5) Step 4: Segment the image using the proposed MAP Fitting
Gaussian Mixture Model
The Expectation-Maximization (EM) algorithm requires that
i
Step 1: Extract the Texture Features
the computation of the conditional expectation value, z j of Texture characterizes local variations of colour or
the hidden variables must be at the Expectation Step (E-step). intensity in an image. The neighbourhood property is a
The E-step computes the expected value using current common similarity among all the texture descriptions.
estimate of parameter and the observed data in the image [16]. Statistical methods consider texture as random phenomena
Due to the application of the MAP estimation of the with identifiable local statistics. So, we can define texture as
parameters { p j } and {θ j } , the mathematical expression for an image feature which is characterized by the spatial
i
distribution of gray values (for X-ray images) or colour
z ij takes the form as given below: patterns (for colour images) in a neighbourhood surrounding
the pixel.
i(t ) i (t ) As the segmentation quality greatly depends on the
(t ) p j ψ ( x |θ j )
z i
= ( )
choice of feature (intensity, colour, texture, and co-ordinate)
¦
j K t
(6) descriptor, in our proposed approach, we have utilized pixel’s
p i ψ ( χ i |θ (j t ) )
l =1 l gray level intensity along with texture feature in the analysis
Next, the maximization of the following log-likelihood of contextually dependent physical phenomena of X-ray
corresponding to the complete data set is performed in the images. Hence, in this research work, we adopt a more context
Maximization Step (M-step) with t denoting the iteration step: sensitive image segmentation technique using wavelet
QMAP (Φ, Γ | Φ ( t ) Γ ( t ) ) (7) transform to produce reliable and accurate segmentation of X-
ray images based on subtle intensity and texture variation.
Although, wavelet transform has made significant
The M-step determines the maximum likelihood estimate of
contribution in several areas of image processing such as
the parameter using the data from the E-step [16]. For each
image enhancement, image compression, and image
parameter, QMAP can be maximized independently, which registration, its use in image segmentation is very limited and
yields the next iteration update equations for parameter of the is gaining attention recently [18-21]. Islam et al. [21] have
component densities μ j . Thus, the problem used wavelet transforms in their image segmentation approach
( t +1)
and [σ j ]
2 ( t +1)
to extract texture features from natural colour images and
of local minima in the EM algorithm has been successfully have reported promising results. However, image analysis in
addressed by using Simulated Annealing. dental biometrics demands a more sensitive texture descriptor
that could provide an efficient way to capture the subtle
III. PROPOSED DENTAL X-RAY IMAGE SEGMENTATION
variations in texture. The Daubechies wavelet transform has
APPROACH
the capability to better represent image semantics such as
In different imaging applications like medical diagnosis, intensity variations and segment configurations [22], [23].
biometrics, security and surveillance, mineral and mining, and Hence, in this research, we have used Daubechies wavelet
material science, where the primary goal is accuracy in image transform as the texture descriptor under Markovian
segmentation, making use of subtle information related to framework for the purpose of image segmentation as it is
306 ISLAM ET AL.

found to be more accurate and sensitive in capturing texture the image. Then the Expectation Maximization (EM)
details in an X-ray image. algorithm is used for estimating the model parameters. In fact,
model parameter estimation, incorporating the neighbourhood
Step 2: Find the Number of Components relationships among pixels and segmentation of pixels are run
Existing approaches are found to be either incapable of simultaneously within MAP estimation under Markovian
handling data with noise or do not perform well in handling framework. The whole process is being done using a
discrete data like image data [7], [10]. Our experimental study particular iterative procedure in an interleaved manner. In
indicates that an unsupervised data clustering approach is doing so, MAP classifies the pixels of the input images into
ideal for dental X-ray images and is found to be more different pixel classes on the basis of intensity and texture
effective in handling noisy image data that have diverse features. Here, the function QMAP plays a crucial role in
varieties of faults and imperfections.
After the extraction of the texture features, in this step, determining the actual membership of a class. In fact, the
intensity and texture features of pixel directly influence the
we input all these features along with pixel’s intensity value
into the Snob program [17] to determine the number of classes maximization of the QMAP function. That means every
present in the X-ray image. We have used the Snob program individual feature has some influence on the maximization
only as a starting point to find the initial number of process of the QMAP function. Some features have greater
components. In general, Snob has been found quite effective
influences while others have minimum effect. If the influence
in identifying the number of components in an image but
of a particular feature is greater than others in the
sometimes it produces spurious results due to its limitations to
handle some imaging artefacts like noise, shadow, and maximization process of the QMAP function for a certain
intensity non-homogeneity. Hence, we refine the number of pixel, then the clustering of that pixel is greatly influenced by
components by determining the cluster ensembles in the next that particular feature.
step.
IV. MODEL SIGNIFICANCE AND EXPERIMENTAL RESULTS
Step 3: Determine the Cluster Ensembles Recognition of a particular object in an image is a task
In this step, we devise a robust unsupervised technique to that is carried out as pixel classification and the most
find the number of components in an image automatically. An significant aspect in pixel classification is the feature
emerging approach like Ensembles of Clusterings could play a extraction. Feature extraction is a process of measurement and
significant role to augment the technique for finding accurate it forms the main role in the process of object recognition or
and authenticated number of components in an image. image segmentation. The performance of image segmentation
Ensembles of clustering are basically a combination of in an experimental study depends strongly on the choice of
multiple clusterings used to arrive at a consolidated clustering feature extraction approach.
without accessing the original data and algorithms. Here, only Our approach in image segmentation is different from
symbolic cluster labels are mutually shared. Multiple previous work in a few major ways. We find in existing
clusterings can be produced either using a single clustering literature, approaches that adopt pre-processing procedures
algorithm by varying the different parameters or using such as gum-line detection, pixel classification and adaptive
different clustering algorithms on the same data set and thresholding are based on classical techniques like
parameters [24], [25]. With Snob as a starting point for getting segmentation using edge, line, contour and region [26-28].
initial clusterings, we base upon Ensembles of Clusterings to The algorithms that are proposed in other previous work are
produce a consolidated clustering. In our experimental exclusively based on high level features using classical
procedure, we produced multiple clusterings for different approaches that are computationally more expensive. They
parameter values and then selected the stable solution found in have not used any stochastic or statistical models that provide
the plot of the number of clusters. The convergence criteria or maximum accuracy. In our novel approach, we have
number of iterations of MRF model is the main parameter that developed a stochastic model to make it efficient for pixel
we have varied to get multiple clusterings. based image segmentation. Though some research has been
conducted in using stochastic model based approach for
Step 4: Segment the Image Using the Proposed MAP Fitting
processing magnetic resonance (MR) images of human organs
Gaussian Mixture Model
for medical diagnosis [29], most of these existing approaches
Next, the four original features, such as intensity / gray are greatly influenced by similarity in intensity. In dental
scale level, and texture, along with the number of classes images, mere similarity in pixel’s intensity does not really
obtained from Ensembles of Clusterings, become the input to mean they belong to the same object. In the front of
our segmentation model as discussed in Section II. background intensity of dental images, many objects of
Accordingly, clustering of pixels is done on the basis of similar intensity may lie, which can be characterized through
homogeneity in intensity and texture properties. As a result, variation of their textures.
regions are obtained where there is a discontinuity in either In our proposed approach, we have used the Daubechies
intensity or texture. wavelet transform for texture feature extraction as it has the
In our segmentation model, the K-means algorithm ability to capture subtle texture information with better
determines the centres of the clusters for each component of
 STOCHASTIC MODEL BASED APPROACH FOR BIOMETRIC IDENTIFICATION 307

computational efficiency. Some researchers have used Image I (A) (B)

directional wavelet transform and image gradient averaging
techniques for enhancing X-ray images [18], [19]. However,
as mentioned before, such intensity-based enhancement
techniques are not suitable for dental images. Wavelet
transforms are yet to be employed as a texture descriptor in
teeth X-ray image segmentation scheme that uses MRF model
Image II (A) (B)
theory and hence our approach is unique. In addition, we have
introduced a robust technique for automatic determination of
the number of pixel classes in images based on Cluster
Ensembles, which can be considered as a novel approach in
the context of finding the number of different segments in
dental X-ray image.
Regarding the experimental study of our proposed Image III (A) (B)
approach, our primary aim is to verify that the algorithm is
able to identify different meaningful segments or regions in
dental X-ray images accurately. The effectiveness of the
algorithm is based on whether the algorithm takes into
account the subtle variations in gray-level intensity and
texture and thereby able to produce segmentation in a precise
manner. While finer image segmentation has a great potential Fig. 1. (A) Original X-ray dental image and (B) Segmented dental image
in many application domains including bio-medical, bio-
technical, material sciences, mining as well as oil and gas Further, a quantitative performance evaluation with
industries, in this research we concentrate on dental X-ray another similar existing method [20] has been done using a set
images as we wish to explore the use of dental image of criteria in the form of an evaluation matrix from previous
segmentation for biometric identification. existing studies [30]. We input the segmentation results of our
In our proposed segmentation approach, we have made proposed approach as well as results from previous existing
use of the fact that gray-level intensity and texture based method for generating the evaluation matrix. The output from
segmentation are capable of producing sharp boundaries and these evaluation criteria at each of the three stages is a set of
homogenous regions in dental images. In fact, individual scores as shown in Fig. 2. The overall scores are determined
objects can be detected more efficiently with the combined based on criteria such as segmentation accuracy and correct
features of gray-level intensity and texture than when either number of objects identified in an image. From Fig. 2, the
the texture or intensity feature is used separately. As a result, higher scores of our proposed approach as compared to
our segmentation approach with combined gray-level intensity existing method clearly confirm that our proposed approach is
and texture outplays others in terms of sharpness and able to segment the images more efficiently. Our experiments
homogeneity. The number of classes has been determined conducted on real-life dental X-rays reveal that the proposed
initially, which has been taken as input for the proposed algorithm is also capable of capturing the finer details of an
segmentation model where the parameter estimation and pixel image in terms of its gray-level intensity and texture
clustering are done iteratively in an interleaved manner accurately. Eventually, these regions/segments that represent
simultaneously during the segmentation process. A series of internal structures of human teeth can be utilized as biometric
image segmentation experiments have been conducted on digital data for identification purposes.
real-life teeth X-rays to evaluate the effectiveness of the
proposed method. Segmented images as sample outputs of our Image I performance results:
model are illustrated in Fig. 1. Proposed Approach Existing Method
In Fig. 1, we present 3 segmented images using our Stage 1 0.8151 0.7305
Stage 2 0.6985 0.6147
proposed model as against their original images. We observe
Final Stage 0.8359 0.6914
that our approach has identified 5, 6, and 4 distinct regions in
Image I, Image II, and Image III respectively, which was Image II: performance results:
difficult to be identified with existing edge-detection or Proposed Approach Existing Method
region-growing methods due to the noise present in the Stage 1 0.8173 0.7458
original data. Hence, our approach has the ability to perform Stage 2 0.7541 0.6879
beyond the classical approaches of image segmentation that Final Stage 0.69723 0.6528
are unable to capture subtle variations in intensity and texture
Image III: performance results:
in dental X-rays, especially with noise. Proposed Approach Existing Method
Stage 1 0.8147 0.7639
Stage 2 0.6987 0.5825
Final Stage 0.6754 0.5769
Fig. 2. Comparative evaluation of proposed approach
308 ISLAM ET AL.

V. CONCLUSIONS AND FUTURE WORK [10] C.S. Wallace, and D..L. Dow, “MML Clustering of Multi-state,
Poisson, Von Mises Circular and Gaussian Distribution”, Statistics and
In this paper, we have proposed a novel dental image Computing, Vol. 10, No. 1, pp. 73-83, 2000.
segmentation approach and have investigated its use in dental [11] G. McLachlan and D. Peel, Finite Mixture Models. New York: John
biometric identification. Our stochastic model based approach Wiley&Sons, 2000.
[12] P. Meinicke and H. Ritter, “Resolution-Based Complexity Control for
utilizes a Markov Random Field model to capture the Gaussian Mixture Models,” Neural Computation, vol. 13, no. 2,
relationships among the neighboring pixels and integrate that pp. 453-475, 2001.
information into the Expectation Maximization model fitting [13] M. Alf`o, L. Nieddu and D. Vicari, “A Finite Mixture Model for Image
Maximum a Posteriori algorithm. The paper has described Segmentation”, Statistics and Computing,, Vol. 18, No. 2, pp. 137 –
150, 2008.
unsupervised means of automatic discovery of classes or [14] S. Geman and D. Geman, “Stochastic Relaxation, Gibbs distribution,
clusters in images to produce precise segmentation of dental and the Bayesian Restoration of Images”, IEEE Transactions on Pattern
X-ray images using intensity and texture information along Analysis and Machine Intelligence, Vol. 6, No. 6, pp. 721–741, 1984.
with neighbourhood relationships among image pixels. Such [15] P. Clifford, “Markov Random Fields in Statistics”, In: Geoffrey
Grimmett and Domnic Welsh (Eds.), Disorder in Physical Systems: A
an approach has provided the required accuracy in Volume in Honour of John M. Hammersley, pp. 19–32. Oxford
segmentation as well as in finding the number of components University Press, 1990.
in an image automatically through the use of Cluster [16] T.K. Moon, “The Expectation Maximization Algorithm”, IEEE Signal
Ensembles and MRF model. In addition, our results are Processing Magazine, Vol. 13, No. 6, pp. 47-60 1996.
[17] D.M Boulton,. and C.S Wallace, “A Program for Numerical
enhanced by the use of Daubechies wavelet transform for Classification”, The Computer Journal, Vol. 13, No.1, pp. 63–69, 1970.
increasing the content sensitivity of the segmentation model, [18] D. Heric and B. Potocnik, “Image Enhancement by Using Directional
thereby addressing even noise in images. Finally, the Wavelet Transform”, Journal of Computing and Information
iterations of the algorithm have been converged using specific Technology- Vol. 14, No. 4, pp. 299–305 .2006.
[19] J. Feng, N. Xiong and B. Shuoben, “X-ray Image Enhancement Based
criterion on model parameters as they were calculated for on Wavelet Transform,”, Proceedings of IEEE Asia-Pacific Services
arriving at the resulting image segments of dental X-rays in an Computing Conference, APSCC, pp. 1568-1573, 2008.
interleaved manner. A pilot experimental study has tested the [20] M. Islam, J. Yearwood and P. Vamplew ”Unsupervised Color Textured
algorithm on real-life X-ray images and has provided Image Segmentation Using Cluster Ensemble and MRF Model” ,
Advances in Computer and Information Sciences and Engineering,
convincing results. Springer Netherlands, ISBN 978-1-4020-8740-0 (Print) 978-1-4020-
Our future work will test the performance of our proposed 8741-7 (Online), pp. 323-328, 2008.
approach described in this paper using large data sets so as to [21] M. Islam, P. Vamplew and Yearwood J., “MRF Model Based
determine the threshold level of noise in image that could still Unsupervised Color Textured Image Segmentation Using
Multidimensional Spatially Variant Finite Mixture Model”,
be catered within the model. Future work also entails in International Joint Conferences on Computer, Information, and
addressing the important research question with regard to the Systems Sciences, and Engineering (CISSE 2008), Bridgeport, USA,
practical implementation of human dental identification in 2008.
real-life biometric applications. [22] I. Daubechies, “Orthogonal Bases for Compactly Supported Wavelets”,
Communications on Pure and Applied Mathematics, Vol. 41, pp. 773-
REFERENCES 789, 1988.
[23] J.Z. Wang, G. Wiederhold, O. Firschein and S.X. Wei, “Content-based
[1] J. D. Woodward, N.M Orlans and P.T Higgins, Biometrics: Identity Image Indexing and Searching Using Daubechies’ Wavelets”,
Assurance in the Information Age, Berkeley, CA: McGraw-Hill/Osborne, International Journal of Digital Library, 1, pp. 311-328, 1997.
2003 [24] A.L.N. Fred, and A.K. Jain, “Data Clustering Using Evidence
[2] J. R. Vacca, Biometric technologies and verification system. Oxford: Accumulation”, Proceedings of the 16th International Conference. on
Butterworth-Heinemann, 2007. Pattern Recognition 'ICPR'02', Quebec, Canada, 2002.
[3] L.R. Long, S. Antania, D.J. Leeb, D.M. Krainakc, and G.R. Thomaa, [25] A. Strehl, and J. Ghosh, “Cluster Ensemble - A Knowledge Reuse
“Biomedical Information from a National Collection of Spine X-rays – Framework for Combining Multiple Partitions”, Journal of Machine
Film to Content-based Retrieval”, Proceedings SPIE , 2003 Learning Research, Vol. 2, pp. 583-617, 2002.
[4] A. K. Jain and H. Chen, ``Matching of Dental X-ray Images for Human [26] H. Chen and A. K. Jain, “ Tooth Contour Extraction for Matching
Identification,'' Pattern Recognition, Vol. 37, No. 7, pp. 1519-1532, Dental Radiographs,” Proceedings of 17th International Conference.
2004. on Pattern Recognition 'ICPR'04,’ Cambridge, UK, 2004.
[5] G. Fahmy, D. M. Nassar, E. Haj Said, H. Chen, O. Nomir, J. Zhou, R. [27] J. Zhou and M. Abdel-Mottaleb, ``Automatic Human Identification
Howell, H. H. Ammar,M. Abdel-Mottaleb, and A. K. Jain, “Toward an Based on Dental X-ray Images,'' Proceedings of SPIE Technologies for
Automated Dental Identification System,” Journal of Electronic Homeland Security and Law Enforcement Conference, the Biometric
Imaging, Vol. 14, No. 4, pp. 1–13, 2005. Technology for Human Identification Conference, Orlando, USA, 2004.
[6] S. Purushotham, and M. Anouncia, “Enhanced Human Identification [28] E. Haj Said, D. M. Nassar, and H. H. Ammar, “Image Segmentation for
System Using Dental Biometrics”, Proceedings of the 10th WSEAS Automated Dental Identification,” SPIE Electronic Imaging, San Jose,
International Conference on Neural Networks, Prague, Czech Republic, CA, 2006.
2009, pp. 120 -125, 2009. [29] Y. Zhang, Michel Brady and S. Smith, ‘‘Segmentation of Brain MR
[7] C.S. Wallace, and D.M. Boulton, “An Information Measure for Images Through a Hidden Markov Random Field Model and the
Classification”, Computer Journal, Vol. 11, No. 2, pp. 185-194, 1968. Expectation Maximization Algorithm,’’ IEEE Transactions on Medical
[8] N. R. Pal and S. K. Pal, “A Review on Image Segmentation Imaging, Vol. 20, No. 1, pp. 45- 57, 2001.
Techniques”, Pattern Recognition, Vol. 26, No. 9, pp. 1277 - 1294, [30] Islam, M., “Unsupervised Color Image Segmentation Using Markov
1993. Random fields Model”, Master by Research Thesis, Graduate School of
[9] Shih, F.Y. and Cheng, S., “Automatic Seeded Region Growing for Information Technology and Mathematical Sciences, University of
Color Image Segmentation”, Image and Vision Computing, Vol. 23, No. Ballarat, Australia, 2008.
10, pp. 844–886, 2005.
 A New Stochastic Model Based Approach for Object
Identification and Segmentation in Textured Color
Image
Mofakharul Islam1 Paul A. Watters2
Internet Commerce and Security Laboratory
Graduate School of Information Technology and Mathematical Sciences
University of Ballarat, Australia
[email protected]
[email protected]

More noise means more uncertainty. Handling uncertainty

Abstract-We investigate and propose a novel stochastic model is the most significant problem in image segmentation.
based approach to implement a robust unsupervised color image Conventional probability theory was the primary mathematical
content understanding technique that segments a color textured model that used to deal with this sort of uncertainty
image into its constituent parts automatically and meaningfully. Researchers have found that probability theory alone is not
The aim of this work is to detection and identification of different
objects in a color image using image segmentation. Image
quite adequate to handle uncertainty especially where the
segments or objects are produced using precise color situation demands more precise handling. As a result, there
information, texture information and neighborhood relationships was a need to augment conventional probability models by
among neighboring image pixels. As a whole, in this particular introducing some additional concept or knowledge to make it
work, the problem we want to investigate is to implement a more efficient and effective. Over the last few decades
robust Maximum a posteriori (MAP) based unsupervised color researchers from the different computing communities
textured image segmentation approach using Cluster Ensembles, proposed various approaches that are quite efficient in
MRF model and Daubechies wavelet transform for identification handling uncertainty. Markov Random Field Model (MRF)
and segmentation of image contents or objects. In addition, theory based approach is one of them. The only difference
Cluster Ensemble has been utilized for introducing a robust
technique for finding the number of components in an image
between MRF model based approaches with other models is
automatically. The experimental results reveal that the proposed that MRF model based approaches consider the neighborhood
model is able to find the accurate number of objects or influences of pixels in addition to their pixel features (color
components in a color image and can produce more accurate and and texture) while other models consider the pixel low level
faithful segmentation of different meaningful objects from features only. This basic difference adds an extra degree of
relatively complex background. Finally, we have compared our segmentation accuracy especially while handling noisy image
results with another similar existing segmentation approach. data. Natural color images are particularly noisy due to the
environment they were produced. Therefore, it is hard to
I. INTRODUCTION develop a robust and faithful unsupervised technique for
automatic determination of number of objects in a color
Naturally, color image segmentation demands well image. Although there are a few existing approaches for
defined borders of different objects in an image. So, there is a determining the number of components in an image
fundamental demand of accuracy. The segmented regions or automatically, none of them has been found robust in all
components should not be further away from the true object situations. As a result, we introduce Cluster Ensembles to get
than one or a few pixels. So, there is a pressing need for an a consolidated clustering for finding the number of
improved image segmentation technique that can segment components in a natural color image using SNOB [2], [3] as a
different components precisely. starting point. Using a finite/classical mixture model in color
Image data have some particular characteristics that image understanding and segmentation is a popular approach
differentiate them from other form of data. Image data may though computationally intensive algorithms are a common
have corrupted values due to the usual limitations or artifacts drawback in such statistical model-based methods. Even then,
of imaging devices. Noisy data, data sparsity, and high finite mixture models have received significant attention from
dimensionality of data create difficulties in image pixel researchers in the last decade. In this work, we introduce a
clustering. As a result, image pixel clustering becomes a Maximum a priori (MAP) classifier based on a finite mixture
harder problem than other form of data. Although there are model framework, where Expectation-Maximization algorithm
some existing algorithms for unsupervised color image has been employed for simultaneous model parameter
segmentation, none of them has been found to be robust in estimation during pixel clustering. Further, the model has been
determining an accurate number of components or segments extended into a multidimensional finite mixture model to
and eventually in producing faithful segmentation. enable the handling of multiple cues/features (intensity, color,

309
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_54, © Springer Science+Business Media B.V. 2010
310 ISLAM AND WATTERS

texture, and coordinates) simultaneously. Further, Markov where, α is called the regularization parameter, the
Random Fields (MRF) model is introduced to capture the normalizing constant Z is called a partition function,
spatial relationship among neighboring pixels in terms of its and VS x denote clique potentials of the label
color and texture features [1]. In doing so, it can efficiently
handle specific situations like recognition of different objects configuration within the neighborhood S x of
in an image which is usually associated with numerous the xth pixel.
artifacts such as noise and intensity inhomogeneity. So, the An MRF is characterized by its local property whereas a
proposed model has some advantageous features, especially in GRF (Gibbs Random Field) is characterized by its global
the case of contextually dependent physical phenomena like property. At this point, we need a global or joint probability
images as it incorporates the outstanding features of the MRF distribution to represent a whole image. The Hammersly-
model without additional computational burden. Cliford theorem establishes the equivalence of these two types
The remainder of this paper proceeds as follows: In of distributions. The theorem states that any conditional
Section II we present an EM Algorithm Fitting MAP distribution has a joint distribution which is Gibbs if the
Classifier followed by the proposed image segmentation following conditions are satisfied;
model using the aforesaid MAP classifier in Section III. In Positivity: Pr( X = x) > 0
Section IV, we present previous work in this specific field.
Section V presents experimental results demonstrating the Locality: Pr( X s = xs | X t = xt , t ≠ s, t ∈ S )
effectiveness and accuracy of the proposed approach and a
= Pr( X s = xs | X t = xt , t ≠ s, t ∈ N s )
discussion on the experimental results and finally in Section
VI, we present our conclusion. where, N s is the neighborhood of site s.
II. STOCHASTIC MODEL BASED MAP Homogeneity: Pr( X s = xs | X t = xt , t ≠ s, t ∈ N s ) is
CLASSIFIER the same for all sites s.
Clique potentials of the neighborhood can be calculated
Let us assume i x is the observation at the xth pixel of an as;

image ( x = 1,....K ), and we have a Gaussian mixture model ¦V Sx (Γ ) = ¦ p (v

l∈K x
x ,l ) (4)
with an unknown number of constituents or components N
where, each component has its own vector of density where the vx ,l denotes the distance between the two
parameter λ y for each feature. label vectors p and p . The function p (v) must
x l

be gradually increasing and non-negative.

So, the probabilities of the xth pixel belonging to the yth A posterior log-density function can be derived from Eq.
class label are (3) as;
ξ yx = P( y | i x ) (1) P (Γ, Ω | I )
(5)
= ¦ x =1 log( xi | Γ, Ω) + log P(Γ)
K
where, i x = observation at the xth pixel.
The Expectation-Maximization (EM) algorithm requires
The probability density function f (i x | Γ, Ω) at any that the computation of the conditional expectation values z y
x

observation i x can be expressed as: of the hidden variables must be at the Expectation step due to
N MAP estimation of the parameters {ξ y } and
x
{λ j } ;
f (i x | Γ, Ω) = ¦ ξ yx Φ (i x | λ y ) (2)
x( t )
y =1 (t ) ξ Φ (i | Ω )
x (t )

where, f (Φ (i x | λ ) is a Gaussian distribution with

y z yx = N
y y
(6)

parameters λy = (μy ,σ y ) , and N is the number of ¦p

m =1
x( t )
m Ψ (i | Γ )x (t )
y

components in the mixture. Then, maximization of the following log-likelihood

Maximum a posteriori (MAP) introduces a prior corresponding to the complete data set is performed in the M-
distributionfor the parameter set that takes into account spatial step;
information based on the Gibbs function or distribution. Suppose,
According to the Hammersley-Clifford theorem, Gibbs K N

distribution takes the following form; U =¦ ¦z x( t )

y {log(ξ yx ) + log(π (i x | λ y ))}
1 x =1 y =1
P (Γ ) = exp(−α ¦ Vsx (Γ) (3) Therefore,
Z
 A NEW STOCHASTIC MODEL BASED APPROACH FOR OBJECT IDENTIFICATION 311

K respectively. Using the combined effect of a YCrCb color

MAP(Γ, Ω | Γt Ωt ) = U − α ¦ ¦ p (v x ,l ) (7) model and Duabechies wavelet transform, our proposed
x =1 l∈S x approach is able to identify and segment different objects in an
where, t = iteration step. image more precisely in a meaningful manner. As a whole,
For each parameter, MAP can be maximized our proposed segmentation approach combines a modified EM
independently, which yields the update equations for algorithm and MRF model based MAP classifier, Daubechies
wavelet transform and YCrCb color space along with a robust
parameter of the component densities μ y(t +1) and [σ y2 ]( t +1) . technique for finding the number components in an image
However, maximization of the function MAPψ with using cluster ensembles to implement a robust unsupervised
respect to the label parameters ξ yx ψdoes not provide a closed object identification and segmentation approach that will
produce accurate and faithful object identification on a wide
form of update equations. In addition, the maximization variety of natural color images. The most significant aspect in
procedure must also take into account the constraints image content understanding is feature extraction. Accuracy in
N
object identification greatly depends on the choice of feature
0 ≤ ξ yx ≤ 1 and ¦ξ
y =1
x
y = 1 . To address these difficulties, we (intensity, color, texture, and co-ordinate) descriptor. The
reason for using YCrCb color space is that the human eye is
take some new measures in the M-step of EM algorithm. We less sensitive to chrominance than luminance. Segmentation
set derivative of Eq. (7) equal to zero to get the expression in techniques can take advantage of this phenomenon and
the following quadratic form; subsample the values of Cb and Cr without significant visual
Suppose, degradation of the basic RGB color model. The texture
E = 4α [ ¦ p(vx ,l )](ξ yx
( t +1)
)2 descriptor plays a significant role in the analysis of
l∈K x contextually dependent physical phenomena like image in
addition to the color descriptor. This project aims to
Therefore,
investigate and implement a more content sensitive image
E − 4α [ ¦ p(vx ,l )ξ yl ](ξ yx
( t +1) (t )
) − z yx = 0 (8) understanding and object identification technique that can
l∈K x produce reliable segmentation of different objects in color
images on the basis of subtle color and texture variation and
their neighborhood relationships. In order to do that we have
In the above equation the neighborhood K x can include introduced Daubechies wavelet transforms for the first time as
pixels with updated label parameter vectors in (t + 1) step, as a texture descriptor under Markovian framework for color
image segmentation purpose. The Daubechies wavelet
well as pixels whose label vectors ξ have not yet been
l
transforms has the ability to capture subtle information about
updated in the t step. the texture while other texture descriptors are vulnerable in
Now, we get two roots of the above equation as under; those specific situations. Recently, approaches using MRF
[¦ l∈K p (vx ,l )] model and Bayesian method have provided answers to various
Suppose, F = x contemporary problems in image segmentation. The
2[¦ l∈K p(vx ,l )] application of a clustering method to image segmentation has
x
the particular characteristics that spatial information should be
Therefore, taken into account. Here in addition to intensity values of
1 pixels, the pixel location must also be used to determine the
[¦l∈K p(vx,l ) + zyx [¦l∈K p(vx,l )]]
t

cluster to which each pixel is assigned. Conceptually, in most

( t +1)
ξ yx =F±
x α x
(9) cases it is desired that the same cluster label be assigned to
2[¦l∈K p(vx,l )] spatially adjacent pixels. In implementing these ideas, the
Bayesian framework provides a natural approach. Pixel’s
x

From the above two roots, we select the root with positive color intensity and texture information are captured by the
sign + as it yields ξ yx ≥ 0 which provides a straight forward likelihood term in Bayesian Networks, while a prior biasing
term captures the spatial location information with the help of
update for the values of label parameters ξ yx of each pixel x at Markov Random Fields (MRF) model.
the M-step of each iteration of the EM algorithm. To satisfy

¦
N A. Color Feature Extraction
the other constraints i.e., 0 ≤ ξ y ≤ 1 and
x
y =1
ξ yx = 1 , we Color is an important dimension of human visual perception
utilize an active set method that use Lagrange multiplies. that allows discrimination and recognition of visual
information. Extracting and matching color features are
III PROPOSED COLOR IMAGE SEGMENTATION relatively easy. In addition to that, color features have been
APPROACH found to be effective in color image segmentation [1], [4], [5],
In this particular work, we have used YCrCb and Daubechies [6]. In this particular work, we prefer YCrCb as it has got
wavelet transforms as color and texture descriptors
312 ISLAM AND WATTERS

some specific advantage in vision and graphics as mentioned like Ensembles of Clusterings can plays a significant role to
earlier. augment the technique for finding a faithful and authenticated
number of components in an image. Ensembles of clustering
B. Texture Feature Extraction are basically a combination of multiple clusterings to get a
Texture characterizes local variations of image color or consolidated clustering without accessing the original data and
intensity. There is no formal or unique definition of texture algorithms. Here, only symbolic cluster labels are mutually
though texture based methods are commonly being used in shared. Multiple clusterings can be produced either using a
computer vision and graphics. Each texture analysis method single clustering algorithm by varying the different parameter
defines texture according to its own need. Texture can be or using different clustering algorithms on same data set and
defined as a local statistical pattern of texture primitives in parameter [11], [12]. We use SNOB as a starting point for
observer's domain of interest. Texture is often describe as getting multiple clusterings based on which Ensembles of
consisting of primitives that are arranged according to Clusterings produce a consolidated clustering. The procedure
placement rule which gives rise to structural methods of we have followed for determining the number of components
texture analysis that explicitly attempts to recover the is as follows: produce multiple clusterings for different
primitives and the replacement rules. On the other hand, parameter values and select the stable solution found in the
statistical methods consider texture as random phenomena plot of the number of clusters. The convergence criteria or
with identifiable local statistics. The neighborhood property is number of iteration of MRF model is the only parameter that
a common similarity among all the texture descriptions. So, we have varied to get multiple clusterings.
we can define texture as an image feature which is
characterized by the gray value or color pattern in a E: Segmentation of Image
neighborhood surrounding the pixel. After extraction of the six colors and texture features, we
We are using Wavelet transform to extract texture feature. put all the features into the Cluster Ensembles where the
Using Wavelet transform is a recent trend in image SNOB acts as a starting point to determine the number of
processing. Although, wavelet transform has made significant classes in the input image on the basis of these color and
contribution in the several areas of image processing such as texture features. Now, the six original features along with the
image enhancement, image compression, and image number of classes to be obtained from Ensembles of
registration, its use in image segmentation is very limited. Clusterings, are the input of our proposed MAP classifier as
Generally, Wavelet transforms performs better than other described in Section 2, where clustering of pixels will be done
texture descriptors like Gabor and MRSAR filters [10], [16]. on the basis of homogeneity in color and texture properties
Gabor filters are found efficient in capturing the strong along with the neighborhood relationships among the
ordered or coarser texture information while MRSAR filters neighboring pixels. As a result, regions are obtained where
works well in capturing weak ordered or finer texture details. there is a discontinuity in either color or texture and
Wavelet transform computation involves recursive filtering neighborhood.
and sub-sampling and each level, it decomposes a 2D signal .
into four sub-bands, which are often referred to as LL, LH, IV. RELATED PREVIOUS WORK
HL, and HH (L=Low, H=High) according to their frequency Recently, we [1] proposed a MRF model based
characteristics. In this particular work, we use Daubechies unsupervised color image segmentation approach, which is
wavelet transforms, a family of wavelet transforms discovered mainly focused on textured color image segmentation using
by Ingrid Daubechies to add more texture sensitiveness into Haar wavelet transform, CIE-Luv color space, and
our proposed approach. It has similar concepts like Haar neighborhood influences of neighboring pixels and found
wavelet but differs in how scaling functions and wavelets are promising results while applied on relatively less complex
defined. images. In this work, first order neighborhood has been
applied to capture the neighborhood influences among the
C. Finding The Number of Components neighboring pixels.
Considering the pros and cons, we have selected SNOB [2], Kato and Pong suggested a model based color image
[3], an unsupervised data clustering approach which has been segmentation approach that aims at combining color and
found more effective in handling noisy color image data that texture using Markov Random Field model theory [4]. The
have a diverse varieties of faults and imperfections. approach relies on Bayesian classification associated with
combinatorial optimization or simulated annealing where all
D. Cluster Ensembles the pixels are classified into different groups to form
In general SNOB has been found more or less effective in segments. The authors have used the perceptually uniform
identifying the number of components in an image but CIE-Luv color space and Gabor filters as color and texture
sometimes it produces spurious results due to its limitations to descriptors respectively. It appears from their experiments that
handle some imaging artifacts like noise, shadow, and they have tested their algorithm on synthetic and simple
intensity inhomogeneity. So, there is a pressing need to devise natural images only. Experimental results reveal that this
a robust unsupervised technique to find the number of approach is able to produce coarser (most prominent objects
components in an image automatically. An emerging approach only) segmentation on very simple images only. Further, no
 A NEW STOCHASTIC MODEL BASED APPROACH FOR OBJECT IDENTIFICATION 313

model parameter estimation and number of components extract different objects from an image with the highest
estimation technique has been suggested by the authors which accuracy.
makes the approach primarily supervised in nature. Regarding the proposed approach, our overall aim was to
In 2003, Kato et al. proposed a similar approach using a find an efficient and appropriate machine vision technique to
multi-layer MRF model for unsupervised segmentation that identify and extract individual meaningful objects in textured
automatically determines the number of components in the color images. The results of the proposed algorithm have been
Gaussian mixture and the associated model parameters [6]. To compared with that of one of our similar previous work [1] in
find out the number of components in the mixture, they have order to make a direct comparison of the segmentation
introduced a mathematical term to the energy function of each accuracy between these approaches. Our proposed approach
layer. The paper does not provide any detail about how it differs with the existing method in four major ways; 1) a
works. Gaussian parameter estimation has been done by using second order neighborhood is employed to capture the
an adaptive segmentation scheme where at every 10th neighborhood relationships among the neighboring pixels
iteration, they simply recomputed the mean values and more efficiently, whereas in our existing method [1], first
covariance matrices based on the current labeling and the order neighborhood was employed, 3) to have better
initial estimates at feature layers are obtained via mean shift performance in texture feature extraction especially in terms
clustering. In this paper, they have used perceptually uniform of strong texture, we use Daubechies wavelet, whereas in our
CIE-Luv color values as color features and a set of Gabor and existing method we employed the Haar wavelet only, 3)
Multi-Resolution Simultaneous Auto-Regressive (MRSAR) as YCrCb color space is employed to get a better human
texture features. perceptual color features, whereas in our existing method, we
Kato and Pong suggested another stochastic model based employed CIE-Luv. Further, in our previous work, MRF
color image segmentation method based on MRF model, model, SVFM model, Simulated Annealing, and EM
which aims at combining color and texture features [5]. The algorithm were loosely coupled. Our new framework uses an
theoretical frame works relies on Bayesian estimation via integrated model to tightly couple these techniques.
combinatorial optimization. This work is basically a The comparison of results reveals that our proposed
comparative study among the segmentation results produced approach outperform the existing similar approach in terms of
by color and texture features independently and in segmentation accuracy and has higher discriminating ability to
combination. The authors have tested their algorithm on a identify disjointed segments or regions.
variety of color images including synthetic images and have
compared their results for supervised and unsupervised
segmentation using color only, texture only and combined
features. No technique for automatic determination of number
of components has been employed; instead they manually set
the number. Using Expectation-Maximization (EM) algorithm,
they have estimated the model parameters only which makes it
semi-unsupervised in nature. So, their claims regarding the
comparison of results between supervised and unsupervised
segmentation is invalid as the algorithm they have utilized for
unsupervised segmentation was not completely unsupervised
in nature.
Yiming et al. [10] suggested another unsupervised color
image segmentation approach based on stochastic model Fig. 1. Segmentation Results with our proposed approach
where they employed Minimum Message Length (MML)
principle for finding the number of components in images and In Figure 2, the existing method has found 2 segments in
Maximum Likelihood Estimate (MLE) for pixel classification. the top image but our proposed approach able to identify 3
This is completely a color feature based approach where segments (Figure 1). If we look carefully to the image, in
neither texture feature nor has neighborhood relationship been existing approach wooden balcony in the image is merged
considered. with the building though there is a significant color and
texture difference in between them, while the proposed
V. DISCUSSSION ON EXPERIMENTAL approach able to identify the wooden balcony as disjoint
RESULTS region or segments. Our approach is able to capture the subtle
To qualitatively evaluate the performance of the proposed similarities in color and texture and able to extract or segment
approach we tested this algorithm on a wide variety of natural the region on the basis of these similarity. In Figure 1, our
images comprising different number of regions or objects and approach has identified 3 different objects in the middle image
get promising results but due to space constraint we would not while the existing approach found 4 segments or objects
be able to present all of them. The images are sourced from (Figure 2). In this case, texture features plays the dominant
Bakerely Segmentation Dataset [11]. role in the segmentation process so ignores the dissimilarity in
Figure 1 and Figure 2 show how our proposed approach color and considers the region as a single segment. In our
takes advantage of the discriminating ability to identify and
314 ISLAM AND WATTERS

proposed approach, we employ a powerful texture descriptor Bottom Image:

(Daubechies wavelet) while the existing approach employed a
comparatively less effective texture descriptor (Haar wavelet). Proposed Existing
Same thing happened to the bottom image. Here our proposed MOD_LEV1 0.7365 0.6924
approach able to identify 3 disjoints objects while on the other MOD_LEV2 0.6428 0.5987
hand the existing approach identified 4 objects. The existing ROS2 0.6133 0.5587
method identified 2 individual segments in the petal based on
Fig. 3. Comparative performance evaluation between the proposed and
the color difference while the proposed approach able to existing approach
identify the petal accurately as it is more sensitive to texture.
VI. CONCLUSION
We proposed a straight-forward, robust and effective
algorithm for identification and extraction of different
meaningful objects in a color textured image. The evaluation
of the experimental results proves the potential of the
proposed approach for unsupervised object identification or
extraction applications.

REFERENCES

[1] M. Islam, J. Yearwood, and P. Vamplew, “Unsupervised Color Textured

Image Segmentation Using Cluster Ensembles and MRF Model”,
Advances in Computer and Information Sciences and Engineering,
Fig. 2. Segmentation Results with the existing methods Springer-Verlag, 2008, ISBN: 978-1-4020- 8740-0, pp. 323-328.
[2] C.S. Wallace, D.M. Boulton, “An informal measure for classification”,
Further, a quantitative performance evaluation has been Computer Journal, Vol. 11(2), pp. 185-194.
[3] C.S. Wallace, D.L. Dow, “MML clustering of multi-state, poisson, von
done using a set of evaluation criteria in the form of an misses circular and gaussian distribution”, Statistics and Computing,
evaluation matrix as discussed in Section 3.5.2 in [12]. The Vol. 10(1), Jan. 2000, pp. 73-83.
inputs for generating the evaluation matrix is the segmentation [4] Z. Kato and T.C. Pong, “A Markov Random Field Image Segmentation
results as obtained from our proposed approach and the Model for Color Textured Images”, Image and Vision Computing, Vol.
24(10), pp. 1103-1114.
existing method. The output from these evaluation criteria is a [5] Z. Kato and T.C. Pong, “A Markov Random Field Image Segmentation
set of scores as determined by the criteria. The final evaluation Model for Combined Color and Textured Features”, W. Starbek (ED),
matrix is generated by combining the two scores as produced Proceedings of ICCAIP, Vol. 2124 of LNCS, Springer, Warsaw,
by two segmentation methods. Highest score determine the pp. 547-554.
[6] G.R. Shivani, P. Manika, and D. Shukhendu, “Unsupervised
best segmentation results in terms of segmentation accuracy Segmentation of Texture Images Using a Combination of Gabor and
and finding the correct number of objects in an image. Wavelet Features”, ICVGIP 2004, Proceedings of the Fourth Indian
In Fig. 3, we present an evaluation matrix for quantitative Conference on Computer Vision, Graphics & Image Processing,
comparison of results as obtained from these two approaches. Kolkata, India, Vol. December 16-18, 2004, pp. 370-375.
[7] A.L.N. Fred and A.K. Jain, “Data Clustering Using Evidence
The higher scores in the performance evaluation matrices Accumulation”, In the Proceedings of the 16th Intl. on Pattern
further confirm that the proposed approach clearly recognition ’ICPR’02’
outperforms the existing method. [8] A. Strehl and J. Ghosh, “Cluster Ensemble - a knowledge reuse
framework for combining multiple partitioning”, Journal of Machine
Learning Research, Vol. 2(2002), pp. 583-617.
Top Image: [9] Randen and J.H. Husoy, “Filtering for Texture Classification: A
Comparative Study”, IEEE Transaction on Pattern Analysis and
Proposed Existing Machine Intelligence, Vol. April, 1999, pp. 291-310. Analysis and
MOD_LEV1 0.7944 0.6361 Machine Intelligences, Vol. 23(8), pp. 800-810
[10] W. Yiming, Y. Xiangyu , and C.K. Luk, “Unsupervised Color Image
MOD_LEV2 0.7309 0.6589 Segmentation based on Gaussian Mixture Model”, In Proceedings of the
ROS2 0.7436 0.6211 2003 Joint Conf. of the 4th Intl. Conf. on Information, Communications
and Signal Processing, Vol. 1(15-18 Dec. 2003), pages: 541-544.
[11] D. Martin, C. Fowlkes, D. Tal, and J. Malik, “A Database of Human
Middle Image: Segmented Natural Images and its Application to Evaluating
Segmentation Algorithms and Measuring Ecological Statistics”, Proc.
Proposed Existing 8th Int’l Conf. Computer Vision, July 2001, Vol. 2, pp. 416-423.
MOD_LEV1 0.8125 0.6987 [12] Islam, M., “Unsupervised Color Image Segmentation Using Markov
Random fields Model”, Master by Research Thesis, Graduate School of
MOD_LEV2 0.7845 0.6352 Information Technology and Mathematical Sciences, University of
ROS2 0.6574 0.6214 Ballarat, Australia, 2008.
An Unsupervised Stochastic Model for Detection and
Identification of Objects in Textured Color Images
Using Segmentation Technique
Mofakharul Islam1 Paul A. Watters2
Internet Commerce and Security Laboratory
Graduate School of Information Technology and Mathematical Sciences
University of Ballarat, Australia
[email protected]
[email protected]

found to be robust in determining accurate number of

Abstract-The process of meaningful image object identification is components or segments.
the critical first step in the extraction of image information for As a rule, color image contains noisier image data.
computer vision and image understanding. The disjoint regions Imaging under different environments makes this problem
correspond to visually distinct objects in a scene. In this more difficult. Therefore, it is hard to develop a robust and
particular work, we investigate and propose a novel stochastic
model based approach to implement a robust unsupervised color
faithful unsupervised technique for automatic determination
image content understanding technique that segments a color and segmentation of number of objects in a color image.
textured image into its constituent parts automatically and The presence of intensity inhomogeneities is perhaps the
meaningfully. The aim of this work is to produce precise primary difficulty with color textured image segmentation.
segmentation of different objects in a color image using color The reason behind these inhomogeneities caused by electro-
information, texture information and neighborhood relationships magnetic interference of the imaging devices (camera and
among neighboring image pixels in terms of their features using sensor). Apart from that, the limitations of image capturing
Markov Random Field (MRF) Model to get the maximum devices lead to further difficulties in getting noise free images
accuracy in segmentation. The evaluation of the results is done in some specific application areas, where security and
through comparison of the segmentation quality and accuracy
with another similar existing method which demonstrates that
surveillance miniature camera/sensor and artificial
the proposed approach outperforms the existing method by illumination sources are being used. A miniature camera and
achieving better segmentation accuracy with faithful artificial unidirectional illumination light sources are not quite
segmentation results. enough for homogenous illumination of an object or interest
and its surroundings. As a result, captured images suffer from
I. INTRODUCTION several varieties of imperfections including white spots due to
reflectance, unclear edges, and shadows which eventually
Feature-based image segmentation has become an results in inhomogeneity.
important research focus in the image segmentation Naturally, image segmentation demands well defined
community. Several authors have suggested different borders of objects as future analysis and processing is
approaches to improve the performance and efficiency of pixel completely based on accurate identification of objects. So,
classification but none of them has been found accurate in all there is a fundamental demand of accuracy. The segmented
cases. For measuring cluster similarity and finding the regions or components should not be further away from the
numbers of clusters each approach has used its own true object than one or a few pixels. So, there is a need for an
perspective which makes it challenging to devise a robust improved image segmentation technique that can segment
clustering approach. For a long time, the statistics, data different objects precisely.
mining, and machine learning communities have been Recently, approaches using Markov Random Fields
investigating to find such a robust approach. (MRF) models and Bayesian methods have provided answers
Image data has some peculiar characteristics that differs to various contemporary problems in image segmentation
entirely it from other form of data. Image data may have [1],[2], [3],[7],[5]. The theoretical framework relies on
corrupted values due to the usual limitations or artifacts of the Bayesian estimation via combinatorial optimization. Different
imaging devices. Noisy data, data sparsity, and high pixels represent different segmented regions in the input
dimensionality of data cause difficulties in image pixel image. These classes are represented by multivariate Gaussian
clustering. As a result, image pixel clustering becomes a distributions.
harder problem than other form of data. Although there are a The remainder of this paper proceeds as follows: In
few existing algorithms for unsupervised color image Section II we describe our proposed stochastic model, which
segmentation, but all of them are based on high level features includes color and texture feature descriptors, a robust tool for
like line, edge, and area. As a result, none of them has been finding the number of components/objects and the stochastic

315
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_55, © Springer Science+Business Media B.V. 2010
316 ISLAM AND WATTERS

MAP classifier that we use for image segmentation purpose. each color plane and interaction between different color
In Section III, we present previous work in this specific field. planes. They proposed an Agglomerative Hierarchical
Experimental results demonstrating the accuracy and Clustering for pixel segmentation. The model parameter has
efficiency of the proposed approach are discussed in Section been estimated using a Maximum pseudolikelihood scheme
IV and finally in Section V, we present our conclusion and but no automatic technique has been used for estimating the
future work. number of mixture components.
II. RELATED PREVIOUS WORK

Sanjay-Gopal and Herbert [1] proposed a novel approach III. PROPOSED IMAGE SEGMENTATION MODEL
for pixel labeling and gray scale image segmentation based on
a Bayesian framework using a finite mixture model where an Our proposed image segmentation model composed of
EM algorithm is used for maximum likelihood estimation of following three major sequential phases;
the pixel labels and the parameters of the mixture densities. 1. Image feature extraction
The Bayesian framework provides us with a natural approach 2. Determination of number of components in the
to implement the idea that the application of clustering image using a robust number of object/component
methods to image segmentation has the particular finding tool
characteristic, which takes spatial information of the pixel into 3. Pixel classification using a stochastic model based
consideration in addition to its intensity values. According to MAP classifier
this formulation, a likelihood term is introduced that is based
exclusively on pixel intensity information while spatial A Image Feature Extraction
location information is captured by a prior biasing term that Color textured image has different features like intensity,
uses MRF model theory. color, texture and coordinates. In our project, we have used
Recently, we [2] proposed a finite model based color and texture features for the segmentation purpose. To
unsupervised color image segmentation approach, which is segment an image into objects, nine features are extracted
mainly focused on textured color image segmentation using from the image. Three features are color features and the rest
Haar wavelet transform, CIE-Luv color space, and are texture features. CIE-Lab color space has been chosen for
neighborhood influences of neighboring pixels and found color feature extraction considering its several advantages as
promising results while applied on relatively less complex discussed in [11]. To obtain the other six features, we apply
images. the Haar wavelet transform to the L components (number of
In [3], a novel approach for segmentation of color pixels) of the image to get three features and Gabor filters
textured images based on Markov Random Fields has been under three different orientations for the rest three features.
proposed where they use CIE-Luv color values as color
features and a set of Gabor filters as texture features. The B Color Feature Extraction
proposed approach is supervised in nature. As such, no model Choice of a suitable color space is one of the main aspects
parameter estimation technique has been used and the number of color feature extraction. Most of the color spaces are three
of components was selected manually. In 2003, Kato et al. [7] dimensional where different dimensions represent the different
proposed an algorithm using a multi-layer MRF model for color components. The most common color space in computer
unsupervised segmentation that automatically estimates the graphics is RGB. Unfortunately, this 3D color space does not
number of components in the Gaussian mixture and the correspond to equal perception of color dissimilarity. As a
associated model parameters. To find out the number of result, alternative color spaces are generated by transformation
components in the mixture, they introduced a mathematical of the RGB color space. The color spaces like HSV, CIE-Luv
term to the energy function of each layer. Gaussian parameter and CIE-Lab can be generated by non-linear transformation of
estimation has been done by using an adaptive segmentation the RGB space. We prefer CIE-Lab as it represents the three
scheme where at every 10th iteration, they simply recomputed characteristics (hue, lightness and saturation) that characterize
the mean values and covariance matrices based on the current the perceptually uniform color efficiently.
labeling and the initial estimates at feature layers are obtained
via mean shift clustering. In this paper, they have used C Texture Feature Extraction
perceptually uniform CIE-Luv color values as color feature The texture features are constructed in such a way that
and set of Gabor and MRSAR filters as texture feature. similar texture represents similar intensities. As a result, a well
Further, Kato and Pong [5] proposed another similar algorithm determined value with some variance is assigned to the pixels
where they used CIE-Luv color values as color feature and a with a given texture. The image textures are being extracted
set of Gabor filters as texture feature. They utilized the EM through the wavelet transform as already mentioned earlier.
algorithm for estimation of the model parameter. They set the Wavelet transform computation involves recursive filtering
number of components (pixel class) manually. In another and subsampling; and each level, it decomposes a 2D signal
approach [8], an unsupervised segmentation algorithm which into four sub-bands, which are often referred to as LL, LH,
uses Gaussian MRF model for color textures. Their model HL, and HH (L=Low, H=High) according to their frequency
characterizes a texture in terms of spatial interaction within characteristics [9].
 AN UNSUPERVISED STOCHASTIC MODEL FOR DETECTION AND IDENTIFICATION 317

Degraded images like images from surveillance camera K

have much noise due to the inherent technical difficulty of the Subject to 0 ≤ rj ≤ 1 and
i
¦r i
j =1,
device. As such, we need some robust texture descriptors to j =1
extract our texture features. In addition to the Haar wavelet
¦ ¦ i( t )
N K
Where, U = r {log(rji ) + log(σ ( xni | θ j ))} ,
transform, we employ multi-channel filtering approach where i =1 j =1 j

the channels are represented by a bank of real-valued, even- t=iteration step, Φ =set of probability vector {r , r ,....r }for
1 2 n

symmetric Gabor filters to extract texture features as it has

been found to be effective in capturing stronger texture pixel i,Ω = set of component parameters {θ 1,θ 2 ,.....θ K },
components. α =regularization parameter, q(li ,h ) = distance between two
We extract three texture features using Haar wavelet
i h i
transform. After a one-level wavelet transforms, a 4x4 block is adjacent label vectors r and r , rj = label parameter,
decomposed into 4 frequency bands, each band containing a
2x2 matrix of coefficients. Suppose the coefficients in the HL xni =ith value of pixel in the nth plane, θ j = parameters
band are Ck +i , Ck ,l +1 , Ck +1,l , C( k +1,l +1) . Then the feature of the {μ j , σ j } , K = number of components, and N = number of
block in the HL band is computed as: pixels. The algorithm iteratively maximizes the objective
1 function (3) using the following equation (4 – 8). The
1 1 1
f =( ¦ ¦C 2
)2
(1) (t )

4 i =0 j =0
k + i ,l + j
conditional expectation value of the hidden variables ρ ij is
The reason for choosing Haar wavelet is that it has better derived as follows;
reflection of texture properties where the coefficients in (t )
rji δ ( xni | θ j )
(t )

i( t )
different frequency bands signal variations in different ρ = (4)
¦ i( t )
j K (t )
directions, such as horizontal, vertical, and diagonal [4]. In
k =1 k
r δ ( xni | θ k )
addition to that, Haar transform require less computation
The density function f ( xn | Φ, Ω) at an observation xn
i i
compared to other wavelet transform with longer filters. We
extract three texture features using Haar wavelet transform. is defined as;
Rest of the three texture features are being extracted using K
Gabor filters under different orientations. Different f ( xni | Φ, Ω) = ¦ rjiδ ( xni | θ j ) (5)
orientations can be obtained by rotating the coordinate system. j =1
In our tests, we used three orientations: 0o ; 60o ; 135o . Where, δ ( xni | θ j ) is a Gaussian distribution with
parameters θ = {μ j , σ j } .
j
D Determination of number of classes
We have tried several data classification approaches to In order to incorporate the neighborhood influences
identify the number of components automatically but most of among the neighborhood pixels, Maximum a priori (MAP)
them has been found inappropriate for handling image data introduces a prior distribution for the parameter set - that takes
properly and need major adjustment through a huge series of into account spatial information based on the Markov Random
trial and error. In that perspective, AutoClass C [10], an Fields (MRF) Model theory. An MRF is characterized by its
unsupervised Bayesian classification approach from NASA local property. At this point, we need a global or joint
has been found effective in automatic determination of the probability distribution to represent a whole image. The
number of classes in the input images on the basis of supplied Hammersly-Cliford theorem establishes the equivalence of
color and texture features. these two types of distributions. The theorem states that any
conditional distribution has a joint distribution which is Gibbs
E Stochastic Model Based MAP Classifier if the following conditions are satisfied;
Let X i = ( x1 , x2 ,....xn ) denote the observation of ith pixel
i i i
Positivity: Pr( X = x) > 0
i
of the image, where x n represents pixel values in the nth Locality: Pr( X s = xs | X t = xt , t ≠ s, t ∈ S )
dimensional plane. So, the probabilities of the ith pixel
belonging to the jth class in the nth plane are given as:
= Pr( X s = xs | X t = xt , t ∈ N s ) , where
rji = P( j | xni ) (2) N s is the neighborhood of sites s.
The objective function of our proposed MAP classifier is Homogeneity: Pr( X s = xs | X t = xt , t ∈ N s ) is the
defined as; same for all sites s.
N
The following Gibbs function takes into account spatial
MAP(Φ, Ω | Φ Ω ) = U − α ¦
t t
¦ q(l i ,h ) (3) information and express it in the form of prior
i =1 h∈Ni
distribution p (Φ ) ;
318 ISLAM AND WATTERS

1 Now, we get two roots of the above equation as under;

p (Φ ) = exp(−U (Φ )) (6)
[¦ h∈N q (li , h )]
Z
Suppose, F = i

2[¦ h∈N q (li , h )]

¦
N
where, U (Φ ) = α C Ni (Φ ) , Z is a normalizing i

i =1 Therefore,
constant, ® is a regularization parameter,
1
CNi (Φ ) denotes the clique potential function of the pixel [¦ h∈N q(li , h ) + ρ ij [¦ h∈N q (li ,h )]]
rji
( t +1)
=F±
i α i

label vectors { p } within the neighborhood N i of the ith 2[¦ h∈N q(li , h )]
h

i
pixel which can be defined as ;
(10)
C N i (Φ ) = ¦ q(l
h∈Ni
i ,h ) (7) From the above two roots, we select the root with positive sign
+ as it yields rj ≥ 0 which provides a straight forward update
i

¦
h 2 K h 2
Where li , h =| r − r | = (r − r ) and N i is
i i
i
j =1 j j for the values of label parameters rj of each pixel i at the M-
the set containing pixels that are either horizontally or step of each iteration of the EM algorithm. To satisfy the other
vertically adjacent to the pixel i. Here, we have adopted
¦
K
the function q (l ) = (1 + l )
−1 −1
from [17], which is more
constraints i.e., 0 ≤ rj ≤ 1 and
i i
j =1 j
r = 1 , different well
robust to outliers. studied mathematical approaches are available. In this
particular work, we utilize an active set method that use
Now, based on the above prior density, a posteriori log- Lagrange multiplies.
density function can be defined as;
N
p (Φ, Ω) = ¦ log f ( xni | Φ, Ω) + log p(Φ) (8) IV.
DISCUSSSION ON EXPERIMENTAL
i =1 RESULTS
Recognition of a particular object in an image is a task
To estimate the parameter rj and
i
θh for MAP, the that is carried out as pixel classification, which in many cases
Expectation-Maximization algorithm needs the conditional follows recognition of the individual object on the image. The
expectation values ρ ij as defined in Eqn. (4) at the most significant aspect in pixel classification is feature
extraction. Feature extraction is a process of measurement that
Expectation step while Maximization step maximizes the performs the main role in the process of object recognition.
objective function of our MAP classifier as defined in Eqn. The performance at segmentation depends strongly on the
(3). choice of feature extraction.
However, maximization of the function MAP with respect In this work, our overall aim was to find an efficient and
i
to the label parameters rj does not provide a closed form of appropriate machine vision technique to identify individual
objects in color textured images, which is capable to take into
update equations. In addition, the maximization procedure
account the subtle variations in color and texture and produce
must also take into account the constraints 0 ≤ rj ≤ 1 and
i
precise segmentation.
A series of image segmentation experiments on a wide
¦
K
r = 1.
i
j =1 j variety of color textured images have been carried out to
To address these difficulties, we take some new measures evaluate and compare the effectiveness of the proposed
in the M-step of EM algorithm. We set derivative of Eq. (8) approach. Since the main contribution of our work is
equal to zero to get the expression in the following quadratic development of a new stochastic model based image
form; segmentation approach, we compared our proposed approach
Suppose, with one of our earlier work that employs a stochastic
approach based spatially variant finite mixture model [2]. Our
E = 4α [¦ h∈N q(li ,h )](rji
( t +1)
)2 proposed approach differs with the existing method in four
i

Therefore, major ways; 1) we have employed a new robust technique [10]

to find out the number of components from image
E = 4α [ ¦ q (li ,h )rji ](rji
( t +1) (t )
) − ρ ij = 0 (9) automatically, which has been found to be robust in degraded
h∈N i situation, 2) a second order neighborhood is employed to
In the above equation the neighborhood N i can include pixels capture the neighborhood relationships among the neighboring
pixels more efficiently, whereas in our existing method [2],
with updated label parameter vectors in (t + 1) step, as well as first order neighborhood was employed, 3) to have better
h
pixels whose label vectors r have not yet been updated in performance in texture feature extraction especially in terms
the t step. of strong ordered texture, we use Gabor filters in addition to
the Haar wavelet filter, whereas in our existing method we
 AN UNSUPERVISED STOCHASTIC MODEL FOR DETECTION AND IDENTIFICATION 319

employed the Haar wavelet only, 4) CIE-Lab color space is

employed to get a better human perceptual color features,
whereas in our existing method, we employed CIE-Lab.
Further, in our previous work, MRF model, SVFM model,
Simulated Annealing, and EM algorithm were loosely
coupled. Our new framework uses an integrated model to
tightly couple these techniques.
We have sourced our images from Berkeley image dataset
[12]. In Figure 1, we observe that, our proposed approach is
able to detect 3 segments in all the experimental images and
segments the objects accordingly in a precise manner. It is
able to handle usual imaging artifacts like intensity
inhomogeneity and noises. Here, neighborhood relationships Fig. 1. Segmentation Results with the proposed approach
and texture play a crucial role in detecting and segmenting the
objects. In the top image, color of the bottom part of the bear’s
body is quite different than that of the top part but continuity
in the neighborhood and textures eventually influence these
regions to become members of a single region. On the other
hand, in Figure 2, we observe that the existing approach found
4 segments instead of 3. Top and bottom part of the bear’s
body are identified as separate objects due to poor
neighborhood relationships among the neighboring pixels and
less sensitivity of the texture features. Further, the sandy beach
and the sea have almost similar color but identified as separate
objects due to the neighborhood influence and texture
sensitivity of the proposed approach. We have seen similar
results in the middle and bottom images with the proposed
approach, where, horses are identified accurately as separate
objects. Grassy land and the yellowish vegetation on top of Fig. 2. Segmentation Results with the existing approach
that are identified as separate objects while on the other hand,
we observe that the existing approach again found 4 segments Further, a quantitative performance evaluation has been
in the middle image. Here, we observe over segmentation in done using a set of evaluation criteria in the form of an
the horse objects due to discontinuity in color ignoring the evaluation matrix as discussed in Section 3.5.2 in [16]. The
similarity in texture. In the bottom image, we observe the inputs for generating the evaluation matrix is the segmentation
similar results with the existing approach though it yields 3 results as obtained from our proposed approach and the
segments like the proposed approach. Here, white regions in existing method. The output from these evaluation criteria is a
the forehead and front right leg merged with the background set of scores as determined by the criteria. The final evaluation
as they find better similarity in color. Similarity in texture and matrix is generated by combining the two scores as produced
the neighborhood relationships among the neighboring pixels by two segmentation methods. Highest score determine the
couldn’t influence them to remain in the same region. best segmentation results in terms of segmentation accuracy
Experimental results show that our proposed approach and finding the correct number of objects in an image.
can achieve better segmentation results utilizing the spatial In Fig. 3, we present an evaluation matrix for quantitative
information of pixels more efficiently than our earlier comparison of results as obtained from these two approaches.
approach [2]. The higher scores in the performance evaluation matrices
However, the main contribution of this paper focuses on further confirm that the proposed approach clearly
how spatial information can be efficiently used by a stochastic outperforms the existing method.
model, and concerns only the fundamental mechanism of the
model. For simplicity, degraded images like images from Top Image:
surveillance camera, aerial images, radar images, and images
in artificial environments are not discussed in this paper. To Proposed Existing
handle those images, more complicated segmentation schemes MOD_LEV1 0.8745 0.6878
are needed, which may further improve the clustering MOD_LEV2 0.7925 0.6394
performance. ROS2 0.8436 0.5296
 320 ISLAM AND WATTERS

Middle Image: development of scalable 1- many and many-many pattern

recognition techniques which would logically sit on top of
Proposed Existing high-performance (and highly-accurate) facial segmentation
MOD_LEV1 0.8421 0.7858 routines, as identified in this paper.
MOD_LEV2 0.8134 0.7294
ROS2 0.6913 0.6301 REFERENCES
Bottom Image: [1] Sanjay-Gopal, S., and Herbert, T. J., “Bayesian Pixel Classification
Using Spatially Variant Finite Mixtures and the Generalized EM
Proposed Existing Algorithm” IEEE Trans. On Image Processing, Vol. 7, Issue No. 7, July
19
MOD_LEV1 0.7952 0.6541 [2] Islam M., Yearwood J., and Vamplew P., ”Unsupervised Color Textured
MOD_LEV2 0.7489 0.6132 Image Segmentation Using Cluster Ensembles and MRF Model”,
ROS2 0.6824 0.5964 Advances in Computer and Information Sciences and Engineering,
Fig. 3. Comparative performance evaluation between the proposed and Springer-Verlag, 2008, ISBN: 978-1-4020-8740-0, pp. 323-328.
existing approach [3] Kato Z., Pong T. C., ”A Markov Random Field Image Segmentation
Model for Combined Color and Textured Features” W. Skarbek (ED),
Proceedings of ICCAIP, Vol. 2124 of LNCS, Springer, Warsaw, 2001,
pp. 547-554
V. CONCLUSION [4] Unser, M., “Classification and Segmentation Using Wavelet Frames”
Our aim was to propose and implement a new stochastic IEEE Trans. Image Processing, Vol. 4, No. 11, pp. 1549-1560, 1995.
model based unsupervised segmentation approach for color [5] Kato Z., Pong T. C., “A Markov Random Field Image Segmentation
Model for Color Textured Images” Image and Vision Computing, Vol.
image that can successfully segment color images with each 24, No.10, pp. 1103– 1114, October 2006.
and every details present in color and texture. We have applied [6] Green, P. J., “Bayesian Reconstruction from Emission Tomography
our approach on a wide variety of natural color images and Data Using a Modified EM Algorithm” IEEE Trans. On Medical
compared our results with one of our existing similar method Imaging, Vol. 9, No. 1, pp. 84-93, 1990
[7] Z. Kato, T.C. Pong, and G. Song, “Unsupervised segmentation of
and have got better results in terms of reliability and accuracy. color textured images using a multi-layer MRF model”, Proceedings of
Future work will focus on improvement of this approach to ICIP, Vol. (1), Spain, (2003) pp. 961-964.
make it enable in handling degraded images like surveillance [8] D.K. Panjwani and G. Healy, “Markov Random Field Models for
camera, aerial images, radar images, and images in artificial Unsupervised Segmentation of Textured Color Images”, IEEE Trans. on
Pattern Analysis and Machine Intelligence, Vol. 17(10).
environments. [9] Chang, T., and Jay Kuo, C. C., “Texture analysis and classification with
Further, this approach could be applied to a wide variety tree-structured wavelet transform” IEEE Trans. Image Proc., Vol. 2,
of color images from other application domains as well. No. 4, pp. 429-441, 1993
Comparison of experimental results in terms of efficiency, [10] Cheeseman, P., Stutz, J., “Bayesian Classification(AutoClass): Theory
and Results” RIACS, NASA.
robustness and segmentation quality could be done with other [11] Ihaka R., ”Colour for Presentation Graphics”, Proceedings on the 3rd
stochastic model based approaches. The ability of International Workshop on Distributed Statistical Computing (DSC
discriminating minor variations in color and texture also 2003)
makes the proposed approach more appropriate in some other [12] Martin D., Fowlkes C., Tal D., and Malik J., ” Database of Human
Segmented Natural Images and its Application to Evaluating
specific application domains like material sciences, mining, Segmentation Algorithms and Measuring Ecological Statistics”, Proc.
metallurgy, bio-medical, bio-informatics etc. Further, 8th Int’l Conf. Computer Vision, July 2001, Vol. 2, pp. 416-423.
efficiency of Haar wavelet transform could be compared with [13] Ho, W.H., Watters, P.A., and Verity, D., ”Are younger people more
other texture descriptors like Gabor and MRSAR. difficult to identify or just a peer-to-peer effect?”, Proceedings of the
12th International Conference on Computer Analysis of Images and
Automatically segmenting objects in natural images is an Patterns, Vienna, Austria, 2007
important first step in biometrics systems that could be more [14] Ho, W.H., Watters, P.A., and Verity, D., ”Robustness of the new owner-
widely used in surveillance operations at sensitive locations, tester approach for face recognition experiments.”, Proceedings of the
such as airports. In this application area, multiple objects must IEEE Computer Society Conference on Computer Vision and Pattern
Recognition – Biometrics Workshop, Minneapolis, USA., 2007
be extracted in real-time, identified as faces, and then matched [15] Ho, W.H., and Watters, P.A. ”A New Performance Evaluation Method
in a 1-1 matching against features stored in templates during for Face Identification Regression on Misidentified Risk.”, Proceedings
enrolment [13]. Recent work had indicated that even if of the IEEE Computer Society Conference on Computer Vision and
segmentation is possible in real-time at sufficient resolution to Pattern Recognition, Minneapolis, USA., 2007.
[16] Islam, M., “Unsupervised Color Image Segmentation Using Markov
enable recognition, scaling-up face matching to the size of an Random fields Model”, Master by Research Thesis, Graduate School of
entire nation would be difficult [14]. However, the Information Technology and Mathematical Sciences, University of
development of new statistical frameworks such as the Ballarat, Australia, 2008.
concept of misidentified risk [15] - will assist in the
 Automated Classification of Sunspot Groups
with Support Vector Machines
Mehmmood A. Abd

Electrical And Computer Engineering Department, University Of Windsor, ON, Canada

Sarab F. Majed

Mathematic Department, University Of Western Ontario, ON, Canada

V. Zharkova

Department of Computing and Mathematics, Bradford University, UK.

Abstract__A new effective technique is presented for automatic group of spots. Sunspot groups can have infinite different
classification of sunspot groups on full disk white light (WL) formations and size, ranging from solo spots to the giant
solar images. This technique is implemented on images taken groups of sunspots with complex structures [1]. Despite such
from the Solar Oscillations Investigation Michelson Doppler complexity and diversity, astronomers have been able to
image (SOI/MDI) aboard the Solar Heliospheric observatory
(SOHO). The technique focuses on employing Support Vector
classify the sunspots classes based on The McIntosh Sunspot
Machines (SVMs) as effective classification tool. In addition to Classification Scheme and modified Zurich Scheme [2]. A
applying SVMs the problem of extracting sunspots and sunspot sunspot is recognized by its two main features ; the central
groups from solar image is solved in an efficient and different dark features ‘umbra’; and the outer lighter area which is
way from the ones previously employed. This technique proceeds called ‘penumbra’ (See Figure 1).Therefore,most
in several consequence phases. The first phase involves solar disk computerized recognition techniques rely on these properties .
image extracting. The second phase involves Binarization and Many manual sunsopts in different formats are produced
smoothing of an extracted solar image disk. The third phase all over the world , such as the Meudon Observatory in France
involves unsupervised segmentation of sunspots groups. This ,the Locarno Solar Observatory in Switzerland ,the Mmount
phase consists of two subphases: a) extracting the spots and, b)
combining together the sunspots that belong to the same group.
Wilson Observatory in the United States Of America and
The fourth phase involves attributes extraction of each sunspots many others . Zurich classification is based on the Locarno
group. The final phase is the classification phase using SVMs, by Catalogue which is considered a primary indicator of solar
applying a one–against–all technique to classify the sunspots activity [2][3]. The Zurich Classification is categories into
groups. The proposed technique has been tested using different seven classes of sunspots groups numbered from A to H [2]
sets of sunspots achieving approximately 88.56% recognition rate .(see Figure 2 and Table 1 ). Because of the effect of the solar
for Zurich classification of sunspot groups. activity on the Earth’s atmosphere, human activity, businesses
and even climate, the automated detection and verification
Index Term__Solar Disk, Sunspots Groups, Support Vector Machines, techniques become essential for a reliable forecast of the
Unsupervised Segmentation.
future effects on human lives and businesses. This requires
stringent requirments on the classification accuracy of any
I. INTRODUCTION
automated recognition system. Many researchers have been
S unspots have been a subject of interest to scientists and
researchers for many years. A sunspot is a region in the
developing automated reliable recognition systems with
varying recognition rate and techniques [3].
sun’s photosphere that is darker by comparison with its
photosphere background and has intense magnetic field. A
sunspot may consist of one or more umbra which often are
surrounded by a less dark area called penumbra. Sunspot
numbers increase and decrease with an irregular cycle with a
length of approximately 11 years [1]. Sunspots can be
observed in the visible continuous spectrum, also known as “
white light ‘. Larger sunspots can be observed by using Call
Fig.. 1. A high-resolution SOHO/MDI image showing sunspots with umbra
K1 absorption line image and H and Call K3 absorption line (black spots) and penumbra (dark gray) above the background (gray).
images . Sunspot shapes are varied from individual spots to a

321
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_56, © Springer Science+Business Media B.V. 2010
322 ABD ET AL.

TABLE 1 The remainder of this paper is organized as follows. Section II

SUNSPOT CLASS DESCRIPTION describes preprocessing, Section III presents the segmentation
technique, Section IV presents feature extraction
technique, Section V presents multi-class SVMs Classification
technique, Section VI discusses the results obtained from the
previous sections and summarizes the conclusions .

II. PREPROCESSING
Preprocessing consists of image acquisitions, edge
detection, noise removing and feature enhancing, and
Binarization. The input images are taken from the solar
Oscillations Investigation Michelson Doppler Imager
(SOI/MDI) aboard of the solar Heliospheric Observatory
(SOHO)[11]. Figure 3 shows an example of the input white
light full disc image .The image has dimensions of 1024x1024
pixels with 16 different intensities. The resolution is 2.6
arces/pixel. Edge detection, noise removing and Binarization
are curried out as that reported by Sarab et al [4].

In this paper a new recognition technique is presented based

using one–against-all SVMs.

The classification system receives its input from reliable

segmentation phase that employed the unsupervised
segmentation method [4]. The objectives of this system is to a)
improve the classification rate of sunspot groups
classification,b) improve segmentation techniques of sunspots
groups, c) employ more reliable attributes to enhance the
classification reliability.
The proposed methodology is described by the following
processing phases :
1. Preprocessing
2. Segmentation technique
3. Feature extraction
4. Multi–Class SVMs Classification

Fig. 3. Full disk image obtained from SOHO/MID on 1/1/2000.

Image binarization converts an image into black and white.

In this study a threshold value, t, is determined automatically
using Otsu’s method [12] .The produced image is a binary
image with white sunspots and black background. Figure 4
illustrates the output that will be feed to the next phase
/segmentation phase.

Morphological processing facilitates the sunspot groups’

attributes extracting. Meaningful shape information of
geometrical objects, measuring typical characteristics area,
bounding box, centroid coordinate, and the length of a
perimeter, is computed. Finding sunspot coordinates (x,y) and
area in pixels ,eases a determination of the sunspots locations
in relation to the solar center . Moreover, area of any sunspot
is utilized to estimate the area in heliographic coordinates.
Fig.2. The Modified Zurich Sunspot Classification System developed by
Patrick McIntosh [2]. These attributes are employed later to classify each group
belong to which class. Next section presents sunspots/ sunspot
groups detection and grouping then extracting.
 AUTOMATED CLASSIFICATION OF SUNSPOT GROUPS 323

III. SEGMENTATION TECHNIQUE four different sunspot group images where A represents the
pores and groups B, C, and D respectively [2].
In this paper an unsupervised segmentation method was
employed based on the similarity measure. The sunspot’s is IV. FEATURE EXTRACTION
extracted /cut from a full disk solar disk image by computing
the bounding rectangle (box) that contains it. In this phase the attributes of each sunspots groups are
extracted and formed into a vector of N (7 x 1) based on the
Modified Zurich classification attributes. After extracting all
the attributes the vector is fed to the SVMs classifier [6], [5].
The features consist of a number of sunspots/umbra, size,
penumbra, penumbra’s position, bipolar, unipolar, and major
longitudinal axis. The seven classes of modified Zurich
classification are used as standard classes in this paper. For
further details see [2]. The classes are numbered as A, B, C,
D, E, F, and H. To make the classification more reliable the
class letters are converted to numbers from 1 to 7. Thus, the
class A is represented by one (1) and B by (2) and so on. The
attributes of each sunspot or sunspots groups are formed into
vector and fed to the SVMs. Multi-class SVMs are employed
in this paper.

Fig. 4. Black and white image produced from the preprocessing phase.

Then an iterative algorithm is employed to estimate the

maximum bounding box that will be re-adjusted iteratively, in
order to combine the group of sunspots together. The images
used are WL of size Nr x Nc with intensities given S(i,j) ={Sij:
1 <= i <= Nr , 1<=j <=Nc } the produced image from the
previous step is masked so that the background is black and
sunspots are white. The bounding box is represented by the
vector V where V=(x,y,xw,yw) ,and (x,y) coordinates of the
left upper corner and ( xw,yw) are the x-width and y-width,
respectively. Then the other corners are represented by
(x,y+yw) (for the upper right corner), (x+xw ,y) (for the lower
left corner )and (x ,y+yw) (the lower right corner).
Fig. 5. Sample of sunspot groups segmentation technique applied on WL
The algorithm starts by checking the size of each bounding image disk SOHO/MID on 25/1/2005.
box to find the one that contains adjacent sunspots. This
technique starts by choosing a seed as the maximum sunspot V. MULTI-CLASS SVMS CLASSIFICATION
area with the maximum bounding box. Then the maximum
bounding box will be adjusted to contain the neighbor boxes. SVMs are basically binary classifiers and, thus, it is not
The segmentation algorithm can be written as follows. If the straightforward to turn them into multi-class (N-class)
initial bounding box has vector V0=(x0,y0,xw0,yw0), and the recognition systems. There are several methods invented to
neighbor box has vector , Vr=(xr,yr,xwr,ywr) , then the construct a multi–class SVM. The most typical method is to
adjustable box has the vector , Vr=(x,y,xw,yw), where construct N SVMs, each of which classifies one class against
x=MIN(x0,xr) , and y=MIN(y0,yr), then xw=MAX (xr+xw- all the other classes. This method is commonly called one-
x,yr+xwr,-x), against-all (1-v-all) [7],[10]. The second method that can be
yw=MAX (yr+ywr-y,yr+ywr-y). used is to combine all possible two-class classifiers, where for
an N-class problem N(N-1)/2 classifiers must be constructed.
The above algorithm is repeated until it finds all spots This is commonly referred to as one-against-one (1-v-1).
contained in each group and fits them into one box. Then the Decision Directed Acyclic Graph (DDAG) and Max Win
area that contains the sunspot group is extracted from the main Algorithm (MWA) are also used to construct multi-class
image and saved as a separate image that represents one SVMs based on (1-v-1). Further details can be found in
sunspot group. Figure 5 shows the segmentation results with [6],[8]. The one-against-all method is used in this study. N
SVMs are constructed, where N is the number of classes. For
324 ABD ET AL.

i=1,2…N, the ith SVM is trained with all the samples in the ith NS
(6)
class considered as positive examples and the samples from all w = ¦a i y i (x i )
other classes considered as negative examples. Given {(x1, y1), i =1

(x2, y2),…,(xk, yk)} as the training data set, where xj ∈ Rn are Where Ns is the number of support vectors and ai Lagragian
the n-dimensional samples (feature vectors) and yj ∈ multiplier for any training point [5].
{1,2,…N}are the corresponding class labels j=1,2,…,k, the ith
SV solves the following problem [6], [9]: Replacing the inner product between Φ in (5) with a function
simplifies things, where in kernel functions are introduced. In
Min
{1
k
w i ,b i , ‡ i
2
(w ) i T
w i
+ C ¦ ‡ ij } this work, Gaussian Radial Basis Functions (GRBF) are
j=1
(1) employed as kernel functions:
(w )  (x ) + b
i T
j
i
≥ 1 − ‡ ij , if y j = i § 2·
(7)
¨ −‰ x −y ¸
(w )  (x ) + b
i T
j
i
≥ 1 + ‡ ij , if y j ≠ i K ( xi , x j ) =e
¨
©
i j ¸
¹,‰ >0
‡ ij ≥ 0, j = 1,........ .k The linear kernel is a special case of GRBF as Kreethi and Lin
have shown in [9]. Sigmoid kernels produce results generally
where 1/2 (Wi)T.Wi is the regularization term (objective comparable to GRBF kernels. However, GRBF has less
function), numerical complexities in comparison to other kernels [7].
GRBF has a control parameter, ‰, which along with the cost
k
factor C are the two tunable parameters in the SVM
C ¦ j = 1
‡ i
j
(2)
optimization problem. As suggested in [9], a useful way to
determine these is to perform grid-search with cross-
validation. A two-stage procedure can also be followed, where
is a cost term (training error) used to constrain the solution in after identifying a good region in the (C, ‰) space, a finer
cases of non-separable data (‡ are ‘slack’ variables introduced search can be conducted which may improve the results
for this purpose [5]), function  is used to map the training ϕ

somewhat in some cases. The best parameters thus determined

data into a higher dimensional space, and C is a penalty factor are used in the classification.
set by the user.
The above equation will be solved for each class VI. RESULTS
i = 1,2,.., N , where the attempt is made to balance between
the regularization term and the training errors. Thus, there are The set of images are collected form SOHO /MID and
N decision functions EGSO catalogue. In this study these images are used as
database, for training and testing, in this study. Twenty images
(w ) 1 T
(x ) + b 1 (3) were used for each class of sunspots. The images are divided
........................... into two groups of ten images.

(w ) ˆ T
(x ) + b ˆ
The attributes are extracted from the images and
formatted into two text files. One of the text files is used as a
and x is said to belong to the class which gives the largest training resource and the other as a test resource. In the final
value for the decision function, i.e., testing stage, ten-fold cross validation is used to determine
good C and gamma parameter values, followed by actual
i=1,...,ˆ w (x) + b
class of x≡ argmax i i
(( )
T
) (4) classification using these values. The recognition rates
achieved are in the range of 80-88 % correct classification.
In practice the dual problem, formulated in terms of Lagrange
multipliers, is solved, i.e., maximize the follows VII. CONCLUSION

1 The system using SVMs as a classification tool in

¦a i − ¦ a i a j y i y j (xi ) ⋅ (x j )
2 i, j Sunspots Groups has shown promising results. The
i (5) classification system relies on the multi-class SVMs
0 ≤ ai ≤ C classification tool and modified Zurich classification .The
system improvement depends on the segmentation phase
images quality. Thus high quality images and with less
¦ai
i yi
distortion will enhance the classification rate. The
segmentation phase output is another important factor that has
equation (5) has the following solution an effect on the classification rate.
 AUTOMATED CLASSIFICATION OF SUNSPOT GROUPS 325

REFERENCES [8] Kreethi, S. S. and Lin C. J., ”Asymptotic behaviors of support vector
machines with gaussian kernel”, Neural Computation, Vol. 15 No.7,
pp. 1667-1689, 2003.
[1] T. Nguyen, C. P. Willis, D. J. Paddon,”Learning sunspot classification,“
[9] C-W. Hsu, C-C. Chang, and C-J. Lin, “A practical guide to support
Fundamental Informaticae XX, vol.72, 2006, pp. 295-309.
vector classification”, [Online]. Available:
[2] P. McIntosh,” Solar Physics,”, pp. 125-251, 1990.
[3] S. Zharkov, V. Zharkova, S. Ipson, A. Benkhalil,”Technique for
www.csie.ntu.edu.tw/~cjlin/papers/guide/guide.pdf.
automated recognition of sunspots on full disk solar,”, EURASIP [10] M. Abd and G. Paschos,” Effective arabic character recognition using
Journal on Applied Signal Processing, vol. 15 , pp. 2573–2584 2005. support vector machines,” In Proc.2007, Innovations and Advanced
[4] S. Majed, M.Abd and V.Zharkova,” Automated detection of sunspots techniques in Computer and Information Sciences and Engineering,
and sunspot groups in full-disk solar images,” ,Accepted in CISSE09 Springer, ISBN 978-14020-6267-4, 2007, PP. 7-11.
[5] Christopher J.C., ”A tutorial on support vector machines for pattern [11]. Solar Heliospheric Observatory (SOHO), (10-9- 2009),[Online].
recognition”, [Online] Available: Available: http://sohowww.nascom.nasa.gov/
http://aya.technion.ac.il/karniel/CMCC/SVM-tutorial.pdf. [12] M. Sezgin and B. Sankur, "Survey over image thresholding techniques
[6] Chih-Wei Hsu and Chih-Jen Lin, ”A comparison of methods for multi- and quantitative performance evaluation," Journal of Electronic Imaging.
class support vector machines”, IEEE Transactions on Neural Networks, Vol.13 (1), pp.146–165, 2004.
Vol. 13,No.2, pp. 415-425,March 2.002.
[7] Eugene A. Borovikov, ”An evaluation of support vector machines as a
pattern recognition tool”, University of Maryland at College Park,
2005., [Online]. Available:www.umiacs.umd.edu/users/yab/
SVMForPatternRecognition/report.pdf,
 A Novel Ballistics Imaging System for Firearm
Identification
Dongguang Li
School of Computer and Security Science, Edith Cowan University, Australia
Email: [email protected]

ABSTRACT makes the system very difficult to use. The firearm

identification system called Fireball [8] was developed by
The efficiency of traditional ballistics imaging system is Edith Cowan University in 1995, and it has been used by
heavily dependent upon the expertise and experience of Australia Police Services for identifying, storing, and
end-user. An intelligent ballistics imaging system is highly retrieving the images of projectiles and cartridge cases. The
demanded to address this issues. This paper presents a novel major limitations of Fireball are that the shape and position
ballistics imaging system which achieves traditional are turned manually and the precision of comparisons relies
functions and many new features including line-scan image on the expertise and experience of the end-user. The major
module, characteristics extraction module and intelligent drawbacks of the existing systems are outlined as follows:
image processing module. Benefit from these features, the Firstly, the identification of firearm strongly depends upon
system can identify firearm more efficiently and effectively. the end-used due to lack high level intelligent image
processing techniques; secondly, the image contrast of
Index Terms— Ballistics system, firearm identification, cartridge cases and projectiles is not good due to failure to
image processing find a satisfactory illuminator; finally, the lack of online
image processing, retrieving and identification does not
1. INTRODUCTION meet the demand of end-user. Therefore, the existing
systems require further improvement and development.
The analysis of marks on the cartridge case and projectile of The automatic ballistics imaging system mainly includes
a fired bullet is very important to identify the firearm from two components, i.e. the acquisition and the correlation
which the bullet is discharged [1, 2]. When a firearm is component [10]. The acquisition component includes
loaded and fired, the characteristic marks are produced on capturing of cartridge cases and projectiles and extracting
the cartridge case and projectile. These marks are significant their features in order to make them analyzable. The
evidences for identifying the suspicious firearm. Over thirty correlation components are responsible for comparing the
different features within theses marks can be referred as a features and organizing the results for the user’s inspection.
‘fingerprint’ for ballistics recognition [3]. Since each The following functions should be involved in ballistics
firearm owns its unique toolmarks [4, 5], it is possible to imaging system [11]:
identify both the type and model of a firearm and each • Evaluate the degree of similarity between two sets of
individual weapon as effectively as human fingerprint features;
identification. • If more than two bullets are involved in a comparison,
Several commercial or testing ballistics imaging systems to organize the results of a set of comparisons in some
have developed by organizations or researchers for decades convenient way;
[6-9]. Integrated Ballistics Identification System (IBIS) [9] • To provide the user with tools to verify the results
is an international ballistics imaging system which has been obtained by the correlation algorithms.
widely used in several countries. Another commercial The paper describes a novel ballistics imaging system for
system called “Bulletproof” was developed by a Canadian firearm identification. The system can intelligently capture,
company named Walsh Automation, which can acquire and store, analyze and retrieve the digital images of cartridge
store images of projectiles and cartridge cases by searching cases and projectiles. The main contribution of the system is
its image database for particular striations on projectiles and described as follows. Firstly, it improves efficiencies in time
cartridge cases. However, it is required to match the and personnel by computerizing imaging system; secondly,
impressed markings or striations by end-user on the the precision of identification has been also improved due to
projectiles and cartridge cases. This kind of limitation the introduction of artificial intelligent techniques; finally,

327
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_57, © Springer Science+Business Media B.V. 2010
328 LI

the online system is convenient for end-user and more • “B” type characteristics were defined as broach series
independent on the client computer. characteristics, with B1 and B2 subclasses that occur on
The rest of the paper is organized as follows: Section 2 the edges of land impressions.
gives an introduction to firearm identification. In Section 3, Generally, Firearm identification is concerned with two
we illustrate the ballistics imaging system. An intelligent types of characteristics, i.e. class characteristics (C type)
cartridge case segmentation algorithm is described in and individual characteristics (A type). While broach series
Section 4. Section 5 concludes this paper. characteristic (B type) is only applied in particular types of
rifles. Class characteristics are useful in determining what
2. FIREARM IDENTIFICATION brand of firearm was used, but not helpful in identifying a
individual firearm. Individual characteristics can directly
Firearm identification aims to provide a link between the match a bullet to a firearm. That is, the match between
suspect firearm and the forensic ballistics specimens based individual characteristics and a bullet can ensure that one
on the depressions, scratches and markings on the certain firearm was the only firearm that could have fired
specimens of cartridge cases and projectiles. These various the bullet. These characteristics are detailed in Table 1.
marks are called ‘toolmark’ and it is thought that the
toolmark is unique to itself [5]. In this regard, it is possible 3. BALLISTICS IMAGING SYSTEM
to identify the suspect firearm by toolmark left on ballistics
specimens. Due to several drawbacks of workstation system including
Firearm identification mainly includes the following low data sharing and security, expensive cost of install and
stages [12]: maintenance, inconvenient application environment etc, we
• The crime scene weapon is examined and test fired for developed an Online Ballistics Imaging System (OBIS)
a test specimen; based upon Browse-Server (BS) structure. Fig. 1 shows the
• The characteristic marks on the cartridge cases and structure of the system. By a computer with internet access
projectiles are found and recorded; and a web browser, users can access OBIS anywhere and
• Features of the characteristic marks are investigated; anytime.
• Comparison of marks between the test specimen and
the crime scene specimen;
• Assessment of the comparison between the test and
crime scene specimens.
• Storage of the characteristic marks for further
comparisons.

Table 1 Characteristics of a Firearm

Individual
Object Class characteristics
characteristics Fig. 1 System structure of OBIS
Caliber
Position of firing pin Striations of OBIS not only achieves the traditional functions
Cartridge
Position of extractor mark firing pin including image capturing, storing, analyzing and retrieving
Case
Position of ejector mark impression
but also provides intelligent image processing. The new
Breach or bolt impression
Caliber functions of OBIS include image acquisition module,
Twist direction of rifling characteristic extraction module, firearm identification
Land mark width Striations of module and intelligent image processing module.
Projectile Groove mark width groove
Degree of the rifling twist impressions 3.1. Image acquisition
Number of lands The digital image acquisition of cartridge cases and
Number of grooves projectiles focuses on the following issues: how to select
and locate the satisfactory illuminator to get high contrast
Charchman divided the characteristics of toolmarks on digital images; how to transform the surface of cylindrical
ballistics specimens into three major categories [13]: projectile into a 2D flat image in order to facilitate
• “C” type characteristics were defined as class identification.
characteristics; As indicated by Smith [8], in order to produce high
• “A” type characteristics were defined as accidental contrast of images, the illuminator must be installed at an
characteristics, which are individual characteristics; incident angle of greater than 45 degrees from the
illuminator to the plane of the head of the cartridge case.
Illuminator from such angles can produce better shadow
 A NOVEL BALLISTICS IMAGING SYSTEM FOR FIREARM IDENTIFICATION 329

cast of the peaks and ridges of striations. The high quality of 4. Repeat step 1, 2, and 3 until the whole surface of
shadow cast makes discernible features for the identification the projectile specimen is scanned.
process. According to Smith’s guideline, we test point 5. Combine all arrays in the buffer one after another
source illuminator and LED ring-light illuminator. Although so that a 2D flat image for the whole surface of
point source illuminator can produce better shadow cast, the the projectile is produced.
shadow cast of the image may vary with the rotational
position of the cartridge case about its central axis. While
the shadow cast of the image is invariant in LED ring-light
illuminator. Therefore, LED ring-light illuminator is
selected. Fig. 2 shows the high contrast image of head of
cartridge case using LED ring-light illuminator.

(a) Image of projectile

(b) 2D image of projectile

Fig. 3 Image of ballistics projectile
Fig. 2 The image of head of cartridge case

The cylindrical shapes of ballistics projectiles (displayed

in Fig. 3(a)) are inherently difficult to identify using
traditional incident light microscope. Maintaining the
quality of image from low magnification microscope is very
difficult because of loss of focus as the ballistics projectile
translated and rotated. To resolve the issue, the line-scan
imaging technique [14] is employed by scanning
consecutive columns of picture information and storing the
data in a frame buffer so that a 2D image of the surface of
the cylindrical projectile is produced (displayed in Fig. 3(b)).
Benefit from the line-scan technique, all points on the 2D Fig. 4 Line-scan imaging system
image are in focus, because the cylindrical projectile is
rotated about an axis of rotation relative to a stationary line 3.2. Characteristic extraction
array of sensor. Thus, during one full rotation of the Since broach series characteristic (B type) is only applied in
cylindrical ballistics projectile, all points on the rotating particular types of rifles, we focus on class characteristics
surface can be captured on the 2D image. and individual characteristics. Class characteristics on
Fig. 4 shows the line-scan imaging system including cartridge case and projectile are measured on corresponding
stepper motor, LED ring-light illuminator, microscope, images acquired by image acquisition module. Due to the
CCD camera and pc with graphic card. The procedure of lack of criteria of individual characteristics, we directly save
line-scan imaging approach is detailed as follows: the corresponding images into Database.
1. With the stepper motor’s every step; To facilitate measurement, we employ a clock face to
2. The CCD camera captures the current image of describe class characteristics on cartridge case. Cartridge
projectile specimen and send the image to PC by cases are divided into two classes, one is centrefire cartridge
graphic card; case, and the other is rimfire cartridge case. For rimfire
3. The pixels on middle column in the image is cartridge case, the firing pin mark position is designated at
extracted and saved consecutively in an array in 12 o’clock and the extractor and ejector mark positions are
the buffer on PC;
330 LI

coded with respect to this position. While, for centrefire methods because of the heavy shadow in the input image.
cartridge case the ejector and extractor mark position are To acquire the satisfactory segmented image of cartridge
orientated as they would be in the firearm. These case, a novel approach based on color invariant properties
characteristics can be easily measured and saved by mouse [15] and geometrical shape of cartridge case is proposed.
click on the graphic user interface. The method is composed of two steps. That is, remove the
It is difficult to accurately measure the class background and shadows using color invariant properties;
characteristics on projectile due to its cylindrical shape. On detect and segment the circular cartridge case using Hough
the contrary, these characteristics can be easily measured on transform [16].
2D images acquired by line-scan imaging system. Fig. 5
shows that these characteristics can be measured by
calculating the difference of coordinates of pixels. For
example, “Land mark width” can be measured using the
following steps: firstly, draw a horizontal line on “Land
mark” in the image by mouse; then calculating the number
of pixels between the starting pixel and the ending pixel of
the line. Clearly, the product multiplying the number by the
length of each pixel is the value of length of “Land mark (a) (b)
width”.

(c) (d)

Fig. 5 Characteristics measurement on 2D image (e) (f)

Fig. 6 Segmentation of cartridge case. (a) Input image,
3.3. Firearm Identification (b) Image removed background and shadows, (c) Binarized
In our system, firearm identification includes two steps: one image, (d) Edged image,(e) Detected cartridge case (white
is to retrieve the short list of suspect firearm by searching circle), (f) Segmented cartridge case
class characteristics saved in database. The other is to make
sure the unique suspect firearm from the short list by 4.1. Remove background and shadows
comparing their high resolution images in detail manually. The color invariant properties are employed to remove the
The short list can be easily generated by graphics query background and shadows. Color invariants are function
interface. Various image techniques are provided to further which describes the color configuration of each pixel on
compare the images in short list. Benefit from these image images discounting shadows, and highlights. These features
techniques, it is easy to find the suspect firearm. These are demonstrated to be invariant to a change in the imaging
techniques are listed as follows: conditions including viewing direction, object’s surface
• Freely zoom, translate and rotate images; orientation and illumination. Color invariants features are
• Transparentize and overlap images; acquired by RGB using the following equations:
• Add or remove “image mask” including grid, clock face. R ( x, y )
C1( x, y ) = arctan
4. INTELLIGENT SEGMENTATION OF max(G ( x, y ), B( x, y )
CARTRIDGE CASE G ( x, y )
C 2( x, y ) = arctan
The segmentation of cartridge case from the input image max( R( x, y ), B( x, y )
(displayed in Fig. 6(a)) is of paramount importance for B ( x, y )
firearm identification. However, the satisfactory result C 3( x, y ) = arctan
max( R( x, y ), G ( x, y )
cannot be achieved using traditional threshold-based
 A NOVEL BALLISTICS IMAGING SYSTEM FOR FIREARM IDENTIFICATION 331

where, R(x, y), G(x, y), and B(x, y) respectively Acknowledgement

represent the red, green and blue components of a pixel in a The ballistics specimens were provided by Australia Police.
image, while the C1(x, y), C2(x, y) and C3(x, y) represent Mr Z. Huang , as a visiting fellow at ECU, helped with digital
corresponding color invariants. image acquisition.
The procedure to remove background and shadows REFERENCES
is detailed as follows: [1] Dongguang Li, "Ballistics Projectile Image Analysis for
1. Select a background whose values of color Firearm Identification," IEEE Transaction on Image Processing,
invariants do not overlap the values of cartridge vol. 15, pp. 2857-2865, 2006.
case. It is well known that most cartridge cases are [2] R. Saferstein (ED), "Forensic Science Handbook," Volume 2.
made by brass. Therefore, they have similar values Englewood Cliffs: Prentice Hall, 1988.
of color invariants. In our research, the range of [3] G.Burrard, "Identification of Firearms and Forensic Ballistics,"
London, U.K:Herbert Jenkins, 1951.
color invariants for cartridge cases
[4] A. Biasotti and J. Murdock, "Criteria for identification or State
is C1 ∈ [0.71,1.11], C2 ∈ [0.4,0.80], C3 ∈ [0,0.80] , of the Art of Firearm and Toolmark Identification," AFTE Journal,
the color with R=54, G=12, B=170 is selected as vol. 16, pp. 16-34, 1984.
background. Its color invariants is C1= 0.31, [5] A. Schwartz, "A Systemic Challenge to the Reliability and
Admissibility of Firearms and Toolmark Identification," The
C2=0.07, C3=1.26.
Columbia Science and Technology Law Review, vol. VI, pp. 1-42,
2. Capture the image of cartridge case based on the 2005.
background selected in step 1. [6] Chenyuan Kou, Cheng-Tan Tung, et al., "FISOFM: Firearms
3. Remove the pixels whose color invariants near to the Identification based on SOFM model of Neural Network," in
values of background. Thus, the background and Security Technology, 1994. Proceedings. Institute of Electrical and
shadows are removed (Fig. 6(b) shows the result). Electronics Engineers 28th Annual 1994 International Carnahan
Conference, Albuquerque, NM, USA, 1994, pp. 120-125.
4.2. Segment cartridge case [7] C.L. Smith, Robinson. M, et al., "Linescan Imaging for the
Although the most background and shadows are positive identification of bal-listics Specimens," in IEEE 34th
Annual 2000 International Carnahan Conference, Ottawa, Ont.,
removed in the Section 4.1, the image of cartridge case
Canada, 2000, pp. 269-275.
isn’t suitable for identification. A few noise pixels [8] C. L. Smith, "Fireball: a forensic ballistics imaging system," in
belong to the background does not be removed, and The Institute of Electrical and Electronics Engineers 31st Annual
the borders of the cartridge case are not smooth. To 1997 International Carnahan Conference, Canberra, 1997, pp. 64-70.
address these issues and accurately segment cartridge [9] IBIS (Integrated Ballistics Identification System).
case, the Hough transform is employed. The method http://www.fti-ibis.com/.
segmenting cartridge case includes the following steps: [10] Natalie G. Carpenter, "An Examination Of A Three-
1. Binarize the image displayed in Fig. 6(b) using Dimensional Automated Firearms Evi-dence Comparison," in
Otsu algorithm (shown in Fig. 6(c)); Department of Criminology College of Arts and Sciences
2. Detect the edges to the binarized image using University of South Florida, April 8, 2004.
[11] Bachrach B., "Development of a 3D-based automated
Sobel operator (shown in Fig. 6(d));
firearms evidence comparison system," Journal of Forensic
3. Detect the circle center and radius of the cartridge Sciences, vol. 47, pp. 1-12, 2002.
case in the edged image using Hough transform [12] C. L. Smith, "Multi-dimensional cluster analysis of class
(shown in Fig. 6(e)); characteristics for ballistics specimen identification," in Security
4. Segment the cartridge case based the circle center Technology, 2001 IEEE 35th International Carnahan Conference,
and radius from the input image (shown in Fig. London, UK, 2001, pp. 115-121.
6(f)). [13] J. Churchman, "The reproduction of characteristics in signatures
of Cooey rifles," RCMP Gazette, vol. 11, pp. 133-140, 1949.
5. CONCLUSION [14] A. Zographos, M. Robinson, et al., "Ballistics Identification
Using Line-Scan Imaging Tech-niques, in Security Technology,
1997. Proceedings. The Institute of Electrical and Electronics
This paper presents a novel ballistics imaging system for Engineers 31st Annual 1997 International Carnahan Conference,
firearm identification which not only involves traditional Canberra, ACT, Australia, " 1997, pp. 82-87.
functions including image capturing, storing, analyzing and [15] T. Gevers, "Color-based object recognition," Pattern
retrieving but also intelligent image processing. The system Recognition, vol. 32, pp. 453-464, 1999.
can identify firearm more efficiently and effectively due to [16] R. O. Duda and P.E. Hart, "Use of the Hough Transformation
the new functions including the line-scan imaging system to Detect Lines and Curves in Pictures," Comm. ACM, vol. 15,
and intelligent image processing. Future work includes pp. 11-15, 1972.
researching the criterion for extracting individual
characteristics and increasing more intelligent analysis for
firearm identification.
 Color Seal Segmentation and Identification
Dongguang Li

School of Computer and Security Science, Edith Cowan University, Australia

E-mail: [email protected]

Abstract—Automatic seal imprint identification system is high- rotate C to make it have the same orientation as R, and com-
ly demanded in oriental countries. Even though several seal pare the corresponding details between them. If all of the de-
identification techniques have been proposed, it is seldom to tails are satisfactory to the technician’s subjective judgment,
find the papers on the recovery of lost seal imprint strokes the candidate seal imprint C is regarded as a genuine one.
caused by superimposition. Most identification methods are
conducted by matching the reference seal imprint and candi-
According to the process mentioned above, the hard-
date one pixel by pixel. In this paper, a robust color seal seg- matching consists of two steps, i.e. segment seal imprint
mentation and identification approach is proposed. This ap- from the input image, and then identify whether the seg-
proach segments the seal imprint from the input image in mented one is genuine or forgery. There are three main is-
terms of the adaptive thresholds. The lost seal imprint strokes sues on computerizing this procedure. Firstly, the input seal
are recovered based on the text stroke width that can be de- imprint is usually superimposed by other information such
tected automatically. In addition, the moment-based seal iden- as signature or background text, consequently, the segmen-
tification is to compare the reference seal imprint and the re- tation of seal imprint is difficulty[5]. Secondly, the differ-
covered one. Experimental results show that the proposed me- ence between the genuine seal imprint and the forgery one is
thod is able to correctly segment seal imprint and efficiently
discriminate the genuine and forgery seal imprint.
small, so the results are very sensitive to noise. Thirdly, it is
very difficult to make the candidate seal imprint have exact
Keywords-Seal segmentation, Seal identification, Image same orientation and size as the reference one.
binarization, Moment invariant Automatic seal segmentation have been studied in
many literatures [6-9]. These methods were mostly imple-
I. INTRODUCTION mented based upon RGB or HSV color space and can ac-
quire satisfactory results. However, none of the methods
Seals are widely used in oriental countries such as Chi- focuses on how to detect and recover losses of seal imprint
nese, Japanese and Korea to identify a person, a social strokes in segmented seals. The losses of seal imprint
group, or an organization etc[1, 2]. Currently, a variety of strokes often exist in segmented seal imprint due to supe-
seals must be manually registered, which relies on the exper- rimposition among the seal, signature and background text.
tise and experience of professional technicians. The process The result of seal identification may be strongly affected by
of seal identification not only consumes so much time, but losses of seal imprint strokes. As a consequence, it is of
also inevitably introduces some misidentification. Therefore, great importance in automatic seal imprint identification.
a seal imprint verification system is highly demanded to re- Several seal identification methods have also been pro-
solve this issue automatically, speedily and reliably. posed for decades [10-14]. These methods mainly include
Seals can be made of various materials e.g. wood, stone, four steps: 1) Rotate the candidate seal imprint to the same
marble, plastics, ivory, crystal, metal. The shape and size of orientation as the reference one; 2) Compute the centroid to
seals are also different. Basically, there are three shapes, i.e., the candidate imprint and reference one; 3) Overlap the ro-
circular, rectangular, and elliptic[3]. Generally, the color of tated candidate seal imprint and the reference one in terms
seal imprint is pure red or pure blue. Seal identification can be of the centroid; 4) Match them pixel by pixel. From these
regarded as a template match between the reference seal im- steps, we can see that the performance and robustness of
print and the candidate one. To exactly match the seals, a seal identification strongly depend upon the precision to
variety of features have to be considered, including the cha- detect the angle of rotation and position of centroid. Owing
racters in the imprint, the scales, distributions, relative spatial to the difficulty in exactly overlapping the input seal imprint
positions, strokes etc. As a result, the matching is called and reference one, the method matching candidate seal im-
“hard-matching” [4]. Currently, hard-matching is usually print and reference one pixel by pixel is difficult to use.
performed by human visual inspection. To describe the Clearly, the exact detection of rotated angle is very compli-
process of seal identification in detail, we define two va- cated and difficult.
riables R (Reference seal imprint) and C (Candidate seal im- In this paper, a robust approach for seal imprint seg-
print). To identify whether C is genuine or forgery, a profes- mentation and identification is proposed. To improve the
sional technician firstly overlap the centroid of R and C, then precision and robustness to segment seal imprint, an adap-

333
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_58, © Springer Science+Business Media B.V. 2010
334 Li

tive thresholding method based on RGB components is pre- seal imprint, signature and background. Seal imprint is ex-
sented. This segmentation method is very suitable for both tracted based on the cluster index. Since K-means method
seal segmentation and extracting a particular color cluster divides the pixels in terms of Euclidean distance, the com-
from an image. The recoveries for the losses of seal imprint pletely different color may be divided into a same cluster.
strokes are considered for minimizing the errors in seal im- For example, the color (R=255, G=0, B=0) and (R=0,
print identification. A moment-based method is employed G=255, B=0) are red and green, but they have the same
because it is independent on the image translation, rotation Euclidean distance to black (R=0, G=0, G=0). This method
and scaling. As a result, the proposed method not only may also get incorrect result when noise is strong. Liang[6]
avoids the complicated match processing but also improves divided the RGB space into eight subspaces corresponding
the robustness and precision to identify seal imprints. The to color index 0, 1, 2 ,…, 7 by dichotomy, and then compute
flowchart of the proposed method is shown in Fig. 1. It con- the color index for each pixels in input image in terms of
sists of five major steps: 1) Segment the seal imprint in equation (2.1), where “[]” represents integer Round opera-
terms of the adaptive thresholds; 2) Recover the losses of tor, r, g, b represent respectively red, green and blue com-
seal imprint strokes in segmented seal imprint; 3) Binarize ponents in color normalized image. Finally, extract the seal
the segmented seal imprint and the reference one using Otsu imprint by index (e.g. the index of red is 1). This method
algorithm[15]; 4) Compute the Hu’s moment invariants[16] may also be easily attacked by noise.
to the segmented seal imprint and the reference one; 5) Index= [b/128] × 4 + [g/128] ×2 +[r/128] (2.1)
Identify them based on Euclidean distance of moment inva- Shih-Hsu [12] proposed a register method in terms of
riants. the feature points. This method includes four major steps: 1)
Feature points extraction; 2) Estimate the rotation and trans-
lation parameters using the point matching technique; 3)
Seal registration using the affine transformation by the esti-
mated parameters; and 4) Fake detection from the registered
seals. However, the seal imprints with different orientation
stamped by the same seal may generate different feature
points that may strongly affect the stability of this algo-
rithm. To rotate the input seal imprint and the reference one
to the same orientation, Yung-Sheng[10] used the contour
analysis to find the principal orientation of a seal imprint,
and then rotate the input seal imprint to the same orientation
with the reference one. The algorithm may misidentify if the
seal imprint does not exist principal orientation such as the
Fig. 1. Seal imprint identification system seal imprint displayed in Fig. 2(b).
In this paper, we employ the Otsu algorithm [15] for
The rest of the paper is organized as follows: Section 2 the adaptive seal segmentation. Otsu is a typical and popular
gives an introduction to related works. In Section 3, we illu- image binary algorithm based on global threshold. This me-
strate the proposed approach. The experimental results are thod divides input data into two classes by threshold. The
presented and analyzed in Section 4. Section 5 concludes basic idea of Otsu is to find an optimal threshold that gives
this paper. the best separation between the two classes in terms of their
values. The threshold can be estimated automatically on an
II. RELATED WORK iterative scheme, in order to find the maximum variance
A variety of seal identification techniques have been between the two classes. It works well in situations where
proposed. In general, the studies mainly focus on the seal there is a reasonably clear valley between the modes of the
segmentation and seal registration. Seal segmentation deals histogram related to the two classes.
with how to exactly extract seal imprint from the input im- Hu’s moment invariants[16] are very effective on identi-
age such as bank check image. Seal registration is to identi- fying the input seal imprint. Moment invariants have been
widely applied to image pattern recognition in a variety of
fy that the input seal imprint is genuine or forgery by com-
applications due to its invariant features on image transla-
paring the input and the reference seal imprint. tion, scaling and rotation. Two-dimensional (p+q)th order
Katsuhiko[9] extracted the signature and seal imprint moments are defined in equation (2.2).
by projecting orthogonally all pixels of a bank check image ∞ ∞
on to an appropriate axis in RGB color space[17]. This me- m pq = ³ ³ x p y q f ( x, y )dxdy
thod may fail if the center of three clusters is all in or near a −∞ −∞ (2.2)
straight line. In [14], the input image is first transformed p, q = 0,1,2,
into HSV (Hue, Saturation, and Intensity-Value ) perceptual To achieve the independence for image translation and
space[18], and then K-means algorithm[19] is employed to rotation, the equation (2.2) is transformed into central mo-
divide the input image into three clusters corresponding to ment shown in equation (2.3)
 COLOR SEAL SEGMENTATION AND IDENTIFICATION 335

∞ ∞ A. Seal Segmentation
μ pq = ³ ³ ( x − x ) p ( y − y ) q f ( x, y)dxdy
(2.3)
−∞ −∞ Seal segmentation aims to extract the seal imprint from
p, q = 0,1,2, the input image and recover the losses of seal imprint
strokes caused by superimposition. It mainly contains the
where x =m10/m00 and y =m01/m00, which are the cen- following steps: 1) Estimate the thresholds to segment the
seal imprint, 2) Segment the seal imprint and text from the
troid of the image. Scale invariance can be obtained by nor-
malization. The normalized central moments are defined in input image, 3) Recover the losses of seal imprint strokes, 4)
equation (2.4) Binarize the segmented seal imprint and reference one.
μ pq 1. Estimate Thresholds
η pq = , γ = ( p + q + 2) / 2, p + q = 2,3, (2.4)
μ 00γ In our demonstration, the color of seal imprint is red,
i.e. the red component of seal imprint is stronger than
Based on normalized central moments, Hu[16] intro- green and blue components. For three color clusters, i.e.
duced seven moment invariants: red (seal imprint), white (background) and black (text), the
R (Red), G (Green) and B (Blue) components are equal for
ϕ1 = η 20 − η 02  white(R=G=B=255) and black(R=G=B=0), The red seal
φ2 = (η 20 −η 02 ) 2 + 4η112  imprint can be easily segmented under the conditions R>G
and R>B. Fig. 3 shows the segmented result only based on
φ3 = (η 30 − 3η12 ) 2 + (3η 21 − μ 03 ) 2  the conditions R>G and R>B. The result shows that so
φ4 = (η 30 + η12 ) 2 + (η 21 + μ 03 ) 2  many pixels belonging to background or text are also meet
the conditions in practice because of various noisy factors
φ5 = (η30 − 3η12)(η30 +η12 )[(η30 +η12)2 − 3(η21 +η03)2 ] such as camera, illumination.
+ (3η21 −η03)(η21 +η03) [3(η30 +η12 )2 − (η21 +η03)2 ]
φ6 = (η 20 −η 02 )[(η 30 +η 12 ) 2 - (η 21+η 03 ) 2 ]
 
+ 4η 11(η 30 +η 12 )(η 21+η 03 )
φ7 = (3η21 −η 03)(η 30+η12)[(η 30+η12) 2 - 3(η 21+η 03) 2 ]
- (η 30-3η12)(η 21+η 03)[(3(η 30+η12) 2 - (η 21+η 03) 2 ]

This seven moment invariants are useful properties of be-
ing unchanged under image scaling, translation and rotation. Fig. 3. Segmented Seal with Noise
III. PROPOSED APPROACH
In general, the shapes of seal may be circular, rectangu-
lar, or elliptic. The colors of seal imprint may be red or blue.
For simplicity, only circular seal and red seal imprint are
selected in our demonstration. The genuine seal imprint is
shown in Fig. 2.(a). Obviously, the image is composed of
three color clusters, i.e. the red seal imprint, the black text
and the white background. Fig. 2.(b) displays the reference
seal imprint in which only includes white background and
red seal imprint. The forgery seal imprint is displayed in Fig.
2.(c).
Fig. 4. Histograms for the values of R/G and R/B

We can found that the ratios of R/G and R/B for noise
pixels are far less than the ratios of seal imprint pixels. It is
possible to find the thresholds of R/G and R/B to divide the
pixels which meet the conditions (R>G and R>B) into two
classes, i.e. the noise pixels and the seal imprint pixels. To
(a) Genuine seal imprint (b) Reference seal imprint (c) Forgery seal imprint
find the optimal thresholds for R/G and R/B, the histo-
Fig. 2. The input images for demonstration
grams for the ratios of R/G and R/B on pixels whose R>G
and R>B are constructed, as shown in Fig. 4. Since both
336 Li

histograms of R/G and R/B have evident bimodality, the 3. Recover the losses of seal imprint strokes
optimal thresholds can be acquired using Otsu algorithm
[15] detailed in Section 2. For Fig. 2(a), the optimal thre- Since the brightness and color vary in superimposition
sholds of R/G and R/B calculated by Otsu are respectively of text and seal imprint inks, the character strokes of seal
2.1 and 1.8. imprint in superimposition may lose in segmented seal im-
print. This may cause incorrect result in seal imprint regis-
2. Segment Seal Imprint and Text tration process. To avoid this, a method for detecting and
After acquiring the thresholds, the seal imprint can be recovering losses of seal imprint strokes is proposed in this
segmented easily. Since the text in input image will be section. Comparing the Fig. 7(a) and Fig. 5(b), we can see
used in subsequent process, the seal imprint and text are that the losses of seal imprint strokes basically take place in
segmented in the same time. The steps of segmentation are superimposition between the seal imprint and text. Further-
detailed in algorithm 1. more, the width of lost stokes is very close to the width of
stokes corresponding text. Accordingly, we apply the fol-
Algorithm 1 Segmenting Seal Imprint and Text from lowing steps to recover the lost strokes of seal imprint.
the Input Image • Estimate the width of the strokes of black text;
Input: an image with red seal imprint, black text and • Recover the lost strokes in seal imprint image in
white background terms of the width of the strokes of black text.
Output: an image of red seal imprint, an image of To estimate the stroke width of black text, we scan
black text each row of black text image (shown in Fig. 5(b)) to calcu-
late the number of each continuous black pixels, and then
i: Estimate the threshold THRG and THRB using method construct the stroke histogram using all of the number of
described above; continuous black pixels. The position of maximum in histo-
ii: Scan input image, get values of the red, green and gram is considered as the stoke width of black text. For the
blue component in each pixel and save in variable image shown in Fig. 5(b), the histogram is displayed in Fig.
r, g and b respectively; 6. Evidently, the stroke width of black text is 4 pixels.
iii: if r/g> THRG and r/b> THRB
The pixel is belong to seal imprint Iseal
Else
The pixel is belong to background or text IBT
End if;
iv: Binarize the pixels in IBT using Otsu algorithm [15]
and then get the text image Itext;
v: Remove the black block in Itext whose size less than
5 pixels;
vi: Return the seal imprint Iseal and black text Itext.
Fig. 6. Histogram of stroke width on black text
The segmented seal imprint is displayed in Fig. 7(a).
And the black text with noise is shown in Fig. 5(a). To re- After getting the stroke width of black text Wstroke, we
move the noise, the size of each black block is computed. can recover the segmented seal imprint using the following
And then remove the small block whose size less than 5 procedures. Scan the black text image shown in Fig. 5(b).
pixels. After this operation, a clear text image is acquired For each black pixel, detect its Wstroke hWstroke neighbor
and showed in Fig. 5(b). pixels, if the pixels located at up and down position or left
and right position of the black pixel are all red, the white
pixel with the same position as the black text image in the
seal imprint image (shown in Fig. 7(a)) is converted into
black. The recovered seal imprint is displayed in Fig. 7(b).

(a) Text with noise (b) Noise-removed text

Fig. 5. Images of black text (a) Seal imprint without strokes recovery (b) Recovered seal imprint

Fig. 7 Segmented seal imprint

 COLOR SEAL SEGMENTATION AND IDENTIFICATION 337

B. Seal Registration D are circular seals (Type 1), Seals E to H are rectangular
Normally, the registration includes two steps: 1) Ac- seals (Type 2), and Seal I to L are elliptic seals (Type 3).
cord the orientation and size of candidate seal imprint with Each type includes two genuine seals and two forgery seals.
the reference one; 2) Overlap the candidate seal imprint and All seals are engraved by one maker, therefore the differ-
the reference one in terms of their centroid, and then match ence between the genuine seal and forgery one are very
them pixel by pixel. Since exact adjustment on the orienta- small. The experimental steps are describes as follows.
tion of candidate seal imprint is very difficult or even im- • Stamp each genuine seal on white paper, and convert it
possible, especially to circular seal, the precision of registra- into a digital image by CANON IXUS 90 IS camera as
tion may be strongly affected by adjusted precision. We a reference seal imprint. Then transform them into bi-
propose a novel moment-based approach to address the dif- nary images and calculate the moment invariants on the
ficulty. The moment invariants are independent of transla- binary images;
tion, rotation and scaling, so the complicated adjustment in • Arbitrarily stamp each of twelve seals on five different
orientation and size can be avoided. documents. And then convert them into digital images
by CANON IXUS 90 IS camera. Thus, sixty candidate
seal imprints are applied to examine the proposed algo-
rithm;
• Segment each seal imprint from input images with seal
imprint and recover losses of seal imprint strokes. Then,
transform them into binary images;
• Calculate the moment invariants on each image from
(a) Genuine seal imprint (b) Reference seal imprint (c) Forgery seal imprint last step.
Fig. 8. Binarized seal imprint • Calculate the Euclidean distance between the reference
seal imprint and corresponding candidate seal imprints.
To minimize the effect of luminance, we transform all • Make sure whether the candidate seal imprint is ge-
seal imprints into binary images with black background and nuine or not in terms of the selected threshold.
white seal imprint. The sample seal imprints are displayed
in Fig. 8. Seal registration is achieved by computing the
seven moment invariants on all seal imprints and Euclidean
distance (shown in equation (3.1)) based on these invariants.
7
Ci − Ri
D=¦ (3.1) (A) Genuine (B) Forgery (C) Genuine (D) Forgery
i =1 Ri
where Ci, Ri respectively present the seven moment in-
variants of candidate seal imprint and reference one. Table 1
displays the results of comparison between the reference
seal imprint and genuine or forgery one. Although it is very
difficult to manually distinguish the reference seal imprint (E) Genuine (F) Forgery (G) Genuine (H) Forgery
from the forgery one (displayed in Fig. 8), the Euclidean
distance between the reference seal imprint and forgery one
is evidently greater than the values between the reference
seal imprint and genuine one. Consequently, the forgery seal
imprint can be easily identified by the Euclidean distance. (I) Genuine (J) Forgery (K) Genuine (L) Forgery
Fig. 9. Twelve seals for experiments
Table 1. Comparison of moment invariants for refer-
ence, genuine and forgery seal imprint Fig. 10 shows the results of recovered seal imprints
segmented from the input image ( only Seal A and B are
displayed). Although the orientation of seal imprints are
different from each other, which not affects the results of
identification due to the moment invariants are independent
to rotation. The experimental results are displayed in Table
IV. EXPERIMENTAL RESULTS 2. From this table, we can see that the Euclidean distance
To evaluate the performance of the proposed seal seg- between genuine and reference seal imprints are all less than
mentation and identification algorithm, extensive experi- 1, and the Euclidean distance between the reference and
ments are conducted based on sample images. Twelve seals forgery seal imprints are all greater than 1.8. In real applica-
are used in the experiments (displayed in Fig. 9). Seals A to tion, the errors to identify a forgery seal imprint as a ge-
338 Li

nuine one is not acceptant. Therefore, in our experience, we Euclidean distance to identify seal imprints. In addition,
select 1 as the threshold to identify whether the seal imprint more experiments will be conducted to check whether the
is genuine or forgery. approach is able to be applied to more extensive conditions.
VI. ACKNOWLEDGEMENT
Mr Z. Huang, as a visiting fellow at ECU, helped with
digital image preparation.
REFERENCES
[1] H. Leung, "Analysis of traditional Chinese seals and synthesis of
personalized seals," in Multimedia and Expo, 2004. ICME '04. 2004
IEEE International Conference on, 2004, pp. 1283-1286 Vol. 2.
[2] J. J. Brault and R. Plamondon, "Segmentation handwritten signatures
at their perceptually important points," IEEE Trans. Pattern. Anal.
Fig. 10. Five seal imprints of A (row 1) and five seal im- Mach. Intell., vol. 15, pp. 953-957, 1993.
prints of B (row 2) [3] Chou Ju-Pei and Chen Hu-Sheng, "Chinese seal identification," in
Security Technology, 1990. Crime Countermeasures, Proceedings.
IEEE 1990 International Carnahan Conference on, 1990, pp. 51-53.
Table 2. Comparison of moment invariants for twelve seals [4] Yung-Sheng Chen, "Computer Processing on the Identification of a
Chinese Seal Image," in Third International Conference on Docu-
ment Analysis and Recognition (ICDAR'95), 1995, pp. 422-425.
[5] Yung-Sheng Chen, "Approach to the Chinese Seal Registration," in
Advances in Communication Systems and Electrical Engineering.
vol. 4, 2008, pp. 529-542.
[6] Liang Cai and Li Mei, "A Robust Registration and Detection Method
for Color Seal Verification," in Advances in Intelligent Computing.
vol. 3644, 2005, pp. 97-106.
[7] Ching-Liang Su, "Chinese-Seal-Print Recognition by Color Image
Our Proposed approach has been implemented using Dilating, Extraction, and Gray Scale Image Geometry Comparison,"
Visual Basic .NET, and running on a PC (CPU P4 2.0G, Journal of Intelligent and Robotic Systems vol. 44, pp. 349-359,
2005.
RAM 1G). The processing time to segment and recover the [8] Aureli Soria-Frisch, "Color Seal Extraction from Documents: Ro-
seal imprint is about 0.9 second for 500*500 color image. bustness through Soft Data Fusion," EURASIP Journal on Applied
And the processing speed for calculating the moment inva- Signal Processing, vol. 2005:13, pp. 2146-2152, 2005.
riants is about 0.8 second. As a result, the whole computa- [9] K. Ueda, "Extraction of signature and seal imprint from bankchecks
by using color information," in Document Analysis and Recognition,
tional time is less than 2 seconds. 1995., Proceedings of the Third International Conference on, 1995,
pp. 665-668 vol. 2.
V. CONCLUSION [10] Yung-Sheng Chen, "Registration of Seal Images Using Contour
In this paper, a robust segmentation and identification Analysis " in Lecture Notes in Computer Science. vol. 2749: Springer
Berlin / Heidelberg, 2003, pp. 255-261.
algorithm for color seal imprint is proposed. This method is [11] Ching-Liang Su, "Edge distance and gray level extraction and orien-
composed of segmentation of seal imprint, recovery of tation invariant transform for Chinese seal recognition," Applied
losses of seal imprint strokes, and identification of seal im- Mathematics and Computation, vol. 193, pp. 325-334, 2007.
print. The segmentation automatically extract seal imprint [12] Shih-Hsu Chang, Fang-Hsuan Cheng, et al., "Automatic seal identifi-
cation by using point matching and fake detection," Pattern
from the color input image in terms of the adaptive thre- Recognition Letters, vol. 20, pp. 1545-1552, 1999.
sholds. To minimize the errors, the losses of seal imprint [13] T . Horiuchi, "Automatic seal verification by evaluating positive
strokes are recovered by detecting the width of black text cost," in Document Analysis and Recognition, 2001. Proceed-
stroke and superimposition pixels between the red seal im- ings. Sixth International Conference on, 2001, pp. 572-576.
[14] K . Ueda, T. M utoh, et al., "Automatic verification system for seal
print strokes and the black text strokes. To avoid compli- imprints on Japanese bankchecks," in Pattern Recognition,
cated adjustment on orientation and size between the refer- 1998. Proceedings. Fourteenth International Conference on,
ence seal imprint and candidate one, the moment-based 1998, pp. 629-632 vol. 1.
technique is applied to seal imprint identification. [15] N . Otsu, "Threshold Selection Method from Gray-Level Histo-
grams," IEEE Transcations on Systems Man and Cybernetics,
Experimental results showed that the proposed seg- vol. 9, pp. 62-66, 1979.
mentation method is able to segment seal imprint from color [16] H u Ming-Kuei, "Visual pattern recognition by moment invariants,"
input image, and the losses of seal imprint strokes can be Information Theory, IRE Transactions, vol. 8, pp. 179-187, 1962.
correctly recovered. Moreover, the identification method [17] Rafael C. Gonzalez and Richard E. Woods, Digital Image
has high performance in exactly discriminating the genuine Processing (Third Edition): Prentice Hall, 2007.
[18] AR . . Smith, " Color gamut Transformation Pairs," Computers
and forgery seal imprint. The computational time is less Graphics, vol. 12, pp. 12-19, 1978.
than 2 seconds on general PC. Therefore, the proposed ap- [19] Y. Linde, A.Buzo, et al., "An algorithm for vector quantizer design,"
proach is fast and feasible. In future, we will find more fea- IEEE Trans. on Communications, vol. 28, pp. 84-95, 1980.
tures on seal imprints, and employ neural network instead of
 Efficient SOPC-based Multicore System Design
using NOC
Vignesh Veerapandian, Xingguo Xiong
Department of Electrical and Computer Engineering, University of Bridgeport, Bridgeport, CT 06604, USA
Email: [email protected], [email protected]

Abstract — Due to the advancement of VLSI (Very Large Scale As a result‚ around 1999 several research groups have
Integrated Circuits) technologies, we can put more cores on a started to investigate systematic approaches to the design of
chip, resulting in the emergence of a multicore embedded the communication part of SOCs. It soon turned out that the
system. This also brings great challenges to the traditional
parallel processing as to how we can improve the performance
of the system with increased number of cores. In this paper, we
meet the new challenges using a novel approach. Specifically,
Problem has to be addressed at all levels from the physical
we propose a SOPC (System on a Programmable Chip) design to the architectural to the operating system and application
based on multicore embedded system. Under our proposed level. Hence‚ the term Network on Chip (NOC) is today
scheme, in addition to conventional processor cores, we used mostly in a very broad meaning‚ encompassing the
introduce dynamically reconfigurable accelerator cores to hardware communication infra-structure‚ the middleware
boost the performance of the system. We have built the and operating system communication services and a design
prototype of the system using FPGAs (Field-Programmable methodology and tools to map applications onto a NOC. All
Gate Arrays). Simulation results demonstrate significant this together can be called a NOC platform. [4] Networks on
system efficiency of the proposed system in terms of
Chip (NOCs) have emerged as a viable option for designing
computation and power consumption. Our approach is to
develop a highly flexible and scalable network design that
scalable communication architectures. For multiprocessor
easily accommodates the various needs. This paper presents System-on-Chips (MPSOCs), on-chip micro networks are
the design of our NOC (Network on Chip) which is a part of used to interconnect the various cores. The main idea with
the platform that we are developing for a reconfigurable NOCs, besides the solutions to the physical issues, is the
system. The major drawback of SOPC based systems lies in the possibility for more cores to communicate simultaneously,
routing of the various on-chip cores. Since it is technically leading to larger on-chip bandwidths. The adoption of NOC
difficult to integrate more than one core on a single chip, we architecture is driven by several forces: from a physical
come across several routing problems which lead to inefficient design viewpoint, in nanometer CMOS technology
functioning. Thus we implemented several NOC based routing
interconnects dominate both performance and dynamic
algorithms which considerably improve accessing speed and
enhance the system efficiency.
power dissipation, as signal propagation in wires across the
chip requires 2 multiple clock cycles. NOC links can reduce
Keywords: Multicore system, System on a Programmable the complexity of designing wires for predictable speed,
Chip (SOPC), Network on Chip (NOC), Multiprocessor power, noise, reliability, etc., thanks to their regular, well
System-on-Chip (MPSOC). controlled structure. From a system design viewpoint, with
the advent of multi-core processor systems, a network is a
I. INTRODUCTION natural architectural choice.

During the 1990s more and more processor cores and large NOC can provide separation between computation and
reusable components have been integrated on a single communication; support modularity and IP reuse via
silicon die‚ which has become known under the label of standard interfaces; handle Synchronization issues; serve as
System-on-Chip (SOC). Buses and point-to-point a platform for system test and hence increase engineering
connections were the main means to connect the productivity.
components, Hence they can be used very cost efficiently.
As silicon technology advances further‚ several problems
related to buses have appeared. [1][5] Buses can efficiently II. DIFFERENTNOCTOPOLOGIES
connect 3-10 communication partners but they do not scale
to higher numbers. The Network-on-Chip (NOC) architecture, as outlined in
Figure 1, provides the communication infrastructure for the
resources. In this way it is possible to develop the hardware
of resources independently as stand-alone blocks and create
the NOC by connecting the blocks as elements in the
network.

339
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_59, © Springer Science+Business Media B.V. 2010
340 VEERAPANDIAN AND XIONG

alternate switch is directly connected making each resource

element reach a lot more elements with minimal number of
hops.

.
Figure.1. Network on chip [7]
A number of different NOC topologies have been proposed. Figure.2. A honey comb structure for NOC [7]
They all have in common that they connect resources to
each other through networks and that information is sent as 2.2 Mesh Topology
packets over the networks [7]. Network on Chip (NOC) has
evolved as an important research topic during the last few NOC is a scalable packet switched communication platform
years. The idea is that scalable switched networks are used for single chip systems. The NOC architecture consists of a
for on-chip communication among processing units, in order mesh of switches together with some resources which are
to cope with design of continuously growing systems. placed on slots formed by the switches.[2] Figure 3 shows
Design complexity promotes reuse of investment in earlier NOC architecture with 16 resources. Each switch is
designs as well as purchase of outside intellectual property connected to four neighboring switches and one resource.
(IP). However, in larger designs, communication among Resources are heterogeneous. A resource can be a processor
components will become a bottleneck using traditional core, a memory block, a FPGA, custom hardware block or
techniques like common buses. NOC is one solution to any other intellectual property (IP) block, which fits into the
address this issue because packet switched communication available slot and complies with the interface with the NOC
can provide higher flexibility, throughput and reusability. switch. We assume switches in NOC have buffers to
To gain full advantage when using this concept in NOC manage data traffic.
architecture design, the size of resources should be similar
and the communication facilities should be homogeneous.

2.1 Honey Comb Technology

In NOC design, the resources communicate with each other

by sending addressed packets of data and routing them to
the destinations by the network of switches [7]. Though
many topologies are possible, we will first discuss about
Honey comb topology. The overall organization is in the
form of a honeycomb, as shown in Figure.2. The resources -
computational, storage and I/O - are organized as nodes of
the hexagon with a local switch at the centre that
interconnects these resources. Hexagons at the periphery
would be primarily for I/O, whereas the ones in the core
would have storage and computational resource. To further
improve the connectivity, switches are directly connected to
their next nearest neighbors, as shown in Figure 2, allowing
any resource to reach 27 additional resources with two hops.
As a last measure to further improve connectivity, every Figure.3. 4×4 NOC switch [2]
 EFFICIENT SOPC-BASED MULTICORE SYSTEM DESIGN USING NOC 341

Every resource has a unique address and is connected to a Figure.4 shows an FPGA design that includes an SOPC
switch in the network via a resource network interface Builder system and custom logic modules. We can integrate
(RNI). The NOC platform defines four protocol layers: the custom logic inside or outside the SOPC Builder system. In
physical layer, the data link layer, the network layer, and the this example, the custom component inside the SOPC
transport layer. The RNI implements all the four layers, Builder system communicates with other modules through
whereas every switch to switch interface implements the an Avalon-MM master interface. The custom logic outside
three of four layers except physical layer. The NOC of the SOPC Builder system is connected to the SOPC
architecture also has a concept of region allows us to handle Builder system through a PIO interface. The SOPC Builder
physically larger resources and can be used to provide fault system includes two SOPC Builder components with
tolerance. A typical NOC architecture will provide a Avalon-ST source and sinks interfaces. The system
scalable communication infrastructure for interconnecting interconnect fabric shown below in Figure.5 connects all of
cores. The area of multi-media is a very suitable candidate the SOPC Builder components using the Avalon-MM or
for using this high computing capacity of NOCs. NOC is a Avalon-ST system interconnects as appropriate.
general paradigm and one needs to specialize a NOC based
architecture for every application area.

III. SOPC BUILDER

SOPC Builder is a powerful system development tool.

SOPC Builder enables us to define and generate a complete
system-on-a-Programmable-chip (SOPC) in much less time
than using traditional, manual integration methods. SOPC
Builder is included as part of the Quartus II software
(www.Altera.com). We used SOPC Builder to create
systems based on the Nios® II processor. [3]

Figure.5. System interconnect fabric

The systems interconnect fabric [3] for memory-mapped

interfaces are a high-bandwidth interconnects structure for
connecting components that use the Avalon® Memory-
Mapped (Avalon-MM) interface. The system interconnect
fabric consumes minimal logic resources and provides
greater flexibility than a typical shared system bus. It is a
cross-connect fabric and not a tri-stated or time domain
multiplexed bus. Here we describe the functions of system
interconnect fabric for memory-mapped interfaces and the
implementation of those functions.

3.1. Chip Planner

The Chip Planner provides a visual display of chip

resources. It can show logic placement, Logic Lock and
custom regions, relative resource usage, detailed routing
Figure.4. System example [3] information, fan-in and fan-out paths between registers, and
delay estimates for paths.
342 VEERAPANDIAN AND XIONG

Layers icon located next to the Task menu. Under

Background Color Map, select I/O Banks.

3.4. Generating fan-in and fan-out Connections

This feature enables us to view the immediate resource that

is the fan-in or fan-out connection for the selected atom. For
example, selecting a logic resource and choosing to view the
immediate fan-in enables us to see the routing resource that
drives the logic resource. We can generate immediate fan-in
and fan-outs for all logic resources and routing resources.
To remove the connections that are displayed, click the
“Clear Connections” icon in the toolbar.

3.5. Highlight Routing

This feature enables us to highlight the routing resources

used for a selected path or connection.

Figure.6. Chip planner tool bar from Quartus II Software 3.6. Delay Calculation
We can view the timing delays for the highlighted
With the Chip Planner, we can view critical path connections when generating connections between
information, physical timing estimates, and routing elements. For example, you can view the delay between two
congestion. We can also perform assignment changes with logic resources or between a logic resource and a routing
the Chip Planner, such as creating and deleting resource resource.
assignments, and post-compilation changes like creating,
moving, and deleting logic cells and I/O atoms. By using the 3.7. Viewing Assignments in the Chip Planner
Chip Planner in conjunction with the Resource Property
Editor, we can change connections between resources and
make post-compilation changes to the properties of logic Location assignments can be viewed by selecting the
cells, I/O elements, PLLs, and RAM and digital signal appropriate layer set from the tool. To view location
processing (DSP) blocks. With the Chip Planner, we can assignments in the Chip Planner, select the Floor plan
view and create assignments for a design floor plan, perform Editing (Assignment) task or any custom task with
power and design analyses, and implement ECOs in a single Assignment editing mode. The Chip Planner shows
tool. location assignments graphically, by displaying assigned
resources in a particular color (gray, by default). We can
3.2. Viewing Routing Congestion create or move an assignment by dragging the selected
resource to a new location.
The Routing Congestion view allows us to determine the
percentage of routing resources used after a compilation. IV. RESULTS
This feature identifies where there is a lack of routing
resources. This information helps us to make decisions Using SOPC Builder in Quartus II tool, we designed and
about design changes that might be necessary to ease the simulated efficient SOPC-based Multicore System, and the
routing congestion and thus meet design requirements. The results are listed in Figure 7-11.
congestion is visually represented by the color and shading Figure 7 shows the screenshot of the developed SOPC
of logic resources. The darker shading represents greater builder system. It is done using SOPC builder, this build
routing resource utilization. We can set a routing congestion system has a design of multicore system with 2 CPU’s,
threshold to identify areas of high routing congestion. After Avalon tri state bridge, flash memory, LCD and PIO’s. It’s
selecting the Routing Utilization layer setting, click on the a FPGA design which includes SOPC builder system and
Routing Congestion icon on the taskbar. custom logic modules. We can integrate custom logic inside
or outside the SOPC builder system. In this design the
3.3. Viewing I/O Banks custom logic modules inside the SOPC builder system
communicates with other modules through an Avalon-MM-
The Chip Planner can show all of the I/O banks of the master interface. The custom logic modules outside of the
device. To see the I/O bank map of the device, click the
 EFFICIENT SOPC-BASED MULTICORE SYSTEM DESIGN USING NOC 343

SOPC builder system is connected to SOPC system through designed system has no errors so in order to test the system
a PIO interface. we compile our system. It shows 100% full compilation
during synthesize of our SOPC system.

Figure.9. Compilation
Figure.7. SOPC builder for system building
The chip planner view of the compiled design is shown in
The block diagram of the symmetric file is shown in Figure Figure 10. In this screenshot we can see the build
8. SOPC builder allows us to design the structure of a components placed in this chip planner. We can view
hardware system. The GUI allows adding components to a critical path information, physical timing estimation and
system configure the components and specify the routing congestion. The Chip Planner uses a hierarchical
connectivity. After adding and parameterize components, zoom viewer that shows various abstraction levels of the
SOPC Builder generates the system interconnect fabric, targeted Altera device. As we increase the zoom level, the
outputs HDL files and .BDF during system generation. level of abstraction decreases, thus revealing more detail
This .BDF file shown in Figure8 represents the top –level about our design.
SOPC system for use in Quartus II.

Figure.8. Block diagram of the symmetric file

The compilation result is shown in Figure 9. Once the

system design is over it need to be verified whether the Figure.10. Chip planner view of compiled design
344 VEERAPANDIAN AND XIONG

The routing utilization is shown in Figure 11. It allows us to bandwidth and reduces routing congestion so that the
determine the percentage of routing resources used after system efficiency is enhanced. Our designed multicore
compilation. This information helps us to make decisions system has two CPU cores which are individually optimized
about design changes which is necessary to ease the routing to the particular computational characteristics of different
congestion to meet the design requirements. The routing application fields, complementing each other to deliver high
congestion is visually represented by the color and shading performance levels with high flexibility at reduced cost.
of logic resources. In Figure 11 we can see some areas are The research focus is shifting from implementation of NOC
dark and some areas are bright. The dark regions represent to investigation of its optimal use. The research problems in
greater routing resource utilization and the bright regions NOC design are identified as synthesis of communication
represent no routing congestion. infrastructure, choice of communication paradigm,
application mapping and optimization. In the future, we will
continue to design an efficient multicore SOPC with
optimized timing constraints, reduced latency and improved
programmability. We will also develop highly embedded,
multi-core systems with more number of cores which in turn
increases the system performance and many applications
can run at the same time.

REFERENCES

[1] Luca Benini, Giovanni De Micheli, “Networks on

Chips: A New SoC Paradigm”, Computer, v.35 n.1,
p.70-78, January 2002.
[2] S. Kumar, A. Jantsch, J.-P. Soininen, M. Forsell, M.
Millberg, J. Oberg, K. Tiensyrja, A. Hemani, “A network
on chip architecture and design methodology”, Proceedings
of IEEE Computer Society Annual Symposium on VLSI,
Figure.11. Routing utilization (Darker areas pp. 105-112, April 2002.
Show dense routing connections) [3] Quartus II Handbook, SOPC Builcer, Version 9.0,
Volume 4, URL:
www.altera.com/literature/hb/qts/qts_qii5v4.pdf
V. CONCLUSIONS AND FUTURE WORK [4] Axel Jantesh and Hannu Tenhumen, “Networks on
Chip”, Kluwer Academic Publications, 2003, Boston, USA.
In this paper, a simple multicore embedded system was [5] Muhammad Ali, Michael Welzl, Sybille Hellebrand, “A
developed and synthesized using Altera SOPC Builder. dynamic routing mechanism for network on chip”,
The synthesize results demonstrate that routing will be the Proceedings of IEEE NORCHIP, Oulu, Finland, 21-22,
greater issue when it comes to the chip design. Using Nov. 2005.
network-on-chip architecture a prototype system based on [6] International Technology Roadmap for Semiconductors
networking was developed to overcome the routing issue. 2003, URL: http://public.itrs.net
As a result, network-on-chip micro networks are used in [7] H. Tenhunen and A. Jantsch, "Networks on Chip",
multicore processors to interconnect the various cores to Springer, 1st edition, ISBN: 1402073925, Jan. 2003.
communicate simultaneously. This leads to larger on-chip
 Architecture and Initial Performance Indicators
of New OPCUA Automation Protocol Stack Java
Implementation
Mikko Salmenperä, Mikko Salonen

Tampere University of Technology, Department of Automation Science and Engineering, P.O.BOX 692, FI 33101
Tampere, Finland Tel: 03-31154203, Fax 03-31152340, email: [email protected]

Abstract. This paper presents OPCUA binary protocol integration of function block based basic automation
from telecommunication point of view by using OSI- information. OPC UA focuses on providing support for
network model as a basis. Presented architecture is based complex data models and flexible address space. OPC UA
directly on OPCUA specifications and Java reference abandons inherent need for Windows operating system and
implementation. Practical interest is focused on OPCUA tight coupling with Windows security settings present in
security and its feasibility for real world applications. The OPC/DCOM. The somewhat complex programming with
impact of enabling security in OPCUA is measured by COM technology is also replaced with modern software
benchmarking performance of current Java default development approach. OPC UA also increases the number of
implementation. possible types of integration solutions for which OPC can be
used by basing on standard technologies to implement Service
Oriented Architecture (SOA), namely web services (WS).
I. INTRODUCTION Even though OPC UA has significantly evolved from
OPC/DCOM in both data model and implementation
architecture backward compatibility has not been neglected.
Integration of industrial information systems has become a
OPC Foundation has realized that clear and working way to
common means to increase the overall efficiency of
emigrate from legacy OPC/DCOM is requirement for OPC
production. Information system integration is even exceeding
UA success. This transformation can be accomplished by
plant and company borders forming a collaborative network
using wrapper technology for which OPC Foundation
between businesses. The current trend of increasing demands
provides sample solutions.
posed for integration calls for unified base on which the
The goal of this paper is to present structure of OPC UA
solutions are built on. This significantly reduces the cost and
communication stack as it has been implemented in JAVA
time of integration in the long term.
reference implementation. Java implementation is closely
One widely adopted solution to evolving needs of industrial
related but not identical to that of .NET/C# implementation
integration is OPC Foundation specifications. The term OPC
developed by OPC Foundation. The performance of OPC UA
was originally acronym of concept OLE (Microsoft Object
protocol stack is also studied based on measurements done
Linking and Embedding) for Process Control. However as
with released version of the Java stack. Migration path to
OLE is no longer relevant technology for OPC UA the
OPC UA is further studied in papers [1] and in book [3] pages
interpretation has been changed to Openness, Productivity
283-293. Capabilities of OPC UA in low level device
and Collaboration. Over the years OPC has become the de
integration are studied in references [4] and [5].
facto standard for industrial integration and process
information sharing.
The original OPC was based on Microsoft DCOM
(Distributed Component Object Model) which is no longer II. BACKGROUND
actively developed by Microsoft. Traditional OPC is also
facing integration needs that it can no longer fulfill due static This paper is based on results and expertise obtained in the
information and communication models based on. These OPC Unified Architecture Java reference implementation
reasons have leaded the OPC Foundation to reinvent OPC (OPCUA Java, 2006-2007) and OPC UA Information model
replacing the core components and technologies with modern, (OPCUA Java 2, 2008 - 2009) projects. The goals of these
vendor independent solutions. Backward compatibility has projects were to contribute to the OPC Foundation’s Early
not been forgotten and OPC UA profiles detail use cases Adopter Program, develop an OPC UA Java reference
similar to those of legacy OPC systems. The new implementation and assess the feasibility of the OPC UA
specification is called OPC Unified Architecture (OPC UA). specification in the Java programming environment.
New characteristics of OPC UA are quite different from old Final results were handed over to the OPC Foundation to be
OPC/DCOM which was meant for straight forward published officially for OPC Foundation members.

345
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_60, © Springer Science+Business Media B.V. 2010
346 SALMENPERÄ AND SALONEN

References [1] and [2] details further results of these projects. good features from existing protocols in order to be able to
The projects were funded by Tekes (Finnish Funding Agency utilize existing security libraries and architectures. This
for Technology and Innovation), as well as Finnish approach also attempts to bypass caveats present in designing
companies Metso Automation, Prosys PMS, Wapice and secure and robust communication protocol from scratch.
Kone. The Technical Research centre of Finland (VTT) and Presented model of OPC UA communication stack is based
Tampere University of Technology (TUT) participated in the on the specification parts [7], [9], [10] and [11], OPC
development of the Java implementation and Helsinki Developers Conference materials [12] and the knowledge
University of Technology (HUT) provided expertise. acquired from above mentioned projects.

OPC UA stack is in itself layered as can be seen from Figure

III. OPC UA PROTOCOL 1. Basic architecture and functionality mimics TCI/IP stack.
Each layer is independent from other layers and provides
certain functionality or service for the application. The lowest
Overall functionality, architecture and complexity of OPC
layer is UA Transport. It binds the stack to either TCP/IP or
UA stack are rather difficult to grasp based on the
SOAP/Web Services protocol. OPC UA application can
specification only due to fragmentation of relevant pieces of
freely choose which transport binding it uses or even use
information into 12 different specification parts. In this
them both. Our current Java implementation supports at the
chapter, a brief analysis of OPC UA stack binary protocol
moment only binary TCP/IP binding.
implementation in Java is given. Architecture used in Java is
UA Secure Conversation layer implements security features
closely related to C# implementation but not identical.
required by the specification. These features include various
encryption and signing methods. This functionality could be
moved into OSI model’s session layer. This would mean
basing security on simple SSL/TLS based communication
used in most other current applications requiring reliable and
encrypted communications. This would simplify the stack
significantly and also increase security as all OPC UA
protocol content would be protected by security. The current
implementation leaves some header fields to plain text even if
encryption is used. On the other hand by implementing
security in the OPC UA stack interoperability can be
promoted.
UA Chunk fragments large messages into smaller chunks
that are then transferred individually over the network. On the
receiver’s side chunks are then assembled. Once the whole
message has been reconstructed it can be passed on to upper
layer in the stack. Advantage of this is that each chunk goes
through the underlying security layers separately. Thus if
used threaded stack each chunk can be encrypted and
transferred independently and in parallel to each other.
Without threading the meaning of chunks is somewhat
diminished. The actual memory usage is largely unaffected by
chunk mechanism as the message must be completely formed
before it can be processed any further.
UA encoder/decoder layer is responsible for transforming
object oriented data stored in program memory into either
Figure 1: OPC UA Protocol stack is located on OSI XML or binary presentation defined in OPC UA
models application layer specification.
UA low level service API is generated service interface
In practice, OPC UA protocol stack is located on OSI (Open through which all data transfer occurs. It consists of service
Systems Interconnection Reference Model) models level 7 – oriented functions defined in the specification. At the moment
Application Layer. OSI model is defined in ITU-T standard if in Java and C# language this API is automatically generated
reference [6]. UA stack contains partly similar functionality from XML description from OPC UA web services. This API
as lower levels of OSI model based network stack. This can be seen as boundary between UA stack and UA
functionality could be moved away from the OPC UA stack application. OPC Foundation is offering Java, C# and ANSI
but then the implementation would be more dependent on C stack to members. Above this API is OPC UA application
TCP/IP part of OSI model. OPC Foundation avoided this kind either a server or a client which is self made or developed by
of dependency and designed the stack to be as independent 3rd party.
and isolated from underlying transport protocols as possible.
However OPC UA stack tries to utilize security and other
 ARCHITECTURE AND INITIAL PERFORMANCE INDICATORS 347

UA high level tool API signifies a need for 3rd party toolkit Policies have been defined in UA specification part seven to
that expedites the development of UA applications. It respond to the information security need for the different
provides abstraction of UA stack and possibly handles routine application fields, as comprehensively as possible. For
functionality such as certificates, sessions and access control. example, the 128-bit encryption algorithms are sufficient to
This layer is no required by specification but rather is secure the normal applications at the moment, but UA also
implicitly present in all UA applications. Above this layer is defines algorithms that use 256-bit encryption algorithms.
UA Vendor application which is the actual client/server Also, more UA Security Policies can be defined to UA
software. according to demands. [3]
In addition to the Security Policies, the level of security in
UA may be tuned using the concept of
IV. OPC UA BINARY PROTOCOL SECURITY FEATURES MessageSecurityMode. MessageSecurityMode defines
whether the messages are signed and encrypted or only
One of the new features for the OPC UA is the security model signed, before sending them to the network. There is also a
that is a built-in property of the new protocol. The main ‘None’ mode defined, meaning that all the security features
requirement for the security model is to offer a flexible directed to messages are turned off. [3]
security approach as OPC UA may be used in such diverging
environments. As a result, the UA applications must be able
to mitigate several threat types and security requirements.
OPC UA defines a secure way to transfer the data between
UA applications where the generality is added with the help
of UA Security Policies. The following describes the security
features that are available when the UA Secure Conversation
(UASC) protocol is used.

A. Secure Channel as security provider Figure 2: UA Open Secure Channel sequence

The security of the messages that use the binary protocol can
be improved by using the UASC protocol. By using the
protocol, messages between the UA applications may be
encrypted and signed according to the predefined encryption
algorithms. As a result, the integrity and confidentiality of
messages are assured.
In OPC UA, the principal factor of the security is the concept
of Secure Channel, which is implemented as an independent
layer to the OPC UA protocol stack. The Secure Channel is a
long-term logical connection between OPC UA client Figure 3: UA Create Session sequence
application and OPC UA server that enables the
confidentiality and integrity of the messages. This secure B. The Security handshake
connection must be opened between the applications before
the actual device and system data may be exchanged between As discussed earlier, the connections between UA
UA applications. The Secure Channel can be opened by applications are secured when the connection is established.
calling an OpenSecureChannel method specified in Secure This part of the UA communication is called “Security
Channel Service Set which defines the methods for the Secure handshake” and it is performed according to the sequence
Channel. The secured channel is formed according to the which is depicted in Figures 2 and 3. The encryption and
security handshake discussed in the next chapter. The signature of messages take place by using asymmetric
messages that are related to the security handshake may be encryption and digital signature algorithms. The Security
both encrypted and signed or they can be sent without Handshake contains several phases: Firstly, the Secure
security procedures. Naturally, the channel is secure only if Channel is created. Secondly, an UA Session is created and
the encryption has been used. Thereby, Secure Channel is thirdly, the Session is activated. [3]
created even if no security or encryption is used. [8][9] The Secure Channel may be constructed using the
The security of the Secure Channel is based on certificates, OpenSecureChannel service. The Session is created and
chosen UA Security Policy and fine-tuning of security level activated by service methods defined in Session Service Set
(Message security mode). The certificates are used to verify described in UA specification part 4 in reference [9].[9]
the parties’ identities. Also, the keys contained in the The OpenSecureChannel request is always made by the UA
certificate can be used for encryption and signature purposes. Client application that first signs the request by using its
The UA Security Policy specifies the algorithms which are private key, after which the request is encrypted with the help
used to encrypt and sign the messages. These Security
348 SALMENPERÄ AND SALONEN

of the public key to be obtained from the receiver's certificate. Instance Certificate can also be used to sign and encrypt the
In this case, the receiver is the UA Server chosen to be messages that use the asymmetric algorithms. [3]
connected to. The Software Certificate is used to identify the particular
When an OpenSecureChannel request arrives at the UA OPC UA product. The Software Certificate contains the
server from the UA client (and the message has been information about the profiles supported by the product. In
encrypted), the server will first decrypt the message using the practice, the OPC UA server returns in the response of the
private key of the server. Also, the integrity of the message is Session service to the UA client program the Software
verified by the server. After that, the server verifies the trust Certificates list on the basis of which the client program can
to the sender by checking the sender’s certificate that is conclude the supported services of the server. When a session
received as one part of the OpenSecureChannel request. is activated, the client program has to transmit its own
Before the UA server may send the response message back to Software Certificates to the UA server. The server can forbid
UA client, the server also processes other parameters of the the activating of the session, if necessary, in case the Software
request and calculates the derived symmetric keys to itself Certificate does not meet the demands of the server. The
with the help of generated nonce. Finally, the server sends a supplier of the UA product always delivers the Software
response to the OpenSecureChannel request. The response Certificates with the product. [3][9]
contains along with other information, an unambiguous The User Certificates can be used, when activating a session,
identifier of the channel determined by the server as well as to identify the user of the UA application. OPC UA also
the nonce of the server that the client program requires in provides other opportunities to identify the user, for example
order to calculate the symmetrical keys. [3][11] the combination of username and password or web service
The level of the information security of the response is the security token instead of the User Certificate. UA also
same as in the request. Correspondingly, the possible provides a possibility to use the client applications
encryption and signature take place like in the case of the anonymously. [3] In that case, the user is not identified at all.
request. After the Secure Channel has been opened between
the UA applications, the possible encryption and signature of V. IMPACT OF ENABLING SECURITY
messages take place with the help of symmetrical algorithms.
However, before an actual use of UA, a Session must be The usage of encryption and signature ascertain the
created and activated. This can be achieved by using the confidentiality and integrity of the messages. However, the
Session Service Set that defines the methods for the Session. encryption and signature of messages require considerably
With the help of the Session, a person may be authenticated more resources than those of the messages that are sent
and authorized to call the actual UA services like Browse, without securing them.
Read or Write. The Session creation and activation part of the The Department of Automation Science and Engineering of
security handshake has been presented in a sequence [Figure Tampere University of Technology (TUT) began a
3]. [3][9] preliminary efficiency testing concerning the security in
February 2009. According to the plan, the purpose was to test
the effect of different encryption algorithms on the capacity
C. Role of Certificates for the Java UA protocol stack. The OPC UA stack
implemented using Java programming language is developed
A digital certificate is a structure that associates a public key in Finland as the part of TEKES funded project, ”OPC
with an identity. As a result, the digital certificates may be Unified Architecture Information Modeling and Java
used to identify a party that may be for example a person, an Implementation”.
application, a product or a machine. A certificate always The tests are performed by sending messages between the UA
contains a public and private key with some other applications. The messages are protected by using the several
identification information depending on a certificate type. The different UA Security Policies and security modes supported
keys combined to the certificate may be used for message by the Java OPC UA Stack. The UA applications function in
encryption and signing. [3] separate virtual machines and the measured data was
OPC UA uses X.509 certificates. The certificates in UA are collected by using the tcpdump program. In tests, the time
used in three different purposes depending on the type of from the client program’s request (first TCP packet sent) to
certificate: to verify the UA application (Application Instance the time the server returns the response (first TCP packet
Certificate), to verify UA product (Software Certificate) and sent) was measured. 20 requests were carried out in one
to identify a user of UA client program (User Certificate). measurement for each message size towards every supported
The usage of the two first mentioned certificate types is security level. Every test was performed three times, resulting
mandatory in UA, whereas the usage of the User Certificate is in total of 60 requests.
optional. [3] Table 1 depicts the averages calculated from the results. In
The Application Instance Certificate of UA is a digital figures 4, 5 and 6 the effect on the capacity of the message
certificate which can be used to identify the individual UA sizes and encryption types is seen. Figure 4 will show
application, thus every separate UA application (installation) measurement results when a payload of 1024 bytes is
requires its own application certificate. The Application transported. In Figure 5, payload was 10240 bytes, and Figure
6 outlines 102400 bytes payload transfer times.
 ARCHITECTURE AND INITIAL PERFORMANCE INDICATORS 349

Table 1: Averages in milliseconds using the 256-bit encryption and signature algorithms. The
Payload Rsa128 Rsa128 RSA256 None Effect difference with regard to the non-secure transmission of the
size in Enc & Sign Enc & of message was about 10-fold. It is true that the message sizes
Bytes Sign Sign security that were tested are large considering the purpose of the
1024 2,0 0,8 1,7 1,0 2X UASC protocol, but the main target in these tests was to see
the effects of the different security policies. As seen from
10240 31 1,8 27 1,7 18 X Table 1, the differences were more distinct when the amount
102400 65 14 88 8,1 11 X of data was increased.

Figure 4: Packet size 1024 bytes

Figure 5: Packet size 10240 bytes

Figure 6: Packet size 102400 bytes

One can perceive from the results how much both the size of VI. CONCLUSIONS AND FUTURE WORK
the message and the encryption algorithm used affects to the
processing times. With the smallest amount of the data, the
OPC foundation has chosen a new approach to integration.
128-bit encryption was the slowest, since the encryption and
The clear and fundamental change of OPC UA compared
decryption of the messages took about twice a time longer
with previous OPC specification is neutrality. OPC UA is no
than processing the message sent without any security. The
longer coupled to any 3rd party technology. It is also platform
biggest payload size (102400 bytes) was the slowest when
and programming language independent. The advantages of
350 SALMENPERÄ AND SALONEN

this choice are significant as OPC UA is targeted for Industrial Informatics (INDIN 2008), Daejeon, Korea, July
integration in industrial environments, where life time of 13-16 , 2008.
applications exceeds conventional office application by 2. Implementing OPC UA in Java; Mikko Salmenperä, Seppo
decades. On the other hand OPC specifications have grown Kuikka, Jyrki Peltoniemi, Tommi Karhela, Marko Mattila,
significantly in both size and complexity. Jouni Aro, Olli Snellman, Esa Salonen; Automaatio 07
OPC UA resides completely on OSI model’s application Seminar; Finnish Society of Automation; 2007; ISBN 978-
layer even though parts of it could have been moved to lower 952-5183-33-7
OSI levels in OSI model. This would have reduced the size in 3. OPC Unified Architecture, Wolfgang Mahnke, Stefan-
OPC UA stack and spared resources for other important new Helmut Leitner, Matthias Damm, 2009, Springer
aspects of OPC UA, such as data model. Security 4. Providing device integration with OPC UA, Hadlich, T.,
implementation is notoriously vulnerable to bugs and need Industrial Informatics, 2006 IEEE International Conference
constant updates and upkeep. It has been said that when on, Aug. 2006.
security protocol passes initial quality assurance and audit it 5. Bender K., Grossmann, D., Danzer, B, Future Device
still needs to be used in a real world for at least five years Integration, V1.1, TUM, ITM, 2007.
before any creditability can be accounted to it. 6. X.200: Information technology - Open Systems
Performance of OPC UA stack binary protocol Interconnection - Basic Reference Model: The basic model.
implementation can be considered good and even the initial ITU-T Recommendation X.200. 1994.
test shown in this paper clearly demonstrate that it provides a 7. OPC Foundation, ‘OPC Unified Architecture Specification
good tool for both vertical and horizontal integration. Effect Part 1: Concepts’, Version: Release 1.01, February 2009
of security becomes apparent with larger messages but as it 8. OPC Foundation, ‘OPC Unified Architecture Specification
can be used only where necessary there should not be major Part 2: Security Model, Version: Release 1.01, February 2009
problems for even embedded environments. 9. OPC Foundation, ‘OPC Unified Architecture Specification
Part 4: Services, Version: Release 1.01, February 2009
10. OPC Foundation, ‘OPC Unified Architecture Specification
REFERENCES Part 5: Information Model’, Version: Release 1.01, February
2009
11. OPC Foundation, ‘OPC Unified Architecture Specification
1. Hannelius, T., Salmenperä, M., Kuikka, S., "Roadmap to
Part 6: Mappings, Version: Release 1.00, February 2009
Adopting OPC UA", in 6th International IEEE Conference on
12. OPC Foundation, DevCon 2007, OPC Unified
Architecture, Conference Materials, OPC Foundation, 2007
 Automatic Segmentation of Cardiac Cavity Images
Using Collinear and Triangle Equation
Riyanto Sigit, Mohd. Marzuki Mustafa, Aini Hussain, Oteh Maskon, and Ika Faizura Mohd Noh
Dept. of Electrical, Electronic and Systems Engineering,
Faculty of Engineering and Built Environment,
Universiti Kebangsaan Malaysia
Bangi, Malaysia
[email protected], [email protected], [email protected]

Abstract- Automatic segmentation of cardiac cavity images used for segmentation of cardiac cavity [8]. Accordingly, we
using collinear and triangle equation algorithms to detect and have devised a technique which involves the use of collinear
reconstruct the boundary of cardiac cavity is afforded in this and triangular equations. Our proposed technique employs the
paper. Firstly, high boost filtering is used to enhance the high high boost filter to enhance the high frequency component
frequency component without affecting the low frequency
component. Next, to eliminate noise and convert the image into
while keeping the low frequency component unaffected. This
binary format, the morphological and thresholding operators are is followed by applying the morphological and thresholding
applied to the image. Then, using negative Laplacian filter operators to the image to eliminate noise and convert it to the
followed by region filtering, the edge detection is performed. binary format. Next, Laplacian filter and region filtering are
Finally, the collinear and triangle equations are used to detect applied in order to perform the edge detection process.
and reconstruct the more precise cavity boundary. Results The final step is the implementation of the collinear and
obtained showed an improvement in the detection boundary of triangle equation which is used to detect and reconstruct an
cardiac cavity which implies the suitability of the proposed improved cardiac activity border. It is therefore, the aim of
technique in performing the boundary detection task of cardiac this paper to present our proposed technique which we believe
cavity from echocardiographic images.
can be used to perform segmentation of cardiac cavity very
Keywords: Cardiac cavity, high boost filter, morphology,
precisely. The following section describes the development of
negative Laplacian, region filter, collinear and triangle equation. the proposed technique further and outcomes of the
implementation are shown in the results and discussion
section.
I. INTRODUCTION
II. PROPOSED AUTOMATIC SEGMENTATION METHOD
The two-dimensional (2D) short axis echocardiography The schematic diagram of the proposed technique is shown
images have been used by medical practitioners to diagnose in Fig.1. As shown, the collinear and triangle equations
patient cardiac condition by studying the cardiac cavity. As implementation represents the last step in our proposed
such, various researches and methods have been conducted to automatic segmentation technique to detect edge and
develop computerized technique to detect cardiac cavity [1-9]. reconstruct the border. Prior to it, the echocardiography image
Although many techniques have been developed, there is still is subjected to a 3-step procedures involving high boost
room for innovation and development of such methods and
algorithms. For cardiac activity detection, researchers have Viewed Image containing cardiac
used either the short axis images [1-6] or long axis images i
[7&8] implemented in semi-automatic mode [1, 4 & 6] or
fully automatic [3 & 7]. In addition, active contour models or High Boost Filtering
snakes have also been used to segment cardiac cavity [5 & 8].
J. W. Klinger et al in [1] have applied segmentation of
Morphology & Thresholding
echocardiography images using mathematical morphology.
On the other hand, Laine and Zong in [2] have used border
identification that depended on the shape modeling and border Negative Laplacian & Region Filtering
reconstruction from a set of images. Detection of left
ventricular endocardium using ternary threshold method was
Collinear & Triangle Equations
introduced by Ohyama et al [3], in [1 & 4] the semi automatic
detection method was employed and the multiple active
contour models for cardiac boundary detection by Chalana et Cardiac cavity boundary detected in
viewed image
al is presented in [5]. In a more recent work, watershed pre-
segmentation snake was used for boundary detection and to Fig.1. Schematic Diagram of the algorithm
find its center point [6 & 7]. Of late, applied radial-search was

351
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_61, © Springer Science+Business Media B.V. 2010
352 SIGIT ET AL.

filtering, morphological and thresholding and a double On the contrary, the closing algorithm is when image A is
filtering using the Laplacian and region filters. dilated and eroded by B. The mathematical notation of the
closing algorithm is shown in equation 2.
A. Cardiac Short Axis Image View
(2)
There are two main standard views used in cardiac imaging
that is the short axis and long axis. In this research, we have D. Negative Laplacian & Region Filter
used the short axis images to test out our proposed
segmentation technique. An example of the image comprising Next, the negative Laplacian filter is implemented. It is a
the left ventricular cardiac cavity from echocardiography is derivative filter that is used to find areas of rapid change (i.e.
shown Fig. 2. edges) in an image. There are different ways to find an
approximate discrete convolution kernel that can approximate
the effect of the Laplacian. A possible kernel is as shown in
Fig.4 below.

0 1 0
1 -4 1
0 1 0
Fig. 4. Kernel used for negative Laplacian.

Following negative Laplacian filtering, the image is then

subjected to region filtering. The main function of region filter
is to eliminate small contours. The region filter scans the
contour and calculates the area of each contour. Regions with
Fig.2. Screen-shot of the short axis left ventricular cardiac
area that is smaller than the pre-determined threshold are
cavity image from echocardiography eliminated from the contour [6]. The threshold value is set to
25 pixels, which was empirically determined.
B. High Boost Filter
E. Collinear & Triangle Equation
The first step in our proposed technique is the
implementation of the high boost filter algorithm [6], which Next, the collinear equation algorithm is used to find
uses the spatial mask shown in Fig. 3. By using the spatial centroids of all contours by keeping and deleting some
mask, the high-boost filter is used to enhance high frequency contours so that the resulted contour is closer to the actual
component without affecting the low frequency components. boundary. All existing contour centroids are computed using
equation 3 shown below.
-0.1111 -0.1111 -0.1111
-0.1111 9.89 -0.1111 ­° ¦n X k ¦n Yk ½°
-0.1111 -0.1111 -0.1111 Centroid (C ) = ® k =1 , k =1 ¾ (3)
°̄ n n °
¿
Fig.3. Mask used for the high boost filter. The collinear equation is applied from center of the
boundary to contour centroids by finding the slope and
C. Morphology and Thresholding intercept using equation 4, 5 and 6.
After image enhancement using high boost filter, the y = wx + b (4)
enhanced image is subjected to the morphological operation
and thresholding. Morphological operation involves the use of n¦ xy − ¦ x ¦ y
Slope ( w) = (5)
the opening and closing algorithms [1]. The main function of n¦ x 2 − (¦ x )
2

the algorithms is to reduce speckle noise in the cardiac cavity

image. The opening algorithm involves eroding image A by B Intercept (b) = y − wx (6)
and dilation by B. The mathematical notation of the opening
algorithm is shown in equation 1. If the slope of the contour intersects with the inner boundary,
the contour is maintaned and removed if it intersects with the
(1)
outer boundary. Fig. 5 shows a triangle, where A, B, C
represents the corners, and a, b, c are the distances between
where and denote erosion and dilation, respectively.
corners.
 AUTOMATIC SEGMENTATION OF CARDIAC CAVITY IMAGES 353

c a

A b C

Fig.5. A triangle

If a, b, and c are known, then the angle of each corner can

be calculated using equation 7 and 8.

a 2 = b 2 + c 2 − 2bc(cos A) (7)
Fig. 6. Sample result from high boost filter implementation
A = a cos((b 2 + c 2 − a 2 ) / 2bc ) (8)

After thresholding operation, the resultant cardiac cavity

image will consist of a closed border and some open borders.
To close the borders one can utilize the small corner angle
using the triangle equation. Accordingly, we utilized the
OpenCV Library, which is a library for computer vision to get
the boundary contour and reconstruct it. To do so, some
modification is required and as such we have made
modification to this library. In the OpenCV library, a contour
that can be stored inside the memory storage is known as a
sequence. A sequence in OpenCV is actually a linked list [7].
The cvFindContours function is used to retrieve
contours from the binary image and to return the number of
retrieved contours. The function cvSeqPop is used to Fig. 7. Sample result after morphological operations with thresholding

remove an element from the sequence and cvSeqPopFront

function, on the other hand, is used to remove an element from
the beginning of the sequence. As such, one can perform
contour cutting by using the cvSeqPop function and
cvSeqPopFront function.
However, to do so it is crucial that the initial center
boundary of the cardiac cavity is determined by using the
cvFindContours function. Once the center is determined,
the next step is to cut the contour radius that is greater than 0.5
of the image width size and then connect two end points to
each other. This process is repeated until all small contours are
eliminated.
Fig.8. Sample result from negative Laplacian filter implementation
III. RESULTS AND DISCUSSION

The result of the high boost filter implementation is

shown in Fig. 6 followed by the results after morphological
operation with thresholding in Fig. 7. In can be seen that high
boost filter implementation yields an enhanced image of the
cardiac cavity which is further improved by using
morphological operations with thresholding. Result from the
negative Laplacian filter implementation is shown in Fig. 8
while result after region filtering is shown in Fig. 9.
Accordingly, Fig. 10 depicts the result from collinear equation
computation where a contour is maintained if the slope of the
contour intersects with the inner boundary and removed if it
intersects with the outer boundary. Fig.9. Sample result from region filter implementation
354 SIGIT ET AL.

All these steps are then validated by testing with images

from a video recording that consists of 9 frames. The typical
image size used is 320 pixels wide and 240 pixels high. Fig.
12 shows the reconstructed cardiac cavity boundaries of all
nine frames. It can be seen our proposed method successfully
detects and trace the boundaries of all cardiac cavity in the
video as it changed from large to small.

IV. CONCLUSION

A practical solution to perform automatic segmentation of

cardiac cavity using collinear and triangle equation has been
described. The proposed method successfully detects and
reconstructs precise boundary of cardiac cavity which can help
Fig.10. Sample result from collinear implementation
in the cardiac investigations by medical practitioners.

ACKNOWLEDGMENT

The authors would like to thank the government of

Malaysia for the funding of this research through research
(a)
grant contract number UKM-GUP-TKP-08-24-080.

REFERENCES
[1] J. W. Klinger, C. L. Vaughan, and T. D. Fraker, “Segmentation of
echocardiographic images using mathematical morphology,” IEEE
Trans.Biomed. Eng., vol. 35, 1988, pp. 925–934.
[2] A. Laine, and X. Zong, “Border Identification of Echocardiograms via
multiscale edge detection and shape modeling,” Proc. of the IEEE Int.
Conf. on Image Processing, vol. 3, Sep. 1996, pp. 287 - 290 vol.3.
[3] W. Ohyama, T. Wakabayashi, F. Kimura, S. Tsuruoka, and K. Sekioka,
(b)
“Automatic Left Ventricular Endocardium Detection in
Echocardiograms Based on Ternary Thresholding Method”, in 15th
International Conference on Pattern Recognition (ICPR’00), Barcelona,
Spain, 2000, pp. 320-323.
[4] Maria do Carmo dos Reis, Adson F. da Rocha, Daniel F. Vasconcelos, etc,
“Semi-Automatic Detection of the Left Ventricular Border “, 30th
Annual International IEEE EMBS Conference Vancouver, British
Columbia, Canada, August 20-24, 2008.
[5] V. Chalana, D. T. Linker, D. R. Haynor and Y. Kim, “A Multiple Active
Contour Model for Cardiac Boundary Detection on Echocardiography
(c) Sequences”, IEEE Trans. on Medical Imaging, vol.15, no. 3, June 1996.
[6] Cheng, J, Foo, S.W. & Krishnan, S. M. “Automatic Detection of Region of
Interest and Center Point of Left Ventricle using Watershed
Segmentation,” IEEE Int. Symposium on Circuits and Systems, vol. 1, n.
2, May 2005, pp. 149-151.
[7] Cheng, J, Foo, S.W. & Krishnan, S. M. “Watershed-Pre-segmented Snake
for Boundary Detection and Tracking of Left Ventricle in
Echocardiographic Images,” IEEE Trans. on Information Technology in
Biomedicine, vol. 10, n. 2, 2006, pp. 414-416.
Fig.11.The step-by-step results to eliminate small contours [8] Lacerda SG, da Rocha AF, Vasconcelos DF, de Carvalho JL, Sene IG Jr &
using the OpenCV library functions. Camapum JF, “Left Ventricle Segmentation in Echocardiography Using
a Radial Search Based Image Processing Algorithm”, 30th Annual
Step-by step results using the triangle equation to eliminate International IEEE EMBS Conference Vancouver, British Columbia,
small contours by small corner angle computation are shown Canada, August 20-24, 2008
in Fig. 11(a), (b) and (c). Figure 11(c) depicts the final cardiac [9] M. Kass, A.Witkin, and D. Terzopoulos, “Snakes: Active contour
models”, presented at the Int. Conf. Computer Vision, ICCV’87,
cavity boundary that has been successfully detected and London, U.K., 1987.
precisely reconstructed. [10] Gary Bradski and Adrian Kaehler, “Learning OpenCV”, O’Reilly Media,
September 2008.
AUTOMATIC SEGMENTATION OF CARDIAC CAVITY IMAGES 355

Fig.12. Frame-by-frame border detection of cardiac cavity in

echocardiograph image
 Both Side More-Output Transmission
with Harmonic Gear
Ing. Jozef HaŠko, PhD.
Faculty of manufacturing Technologies,
TU Košice with seat in Prešov, Štúrova 31,
08001 Prešov, Slovak republic
[email protected]

Abstract - The paper presents the design of the I.I THE FIRST ALTERNATIVE
principle of multiple output both side differential
transmissions with integrated harmonic transmission. The Transmission of revolutions and performance is
presented principle enables more coincidental coaxial possible through the input eccentric shaft I, where
outputs with the alternative of input and output fulcrum wheel „2“ and „7” is located with gearing z2 and
interchange. z9 and firmly connected gearing z5 and z7, meshing with
gearing z6 eventually z8 connected with the output shafts
INTRODUCTION II, V.
The next output is possible either through the
Department of technological equipment design of flexible member (disk wheel) „3“ to the output shaft III,
the FVT of the TU in Košice within the process of or through the crown wheel „4“ connected with the output
searching for alternatives to the so-called high precision shaft IV. The second output is done through the flexible
transmissions of the constructions based on cycloidal member „3“ with the internal gear z3a and external gear
teeth with the relatively complicated transmission of z3b.
movement from satellite to centric has aimed it research at Thus, gearing z2 and z9 of the wheel „2“ and „7”
the new principles for design with relatively simple meshes with internal gear z3a of the flexible member „3“
transmission of movement as well as presented and consequently the flexible member by means of
transmissions regarding the number and range of external gear z3b meshes with the internal gear z4 of the
transmission ratios. non-flexible member of the crown wheel„4“. If the crown
The principle of integrated harmonic transmission wheel „4“ is braked, corresponding performance of the
with the multiple both side outputs has been suggested. third output is transmitted to the shaft III by means of the
The paper also presents the kinematic analysis of flexible member „3“. If the flexible member „3“ is braked
the designed construction with transmission ratios for the , performance is transmitted by means of the member „4“
certain levels. to the shaft IV.

I.II THE SECOND ALTERNATIVE

I. PRINCIPLE OF MULTIPLE BOTH SIDE OUTPUT
TRANSMISSION WITH INTEGRATED To transmit revolutions and performance, the
HARMONIC TRANSMISSION driveshaft II can be used. Then shafts I, III, V can be the
input members, if the member „4“ eventually I, IV, V is
Modern standard harmonic transmissions are
braked, if the member „3“ is braked. Analogically it is
usually designed as „one-way“ with the only input and
possible to determine output shafts, if shaft V is
output without the possibility of their interchange.
a driveshaft.
The suggested multiple output transmission with Figure 2 shows kinematic ratios between
integrated harmonic transmission is considered to be the memebers „2“ a „3“ and consequently transmission ratios
new design with the variants of one input and three for the first and the second alternatives have been derived.
outputs with one braking of one of the transmission
members. With the two levels of latitute, the transmission
can operate as a differential. Output shafts are possible on II. TRANSMISSION RATIOS FOR THE FIRST
both side. ALTERNATIVE – SHAFT I IS A DRIVESHAFT
From the design principle it follows that
alternative transmission from input shaft to the three II.I TRANSMISSION RATIOS WHEN MEMBER „3“
output shafts (members) is achieved by means of one one- IS BRAKED, n3 = 0
level transmission integrally interconnected with the two
two-level transmissions within one construction of the a) Transmission ratio between shaft I and gear wheel „2“
whole transmission mechanism (fig 1). Transmission of if the flexible member „3“ is braked
revolutions and performance for the suggested design is The analogue relations are also between shaft I and gear
possible in two alternatives with the possible wheel „7”.
determination of corresponding transmission ratios.

357
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_62, © Springer Science+Business Media B.V. 2010
358 HAL’KO

From the kinematic ratios shown in fig.2 it follows that v A 2O = v A 2 E + vE O = v A3O (2)
for the velocity vector at point „A“ the following is
valid: From the equation (1) it follows that at the point A the
following is valid:
v A 2 O = v A 3O (1)
ω A 2 O = ω A3O (3)
After substitution the velocity vectors at the point A
v A3O = ω A3O . r3 ae , v A 2 E = ω 2 E .r2 , v E O = ω E O . re
and if we take into consideration the supposed direction of
revolution regarding fig. 2 the following is valid:
ω A 3 O . r 3ae = ω 2 E . r 2 + ω E O . r e (4)
Regarding the fact that body „3“ is in the part of gearing
z3a , z3b flexible, ω A 3 O it does not always equal the angle
velocity of the body „3“, thus ω A3O ≠ ω 3O . Angle
velocity of the non-deformed body „3“ equals angle
velocity of the input shaft III. With constant ω I = ω EO

Fig. 1 Principle integrated harmonic gears of mechanism

also ω 3O = constant. When deriving the transmission
ratio ω 3O between shaft I and wheel „2“ it is necessary
(Two step gear – authorship certificate Ing. Jozef HaŠko,
PhD., number 3937, Slovakia) to follow from circumferential velocity v A′3 O at point
A′ on the pitch circle of gearing z3a of the non-
deformed member 3. Thus v A′3 O = ω A′3O r3 a , with
ω A′3O = ω3O .
If we consider that body „3“ (flexible member) is
braked in the body of transmission then when deriving the
transmission ratio between input shaft and wheel „2“ we
follow form the fact that gear wheel „2“ when shaft
I with the certain slew angle performs the same mesh
trajectory of deformed gear member „3“ by gearing z3a as
in case of non-deformed member „3“ with the other
corresponding eccentricity re′ , that is smaller than

Fig. 2 Kinematic scheme eccentricity re by a value of deformation Δr (see fig.

2). Based on the above mentioned, it is possible to write
At point „A“ the pitch circle of the gearing z2 of the the following for the velocity vector at the point A′ ,
member „2“ touches the pitch ellipse of gearing z3a of which represents the meshing the gear wheel „2“ with
the deformed member „3“, basic shape of which is circle non-deformed member „3“ with the gearing z3a :
with pitch circle radius r3a, of the gearing z3a. At the
same time, meshing of gearing z4 of the member „4“ v A′ 2 O = v A′ 2 E + v E ′′o = v A′ 3 O (5)
with gearing z3b of the member „3“ at point „B“ occurs. after substitution:
At this point the pitch circle touches gearing z4 with the ω A′ 3 O . r 3a = ω 2 E . r 2 + ω E ′ O . r e′
pitch ellipse of the gearing z3b of the deformed member
and the following is valid:
„3“. It follows that big (main) axle of pitch ellipse of
gearing z3a at the point „A“ has the size r3ae. It follows r e′ = r 3 a − r 2 , ω 2 E = ω 2 , ω E ′ O = ω E O = ω1
from the above as well as from fig. 2 that: after substitution the following is valid:
r3a e = re + r2 r3a e = r3a f ωI r2
where: − = (6)
ω 2 r3a − r 2
re - axle eccentricity of the member „2“ rotation
ωI
r 2 - radius of pitch circle of gearing z2 of the member „2“. Expression is the searched transmission ratio. Let us
Radius r3ae is smaller than r4 by a value h (see fig. 1)
ω2
Thus: h = r 4 − r3a e denote it by u I3− 2 , then we can write:
Value h usually follows from the body „3“ construction r2 r2
3
based on pre-determined requirements for a transmission. u 3
I −2 =− or u I −2 = (7)
From the vector kinematics for velocity at the point „A 2“ r 3a − r2 r2 − r 3a
regarding point „O“ the following is valid:
 BOTH SIDE MORE-OUTPUT TRANSMISSION WITH HARMONIC GEAR 359

where u 3
I −2 - is transmission ratio between shaft If ω5 = ω 2 (gearings z 5 and z 2 are firmly gripped on
I and wheel „2“ when member „3“ is braked. the wheel „2“) then we can substitute from the equation
m z 3a m z2 (9) ω 2 = ω 5 to the equation (12). From the transmission
If we substitute r 3 a = and r2 = , (m- design it follows that angle velocity of the shaft V equals
2 2
3 the angle velocity of the gear wheel z 4 , because of the
module gearing z2,z3a) then transmission ratio u I − 2 is
as follows:
firm connection with the shaft „V“. Thus ω 6 = ωV .
3
z2 ωI After substitution we will obtain:
u I −2 = = (8)
z 2 − z 3a ω2 § r −r ·
r6ωV = r5 ¨¨ ω i 2 3a ¸¸ + (r6 − r5 )ω I
b) Transmission ratio between the shafts I and IV when © r2 ¹
member „3“ is braked. ωI
Ratio is the transmission ratio between shaft
Output shaft V has internal gear wheel. ωV
From the equation (6) and (8) for kinematic ratios of gear I and V when member „3“ is braked. Let us denote it by
wheels z 2 and z 3 of the harmonic transmission for angle u I3−V and angle velocity ωV by ωV3 (index 3 – braked
velocity of the internal gear wheel „2“ it follows member „3“ , and consequently we will obtain the
r1 − r 2 following:
ω2 = ωI (9) ωI r6 r2
r2 u I3−V = = (13)
ωV r5 (r2 − r3a ) + r2 (r6 − r5 )
3
If annual gears z 2 and z8 are on the same body of the
After the radius through teeth number has been defined:
gear wheel the following is valid: ω 2 = ω8 . ωI z6 z 2
Eccentricity e at the same time equals the radius of u I3−V = = (14)
revolution of the wheel axis „2“ (of the point E) around ωV z 5 ( z 2 − z 3a ) + z 2 ( z 6 − z 5 )
3

the point „O“ of the central axis of the transmission.

Thus: e = re = r6 − r5 = r3a − r2 . (10) According to the equation (14) is to achieve large size
3
gear ratio u I − IV . Normally is possible to achieve then
From fig. 3 it follows : v C 6O = v 6 E + v EO , (11)
divider the equation (14) equal zero. Then
At the same time the following is valid: vC 6O = vC 5O ,
u I 3− IV = ∞ (input shaft “I” turn, output shaft “V” stay).
then vC 6O = r6ω 6 , vC 6 E = r3ω 5 , v EO = re ω EO , and Transmission ratios between the shaft I and II with the
ω EO = ω I . corresponding numbers of teeth for gearing are of the
analogical form.
After substitution and regarding the selected revolution
according to fig. 3 we will obtain:
c) Transmission ratio between the shaft I and IV when
r6ω 6 = r5ω 5 + reω I (12) member „3“ is braked.

For this transmission, analogical relationship known form

harmonic transmissions can be derived. Thus, the
following is valid:
z4
u I3− IV = (15)
z 4 − z 3b

II.II TRANSMISSION RATIOS WHEN MEMBER „4“

IS BRAKED AND n4 =0.

a) Transmission ratio between the shaft I and III

For the transmission ratio the following relationship can
be derived:
ωI z 3b
u I4− III = = (16)
ω III z 3b − z 4
4

Fig. 3 Kinematic scheme of the second level transmission. Output shaft

with the external gear wheel.
360 HAL’KO

4
where ω III - is the angle velocity of the shaft III when III.II. TRANSMISSION RATIOS WHEN MEMBER „3“
IS BRAKED AND n3 = 0
member„4“ is braked.
a) Transmission ratio between the shafts II and I
b) Transmission ratio between the shaft I and V
For the angle velocity of the shaft V the following is 1
u II3 − I =
valid: ω =ω +ω
V
4
V
3 4
III (17) u I3− II
where b) Transmission ratio between the shafts II and IV
ωV4 - is the angle velocity of the shaft V when member u I3− IV
u II3 − IV =
„4“ is braked u I3− II
ω V
3
- is the angle velocity of the shaft V when member c) Transmission ratio between the shafts II and V
„3“ is braked u I3 − V
Let us denote transmission ratio between the shaft I and u II3 − V =
V when member „4„ is braked by u I 4− V , then u I3 − II

ωI ω CONCLUSION
u I 4− V = = 3 I 4 (18)
ωV ωV + ω III
4
The objective of the paper is to present
principally new design of multiple output both side
After expression ωV from
3
the equation (14) and ω III4 transmission with the possibility of input and output
from the equation (16) and after substitution to the interchange with the alternatives of the possible three
equation (18) we will obtain: outputs from the given transmission mechanism. Based on
z 6 z 2 z 3b the kinematic scheme, corresponding transmission ratios
u I4−V = have been derived. Within the further research, force
[z 5 (z 2 − z 3a ) + z 2 (z 6 − z 5 )]z 3b + (z 3b − z 4 )z 6 z 2 ratios analysis and possible loadings of the mentioned
Relationship foR transmission ratio u I 4− II between the mechanism, eventually the other corresponding solutions
[1], [4], [9] will be given great attention. This
shafts I and II when member 4 is braked is of the
contribution forms a part of the solution of the grant task
analogical form , although when the corresponding
VEGA 1/4156/07 and KEGA 3/6279/08.
number of teeth of the gearing are substituted.

III. TRANSMISSION RATIOS FOR THE SECOND REFERENCES

ALTERNATIVE – SHAFT II IS THE DRIVESHAFT
1. HA‹KO, J. – KLIMO, V.: Dvojstup’ový viacvýstupový prevod. In:
Sborník 45. mezinárodní konference kateder ástí a mechanism
III.I. IF THE WHEEL „4“ IS BRAKED AND stroj : 7. – 9. zá í 2004, Blansko, eská republika, Brno: VUT,
n4 = 0. 2004, 6 s., ISBN 80-214-2702-7
2. HOMIŠIN, J. et al.: Základy strojného inžinierstva. Košice: TU-SjF,
a) Transmission ratio between the shafts II and I 2001, 253 s., ISBN 80-7099-661-7.
1 3. KLIMO, V. – MANTI, M.: Some Problems on Micromechanism′s
u II4 − I = 4
Actuators. In: Acta Mechanica Slovaca., ro. 8, . 1, 2004, s. 69-74.
u I − II 4. PAŠKO, J. – PAVLENKO, S. – KISLÍKOVÁ, I.: Pružnos
a pevnos 2: Teória a príklady. Prešov: Fakulta výrobných
b) Transmission ratio between the shafts II and III technológií TU v Košiciach so sídlom v Prešove, 2003, 154 s., ISBN
u I4− III 80-7099-983-7
u II4 − III = 5. VOJTKO, I. – MATIJA, R.: Meranie a zhodnocovanie vibrácií. In:
u I4− II Nové smery vo výrobných technológiách 2006: 8. medzinárodná
vedecká konferencia, Prešov, 22. – 23. 6. 2006: Zborník referátov,
c) Transmission ratio between the shafts II and V Košice: FVT TU, 2006, s. 503 – 508, ISBN 80-8073-554-9
u I4− V 6. BATEŠKOVÁ, Eva - MAŠENIK, Jozef - NOVÁKOVÁ, Martina:
u II4 − V = 4
Imagination support in depicting incisions and sections by means of
u PC. In: MOSIS '09, Czech republic. Ostrava: MARQ, 2009. p. 203-
I − II
206. ISBN 978-80-86840-45-1.
7. Firemná literatúra fa. The MathWorks Inc. – MATLAB, Simulink,
Simscape
 Intelligent Plasma Monitoring Systems Based
on Optical Methods and Chromatic Modulation
Dimitrios Tomtsis,
Technical and Educational Institute of West Macedonia,
Department of Applied Information Systems in Business and Economy,
Grevena, 51100, Greece
Abstract—In this work an intelligent optical monitoring system chosen to be close enough together to ensure that spurious
is presented the operating principle of which is based on natural signals affect both wavelengths equally. However, the
biological and optical systems. The monitoring system perceives modulator needs to affect only one of the wavelengths so
its environment thought optical sensors for goal directed actions
leading to conflicting demands. Such wavelength modulating
even in dynamic changing environments and exhibits
unsupervised and autonomous operation. Chromatic modulation systems may be constructed using spectrometers or by using
methods are applied extensively to provide information two narrowband optical filters in conjunction with two
compression in a form suitable for conventional image processing photodetectors. Systems which make use of a spectrometer are
techniques. Chromaticity changes in a number of chromatic expensive, optically inefficient and require excessive data
parameters are related to changes in physical plasma processing. Systems using filters and photodetectors are
characteristics and properties (e.g. gas composition). An
wasteful of optical power.
extensive range of parameters can be monitored with the same
system so leading to the realization of a unified intelligent Many of the difficulties inherent in the spectral or two
measurement philosophy. wavelength monitoring methods may be overcome using a
chromatic modulation method. A number of sensor systems
I. INTRODUCTION based upon this approach have been developed [1] and shown
Intelligent systems use ideas and get inspiration from to possess attractive advantages.
natural systems and build on both established and novel Taking into account the above considerations, it is desirable
techniques from a number of scientific fields such as computer to develop a general purpose intelligent monitoring system the
vision, machine learning and artificial intelligence. Intelligent operation of which is not limited to a single task or application
systems may perceive their environment through sensors for but it is reusable. In this work an intelligent optical monitoring
goal-directed actions even in dynamic situations while system is presented which can be used in many different
learning and functioning autonomously in order to cope with applications. The systems operation is based on chromatic
changing environments or inaccurate a-priori knowledge. modulation.
Many intelligent optical systems incorporate those
characteristics but are highly application dependent. Some
systems are stand-alone applications which solve a specific II. BIOLOGICAL CHROMATIC PROCESSING

measurement or detection problem, while others constitute a An aspect of the science of photic fields relates to the
sub-system of a larger design. The specific implementation of interaction of the photic field power P(), which in general
an intelligent optical system also depends on if its may be a complex function of wavelength  with matter of
functionality is pre-specified or if some part of it can be different kinds, via its responsivity R(), to produce a physical
learned or modified during operation. or chemical effect V [2] according to:
In many intelligent optical monitoring systems intensity
modulation offers the advantages of inherent simplicity. ª º
m

However, conventional methods involving absolute intensity V = α 1 « ³ P ( λ ) M ( λ ) R ( λ )T ( λ ) d λ » (1)

have associated problems. The most basic intensity ¬λ ¼
monitoring systems use only a single photodiode to produce
an output but these systems tend to be sensitive to spurious Where 1 is essentially a constant of proportionality, T() is
changes in intensity resulting from variations in the light the transmissivity of the propagating medium and M() is an
source or other components within the system. Removing externally imposed modulation. Eq. 1 describes the function
these spurious effects is difficult and leads to complicated and of many diverse natural systems such as the production of
expensive systems. chlorophyll by photosynthesis, radiation effects on animals
Wavelength monitoring systems attempt to deduce the state and animal vision. In the later case, more detailed visual
of a system by taking the ratio of intensities at two different information is obtained via biological receptors having
wavelengths. In principle, the two wavelengths should be different responsivities R1(), R2(), R3().

361
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_63, © Springer Science+Business Media B.V. 2010
362 TOMTSIS

three chromatic parameters, namely dominant wavelength (H),

Light Source intensity (L) and degree of monochromaticity or spectral
width (S) which are defined as follows:

Computer ª § V i − V min
H = 120 « m i + ¨
·º
¸»
¨ V + V − 2V ¸ (4)
¬« © i j min ¹ ¼»
Sensor
Camera a) with mi=0 for i=1, j=2, Vmin=V3, b) with mi=1 for i=2,
j=3, Vmin=V1, c) with mi=2 for i=3, j=1, Vmin=V2

Fig. 1. Graphical illustration of the configuration of the system.

ªV + V min º
L = 100 « max » (5)
¬ 2 ¼
The significance of the above considerations for intelligent
optical monitoring systems is that (1) also describes the
detection of optical signals when R() is the responsivity of ª V max − V min º
S = 100 « » (6)
the photodetector, V is the output voltage of the processing ¬ 200 m 2 − m 3 (V max + V )
min ¼
electronics, T() is the transmissivity of the propagating
system (such as an optical fiber) and 1 is a constant of
with m2=0, m3=-1 for L≤50 and m2=m3=1 for L>50. Vmin
proportionality.
and Vmax are the minimum and maximum detector outputs.
In general, chromatic changes can be monitored by a
For the special case of Rn() corresponding to the
number (n) of detectors with overlapping spectral responses.
responsivities of the human eye, H, L and S become the Hue,
The output of each detector may then be expressed as [3]:
Lightness and Saturation of color science [7]. The color model
representing this mathematical formalism is generally called
Vn = ³ P (λ ) Rn ( λ ) dλ (2) HLS [8].

where P(λ) is the spectral power distribution in the optical III. EXPERIMENTAL SYSTEM
signal and Rn(λ) is the wavelength responsivity of the nth
detector and  is the wavelength. The color model The optical monitoring system is computer based and is
representing this mathematical formalism is generally called broken down into three distinct and important elements: a
RGB and it is widely used in self-luminous display camera, a color frame grabber card (CFG), and a host personal
technologies [4]. computer.
Each detector output may also be intensity normalised The camera converts the spatially distributed optical
according to: information into analog electric signals which are accepted by the
CFG. The signal is then digitised and stored in the card's frame
memory. The C.F.G. card is the interface between the camera
u n = V n / ¦ n VT (3) and the host personal computer which is used to access and
program the card's registers, frame memory, and buffers. Fig. 1
and chromaticity maps may be formed in terms of the illustrates graphically the configuration of the system.
coordinates u1, u2, … u(n-1) (un becomes redundant since Although many measurands may be fully specified by
¦ n uT = 1 ). distimulus detection [9] and dominant wavelength alone, others
The special case of n=3 leads to a two-dimensional require tri-stimulus detection [10] to completely and
chromaticity map (u1:u2) on which changes in optical signals unambiguously determine their state. The suggested intelligent
may be traced. The special case when R1(), R2(), R3() optical monitoring system is tri-stimulus in nature and its
correspond to the responsivities of the human eye leads to the hardware and software components are fully described in the
chromaticity map reducing to the CIE diagram of color following sections. The built-in intelligent nature of the
science [5], [6]. monitoring system results from its ability to automatically
The color model representing this mathematical formalism compensate for any lightness variation as well as its ability to
is called LXY and provides the relative magnitudes of the decide the gamma correction process based on the surrounding
tristimulus values (i.e. x=u1; y=u2; z=u3). environment and the event being monitored.
Although the above method of displaying chromatic A. Optical Frequency Response
information has proved the information compression
The optical frequency response of the CCD camera was
capabilities of the chromatic approach, it does not easily lead
measured using the optical monitoring system and the
to the identification of signal changes in terms of fundamental
apparatus shown in fig. 2.
signal properties. An alternative approach to the processing of
the signals from such chromatic detectors overcomes this
limitation. For the tristimulus case (n=3) this approach utilizes
 INTELLIGENT PLASMA MONITORING SYSTEMS 363

C o n ca v e R e fle c ti n g Light S ource 200

D i ffr a c ti o n G r at in g

Half Silvered C am era 150

Mirror

Output (a.u)
Green
100
Blue
A lign m en t
S l its O sc illoscope 50
Red

R o ta tab le
S tand
O SA 0
420 440 460 480 500 520 540 560 580 600 620 640 660
Wavelength (nm)
Fig. 2. Apparatus used for the optical frequency response of the color CCD
camera. Fig. 3. Optical frequency response curves of the Red, Green and Blue
camera detectors (tri-stimulus detection).
The required part of the spectrum was selected with the The procedure adopted in this work is similar to the Von
alignment slits and was directed to both the camera and the Kries transformation [11] with the difference being that
optical spectrum analyser (OSA) using the half silvered instead of using a white color (or greys of the same
mirror. The oscilloscope was used to display information chromaticity) as a reference, any carefully selected color can
about the light intensity as a function of the selected be used for the same purpose. In this procedure if a stimulus
wavelength incident to the camera and the OSA. Different gives rise to the r,g, and b color responses then the resulting
wavelengths were selected by rotating the diffraction grating responses will depend on r/rc, g/gc, b/bc, where rc, gc, and bc
on the rotating stand. The resulting information was then are the chromatic coordinates of the reference color.
processed by the host personal computer to provide color For a stimulus in the state considered to have the same color
quantification that correlated with the wavelength monitored appearance as the same stimulus in a reference state it is
by the OSA. necessary that:
The CCD array response of the color CCD camera was
measured in a dark environment to avoid the unwanted effects r / rc = r ' / rc' , g / g c = g ' / g c' , b / bc = b ' / bc' (7)
of external lighting. The camera was used with its focus and
aperture lens removed so that the light reflected from the
where r’c, g’c and b’c are the color responses of the reference
grating was incident directly on the CCD element.
color in the reference state. Hence
Measurements were taken at 25 nm intervals.
The calibration results of the camera are shown in fig. 3.
The obtained frequency response curves show that the CCD r ' = ( rc' / rc ) r , g ' = ( g c' / g c ) g , b ' = ( bc' / bc )b (8)
sensor is composed of three detector elements (tri-stimulus
detection), with peak responses at 595, 525 and 445 nm for The resulting adapted variables r’, g’, b’, obtained from (8),
the red, green and blue detectors respectively, and spectral can then be used to calculate the LXY tristimulus values as well
bandwidth 170 nm approximately for each detector. as their HLS representation.
The frequency response curve of each detector is utilised by Hence, a calibration method for correcting the variations in
eq. 1 and corresponds to the Rn(λ) wavelength responsivity RGB chromatic values was developed which concentrated on
parameter. removing the RGB errors introduced by non-uniform
illumination and external light fluctuations of the external
B. Chromatic Calibration environment.
The chromatic information received by the intelligent The RGB values, the chromatic coordinates X, Y and the
monitoring system depends on the lighting conditions of the corresponding HLS values of a custom color patch were
surrounding environment. This means that the chromatic co- recorded under controlled laboratory lighting conditions [12],
ordinates of an object under daylight conditions will be which have been used as reference information for any further
different than those under artificial light conditions. Other processing.
parameters that affect the chromatic measurements are
lighting fluctuations, secondary light sources and shadowing. C. Gamma compensation
In a chromatic based system, this ambiguity which is involved An important characteristic of the human visual system is
with the measurement of an objects chromaticity makes it that it is sensitive to ratios of intensity levels rather than to
difficult to relate the optical emissions received from an object to absolute values of intensity. Therefore, the intensity levels
the measurand. The RGB, Lxy and HLS color models are should be placed logarithmically rather than linearly to
designed to be used with applications where the lighting achieve equal steps in brightness [13]. Gamma compensation is
conditions are not too different from daylight. Under different a facility available on most CCD cameras and it is applied to the
lighting conditions alternative procedures are usually adopted R, G, and B detector responses to simulate the non-linear
and incorporated into these models to allow for lightness
adaptation.
364 TOMTSIS

0,7

0,6

0,5
Chromaticity Y

0,4
Gamma Off
0,3

0,2

0,1 Gamma On

0
0,3 0,4 0,5 0,6 0,7 0,8 0,9 1
Chromaticity X

Fig. 4. Graph showing the resulting chromatic variation when gamma

compensation is activated (full curve) and deactivated dashed curve.

function of the pupil of the human eye to provide high

amplification of low intensity light and reduced gain for high
intensity light and hence to provide larger intensity bandwidth
for a given aperture setting. The effect of gamma
compensation was investigated using the optical monitoring
system of fig. 1, but in this case the sensor was replaced with
a white card to provide a source of diffused white light. The X Fig. 5. Experimental apparatus for monitoring the electrical plasma.
and Y chromatic coordinates were then evaluated for different
aperture settings with the gamma compensation deactivated. mathematical analysis and considerable experimental effort
This procedure was then repeated with the gamma required to model their complex characteristics [14]. For
compensation activated. The results are shown in fig. 4. example, the arc plasma column has properties which vary
It is evident from this chromaticity graph that the chromatic with time and with radial, axial, and azimuthal coordinates.
changes due to gamma compensation are increased from Phenomena at the arc electrodes are also complex and
{X,Y}={0.65, 0.27} to {X,Y}={0.85, 0.17}. This means that interactions with the plasma column as well as the interaction
this facility can be used as a preprocessing tool in order to of the arc with its environment add to the complications [15].
increase the contrast around an objects boundary and hence to Investigation of the electric arc phenomena has led to the
enhance an algorithmic process. identification of independent conditions which are simplified
by eliminating some of the complexities described above, and
perform detailed, highly accurate and localized measurements
IV. PLASMA MONITORING
on that condition [16]. However, these measurements are
In general, electric arcs consist of three physically distinct made at the expense of determining how the overall electric
regions which correspond to two electrode regions and an arc and its environment behave and respond.
interconnecting plasma column. The relative importance of Spectroscopic detection and analysis techniques were also
the three regions in governing the overall behavior of the arc used in plasma monitoring applications but they are
depends upon a number of factors which include the length of cumbersome and slow for on-line process control.
the arc gap, the type of arc and the electric power dissipated in Often, such detailed knowledge is unnecessary for such
the discharge. process applications and all that is required is the
Electric arcs may be subdivided into two main categories identification of particular spectral signatures which are
which are axis-symmetric and non axis-symmetric arcs. The associated with certain quantities or parameters.
axis-symmetric arc column burns symmetrically along the
A. Experimental Results
inter-electrode axis so providing good cylindrical symmetry,
which is ideal not only for theoretical modeling but also for Operational conditions in the plasma of such discharges
experimentation with sophisticated diagnostic techniques. A need to be kept within specified tolerances and these
form of axis-symmetric arc without any boundary constraints conditions may be indicated by the spectral composition of the
is the ‘free burning arc’ and is representative of furnace type emission from the discharges.
arcs. Non axis-symmetric arc columns are less amenable to The experimental arrangement for producing the plasma is
diagnosis because of the reduced cylindrical symmetry they shown in fig. 5. The discharge was DC in nature being
provide. maintained by a high voltage supply across the discharge
Monitoring electric arc plasma is a difficult process because electrodes. The optical monitoring system of fig. 1 was
of the complexity of the phenomena which govern arc monitoring directly the discharge through a window at the side
behavior. The physics of these processes and properties of the of the apparatus. A white light source was not used since the
electric arcs are not fully understood due to the powerful
 INTELLIGENT PLASMA MONITORING SYSTEMS 365

Fig. 8. Variation in the dominant wavelength (diamond) and intensity

(square) of a plasma as a function of hydrogen concentration (x-axis).

number of spectral lines in the wavelength range 300-475 nm,

hydrogen from 300-650 nm and nitrogen from 300-500 nm
and 620-900 nm. For hydrogen and argon there is continuous
emission apparent at the longer wavelengths.
The chromaticity values measured with the tri-stimulus
system of fig. 1 are shown on the chromaticity diagram of fig.
7. The dominant wavelengths are 820, 920 and 875 nm for the
argon, hydrogen and nitrogen respectively. The horse-shoe
shaped chromatic boundary is also shown on the same figure.
A set of experimental results demonstrating the effect on
Fig. 6. Spectra of gas discharges used for plasma monitoring and processing. dominant wavelength and on intensity of changing the
(a) Top: Argon plasma, (b) Middle: Hydrogen plasma, (c) Bottom: Nitrogen composition of the gases forming a DC plasma are presented
plasma.
in fig. 8. The results show the possible utilization of the
arcing event by itself produced the necessary illumination. chromatic approach for discriminating correct and incorrect
Neutral density filters were used to reduce the illumination of process conditions.
the plasma and to ensure that the operation of the camera was
in the correct range by avoiding saturation of the CCD
elements, and also to provide optical information concerning V. CONCLUSION
emission from the weakly emitting peripheral regions of the The chromatic approach offers considerable scope for
plasma column. intelligent parameter monitoring with a wide range of
Both spectroscopic and chromatic measurements were principles and different optical sensing elements. Hence, there
taken and the emission from three commonly used gases for is the possibility of realizing an intelligent unified and
plasma processing (argon, hydrogen and nitrogen) reusable monitoring system whereby an extensive range of
investigated. The spectral signatures for the three gases are parameters are monitored using common forms of optical
shown in fig. 6. The argon discharge is characterized by a detection and signal processing methods in combination with
various sensing elements.

REFERENCES
[1] D. Tomtsis, V. Kodogiannis, “Optical pH Measurement Using
Chromatic Modulation”, International Conference of Computational
Methods in Sciences and Engineering, ICCMSE 2003 Kastoria, Greece
12-16 September 2003.
[2] P. Moon and D.E. Spencer, “The Photic Field”, Cambridge, MA:MIT,
1981.
[3] D. Tomtsis, V. Kodogiannis, “Digital Image Processing of Axis-
Symmetric Electric Arc Plasma Based on Chromatic Modulation
Methods”, 6th WSES /IEEE Multi-conference CSCC-MCP-MCME
2002, Crete, July 2002.
[4] D. Tomtsis, K. Sapalidis, “A CCD-Based Tristimulus Colorimeter for
Fig. 7. Location on a chromaticity diagram of the emission from the plasma Fast and Improved Measurement of Chromaticity Coordinates Of
of Argon, Hydrogen and Nitrogen gases (diamond) and the white light point Displays by Matrix Correction Methods”, WSEAS Transactions on
(square). Information Science and Applications, Issue 6, Volume 1, December
2004, pp. 1606-1611, ISSN: 1790-0832.
[5] CIE, Colorimetry Publication 15, 1971.
366 TOMTSIS

[6] CIE, Colorimetry Publication 15.2, 1986a [11] R.W. Hunt, “Measuring Color”, J. Wiley and Sons, New York, 1987.
[7] P.C. Russell, J. Cosgrave, D. Tomtsis, A. Vourdas, L. Stergioulas and [12] C. S. McCamy, H. Marcus and J.G. Davidson, “A color-rendition chart”,
G.R. Jones, “Extraction of Information from Acoustic Vibration Signals Soc. Photog. Scientists Engineers, vol. 2, no 3, 1976, pp. 95-99.
Using Gabor Transform Type Devices”, Measurement Science and [13] J.D. Foley, A. Van Dam, K.S. Feiner and J.F. Hughes, “Computer
Technology, Vol. 9, August 1998, pp. 1282-1290. Graphics: Principles and Practice”, Addison-Wesley, 1996.
[8] J.D. Foley, A.V. Dam, S.K. Feiner, J.F. Hughes JF, “Computer graphics: [14] G.R. Jones, “High Pressure Arcs in Industrial Devices: Diagnostic and
Principles and practice”, 1995, Addison-Wesley. Monitoring Techniques”, Cambridge University Press, Cambridge,
[9] D. Tomtsis, V. Kodogiannis, P. Chountas, “Distimulus Chromatic 1988.
Measurement Systems”, WSEAS Transactions on Circuits and Systems, [15] G.R. Jones and M.T.C. Fang, “The Physics of High Power Arcs”, Rep.
Vol. 3, No. 2, April 2004, pp. 211-214, 2004. ISSN: 1109-2734. Prog. Phys., 43, pp. 1415-1465, 1980.
[10] D. Tomtsis, K. Sapalidis, C. Katsanos, “Electronic Tristimulus [16] C.H. Flurscheim, “Power Circuit Breaker Theory and Design”, Peter
Chromatic Measurement Systems”, WSEAS Transactions on Circuits Peregrinus Ltd., Herts, 1975.
and Systems, Issue 9, Volume 3, November 2004, pp. 1835-1840, ISSN
1109-2734.
 A Neoteric Chaotic & Adaptive Watermarking Scheme for
Colored Images
Navita Srivastava* & Achintya Singhal †
*
Department of Computer Science †Department of Computer Science
Awadesh Pratap Singh University, Rewa Banaras Hindu University, Varanasi
E-mail: [email protected] E-mail: [email protected]
Abstract- Today users are enjoying the convenience and This paper uses an image oriented chaotic factor for hiding
advantages of advancement in technology. At present both text and colored logo watermarks in the targeted
technology provides high quality multimedia processing, multimedia data as per algorithm outlined in [5]. The
acquisition, flexibility, and reliability at significantly lower effective size of the image watermark is significantly reduced
costs and ease. Digital multimedia documents authentication before embedding. The proposed algorithm extracts a
is thus one of the most important and investigated security graceful watermark from the authenticated image which is
applications. Watermarking schemes for the purpose has resilient to high quality JPEG compression, image format
become evergreen research area. This paper outlines the transformations like PNG, TIFF etc and image enhancements
previous work carried before by the authors and then like color correction and filtering like high pass, Gaussian
proposes an image adaptive chaotic algorithm for noise etc. This is obtained by avoiding the lossless entropy
watermarking of digital mages. The algorithm is tested coding step of common image compressing algorithms. The
against both image and text watermarks and found resilient paper is organized into sections: section II describes the
to compression and filtering. watermark embedding and extraction process followed by
results and discussion in section III and conclusions in
Keywords: watermarking; data hiding; image authentication; section IV.
multimedia security; chaotic, henon map. II. The Process
I. Introduction A. The background
Quality work entails a lot of hard work, coupled with The watermark insertion process is done in transform
talent and use of intricate expensive equipments. But, the domain using Discrete Cosine Transformation (DCT) where
declining technological barriers, as a means to protect a reversible, linear transform action is used to map an image
intellectual property, have emerged as the greatest threat. It is into a set of transformed coefficients, which are then
therefore obligatory to provide copyright protection for such quantized and coded. The watermark insertion strength is
work and retain confidence of brainpower. Cryptography calculated chaotically depending on the host image block
tools available provide digital content integrity verification pixel value. Henon map is used.
bit by bit but multimedia content authentication requirement For image processing, the definition of two-dimensional
is much different from it where by content enhancement/ DCT, block-sized n×n, is given by [23]:
worsening is allowed, without compromising the semantic ª f ( j, k ) ∝ (u ) ∝ (v) × º
n −1 n −1
« »
content. This resulted use of watermarking techniques to hide F (u, v) = ¦¦ « ª ( 2 j + 1) uπ º ª ( 2k + 1) vπ º »
j =0 k = 0 cos « » cos « »»
the authentication information just inside the digital content « 2n 2n
¬ ¬ ¼ ¬ ¼¼
because watermark becomes the part of the multimedia data
and can be recovered to prove authenticity but can’t be ­° 1
u ,v = 0
removed from it. The vital condition for watermarking is that
the embedding technique should be such that it should
∝ (u ), ∝ (v) = ® n
2
produce the same results (multimedia document) even after °̄ n
u ,v = 1, 2 ,...,n −1

the embedding. for u, v = 0,1, … , n;

Most watermark embedding processes are performed in
either spatial domain or transform domain. In spatial domain Here each transformed coefficient in an n×n DCT block is
watermarking schemes [24], [25], the watermark image is denoted as F(u,v).
directly embedded into the host image by changing its pixel The original image, F and watermark, W is DCT
values. Both the insertion and extraction processes are transformed; F is watermarked by selecting suitable image
relatively simple compared to transform domain DCT blocks and incorporating the DCT coefficients of W
watermarking schemes. However, it is more difficult for with DCT coefficient of F, leaving behind the DC value [10].
spatial domain watermarks to achieve imperceptibility DCT transformation provides resistance against JPEG
because of the need to embed high intensity digital compression and since it is based on HVS, the embedding in
watermark for robustness tends to degrade the image visual the domain helps in achieving imperceptibility.
quality. In transform domain watermarking schemes [6], [7], The watermark embedding strength is decided chaotically
[8], [9], [11], [13], [15], [17], and [20], transform domain for the selected block using Henon map. Chaotic systems are
coefficients of the host image are modulated by the those whose state evolves with time – that may exhibit
watermark information hence makes highly difficult to dynamics that are highly sensitive to initial conditions. Even
extract the watermark for the attacker and also achieve high though chaotic systems are deterministic dynamical systems,
quality watermarked image. but sensitivity to initial conditions results the behavior of

367
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_64, © Springer Science+Business Media B.V. 2010
368 SRIVASTAVA AND SINGHAL

chaotic systems appears to be random. In general, the chaotic Edge detection algorithm mark the points in the digital
model system is given as: image at which the luminous intensity changes sharply.
x ( n) = f ( x ( n − 1)) Sharp changes in image properties usually reflect important
events and changes in properties of the world. Hence,
where x(n) is a chaotic sequence generated by the
generating an edge map of the image and breaking it into
nonlinear map f (⋅), x(0) is the initial condition. equi-sized blocks (8x8), gives blocks with high activity
We propose the use of model introduced by French blocks as discussed in [1], [2], and [3]. Also, the image is
astronomer Michel Hénon called Hénon Map [21]. The quantized using DCT and JPEG quantization matrix. This
Hénon map is a discrete-time dynamical system and takes a helps in identifying blocks resistant to JPEG compression
point (x, y) in the plane and maps it to a new point and also high activity blocks. Common blocks from the two
processes are then selected.
x n +1 = y n + 1 − ax n2
B. The Embedding Process
y n +1 = bx n The graphical representation of the embedding process is
The map depends on two parameters, a and b. For values given in figure 3. The original image and the watermark is
of a and b the map may be chaotic, intermittent, or converge DCT transformed and the blocks suitable for watermarking
to a periodic orbit. But, generally, for smaller iterations, the are segregated as discussed above. For each suitable block,
model has chaotic behavior. The figure 1 shows the Henon watermark embedding strength factor is calculated using
plot generated for a particular image block. Figure 2 shows Hénon map. The watermark to be incorporated is also DCT
the incorporation strength factor generated for each pixel, to transformed. Incase of image watermarks, the watermark is
be used for watermarking in the block (8 X 8). lossy compressed and doubly DCT transformed before
embedding. With the calculated strength factor and the
provided watermark, the watermark is embedded as
discussed in [2].

Figure 1: Henon plot generated for a block and

corresponding histogram in x-y and y-x direction
Figure 3: Watermark embedding process
C. The Extraction Process
The graphical representation of the extraction process is
given in figure 4. This is a reserve process of embedding and
requires original image and the image to be authenticated.
The watermarked image will yield incorporated watermark
else no significant result can be computed.

Figure 2: Showing chaotic strength factor for each pixel

The blocks suitable for watermarking and shedding the
featureless blocks is computed using two techniques in
conjunction i.e. edge detection and quantization matrix.
Since an image is 2-dimensional representation of an object
capturing its characteristics against some environment or
background, it can be segregated into areas representing
important parts and areas representing flat and featureless
regions.
Figure 4: Watermark extraction process
 A NEOTERIC CHAOTIC & ADAPTIVE WATERMARKING SCHEME 369

III. Results The calculation of NC [19], Q [26], PSNR and PSNR-

The results show the original image (figure 5, 13) along HVS [18] gives the image quality index. The value of PSNR
with its edge map (figure 6), the watermarked image (figure above 35dB is considered to very good. The PSNR value
7, 14), and the watermark used. The proposed scheme is below 10dB is not good and below 5dB is unacceptable. The
tested with both the colored watermark (figure 8) and the text values of NC and Q lies in the range of 0 and 1 and of
watermark (figure 15). The multiple copies of extracted PSNR-HVS lie in the range of 0-100db. Higher the value,
watermarks (figure 9-12 and figure 16-19) are shown. better are the results. Computation of NC, Q, PSNR and
PSNR-HVS of extracted watermark is shown.
.
Results with colored image as watermark

Figure 8:
Watermark Used

Figure 5:Original Figure 6:Edge Map of Figure 7:Watermarked Image

Colored Image the image (240x240) NC=0.9999, Q=0.9980, PSNR=37.458db,
(240x240) PSNR-HVS=86.656db

Figure 9: Watermark extracted from the original watermarked image

Extracted Watermark 1 2 3 4 5 6
NC 0.92089 0.9911 0.95733 0.99972 0.98248 0.9999
Q 0.78094 0.97751 0.65726 0.97407 0.94143 0.97479
PSNR 17.453 27.632 14.466 26.846 23.490 26.917
PSNR-HVS 62.223 76.584 58.958 74.708 69.103 74.488

Figure 10: Watermark extracted JPEG compressed (100% quality) watermarked image
Extracted Watermark 1 2 3 4 5 6
NC 0.9210 0.99323 0.96215 0.9997 0.97752 0.99652
Q 0.74085 0.96754 0.65941 0.95702 0.91934 0.95946
PSNR 16.578 26.03 14.503 24.617 22.035 24.827
PSNR-HVS 61.252 73.772 59.024 71.071 67.342 71.357

Figure 11: Watermark extracted from JPEG compressed (70% quality) watermarked image
Extracted Watermark 1 2 3 4 5 6
NC 0.85414 0.99265 0.94215 0.99247 0.98831 0.99541
Q 0.51453 0.8469 0.49458 0.84481 0.82711 0.86554
PSNR 13.167 18.965 11.979 18.858 18.269 19.050
PSNR-HVS 57.889 64.311 56.7 64.124 63.304 64.278

Figure 12: Watermark extracted from JPEG compressed (25% quality) watermarked image
Extracted
1 2 3 4 5 6
Watermark
NC 0.96133 0.88003 0.90673 0.93168 0.90451 0.9888
Q 0.38554 0.27369 0.26757 0.35738 0.31649 0.46923
PSNR 9.8404 9.6085 8.6569 10.037 9.5879 10.812
PSNR-HVS 54.496 54.391 53.578 55.044 54.305 55.757
370 SRIVASTAVA AND SINGHAL

Image Quality Values with JPEG Compression

Image Image Extracted Watermark*
Quality Q NC PSNR PSNR HVS Q NC PSNR PSNR HVS
25 0.97287 0.99578 32.8936 82.3866 0.37312 0.92453 9.7836 54.5583
30 0.97543 0.99608 33.2570 82.9433 0.42885 0.93820 10.5597 55.4103
35 0.97776 0.99652 33.6040 83.4483 0.45301 0.93008 11.1291 55.9846
40 0.98031 0.99675 34.0080 83.9406 0.54114 0.96950 12.0780 56.9800
45 0.98325 0.99720 34.5336 84.5243 0.60567 0.98016 12.9420 57.8166
50 0.98595 0.99762 35.1153 85.1850 0.71419 0.99179 14.8720 59.8380
55 0.98868 0.99858 36.1213 88.5916 0.73065 0.98468 15.4103 60.2820
60 0.98971 0.99874 36.4823 89.2483 0.76950 0.98489 16.2626 61.1966
65 0.99133 0.99899 37.1650 90.6863 0.80732 0.98413 17.3220 62.1320
70 0.99282 0.99927 37.9546 92.2746 0.86967 0.99053 19.0036 63.9820
75 0.99397 0.99948 38.7236 93.8706 0.92029 0.99159 21.4696 66.5353
80 0.99530 0.99973 39.9446 95.9230 0.95746 0.99892 24.2180 69.3803
85 0.99586 0.99989 40.5943 97.9793 0.96939 0.99954 25.8053 71.0510
90 0.99687 0.99997 42.3833 98.4450 0.97823 0.99980 27.3986 72.5283
95 0.99749 0.99999 44.4423 99.5000 0.98510 0.99956 29.2146 74.5810
100 0.99782 1.00000 46.5123 99.9970 0.98599 0.99990 29.2693 74.3253
* Extracted watermark quality measure is calculated against the watermark recovered from original watermarked image.

Results with text as watermark

THIS WORK BELONGS TO
ACHINTYA SINGHAL &
NAVITA SRIVASTAVA,
COMPUTER SCIENCE
DEPARTMENT,
AWADHESH PRATAP
SINGH UNIVERSITY REWA
MP
Figure 13: Original Colored Figure 14: Watermarked Image
Image (240x240) NC=0.9999, Q=0.99379, Figure 15: Text Watermark
PSNR=32.579db, PSNR-HVS=82.632db Used

Figure 16: Watermark extracted from the original Figure 17: Watermark extracted from JPEG compressed
watermarked image (100% quality) watermarked image
 A NEOTERIC CHAOTIC & ADAPTIVE WATERMARKING SCHEME 371

IV. Conclusion
In this proposed watermarking scheme, the strength factor
is computed chaotically and dynamically for each block and
each pixel wherever the insertions have to be made. Further,
the watermark is embedded only in sensitive areas of the
image thus securing the actual concerned content of the
image. For identifying sensitive areas, image feature
extraction using edge detection and JPEG quantization
matrix is deployed. Moreover, the concept exploited only the
Y component of YCbCr color space for watermarking as Cb
and Cr components are more prone to loss during JPEG
compression as our human visual system is dull for changes
in these components [22]. The quality of extracted
watermark (colored logo) from the unhampered image
deteriorated in comparison to the algorithms proposed in [1],
[2], [3], and [4] due to compression of the watermark used.
Figure 18: Watermark extracted from JPEG compressed But, the compression helped in achieving more insertions in
(91% quality) watermarked image lesser space thus retrieval in case of cropping is more
promising.
The concept of double DCT introduced for colored
watermarks to incorporate greater sustainability and provides
cryptic sense to blind attacks being made. The algorithm
proved to be generic and the results depict good sustenance
to compression too with both image and text watermarks.
The quality metrics of the extracted watermark depicted
better results than previous algorithms [1], [2], [3], and [4].
Adobe Photoshop CS and Adobe ImageReady CS have been
used for JPEG compression and various attacks.
The values of NC, Q, PSNR, PSNR-HVS reflect the
imperceptibility of the watermark as well as the quality of
retrieved watermark samples. The use of Hénon map
guarantees that the watermarking strength for each pixel
becomes inadvertent. Redundant inclusions and chaotic
strength factor provides significant robustness and helps in
obeying Kerckhoff’s principle [27].
Figure 19: Watermark extracted from JPEG compressed V. References
(85% quality) watermarked image [1] A. Singhal, N. Srivastava and R. Mishra, “Edge sensitive
The scheme, with colored watermark, shows excellent frequency domain watermarking”, in Proc. International
resistance to various attacks like JPEG compression up to Information Technology Conference, IITC, Colombo, Sri
25% quality, color correction, brightness and contrast Lanka, pp. 127-134, Oct. 2006.
alterations, diffusion, high pass filter, uniform and Gaussian [2] A. Singhal, N. Srivastava and R. Mishra, “Hiding
noise, rotations with 90 and 180 degrees etc. Using text as a signature in colored image”, in Proc. IEEE International
watermark has got steganographic value more than digital Symposium on Communication and Information
watermarking. Its prime objective is to convey the hidden Technologies, ISCIT, Bangkok, Thailand, pp. 446-450, Oct.
message to the receiver of the file. Its sustenance even with 2006.
JPEG compression increases it application. [3] A. Singhal, N. Srivastava and R. Mishra, “Robust and
The proposed algorithm used meaningful colored sensitive watermarking scheme for Colored Images”, in
watermark as well as text watermark in contrast to most of Proc. 41st annual convention CSI, Kolkata,, India, pp. 13-17,
the previous work where either random number bit stream Nov. 2006.
[14], or some known bit stream [16], or binary image [9]. [4] A. Singhal and N. Srivastava, “Digital Watermarking of
The algorithm incorporated 8 pixels per selected block (8x8) Colored Images Using DCT-HSI Transformation”, in Proc.
for colored watermarks and 8 bits per selected block (8x8) 40th annual convention CSI, Hyderabad, India, pp. 201-204,
for text watermarks which is significantly high in Nov. 2005.
comparison to 3 bits per block (8x8) in [17], first three AC
[5] A. Singhal and N. Srivastava, “Chaotic Watermarking
components of selected block (8x8) in [14] and 3840 pixels
Scheme for Colored Images”, The Icfai University Journal of
from 768x512 size image [16]. Further, the use of chaotic
Information Technology, vol. 4, no. 4, pp. 23-33, Dec. 2008.
technique helped in attaining so called randomness in
strength factor in comparison to fixed strength factor [9], [6] C. H. Chang, M. Zhang, and Z. Ye, “A content-dependent
[12], [14], and [16]. robust and fragile watermarking scheme”, in Proc. 2nd
International Conference on Visualization, Imaging and
Image Processing, pp. 201–206, Sep. 2002.
372 SRIVASTAVA AND SINGHAL

[7] C. H. Chang, Z. Ye, and M. Zhang, “Fuzzy-ART based Proceedings on Vision, Signal and Image Processing, vol.
digital watermarking scheme”, in Proc. IEEE Asia Pacific 143, no. 4, pp. 250-256, Aug 1996.
Conference on Circuits and Systems, APCCAS, vol. 1, [17] J. Zhao and E. Koch, “Embedding robust labels into
pp. 425–426, Dec. 2002. images for copyright protection”, in Proc. Int. Congress IPR
[8] C. Rey and J. L. Dujelay, “Blind detection of malicious for Specialized Information, Knowledge and New
alterations on still images using robust watermarks”, in Proc. Technologies, Vienna, Austria, pp. 242–251, Aug. 1995.
IEE Secure Images and Image Authentication Colloquium, [18] K. Egiazarian., J. Astola, N. Ponomarenko, V. Lukin, F.
London, U.K., pp. 7/1–7/6, Apr. 2000. Battisti, and M. Carli, “New full-reference quality metrics
[9] C. T. Hsu and J. L. Wu, “Hidden digital watermarks in based on HVS”, in Proc. Second International Workshop on
images”, IEEE Trans. Image Process., vol. 8, no. 1, pp. 58–68, Video Processing and Quality Metrics, Scottsdale, 4 p, 2006.
Jan. 1999. [19] M. A. Suhail and M. S. Obaidat, “A robust digital
[10] Chun-Shien Lu, “Multimedia security: steganography watermarking technique”, IEEE, pp. 629–632, 2000.
and digital watermarking Techniques for Protection of [20] M. Barni, F. Bartolini, V. Cappellini, and A. Piva, “A
Intellectual Property”, Idea Group Publishing, 2004. DCT-domain system for robust image watermarking”, Signal
[11] E. Izquierdo, “Using invariant image features for Processing, vol. 66, no. 3, pp. 357–372, 1998.
synchronization in spread spectrum image watermarking”, [21] M. Henon, “A two-dimensional mapping with a strange
EURASIP Journal of Applied Signal Processing, vol. 4, attractor”, Communications in Mathematical Physics, vol.
pp. 410–417, 2002. 50, pp. 69-77, 1976.
[12] H. Inoue, A. Miyazaki, and T. Katsura, “Wavelet-based [22] M. Ejim and A. Miyazaki, “A wavelet-based
watermarking for tamper proofing of still images”, watermarking for digital images and video”, in Proc.
International Conference on Image Processing Proceedings, International Conference on Image Processing, ICIP, vol. 3,
ICIP 2000, pp. 88-91, Oct 2000. pp. 678-681, Oct. 2000.
[13] J. Cox, J. Kilian, F. T. Leighton, and T. Shamoon, [23] R. C. Gonzalez and R. E. Woods, “Digital Image
“Secure spread spectrum watermarking for multimedia”, Processing”, Pearsons Higher Education 2/E, 2002.
IEEE Transactions on Image Processing, vol. 6, no. 12, pp. [24] R. G. Schyndel, A. Z. Tirkel, and C. F. Osborne, “A
1673–1687, Dec. 1997. digital watermark”, in Proc. IEEE International Conference
[14] J. Huang, and Y. Q. Shi. “Adaptive image watermarking on Image Processing, vol. 2, pp. 86–90, Nov. 1994.
scheme based on visual masking”, Electronics Letters, vol. [25] W. Bender, D. Gruhl, N. Morimoto, and A. Lu,
34, no. 8, pp. 748-750, Apr 1998. “Techniques for data hiding”, IBM System Journal, vol. 35,
[15] J. J. Eggers and J. K. Su, “A blind watermarking scheme no. 3&4, pp. 313–336, 1996.
based on structured codebooks”, in Proc. IEE Secure Images [26] Z. Wang and A. C. Bovik, “A universal image quality
and Image Authentication Colloq., London, U.K., pp. 4/1– index”, IEEE Signal Processing Letters, XX, 2002.
4/21, 2000.
[27] http://en.wikipedia.org/wiki/Kerckhoffs'_principle
[16] J. J. K. Ó. Ruanaidh, W. J. Dowling, and F. M. Boland,
“Watermarking digital images for copyright protection”, IEE
 Rotary Transducer
Ilie Borcosi1, Nicolae Antonie2, Alina Dinca3
“Constantin Brancusi” University
Calea Eroilor, no.30, Targu-Jiu, Romania
1
[email protected], [email protected], [email protected]

Abstract – Using the speed voltage generators to measure the formed of the following components: emitters (LED),
rotary of a driving system is an advantageous solution for the receivers (photodetectors), rotating codewheel, stationary
analogical structures of the command circuit. For the numerical mask and the electronic circuit for the handling of the signals.
systems we will need to use an analogical-digital converter and The two wheels (rotating and stationary) are divided in the
for this reason these will be more useful and economical using the
numerical transducer. Trough this paper we propose a numerical
opaque and transparent zones. When the transparent zones of
rotary transducer which offers advantageous price/ both wheels are superposed, the receivers get the light flux
performances report. from the emitters and transmit an impulse to the output. It is a
The purpose of this application is to diagnose a process relative transducer that gives a number of specified impulses
automatically, with no human analysis involved. for one rotation of its ax, which depends on the number of the
transparent zones. This impulse rotation number will lead to
I. INTRODUCTION the transducer resolution that can be considered like an
analogical-digital converter of the position.
The control of a driving system of physical process
For a number of N rotation impulses the equivalent
presumes the obtaining of information about it, information
parameters of the position transducer will be:
that need to be worked on, memorized or communicated to a
-the variation beach of the output: N (bits);
higher level. The physical process is determined by physical
-the transducer resolution: [log2N] (bits);
values that can be transformed to electrical signals (usually
analogically by nature) by the transducer obtaining in this way -the precision of the transducer: 2π/N [rad/bit].
information about the process. We can work on this signals Because the encoder doesn’t give information about the
absolute position, we will use a special interface, basically a
using analogical technique or numerical technique case in
witch the analogical signals given by the transducers are numerical circuit, to count the impulses given by the encoder.
converted into numerical signals using special electronic At the same time, because using only one train of impulses
doesn’t allow the obtaining of the rotation sense, the encoder
circuits named data acquisition systems.
The speed transducer or rotary transducer will give give 2 trains of rectangular impulses at 90 degrees one against
information about the speed of the motor during an application the other. A sense detection circuit can be designed so that it
could, based on the notification of this difference phase
with the diving system. Usually we use a speed voltage
generator of direct current to give the information about the (advance or delay with 90 degrees of one signal against the
necessary speed. Because this transducer has an analogical other) to determine incrementing and decrementing of the
position info.
voltage at its output, proportional to the speed of the motor,
we have to use an analogical – digital converter if we want to Because this transducer is relative, which doesn’t point to
obtain the numerical value of this signal, used in a numerical the position modification facing the starting position of the
system. If we know the constant of the speed voltage generator, system, the encoders give another numerical signal, the also
kTG [V/rad/s], and the maximum value of the speed of the called zero signal of the encoder. This signal becomes active
giving a pulse by the length of an encoder step just one time
motor, Ωmax [rad/s], we can determine the maximum value of
on the rotation and it can be also used for the initialization
the tension at the output of the speed voltage generator which
after putting it under supply voltage so we can set an absolute
needs to be measured, VGmax = kG * Ωmax [V]. Usually, the position info of the system.
connecting scheme of the output of the speed voltage In the figure 1 we can see the shape of the signals given by
generator to the entrance of the analogical – digital converter
the encoder (A and B the two measurement canals; Z the
has to be adjusted (by using a simple resistive divisor) used to initial position canal) [2]. It is possible the using of each one
adapt the variation beach of this signal to the accepted values of the positive and negative edge of the signals A and B of the
beach from the entrance of the analogical-digital converter
encoder which allows the increasing by 4 times the precision
(usually +/- 10 V or +/- 5 V). The converter resolution in the of the encoder by implementing some adequate decoding
driving system applications is usually 10-12 bits. circuits.
The position transducer makes the measurement of the
driving system position. An advantageous solution is offered
by the utilization of the incremental optical encoder which
eliminated the analog digital converters [1]. The encoder is

373
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_65, © Springer Science+Business Media B.V. 2010
374 BORCOSI ET AL.

Fig.1. The signals generated by the encoder to output.

Fig. 3. The circuit for handling the sensors signals A, B, C.

II. SYSTEM DESCRIPTION
A. Presenting the system
In the figure 2 is presented a structure for the revolution
transducer designed and made by us. It is formed from a disk
(figure 2.a), placed on the motors ax, disk on which is made an
orifice Or [3]. This orifice allows the passing of the bright flux,
from a transmitter to a receiver set face to face on a scale in
the figure 2.b. These are sorted in a way so that the orifice Or
to pass through each group emitter – receiver.

a)

b)
Fig. 4. The input signals (Vosen) and output (Vc) from the handling circuit.
Shaft
B. System needs
B A The determination of the sense and the value of rotation are
C made by implementing the logical diagram from figure 5.
Or c)
According to the logical diagram, first we attribute to the
rotation n and the variable v1 value zero, after that we check
Fig. 2. The element component of the transducer the A sensor. If A=0 we verify C, and if C=0 then we verify B.
In case that A=1 logical and C=1 logical we set the counter
We use infrared sensors so that the natural light doesn’t (which will start to implement the impulses with fi frequency)
influence the work. The three sensors A, B, C (figure 2.c) and the control variable v1 (which is used to approve the
transmit at the output Vosen=1 logical when they are evidence after a complete rotation). If follows B=1 and then
illuminated. The signal from the output of the sensors (Vosen) C=1 we made a complete rotation (v1=1) and because of these
is applied to a handling circuit (derivation) made with Schmitt succession of activating the sensors we can tell that the
Trigger inverters (figure 3) to eliminate the errors which transducer disk makes a rotation to the right, we calculate the
appear in the cases of slow fronts (at small rotations). In this value of the revolution n (we process the value to the Temp
way, at the illumination of every receiver, the handling circuit counter). After that we reset the contain of the counter and
generates a short impulse (the signal VC from the figure 3 prepare for a new rotation. If the succession of illumination of
which correspond to the C sensor). the sensors is C=1, A=1, B=1 and then C=1 it means that the
transducer makes a revolution to the left. When one of the
The A and B sensors are used to determine the rotation sense, successions of a rotation sense (left or right) is not kept, we
and the C sensor which obtains the value of the rotation which reset the counter and verify which the new succession is. To
sets up a counter which increments the tact impulses. verify the eventual stopping of the motor in the countering
The rotary will be determined using the formula: time, we inserted a procedure in the figure 6, where the current
n = (fi/ Tc)*60 = 1/(Ti*Tc)*60 [circ/min], value can be compared to the reference value T. This value (T)
where fi=1/Ti represents the frequency of the impulses tact represents the maximum time interval as long as the counter
applied to the counter and Tc the value of the counter (number can increment (Tcmax-preset=Tempmax-preset), between the activity
of increment impulses which are equivalent to a temporization of two sensors.
– Temp).
 ROTARY TRANSDUCER 375

START

v1=0
n=0

a
1
*
0
c A

1 0
C * C 0

0 1 Reset counter
1 0 Set counter 1 B 0
v1 B
v1=1
Reset counter Reset counter
Set counter
v1=0 v1=0
v1=1 1
B
0
* d
Right rotation d 1 0
1 0
Processing Temp (n)
Reset counter
A C
*
Set counter 1 A 0
b a

0 1
1 0 v1
B * Left rotation
Set counter
v1=1 Processing Temp (n)
1 C 0 Reset counter
Set counter
Reset contor
v1=0

b 1 A 0 *
1 0
C
*
Reset counter
1 0 v1=0
d B
c
Fig. 5. The logical diagram.

*
It is chosen so that we can measure the rotation n• nmin-preset, 1 0
where nmin-preset is better to have a small value, resulting for Temp <T
Tcmax-preset= Tempmax-preset a bigger value. We consider that the Reset
motor is stopped if Temp>T (n< nmin-preset) when the counter is counter
reset and it’s attributed the variable v1 and the rotation n has
zero value. *
Fig.6. The verification procedure for stopping the motor.
376 BORCOSI ET AL.

Microcontroller Display

P1.1
P1.2
P1.3
Date
VA
VB INT0 Address
VC

RX TX
Level Communication
Up Module
Fig.7. The processing circuits with microcontroller.

C. Making the system -it can be used for monitoring rotation by keeping the dates
Handling the signals given by the circuit in fig. 3 (VA, VB, in a data base or by showing it;
VC) which shows the sensors’ state, it can be made with the -it does need an A/D converter, in the case of a numerical
help of a microcontroller or the calculator through control system, can be cuter the handling circuit with a
implementation of the logical diagram in fig.5, according to microcontroller, its function will be implemented in the
this example we used a microcontroller type AT89C2051 control and command system. Actually we acknowledge a
(according to the figure 7) [4]...[6]. trend to realize the numerical exclusive systems with alone
The signals VA, VB, VC are connected to the specific input to microcontroller witch provides the all control and command
the external interruption INT0, and at the same time, one functions;
separately to another inputs P1.1, P1.2 and P1.3 which -on the disk we can make a lot of orifices. The real rotation
determine if the signal is activated. The connecting to the will be equal to the calculated rotation multiplied to the
INT0 input should be done using a logical gate OR which adds orifices number. Is preferred that the number of orifices to be
the three signals, but, considering that the interruption a multiple of 2, and the distance between them must not be
becomes active on the negative edge then the signals must be bigger than the distance between the A and B sensors;
negated and applying the principals of Boolean algebra in the - we used this rotation transducer and we recommended it to
place of OR gate and using the NAND gate. The timer/counter be used in drive systems with the DC motor controlled by a
function is made by counting the impulses with a frequency computer (using the parallel port) which has a single external
equal to the oscillator frequency divided by 12. The interrupt input. To this input we connect more signals for real
temporization frequency choosing the oscillator frequency time control, for example, synchronization input (for control
fosc=12 MHz is fi=1MHz and the period is Ti = 1μs. The in phase) and speed transducer outputs. The proposed speed
timer/counter circuit is on 16 bits. We can temporize transducer will enable rarely the interrupt input (for each of its
maximum: 216*Ti=65536*1μs=65536μs – 65ms. signals, once at a complete rotation), compared with an encoder,
It is preferred to obtain a temporization of 50 ms, other wise at which the activation depends on the number of pulses.
if we want a bigger temporization which is easier to multiply. REFERENCES
These value of temporization result if we load the [1] Fl.Grofu, L. Popescu, Digital system for optical incremental encoder
temporization circuit with the initial value 15536, this connection, U.P.B. Sci. Bull., Series c, vol. 69, no. 3, 2007.
[2] Rotary Encoder, http://hades.mech.northwestern.edu/wiki/index.php.
temporization until the value 65536 and each time are touched /Rotary_Encoder.
these value a counter is incremented. [3] O. Olaru, I. Borcosi, M.C. Popescu, O. Tulpan, N. Antonie, The
To detect the stop of the motor in the timer/counter time we revolution measurement of a d.c. machine,The XIIth Edition of the
considered nmin preset=0,1 rot/s, which means that if in 10 International Scientific Conference of the Engineering Faculty,
November 23rd-24th 2007, Tg.-Jiu.
seconds a full rotation is not made, it means that the motor is [4] L. Toma, Sisteme de achizitie si prelucrare numerica a datelor, Ed. de
stopped and the timer/counter circuit is cleared, resulting for T Vest, Timisoara, 1996.
the value of 10 s. The temporization value is 10 seconds and is [5] D. Marinescu, Microcontrolerul 80C32 Manual de utilizare, Editura
obtained by measuring 20 times 50 ms. Tehnica, 1998.
[6] AT89C2051 and AT89C52 datasheet, available at: http://www.atmel.com.
III. CONCLUSIONS [7] Fl. Grofu, L. Popescu, C. Cercel, Acquisition System for Monitoring
Vibration,, CISSE 07 - IETA 07, University of Bridgeport, USA,
The measurement system of rotation designed and made by December 03-12, 2007, Published by Springer Press.
us has the following advantages: [8] I. Borcoi, O. Olaru, M.C. Popescu, A. Dinca, N. Antonie, M. Ionescu,
-it's easy and simple to make and offers advantageous report Device with Analogical Circuits for Protection to the Lack of the Pulse
price/performances because it has the easier processing digital for the Three-Phase Rectifiers in Electrical Drive, INTERNATIONAL
JOURNAL NAUN, Issue 4, Volume 2, December 2008.
system (a single microcontroller) than optical incremental
encoder [1] ;
 Regression Analysis about Humidity Elimination
from Diesel Fuel Via Bioorganic Compounds to
Increase Antifouling Action
C.G. Tsanaktsidis1, N. Sariannidis2 and S.G. Christidis1
1
Laboratory of Qualitative Fuel Control
1
Department of Pollution Control and Technologies/ 2Department of Financial Applications
Technological Education Institute of Western Macedonia,
Kila, Kozani –Greece 50100

Abstract: In this paper we try to export a regression equation The selection of aspartic acid as a raw material was supported
which describes the variation of humidity of Diesel fuel with by the variation of humidity degree that these particle groups
time. A hydrophilic polymer is used to eliminate humidity from show in a water environment.[15]
diesel fuel and the regression analysis shows that the maximum
decrease of humidity in an hour it is of about ~ 35%.
Keywords: Regression Analysis, Diesel Fuel, Humidity Elimination
Bioorganic Compounds, Hydrophilic Polymer Antifouling Action.

I. INTRODUCTION

The quality control of alternative liquid fuel aims at finding

out if fuel is appropriate so that the decrease in the pollution
of the environment can be secured, as well as the right
function of combustion engine and the maximum output.
One of the most important properties of fuels is the water
presence, which is undesirable because it causes corrosion,
can cause ice blockage in the power grid that can be formed
in low temperatures and last it can be divided during the
warming-up and cause flame extinguishing from the produced
steam (water vapour).
In this paper it is attempted to describe the modification of
humidity of diesel fuel via an equation. For this aim we
introduce in different samples of diesel fuel the hydrophilic
polymer TPA (which has the property to retain water
molecules) and its synthesis procedure presented by a suite of
software, ChemOffice.
The values of humidity is recorded in the modification table
by time and with these results is exported the regression
equation [1-4] which describes the variation.
Fig 1. Synthesis of Hydrophilic Polymer, Thermal Polyaspartate Anion (TPA).
It is stressed that the volume of the fuel and the mass of the
polymer maintain stable.

B. Humidity Elimination of Diesel Fuel via TPA.

II. EXPERIMENTAL AND STATISTICAL PROCESS
At first the humidity of Diesel fuels is determined with the
A. Synthesis of Hydrophilic Polymer, Thermal Polyaspartate Anion (TPA) method ASTM D-1744[13] via an automatic potentiometer
[9,10,11,16]
titrator Karl-Fisher.
Next, 0.1 g of TPA is introduced in samples of Diesel 20 mL
The synthesis of polymer is held through the condensation of volume each one. Then, TPA is eliminated through
D,L aspartic acid (1.5g) towards the equivalent filtration.[14]
polysuccinimide [12] and the hydrolysis of the latter towards the The values of humidity are recorded every 30 minutes in
final product of this reaction, the thermal polyaspartate anion Table I.
(TPA).

377
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_66, © Springer Science+Business Media B.V. 2010
378 TSANAKTSIDIS ET AL.

Volume of Diesel (mL) + Mass of TPA (g)

Time
(t)
Humidity (mg/g) values shows that the maximum decrease of humidity in an
hour it is of about ~ 35%.
20 + 0 0 0.0650

20 + 0.1 30 0.0635

20 + 0.1 60 0.0489

20 + 0.1 90 0.0520 Standardized Residuals Squared Standardized Residuals

20 + 0.1 120 0.0600

Auto Partial Auto Partial
20 + 0.1 150 0.0478 Lags Correl Correl- LB(n) Lags Correl Correl- LB(n)
- ation ation -ation ation
20 + 0.1 180 0.0492

20 + 0.1 210 0.0455 1 0.193 0.193 0.4629 1 -0.286 -0.286 1.0149

20 + 0.1 240 0.0630 2 0.048 0.011 0.4951 2 -0.28 -0.394 2.1206

3 -0.295 -0.318 1.9330 3 -0.086 -0.400 2.2427

Table I. Humidity Values by Time of Diesel Fuels with TPA.
4 -0.376 -0.301 4.7266 4 0.263 -0.083 3.6170

5 -0.217 -0.104 5.8903 5 -0.117 -0.249 3.9540

It must be referred that 20 mL is the demanded quantity of 6 -0.074 -0.095 6.0702 6 -0.021 -0.151 3.9683

Diesel fuel for the determination of humidity according to 7 0.147 -0.006 7.1333 12 0.031 -0.096 4.0153
method ASTM D-1744, while the quantity 0.1 g of polymer is 8 0.074 -0.158 7.6720 24 -0.005 -0.172 4.0176
defined from the limits of chemical additives in liquid fuels
which is 1-4 Kg/tn. Table II. LB Test for Residuals and Squared Residuals from the Regression Estimation.
Note: LB(n) are the n-lag Ljung-Box statistics for the residual series. LB(n) follows chi-square
C. Statistical Empirical Findings variable with n degree of freedom; the series of residual contains 9 observations .

The preliminary statistic analysis of sequences of humidity Y

and time T (see Table I) and especially the relevant scatter
plot have shown that the modifications of Y, probably
Squared Residuals lag(-1) Squared Residuals lag(-2) F-statistic
determined by the cosine of T. Specifically, in this paper we
have examined the regression model ˜i=b0+b1Sin(Ti)+ui in -1.533816 -0.40 0.573

order to investigate the relation between humidity of Diesel (-0.84) (-0.87)

fuel and the time, when the volume of the fuel and the mass
of the polymer maintain stable.
Table III. ARCH–LM Test for Squared Residuals.
Some diagnostic tests were performed to establish goodness Notes: Figures in parentheses are t-statistics.
of fit and appropriateness of the model. First, it was examined
whether the standardized residuals and squared standardized
residuals of the estimated model are free from serial
correlation. The results show (see Table II) that the LB b0 b1 Adjusted R2
statistics for the standardized residuals and standardized 0.055* 0.006667** 0.37
squared residuals are not significant [8]. In addition, the
(0.001983) (0.002795)
independence of the standardized residuals is confirmed by
the Durbin-Watson statistics 1.44 [7]. Also, in the Table IV the
ARCH–LM Test concerning two lags in the residuals (N*R2 Table IV. Mean Equations ˜i=b0+b1Sin(Ti)+ui.
= 1.55) showed that the variance does not exhibit Notes: Standards errors are shown in parentheses.
heteroskedasticity [6] that has been confirmed by the *indicates statistical significance at the 1% level.
application of White test (N*R2 = 2.27). Finally, with the **indicates statistical significance at the 5% level.
application of the Jarque-Bera test (1.27) it was found that the
normality of the residuals ui cannot be rejected [5].
III. CONCLUSIONS
Table IV presents results of the regression. The adjusted R2
(0.37) indicates that modifications of the humidity of fuel (Y) The study investigates how the changes of the fuel’s humidity
affected considerably by the Time (T) that the hydrophilic were influenced by the Time (T) that the hydrophilic polymer
polymer š remains at the fuel. Specifically, the positive š remains at the fuel using a regression model. More
coefficient b1 (0.00667) of cosine time and its statistical precisely, it was found that the change of humidity should be
significance is interpreted as the fuel’s humidity is decreased described via regression equation at which the dependent
periodically with time when the fuel’s volume and the variable is the changes of the humidity and the independent
polymer’s mass that is introduced in fuel are constant. The variable is the cosine of the time that the hydrophilic polymer
periodicity of the phenomenon via the max-min of humidity š remains at the fuel.
 REGRESSION ANALYSIS ABOUT HUMIDITY ELIMINATION 379

The analysis of the regression estimation showed that: the [5] Jarque C. M, Bera A. K. (1987) A test of normality of observations and
regression residuals. International Statistical Review 55: 163-172
decrease of humidity is changed periodically with time and
comes to its maximum value at the first hour. [6] Engle R. F. Autoregressive conditional (1982) Heteroscedasticity with
The use of equation can predict the development of humidity estimates of the variance of United Kingdom inflation. Econometrica 50
values for important time periods and gives the moment for (4): 1987-1007.
the filtering of fuel so that the humidity is removed in the [7] Durbin J, Watson G.S. (1950) Testing for serial correlation in least
biggest degree. The elimination of humidity makes the fuel square regression. Biometrika 37: 409-428
used in the car machines and gives combustion with less
pollution for the environment. [8] Ljung G. M, Box G.E.P. (1978) On a measure of lack of fit in time series
models. Biometrika 65: 297-303
It is necessary to study the description via equations of the
phenomenon associates with volume of fuel, mass of polymer [9] The Presidential Green Chemistry Challenge Awards Program:
in order to export a total equation. Summary of 1996 Award Entries and Recipients; EPA744-K-96-001;
The result of elimination of humidity from fuel gives the U.S. Government Printing Office: Washington, DC, 1996
possibility to be rendered as antipollution at his use, so the [10] J. Clark and D. Macquarrie Handbook of GREEN CHEMISTRY AND
analysis of economic cost of the whole process is required TECHNOLOGY 2002
when it is applied in the industrial production.
Collaboration of sectors of science, organic chemistry, [11] Wheeler, A.P, Koskan L.P “Large scale thermally synthesized
polyaspartate as a substitute in polymer applications” Mat. Res. Soc
technology of fuels, statistical analysis for the export of Symp. Proc, 1993, 292,277.
conclusions in the qualitative control of fuels.
[12] J. Mosig, C.H Gooding, A.P Wheeler, “Kinetic and thermal
characterization on the hydrolysis of polysuccinimide” Ind. Enf. Chem.
REFERENCES Res. 1997. 36. 2163-2170

[1] Chuvieco E.; Riaño D.; Aguado I.; Cocero D., Estimation of fuel moisture [13] ASTM D1744-92 Standard Test Method for Determination of Water in
content from multitemporal analysis of Landsat Thematic Mapper Liquid Petroleum Products by Karl Fischer Reagent (Withdrawn 2000)
reflectance data: applications in fire danger assessment, publisher
Taylor and Francis Ltd, International Journal of Remote Sensing, Vol. [14] ASTM D2276 - 06 Standard Test Method for Particulate Contaminant
23, No 11, 2002 , pp. 2145-2162(18). In Aviation Fuel by Line Sampling ,2006

[2] MINGHOU XU*, YAOGUO FAN, JIANWEI YUAN, CHANGDONG [15] .ˆ. roganis, C.Tsanaktsidis and I.P. Gerothanassis, 14N NMR
SHENG AND HONG YAO, A SIMPLIFIED FUEL-NOx MODEL relaxation times of several protein amino acids in aqueous solution-
BASED ON REGRESSION ANALYSIS, INTERNATIONAL JOURNAL OF comparison with 17O NMR data and estimation of the relative hydration
ENERGY RESEARCH, Int. J. Energy Res., 23, 157Ð168 (1999) numbers in the cationic and zwitterionic forms, J. Magn.Reson.,
2003,164, 294
[3] David Carter, Daniel A. Rogers, Betty J. Simkins, Does Fuel Hedging
Make Economic Sense? The Case of the US Airline Industry, AFA 2004 [16] S.G. Christidis, A.I Maroulis, K.Chatziantoniou -Marouli ,
San diego Meetings, September 16, 2002. A.Tsiomlektsis, Aristotle University of Thessaloniki. «The experiment
in teaching of Chemistry. Green Approach. Composition and
Applications of Thermal Polyasparte Anion.». Proceedings of 20th Pan-
[4] Gerfried Cebrat, Avraam Karagiannidis and Agis Papadopoulos, Hellenic Conference of Chemistry. Ioannina September 2005.
Proposing intelligent alternative propulsion concepts
contributing to higher CO2 savings with first generation
biofuels, Management of Environmental Quality: An International
Journal, Vol. 19 No. 6, 2008, pp. 740-749
 Optimal Threshold for Locating Targets Within
a Surveillance Region Using a
Binary Sensor Network
Zhenxing Luo and Thomas C. Jannett
Department of Electrical and Computer Engineering
The University of Alabama at Birmingham
Birmingham, AL 35294

Abstract - This paper considers the design of the optimal thre- optimal threshold for binary sensors detecting targets having a
shold for locating targets within a specified surveillance range range of powers and locations within a specified surveillance
using a network of binary sensors. The threshold was determined region. The paper is organized as follows. Since this paper
by minimizing a cost function representing the summation of the
variances of estimation errors for the x and y coordinates of the extends the work in [1], Section II closely follows the presen-
target. The cost function was evaluated using multiple integra- tation and notation in [1]. Section II presents the sensor mod-
tion over the specified ranges for target power and target loca- el, exemplary network layout, and maximum likelihood esti-
tion. A geometrical interpretation of the optimal thresholds for mator for quantized data. Section II also reviews a method for
targets with known power and location is presented to offer in- designing optimal quantization thresholds for targets with
sight into the problem. The optimal threshold was validated in known power and location. Section III provides a geometrical
Monte Carlo simulations using a field of sensors having a uni-
form grid layout. In simulations, the summation of the variances interpretation of the optimal thresholds for targets with known
of the estimation errors achieved using the optimal threshold power and location. Section IV develops the optimal threshold
approached the minimum of the cost function. for binary sensors locating targets having a range of powers
and locations within a specified surveillance region. Methods
I. INTRODUCTION for computing the optimal threshold are presented and demon-

T ARGET localization is an important and challenging prob- strated. The optimal threshold is validated and evaluated in
lem in sensor networks [1]-[4]. In a network or network Monte Carlo simulations. Discussion and conclusions are pre-
region, sensors transmit data to a central processing node or sented in Sections V and VI.
fusion center. For the target localization problem, the
processing node receives data from each sensor and estimates II. PROBLEM FORMULATION
the target location. Some methods are based on time differ- We consider an exemplary field of N = 441 sensors un-
ence of arrival (TDOA) [5], [6]. However these methods re- iformly deployed with sensors placed 9 m apart in each of the
quire accurate timing, which is not feasible with inexpensive x and y directions (Fig. 1). Many ideas presented in this paper
sensors. Energy-based methods, which use signal strength apply for other sensor deployments, including fields in which
information, are more practical in real situations [7]-[10]. sensor are placed at random locations.
Much recent work has been focused on maximum likelihood The target emits a signal, with the signal intensity decreasing
target localization in networks of inexpensive sensors that as the distance from the target increases. Following the pres-
have limited energy and communications resources [1], [9]. entation in [1], we consider a signal power model
When resources are limited, it is desirable for the sensors to
transmit quantized data to the processing node [1], [11]. A Gi P0'
maximum likelihood estimator that uses quantized data and ai2 = (1)
( di d0 )
n

the Cramer-Rao lower bound (CRLB) for this estimator are

derived in [1]. The CRLB is the theoretical performance
bound for the variances of the estimation errors achieved us- where ai is the signal amplitude at the ith sensor, Gi is the
ing the maximum likelihood estimator. The CRLB has been gain of the ith sensor, and P0' is the power emitted by the tar-
used as the basis for several methods used to control the net-
get measured at a reference distance d 0 . The Euclidean dis-
work localization performance, including designing detection
thresholds, evaluating network layout, and reconfiguring a tance between the target and the ith sensor is
network after loss of resources [1], [12].
In this paper, we extend the work in [1] by developing an d i = (xi - xt ) 2 + (yi - yt ) 2 (2)

381
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_67, © Springer Science+Business Media B.V. 2010
382 LUO AND JANNETT

100
­ 0 −∞ < si < ηi1
° 1 ηi1 < si < ηi 2
80
°°
Di = ®   . (6)
60
°L − 2 ηi ( L − 2) < si < ηi ( L −1)
°
40
°̄ L − 1 ηi ( L −1) < si < ∞
Y-coordinate (m)

20

0
The probability that Di takes specific value l is

-20
)& η − ai η −a
pil (ηi , θ ) = Q ( il ) − Q ( i ( l +1) i ) (0 ≤ l ≤ L − 1) (7)
-40 σ σ
-60
where Q (⋅) is the complementary distribution function of the
-80 standard Gaussian distribution
-100 2
-100 -80 -60 -40 -20 0 20 40 60 80 100 ∞ 1 − t2
X-coordinate (m) Q ( x) = ³ e dt . (8)
Fig. 1. Sensor field layout showing 441 sensors placed 9 m apart.
x
2π

where (xi , yi ) and (xt , yt ) are the coordinates of sensor i and After collecting the data, D , the processing node estimates
the target, respectively, and n is the power decay exponent. the parameter vector θ = [ P0 xt yt ]T by maximizing the log-
We assume that Gi = G for i = 1,...,N , such that P0 = GP0' . In likelihood function of D
this paper, we use n = 2 and we assume that the target is at N L-1
&
least d 0 = 1 meters away from any sensor at all times, result- lnp(D  ) = ¦¦ (Di - l)ln [ pil (i ,)] (9)
ing the simplified signal power model
i=1 l=0

where
P0
ai2 = . (3) ­1, x = 0
d i2 (x) = ® . (10)
¯0, x = 0
The measured signal at the ith sensor is modeled as
In summary, the maximum likelihood estimate of
si = ai + wi (4) θ = [ P0 xt yt ]T is determined through the optimization prob-
lem:
for i = 1,...,N where wi is a Gaussian noise
max lnp ( D  ) . (11)


wi ~ N ( 0, 2
). (5)
The maximum likelihood estimate (11) and the corresponding
Fisher Information matrix (FIM) and CRLB for the estimation
At each sensor, the measured signal si is quantized and
error variance are derived in [1], which also addresses the
transmitted to the processing node. design of sensor decision thresholds based on the CRLB using
As described in [1], the processing node receives quantized both optimization and heuristic methods.
multi-bit data from each sensor. The received data are In this paper, we considered binary decisions made at each
D = {Di : i = 1,...,N } , where Di can take on any discrete val- sensor that yield binary data
ue from 0 to 2 M - 1 . Using the definition L = 2 M , the set of
quantization thresholds for the ith sensor is ­0 −∞ < si < η
& Di = ®
 i = [i0 ,i1 ,...,iL ] , where ηi 0 = −∞ and ηiL = ∞ . The ith sen- ¯1 η < si < ∞
. (12)
sor supplies quantized data
 OPTIMAL THRESHOLD FOR LOCATING TARGETS 383

A. Optimal Quantization Thresholds for Fixed Power and in a surveillance region and then examine the patterns of sen-
Fixed Target Location sors that would fire if the optimal thresholds are applied.
The CRLB for the ML estimate (11) is a function of the For P0 =10000 and binary decisions made at each sensor, op-
quantization thresholds and was used in [1] as the basis for the timal thresholds were determined over a subsection of the
cost function field (Fig. 1) represented by a 46 by 46 grid of target locations
with x = 0.0, 0.2,…, 9.0 and y = 0.0, 0.2,…, 9.0. For this grid
& & & & &
&
V(  ) =
() ()
J 11  [J 33  + J 22 (& ) ] - J (  ) - J (  )
2
13
2
12
(13)
of target locations, the optimal threshold (14) varies symmetr-
ically with target location from a minimum of 6.89 to a maxi-
J () mum of 11.24 (Fig. 2). For a target located far away from any
sensor, the threshold is lower than for a target located near a
which represents the summation of the variances of the loca- sensor. The optimal threshold sometimes changes abruptly
tion estimation errors in the x and y directions. In (13), J ij is from one target location to another.
the element on the ith row and jth column of the FIM, J .
Expressions for J ij are presented in [1] and will not be repro-
duced here. The cost (13) is an obvious choice for designing a
threshold that provides high accuracy in the estimation of the
target location, although the resulting estimate of the power,
P0 , may not be accurate. For a known P0 the optimal thre-

Threshold
10

shold for a target located at specific coordinates (xt , yt ) is the 8

threshold that is the problem solution

5
& 5
min

()
& V  . (14) 0 0

-5 -5
X-coordinate (m)
For a specified target location and power, this method gives Y-coordinate (m)

the threshold that is optimal for use at all sensors in the field.
a) Optimal thresholds over a field subsection having
However, the method does not address the practical problem vertices [-9, -9], [9, -9], [-9, 9], and [9, 9].
of choosing the threshold needed to locate targets having un-
known power and unknown positions within a surveillance
region. Since the method for threshold optimization (14) de-
pends on knowledge of the parameters θ = [ P0 xt yt ]T to be
estimated, different thresholds would be required for different
target locations within a surveillance range of interest, and the
thresholds are optimal only for a specific P0 . However, the
Threshold

parameters are not known ahead of time, so the method cannot 10

be used in practice. Despite its practical limitations, the op- 8

timal threshold (14) serves as a valuable benchmark for eva- 8

luating thresholds developed using other methods [1]. 8
6
In this paper, the optimal threshold (14) is studied further. 6
In the next section, we present a geometrical interpretation of 4
4
the patterns of sensors that fire at the optimal thresholds for 2 2
demonstrative target locations. The geometrical interpretation Y-coordinate (m) X-coordinate (m)
0 0
provides insight into the threshold design problem and the
limitations of the performance of the optimal threshold (14). b) Detailed view of optimal thresholds over a smaller field subsection having
vertices [0, 0], [0, 9], [9, 0], and [9, 9].
III. GEOMETRICAL INTERPRETATION OF THE OPTIMAL
Fig. 2. Optimal thresholds computed using (14) for the corresponding target
THRESHOLDS FOR FIXED POWER AND
locations over subsections of the field of Fig. 1 (P0=10000, binary decisions,
TARGET LOCATION σ = 1).
The optimal thresholds computed using (13) and (14) are
valid only for a specified target location. In this section, we Next, we choose three demonstrative target locations for
plot the optimal thresholds over a grid of target locations with- which the optimal thresholds used by sensors to generate bi-
384 LUO AND JANNETT

nary decisions are at relatively low, medium, and high levels. range of target locations, then small changes in target location
We demonstrate the impact of the different thresholds by will not produce different firing patterns, resulting in poor
showing the patterns of sensors that fire (detect a target) for identifiability (uniqueness of solution) and possibly, biased
these thresholds at P0 =10000 (arbitrary units) in the noise estimates. If the power is significantly lower than the power
free case (σ = 0). For a target located in the center of a grid of for which the threshold is optimal (power mismatch), few or
four sensors, the target is located as far away as possible from no sensors may fire, resulting in large target location estima-
any of the four sensors, the threshold is low, and four sensors tion errors [1].
fire (Fig. 3a). For a target located an equal distance between Finally, we examine the consequences of applying subop-
two sensors, the optimal threshold is at an intermediate level timal thresholds. If the applied threshold is higher than the
and six sensors fire (Fig. 3b). For a target located close to a optimal threshold, few or no sensors may fire, resulting in
sensor, the threshold is relatively high and two sensors fire large estimation errors. If the applied threshold is lower than
(Fig. 3c). the optimal threshold, more sensors may fire than with the
optimal threshold, perhaps with little consequence. For target
30
locations near the border of the sensor field, a low threshold
Sensor Field Y-coordinate (m)

25
that results in a large pattern of sensors firing may produce an
20 asymmetric firing pattern that introduces estimation errors.
15 Further study of the log-likelihood function (9) and FIM or
10
CRLB may provide insight into the relationship between per-
formance, thresholds, and sensor firing patterns.
5

0
0 10 20 30
IV. OPTIMAL THRESHOLDS FOR A RANGE OF POWERS
Sensor Field X-coordinate (m) OVER A SURVEILLANCE REGION
a) Target location (13.5,13.5) and optimal threshold 6.89.
The minimization in (14) provides a threshold that is optim-
30
al only for a known P0 and specified target location. However,
Sensor Field Y-coordinate (m)

25 our goal is to locate targets having unknown power and un-

20 known positions within a surveillance region. In [1], the idea
15
of integrating (13) over a surveillance region and range of
powers was presented, but was considered impractical due to
10
the prospect of a prohibitive computational load. In this paper,
5 we use this idea for determining an optimal threshold for lo-
0 cate targets having unknown power and unknown positions
0 10 20 30
Sensor Field X-coordinate (m) within a surveillance region. First, we assume that the power
b) Target location (18, 13.5) and optimal threshold 9.75. is uniformly distributed within a specified interval, and we
30 assume that the target locations of interest are uniformly dis-
Sensor Field Y-coordinate (m)

25 tributed within a specified surveillance region. Next, we cal-

20
culate an average cost function by employing multiple integra-
tion of (13) over the power interval and over the target loca-
15
tions within the surveillance region. In this approach, the ele-
10 ments of the FIM are understood to be a function of the thre-
5 shold and the parameter vector θ = [ P0 xt yt ]T , and the cost
0
0 10 20 30
& & & & &
c) Target location (10, 18) and optimal threshold 11.38.
Sensor Field X-coordinate (m) &
V (η ,θ ) = ³³³
( ) ( )
J 11 η ,θ [ J 33 η ,θ + J 22 (η&,θ )] − J (η,θ ) − J (η,θ ) dx dy dP
2
13
2
12

(η,θ )
t t 0
J
Fig. 3. Demonstrative target locations (star) and corresponding patterns of
fired sensors (solid circles) that occur for the optimal thresholds found using (15)
(14) if there is no measurement noise. For a), b) and c) the lower left sensor is computed by integrating over  .
corresponds to location (0,0) in the field of Fig. 1. (P0=10000, binary deci- The optimal threshold (14) designed for a specific power
sions, σ = 0).
performs poorly for mismatched P0 values [1]. When the ac-
Inspection of the sensor firing patterns for the optimal thre- tual power is lower than the value of P0 used to calculate the
sholds (Fig. 3) facilitates assessment of the impacts of noise optimal threshold, few or no sensors fire, and the location
and power mismatch. Noise may make more or fewer sensors &
fire, resulting in asymmetric sensor firing patterns that pro- ()
variance V  computed using (13) becomes very high.
duce estimation errors. If the thresholds are the same over a Therefore, we expect that the cost (15) will be dominated by
 OPTIMAL THRESHOLD FOR LOCATING TARGETS 385

the contributions of the low target powers. In a uniform sensor veillance region. The cost (16) was evaluated over a P0 set
field, for a fixed power, the optimal thresholds (Fig. 2) found [8000 10000 12000] by exploiting symmetry to approximately
using (14) for specific target locations and the corresponding evaluate (16) for the surveillance region based on the grid of
&
()
values of V  follow a symmetric pattern within the field. 81 target positions described earlier (Table I). The minimiza-
tion (17) was carried out using exhaustive search. In consider-
An understanding of the sensitivity of the optimal threshold to ation of Table II in [1], the cost was expected to be dominated
mismatched low P0 values and recognition of the symmetry in by the contributions from the low powers within the P0 inter-
&
()
V  suggest that approximating the cost (15) as val. The cost was evaluated for P0 within the interval from
8000 to 12000 using a P0 set having only three elements be-
& & & & & cause using more elements would be expected to increase the
&
V ( ,) = ¦¦¦
( ) ( )
J 11 , [J 33 , + J 22 ( ,
&
) ] - J ( , ) - J ( , )
2
13
2
12
(16)
computational load without much improvement in the solu-
xt yt p0 J ( , ) tion. In addition, for P0 within the interval from 9000 to
11000, the cost (16) was minimized using the same method
for a P0 set [9000 10000 11000] (Table I).
may give useful results and avoid a prohibitive computational
load in finding the optimal thresholds that are the solutions to B. Evaluation of Thresholds in Monte Carlo Simulations
& The optimal thresholds were evaluated in Monte Carlo si-
( )
& V , θ .
min (17) mulations (1000 runs, binary decisions, n=2, σ = 1) in which
targets were placed at random locations uniformly distributed


over the surveillance region represented by the square having

A. Computation of the Optimal Thresholds vertices [-36,-36], [36, -36], [-36, 36], and [36, 36]. Simula-
Optimal thresholds were computed for exemplary cases that tions were initially performed for fixed P0 values of 8000,
demonstrate the use of (16) and (17). First, the optimal thre- 10000, and 12000 with the thresholds shown in Table II. Op-
shold was found for targets having a fixed power and loca- timum thresholds matching the power or power intervals were
tions within a surveillance region having vertices [-36,-36], studied. We also studied a threshold mismatch in which a
[36,-36], [-36,36], and [36,36] in the field of Fig. 1. Second, suboptimal threshold was applied. The suboptimal threshold
the optimal threshold was determined for targets that have P0 was the lowest of the optimal threshold values found for any
within a specified interval and locations within the surveil- target location within the surveillance region.
lance region. Binary decisions were employed and σ = 1 in all
TABLE II
cases. RESULTS OF MONTE CARLO SIMULATIONS EVALUATING THRESHOLDS
The optimal threshold for a fixed P0 = 8000 was determined AT DIFFERENT POWER LEVELS OVER A SURVEILLANCE REGION
for target locations within the surveillance region. The sym- Optimal Cost at Optimal Sum of Location
metry of the thresholds exhibited in Fig. 2 was exploited to Power, Applied Threshold, Estimation Error
Threshold, &
P0 Threshold Variances Using
approximately evaluate (16) for the surveillance region based  V ( ,)
Applied Threshold
on a grid of 81 target positions for x = 1,2,…9 and y =
1,2,…9. The minimization (17) was performed using exhaus- 8000 7.11 7.11 4.38 4.84

tive search. Optimal thresholds were also determined using the

10000 7.91 7.91 4.05 4.21
same method for P0 = 10000 and for P0 = 12000 for the same
surveillance region (Table I).
12000 8.66 8.66 3.85 4.18

TABLE I Random
OPTIMAL THRESHOLD AND COST VERSUS THE SET OF POWERS USED FOR - 7.481 - 4.92
(8000-12000)
OPTIMIZATION OVER A SURVEILLANCE REGION
& 10000 7.91 6.892 4.05 4.73
Power, P0 Optimal Threshold,  Cost, V ( ,)
1
8000 7.11 4.38 7.48 is the optimal threshold found for P0 set [8000 10000 12000] in (16).
2
6.89 is the lowest of the optimal threshold values found for any target loca-
10000 7.91 4.05 tion within the surveillance region (Fig. 2a).

12000 8.66 3.85

Additional simulations were performed for random P0 un-
[9000 10000 11000] 7.80 4.14 iformly distributed within the interval 8000 to 12000. The sum
[8000 10000 12000] 7.48 4.30 of the experimental location estimation error variances for
estimates of the target position (xt , yt ) was computed and
The optimal threshold for P0 within the interval from 8000 compared to the minimum cost (16) at the optimal threshold
&
to 12000 was determined for target locations within the sur-
& V , θ
min

( )
(Table II).
386 LUO AND JANNETT

V. DISCUSSION A geometrical interpretation of the optimal thresholds that

The optimal thresholds determined for fixed P0 levels of depend on knowledge of the target location and power was
8000, 1000, and 12000 increase with P0 (Table I). The optimal presented for a sensor field with sensors placed in a uniform
threshold determined for the P0 set [8000 10000 12000] was grid pattern.
less than the threshold for the set midpoint P0 = 10000, but The optimal thresholds were found through minimization
was closer to the optimal threshold for the set midpoint P0 = accomplished by approximate evaluation of (16) over a finite
10000 than to the optimal threshold for P0 = 8000. The optim- set of target locations (grid) within the surveillance region
al threshold determined for the P0 set [9000 10000 11000] considered, and over a very coarse set of power values within
was also less than the threshold for the set midpoint P0 = the power interval considered. At the expense of a greater
10000. computational effort, better approximations and use of a finer
The costs at the optimal thresholds for fixed P0 levels of grid might be employed to produce more accurate values for
8000, 10000, and 12000 decrease as the power level used for the optimal thresholds. Other future work could consider re-
optimization increases (Table I). The cost at the optimal thre- finements in the methods used in computing (15)-(17), exten-
shold determined for the P0 set [8000 10000 12000] was sion for multi-bit quantized data, detailed performance analy-
greater than the cost at the threshold for P0 = 10000, and was sis, and random sensor layouts.
closer to the cost for the optimal threshold for P0 = 8000 than
to the cost at the optimal threshold for P0 = 10000. The cost at REFERENCES
the optimal threshold determined for the P0 set [9000 10000 [1] R. X. Niu and P. K. Varshney, “Target location estimation in sensor
networks with quantized data,” IEEE Trans. Signal Processing, vol. 54,
11000] was greater than the cost at the threshold for P0 = pp. 4519-4528, Dec. 2006.
10000. These results confirm that the costs, and corresponding [2] C. Cevher, F. D. Marco, and G. B. Richard, “Distributed target localiza-
optimal thresholds, were dominated by the contributions from tion via spatial sparsity,” Proc.16th European Signal Processing Confe-
the lower P0 values. rence (EUSIPCO), Laussane, Switzerland, 2008.
[3] Y. Zou and K. Chakrabarty, “Sensor deployment and target localization
Despite the approximations and the coarseness of the repre- in distributed sensor networks,” ACM Trans. Embed. Comput. Systems,
sentation of the surveillance region and power interval used in vol. 3, issue 1, pp. 61-91, Feb. 2004.
evaluating (16) in the minimization used to find the optimal [4] K. Chakrabarty, S. S. Iyengar, H. Qi, and E. Cho, “Grid coverage for
surveillance and target location in distributed sensor networks,” IEEE
thresholds, the optimal thresholds gave good performance Trans. Computers, vol.51, no.12, pp.1448-1453, Dec. 2002.
(Table II). In simulations, the sum of the error variances [5] K. H. Yang, G. Wang, and Z.Q. Luo, "Efficient convex relaxation me-
achieved using the optimal thresholds approached the mini- thods for robust target localization by a sensor network using time dif-
ferences of arrivals," IEEE Trans. Signal Processing, vol. 57, pp. 2775-
mum cost for all P0 values considered. When a threshold 2784, July 2009.
mismatch, in which the applied threshold was not the optimal [6] E. Weinstein, "Optimal source localization and tracking from passive
threshold, was studied for P0 = 10000, the sum of the error array measurements,” IEEE Trans. Acoust., Speech, Signal Process.,
vol.30, pp. 69-76, Feb. 1982.
variances achieved (Table II, line 5) slightly exceeded the sum [7] D. Li, K. D. Wong, Y.H.Hu, and A. N. Sayeed, "Detection, classifica-
of the error variances achieved using the optimal threshold tion, and tracking of targets," IEEE Signal Process. Mag., vol.19, no. 3,
(Table II, line 2). pp. 17-29, Mar. 2002.
[8] D. Li and Y. H. Hu, "Energy based collaborative source localization
using acoustic microsensor array," EURASIP J.Appl. Signal Process.,
VI. CONCLUSION no. 4, pp.321-337, 2003.
This paper addresses the important problem of decision [9] X. Sheng and Y. H. Hu, "Maximum likelihood multiple-source localiza-
tion using acoustic energy measurements with wireless sensor net-
thresholds for target localization. In this paper, the work of [1] works," IEEE Trans. Signal Processing, vol.53, no.1, pp. 44-53, Jan.
was extended by developing and demonstrating a method for 2005.
finding the optimal threshold used to make binary sensor deci- [10] N. Patwari and A. O. Hero, "Using proximity and quantized RSS for
sensor localization in wireless Sensor Networks,” in Proc. 2nd Int. ACM
sions for targets having power and location within specified Workshop on Wireless Sensor Networks and Applications, San Diego,
intervals. Symmetry was exploited to find the optimal thre- CA, Sep. 2003, pp.20-29.
sholds using minimization performed over a subset of the sur- [11] N. Katenka, E. Levina, and G. Michailidis, “Robust target localization
from binary decisions in wireless sensor networks,” Technometrics, vol.
veillance region, which reduced the computational effort ex- 50, no. 4, pp. 448-461, Nov. 2008.
pended in finding the optimal thresholds. The optimal thre- [12] P. P. Joshi and T. C. Jannett, "Performance-guided reconfiguration of
shold resulted in target localization performance approaching wireless sensor networks that use binary data for target localization,"
the minimum cost. The new method for designing the optimal Third International Conference on Information Technology, New Gen-
erations, 2006, ITNG 2006, pp. 562-565.
threshold for targets having power and location within speci-
fied intervals could be useful in designing a sensor field to
meet target localization performance requirements.
 A Proposal of a Nonlinear Observer Applied to the Heat
Exchange Phenomena in a Double Pipe Heat Exchanger
J. F. Briones1, M.A. Paz1, J.L. Gallegos1, J.I.O. Rodriguez1, S.C. Garibo2
1
Department of Electronic Engineering, Universidad Politecnica de Aguascalientes, Aguascalientes., Mexico
2
Department of Electrical Engineering, Instituto Tecnologico de Aguascalientes, Aguscalientes, Mexico
Phone (449) 442-1400 Fax (449) 442-1419 E-mail: [email protected]

These variables must be considered as part of heat exchange

Abstract –– The aim of this paper is to show the heat equipment design criteria [1,9], and also they mean
exchange phenomena applied in a double pipe heat exchanger uncertainties that can be estimated by assumptions at the
pilot plant scale, by considering a macroscopic dynamic model, moment when dynamic behavior of the equipment is
that describes the evolution of the temperature of hot and cold
fluids in the equipment. The proposed dynamic model is
formulated, together with other kind of variables, associated
function of mean effective temperature difference and of the with the use of the equipment, like the fouling factor that
thermodynamic and transport properties, which are of strictly decreases the efficiency of the equipment. The way this
uncertain nature. These uncertainties can be isolated with the factor begins to form is uncertain, also the grade of
use of mathematical tools of differential geometry, based on Lie affectation to the equipment through the time.
algebra and nonlinear dynamic observer design that uses only
the information related with mass fluid flows and the mass Regarding the dynamic model used to describe the states
contained in the equipment. The numerical simulations (temperatures) of the equipment, this model has a high
represent the heat exchanger physical behavior, and these nonlinear nature, because it involves different theoretical-
results can be used to formulate robust control strategies
physically feasible.
experimental correlations used to calculate certain indicators
such as: (i) the overall heat transfer coefficient; (ii) the mean
Keywords –– Nonlinear observer, differential geometry, effective temperature difference; (iii) and the heat capacity
double pipe heat exchanger. of the fluids used in the equipment.

I. INTRODUCTION Finally, it is necessary to mention that other proposals

related with the design of observers applied to this kind of
Because of their numerous applications in industrial equipment have been advance (see [5,6,7]). In those cases,
processes, heat exchangers have been the subject of many the proposed observers use the mean effective temperature
studies including, among others: steady-state, transient, and difference and overall heat transfer coefficient, and in this
frequency response analysis; open-loop qualitative behavior work we will present a proposal of a nonlinear observer, that
characterization; fault diagnosis/detection; numerical can describe, with a minimal margin of error, the dynamic
simulation; feedback control; parameter identification; and behavior of the equipment temperatures, only by using the
state reconstruction. Among these topics, the two lastly information reflected by the flows of fluid and by the mass
mentioned have played an important role in the formulation contained inside the equipment; and without using the mean
of solution that help cope with the operation conditions effective temperature difference nor the overall heat
imposed to current industrial processes [5]. transfer; indeed, the designed nonlinear observer can
estimate these parameters as uncertainties that can be
Particularly, the heat transfer phenomena involve different reconstructed by the measuring of temperatures, since these
thermodynamic properties like: thermal dispersion parameters have a strong dependence on the temperature, as
coefficients by conduction and convention (e.g. film has been reported in [8].
coefficients) and heat capacities associated with the fluids
contained in the heat exchanger equipment, also the II. DESIGN OF THE DYNAMIC MODEL FOR THE
transport properties that affect the heat conduction by DESCRIPTION OF HEAT TRANSFER.
convection, specifically the boundary layer and viscosity of
the fluids [8]. The dynamic model presented in this paper is
formulated from a macroscopic heat transfer balance that
Other variables should be considered in the design of the involves the parameters described in the past section.
equipment, such as: equipment geometry and material used
in the equipment production; heat transfer regime The proposed dynamic model is validated with the
(concurrent and countercurrent fluid flows); and the spatial experimental information obtained from a double pipe heat
configuration (e.g. double pipe or tube and shell heat exchanger pilot plant with concentric pipes [2]:
exchanger).

387
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_68, © Springer Science+Business Media B.V. 2010
388 BRIONES ET AL.

2 § AU ·
x1 = ¨ F1 ( x1e − x1 ) − i i Δx ¸¸ I/P TC
M 1 ¨© Cp1 ¹ (1)
hot fluid

2 § AU ·
x 2 = ¨¨ F2 ( x 2 e − x 2 ) + o o Δx ¸¸
M2 © Cp 2 ¹ controlled flow
TT

Where x∈ℜ2 is a vector space that represents the

cold fluid heated liquid

temperatures in the heat exchanger; M is the mass contained Heat Exchanger

in the equipment (i.e. the mass distributed in the equipment to process TE
that can be calculated from the density relation -considered Process Variable:
known- using the volume as the space that contains the heat hot fluid

or cold fluid in the inner tube or the annular space); F is the

temperature
Fig. 1. Instrumentation diagram of the double pipe heat exchanger pilot
mass fluid flow; A is the heat transfer area; U is the overall plant
heat transfer coefficient; Cp is the heat capacity of the fluid;
and Δx is the mean effective temperature difference, that IV. PROPOSED NONLINEAR OBSERVER.
defines in the system (1) the heat transfer regime
(concurrent and countercurrent fluid flows). This design of a nonlinear observer uses the next
proposal affine system:
The subscript 1 represents the heat fluid; 2 represents the
cold fluid; e represent the conditions of both fluids at the x = f ( x ) + g ( x )u
entrance of the equipment; i and o represents the reference (2)
point in the equipment for the calculus of the property (i is y = h( x)
the inner tube surface and o is the outer surface of the inner
pipe). Where y=h(x)=[x1,x2]∈ℜ2 is the set of temperatures to be
operated in the heat transfer, and u has been chosen as the
III. DOUBLE PIPE HEAT EXCHANGER PILOT PLANT. mean effective temperature difference (Δx), because this
term commonly is estimated as a logarithmic or arithmetic
The double pipe heat exchanger pilot plant contains an mean, but strictly this mean is uncertain inside the
inner copper pipe and an outer steel pipe that has to be equipment and it is the main one among the others, like the
isolated, in order to avoid heat losses in to the environment. heat capacity and the overall heat transfer. The vector maps
f(x) and g(x) are defined by the system (1) as:
The cooper pipe has an inner diameter of 11mm and an
outer diameter of 12mm; the outer pipe has an inner 2
( x1e − x1 )
diameter of 26mm. f1 ( x) M1
f ( x) = =
f 2 ( x) 2
The effective longitude for heat transfer in the equipment is ( x2e − x2 ) (3)
M2
7.62m. The time necessary to reach the equilibrium between
the temperatures is 5 to 30 seconds. − 2 AiU i
g1 ( x) M 1Cp1
g ( x) = =
The equipment has a set of valves that allows controlling the g 2 ( x) 2 AoU o
flow of fluids and that predisposes the regime with which M 2 Cp 2
the heat transfer will be used in the heat exchanger
(concurrent and countercurrent). If y=h(x)=[x1,x2], it is easier to show that the mean
difference temperature (independently from the heat transfer
This set of valves determines the way in which the fluids are regime) is a function of y, and it allows for the uncertainty to
going to flow: hot water circulating inside the inner cooper be estimated, but before, it is necessary to establish the
pipe and cold water circulating in the annular space between mathematical relationship between y and u.
the pipes.
Using Lie’s algebra and the definition of the relative degree
The temperatures and the flows of both fluids are registered (ρ) [3], considering this degree the minimal whole number
at the entrance and at the exit of the equipment to calculate: of derivates applied with Lie’s algebra necessary to establish
(i) thermodynamic and transport properties of the fluids; (ii) the mathematical relationship between y and the vector
and the characteristic parameters related with the heat maps f(x) and g(x), that is, LgLf kh(x)=0 and LgLfρ-1h(x)0,
exchanger and the equipment (see Figure 1). where k<(ρ-1), this combination must be validated for all
x∈ℜ 2.
 A PROPOSAL OF A NONLINEAR OBSERVER 389

dxˆ 2F
The L operator is declared by [3]: 1= 1 ( x1e − xˆ1 ) +
dt M
1
n ∂h( x )
f ( x); L h ( x ) = L §¨ L h( x ) ·¸( x )
k k −1
L h( x) = ¦ f© f M ( 2 F k ( x − xˆ ) + M k ( x − xˆ ))
f
i = 1 ∂xi
i f ¹ 1 2 1 1 1 2 2 2 2
n ∂ § n ∂h( x )
¨ ¦
· 2 ( F M − F (t ) M )
L L h( x) = ¦ g ( x ) ¸ f ( x) 2 1 1 2
f g ¨
j = 1 ∂x j © i = 1 ∂xi
i ¸ i
¹ (5)
dxˆ 2F
2= 2 ( x2e − x2 ) +
Thus the relative degree of the system (1) is equal to 2, and dt M
due to its coincidence with the number of states in the 2
system (1), then the system (1) has a complete relative M ( 2 F k ( x − xˆ ) + M k ( x − xˆ ))
2 1 1 1 1 1 2 2 2
degree and this guaranties that y can be used to describe the
uncertain term and that there are no unmodeled dynamics. 2( F M − F M )
1 2 2 1

The consideration of a complete relative degree for the V. ANALYSIS OF THE DYNAMIC SYSTEM.
system (1) guaranties the existence of a stable nonlinear
observer, that will converge in its dynamic behavior with the The analysis consists of the comparison of the
system (1) when time has an infinite limit [4]: numerical simulation between the dynamic behavior of the
systems (1) and (5) in concurrent and countercurrent fluid
flows.
dxˆ −1
= f ( xˆ ) + g ( xˆ )u + [Q ( xˆ )] K [ y − h( xˆ )]
dt (4) In the system (5), the observation gains locate the
y = h( x) characteristic polynomial poles in -1, where this
characteristic polynomial is defined by the error system
formed by the difference between the systems (5) and (1),
Where the superscript ˆ denotes the observed states and
and it allows for the dynamic behavior of the system (5) to
K=[k1,k2]T∈ℜ2 is the set of observer gains with magnitude tend to the dynamic behavior of system (1), when the time
and sign such as the states from system (4) tend to the states tends to infinity.
from the system (1) when time tends to infinity and the
Jacobian matrix Q(x) is defined by: Apart from what has been mentioned before, it is necessary
to consider only known nominal values of the parameters
h( x) involved in the systems (1) and (5), and they are calculated
L h( x) at the heat exchange initial conditions in the equipment.
f These values are defined in Table 1.
Q( x) =

TABLE I
Ln − 1 ( x ) NOMINAL VALUES USED IN THE NUMERCIAL SIMULATIONS IN
f
CONCURRENT AND COUNTERCURRENT HEAT TRANSFER
Parameters Heat exchanger regime
However, with the use of the vector maps f(x) and g(x) given Concurrent Countercurrent
by the system (3) onto the system (4), besides the Jacobian Ui [=] kcal/m2 ºC hr 115.033 100.244
matrix above defined in terms of the Lie algebra and Uo[=] kcal/ m2 ºC hr 103.286 89.653
Ai[=]m2 0.2736
executing the rules associated to this kind of algebra, the Ao[=]m2 0.3040
nonlinear observer is: F1[=]kg/hr 11.279
M1[=]kg 0.767
M2[=]kg 3.258
Cp1[=]kcal/kg ºC 0.999
Cp2[=]kcal/kg ºC 0.997
x1e[=]ºC 90
x2e[=]ºC 29
390 BRIONES ET AL.

300 (a
90 (a
Nominal System 240
80 mean temperature difference
Observer System 180
70 calculated by the observer in the inner tube
120
60 60
Temperature (°C)

50 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0

Temperature (°C)
40
300
30
240
(b
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 180 mean temperature difference
50 calculated by the observer in the annular space
120
(b
45 60
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
40
Nominal System 60
35 (c
Observer System 50
30 40 mean temperature
30 difference calculated by formula
25
20
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
Time (hr) 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
Time (hr)
Fig. 2. Numerical simulation of heat exchange in countercurrent regime. a) Fig. 5. Numerical estimation of mean temperature difference. a) reference
Hot fluid temperature. b) Cold fluid temperature. by observation in the inner tube. b) reference by observation in the annular
space. c) definition of logarithmic mean temperature difference.
180 (a
150 mean temperature difference
120 calculated by the observer in the inner tube Figure 2 and Figure 4 show that system (1) and (5) have an
90
60 equal dynamic behavior in terms of temperature evolution
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 over time, despite the fact that system (5) uses only mass
and flow data from the equipment, meanwhile system (1)
Temperature (°C)

180 (b
150 mean temperature difference needs more information, particularly the mean effective
120
90
calculated by the observer in the annular space
temperature difference and the overall heat transfer
60
coefficient.
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0

60
(c However, the nonlinear observer (5) has a better description
50
40 mean temperature of hot temperature than cold temperature regardless of the
30
20
difference calculated by formula
heat exchange regime, because the difference between the
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
real and the estimated hot temperature is practically zero,
Time (hr) meanwhile the difference between the real and estimated
Fig. 3. Numerical estimation of mean temperature difference. a) reference cold temperature is 1°C.
by observation in the inner tube. b) reference by observation in the annular
space. c) definition of logarithmic mean temperature difference.
Thus the nonlinear observer (5) is a promissory dynamic
90
Nominal System
(a model in terms of describing the hot temperature in the
80
70
Observer System double pipe heat exchanger using a minimum of
60 information, and in physical applications, the hot
50 temperature has more importance than the cold temperature,
40
because the first one is commonly the variable that is
Temperature (°C)

30
intended to control in a process.
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
50
(b
45 Figure 3 and Figure 5 show the dynamic evolution of the
40 mean temperature difference, this way, the observer system
35
Nominal System describes the mean temperature difference in the equipment,
Observer System
30
regardless of heat exchanger regime, and this information is
25
complementary to the measure of the maximum quantity of
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 heat that the fluids can exchange.
Time (hr)

Fig. 4. Numerical simulation of heat exchange in concurrent regime. a) Hot This is an extra value, because this kind of nonlinear
fluid temperature. b) Cold fluid temperature. observer allows user to isolate the principal uncertainties in
the heat transfer process (the mean effective temperature
difference and the overall heat transfer coefficient), and
eventually describes the properties of the equipment.
 A PROPOSAL OF A NONLINEAR OBSERVER 391

So, even if there is a logarithmic mean effective temperature Even more, this nonlinear observer allows the user to
difference (Figure 5c), this parameter only describes the estimate the temperature inside the equipment and uses this
capacity of temperature that can be transferred between estimation for other applications besides control process
fluids (approximately 20°C), but the estimation of this (e.g. improve de the design of double pipe heat exchanger),
parameter obtained from the nonlinear observer, describes and thus, it has with wide possibilities of being physically
the internal temperature of the equipment (about 80°C), and implemented.
this information is more useful if one of the process
objectives is controlling the inner temperature in the REFERENCES
equipment, to avoid the formation of fouling in the inner
tube and the annular space and this measure is very [1] Yunus A. Çengel, Heat and Mass Transfer – A Practical
Approach. Singapore, McGraw-Hill, 2006, pp.619-626.
important, because it is much too difficult to install a set of [2] Y.S.N. Malleswararao and M. Chidambaram, “Nonlinear
temperature sensors inside the equipment to obtain this controllers for a heat exchanger”, J. Process Contr.Vol. 1, No.2,
information, which can be estimated by the proposed pp. 17–21, 1992.
nonlinear observer. [3] Alberto Isidori, Nonlinear control systems: An introduction,
Germany, Springer, 1989, pp. 145-150
[4] H. Murata et. al., “Control of inverted pendulum-wheeled cart
VI. CONCLUSIONS system using nonlinear observers”, SICE’95 Proceedings of the
34th SICE Annual Conference. International Session Papers, Vol.
The proposed nonlinear observer is a promissory dynamic 26-28, pp. 1133-1136, July 1995.
[5] A. Zavala-Rio et. al., “Bounded positive control for double-pipe
model to be used in the synthesis of advance control heat exchangers”, Control Engineering Practice, Vol. 17, No. 1,
strategies, in which the only variables known are flows and pp.136-145, January 2009.
masses in the equipment. [6] C.-M. Astorga-Zaragoza et. al., “Performance monitoring of heat
exchangers via adaptive observers”, Measurement, Vol. 40, No.
4, pp. 392-405, May 2007.
While there are other observers previously proposed [7] C.-M. Astorga-Zaragoza et. al., “Observer-based monitoring of
([5,6,7]), they use two critical parameters (the mean heat exchangers”, ISA Transactions, Vol. 47, No. 1, pp. 15-24,
effective temperature difference and the overall heat transfer January 2008.
coefficient), that can be estimated from experimental [8] Prabhata K. Swamee et. al. “Optimum design of double pipe heat
exchanger”, International Journal of Heat and Mass Transfer ,
correlations with some margin of error. The nonlinear Vol. 51, pp. 2260–2266, 2008.
observer hereby proposed does not use them and it even [9] T.W: Fraser Russell, et. al., Mass and Heat Transfer – Analysis
estimates them with a minimal margin of error than the of mass contactors and heat exchangers, United States of
experimental correlations estimate. America, Cambridge University Press, 2008, pp. 90-102.
 Vibration Signal Processing
by Deconvolution Method
Popescu Ion Marian
University Constantin Brâncu i Târgu-Jiu, Romania, [email protected]
Runceanu Adrian
University Constantin Brâncu i Târgu-Jiu, Romania, [email protected]

Abstract. This paper presents a method of processing vibration

signal provided by a piezoelectric sensor. The main purpose is to
identify the actual vibration signal of a mechanical structure, and
development of estimation methods with a certain precision for
the real signal provided by sensor signal processing, using the
deconvolution method. Basically, knowing the sensor output
signal, knowing mathematical model of the sensor (with a certain
precision of calculation) we are trying to find the input signal
sensor, which is a real vibration signal.

I. INTRODUCTION

A very important method for predictive detection is the

measurement of defects in the mechanical vibrations that occur
in these, relatively high accuracy. Fig.1.Experimental platform for vibration acquisitions
By analyzing the acquired signals can develop efficient
Acquisition system for signal processing from vibration
algorithms to detect a possible future failure even before they
sensors is designed around a core with microcontroller, that has
appear. To develop such a system was initially developed in an
an acquisition speed to around 400,000 Samples/second and
experimental model consisting of a DC motor which rotates a
the local store of a number of samples corresponding to an
shaft with two bearings, one new and one with visible defects.
observation period of 2 ÷ 3 seconds.
Structure experimental model was developed is shown in Fig.1.
The structure of the acquisition system is shown in Fig.2.
Vibrations are measured using piezoelectric sensors {4}
whose output signal is applied to a signal conditioning circuit
{3} to be then converted into digital signal using an acquisition
system {2}.
Digital signal is saved and sent to a computer system {1} for
processing and obtaining of information concerning the status
of the system under observation.
In the stand the bearings under observation are connected on
the same shaft that is rotated by a DC motor with speed up to
1500rpm, motor which is controlled by an electronic module.
In this application, was designed and developed a high-speed
acquisition system to acquire a large number of samples, store
them in the local memory, then to transmit the communication
channel to a computer system at a much lower speed compared
to the speed of acquisition.
This system allows off-line processing of the signal acquired
by the computer system (1) but has the advantage to make a
distributed acquisition system, that is placed near the
mechanical system, then the remote can transmit digital data Fig.2.The structure of the acquisition system of vibration
through a standard communication channel.
The acquisition system includes a 12-bit analog/digital
converter, a 2MØ static RAM, a standard serial port RS232

393
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_69, © Springer Science+Business Media B.V. 2010
394 MARIAN AND ADRIAN

that is the interface for programming and communications over or

short distances, a serial interface RS485 for long distances. .. . ..
To qualify for the rate of acquisition was used for DS89C420 m x r + c x r + kx r = − m x1 (2)
microcontroller core, which offers good performance and is Using these notations and equation solving earlier[4], obtain
compatible with standard microcontrollers family Intel8051. the relationship:
This microcontroller has an improved structure and execute 2
§ω ·
almost any instruction from the instruction set of ¨¨ ¸¸
microcontroller Intel8051 up to twelve times faster at same Xr
= © p¹ (3)
clock frequency of the system. X1 2
ª § ω ·2 º § ω·
2
«1 − ¨¨ ¸¸ » + ¨¨ 2ξ ¸¸
II. MODELING THE VIBRATION TRANSDUCER « © p¹ » © p¹
¬ ¼
ω
Overall construction vibration transducer included 2 2ξ
elements: a mechanical element, which follows the motion of p
θ = arctg 2
(4)
the vibrating structure and a sensor, which converts relative §ω ·
motion into an electrical signal output. 1 − ¨¨ ¸¸
Vibration transducers type seismic devices consist of a mass- © p¹
spring-damper system, connected by a fixed support structure k c
where: p = , ξ= , ccr = 2 km
of the vibration and measuring the relative movement of mass m ccr
to support.
Relationship (3) is simulated with the software package Matlab
Basics elements of a seismic sensor are shown in Fig.3. 7.1 in Fig.4. In this figure both curves tend to value Xr/X1=1
to frequencies ‚>>p. In part III of graph X r ≅ X 1 , so the
xr=Xrsin(‚t-) relative motion between the seismic mass and support is the
same movement support.
In Fig.5. will result in this domain, for a weak damping,
S phase angle is θ ≅ 1800 , so the mass m and support vibrating
in opposition . Mass m is a fixed point to which the support
device has a harmonic motion equal to that structure. If the
transducer is moving, the signal is proportional to the
displacement of the structure studied, and is called vibrometer
x1=X1sin(‚t) seismic transducer.

Fig.3. The principle of a vibration transducer

Support device S is connected rigidly to the structure of
mechanical vibration. Mass m is attached to transducer support
through spring k and damper c. We will consider that the
support device executes a harmonic motion:
x1=X1sin(‚t)
Relative movement of the mass to the support is the response
of the transducer:
xr=Xrsin(‚t-)
Mass displacement relative to a fixed reference point in space
.. .. ..
is: x=x1+xr and the corresponding acceleration is x = x1 + x r . Fig.4. Amplitude-frequency characteristic
Applied forces acting on the mass of spring and damper, Because the natural frequency of the "mass-elastic suspension"
.
must be much smaller than the measured vibration frequency,
proportional with xr-movement and respectively x r -speed, displacement sensors have a "soft" suspension. This makes
which are elastic force that viscous damping force. the relative movement of the seismic mass is greater at lower
The equation of motion of mass m is therefore: frequencies. Consequently, displacement sensors have
§ .. .. · . relatively large dimensions and weight, to be taken into
m¨¨ x1 + x r ¸¸ + c x r + kxr = 0 (1) account when measuring small mechanical structures, whose
© ¹ response can be so greatly influenced.
 VIBRATION SIGNAL PROCESSING BY DECONVOLUTION METHOD 395

Due to mechanical solicitation, the internal electrical

echilibrium of the piezoelectric materials is disturbed, and to
restore him, on the 2 sides opposite is deposited electric
charges.
It produces a variable voltage (dependent electrical capacity
of the system), proportional to the force, so the acceleration of
mass, which can be measured.
1- elastic tensioning
element
2- seismic mass
3- piezoelectric element
Fig.5.Phase-frequency characteristic 4- the sensor base
5- Electrical connections
To frequencies ω << p (interval I of Fig.4.) relationship (3) 6- case
2
becomes:
Xr §ω ·
≅¨ ¸
X 1 ¨© p ¸¹
1
so that we have X r = 2 ω 2 X 1
p
( ) Fig.6.Piezoelectric vibration transducer
For realizing the conditions for the experimental model was
where the expression in brackets is the amplitude of the
chosen MAQ36 accelerometer, because has its own natural
absolute acceleration of the box.
frequency of such large and useful field measurement of the
Movement of the seismic mass is in phase with that of the vibration frequency is higher.
support. If one neglects damping, spring deflection (i.e. the MAQ36 produced by SENSOTEC company, is a miniature
sensor response) is proportional to the force of inertia of the accelerometer, the output signal is in charge, designed for use
mass and proportional to the acceleration of mass. Results in in industrial applications and automation, including
this operating mode, output signal of a seismic displacement experimental models that require little mounting space and
transducer is proportional with acceleration structure studied high natural frequency. MAQ36 is a self generating
(p2 is a constant device). We can say the device works as a piezoelectric transducer, which has no internal electronics and
seismic accelerometer. If the transducer is the speed at the requires no external power for operation.
same operating conditions, device measures the acceleration of
The accelerometer should be connected to an charge
the order 2 (“jerk”) of the structure. Unlike displacement
amplifier, placed as near as possible to the transducer in the
sensor, seismic accelerometer has a relatively strong spring, so
measuring chain. MAQ36 is an compression accelerometer,
a relatively high natural frequency and a total weight much
which due to small size ensures a high own natural frequency.
smaller than the vibrometers. To frequencies ω ≅ p , seismic
Construction of its dimensions are shown in Fig.7.
device working in resonance zone II, the indications of this
become very large and the phenomenon is used only to
measure the frequency.
Generaly, vibratory motion of the structure studied is not
harmonic, as assumed above. When measuring a not harmonic
movement, the sensor response to different spectral
components is different.
As I shown, the presence of damping, amplification Xr/X1
ω
and phase angle , varies with the report , so that the device
p Fig.7.Accelerometer MAQ36
changes the report of amplitude and phase between different
harmonics, making the resultant movement is distorted by the The main features of the accelerometer (to the catalog) are:
real movement.
- Sensitivity 5pC/g ±10%
As a result, seismic devices with high damping give rise to
distortions when measuring not harmonic vibration. - The frequency of work 1Hz until 30KHz
- Natural-frequency assembly 45KHz
The most commonly used accelerometers used as
- Dynamic range ±2000g
piezoelectric transducers.
- Sensitivity to temperature 0.08/›C
Principle scheme of a seismic sensor, with piezoelectric - Transverse sensitivity < 5%
crystal used as an elastic element, is shown in fig.6. - Amplitude linearity better than 1%
Due to high rigidity of crystals, natural frequency of the - The temperature range -90›C +900›C
sensor is high, so it can be used only as an accelerometer. When - Capacity 300pF
attached to a body that vibrate with the acceleration a, seismic - Weight 10 grams
mass m acting with a force F = ma on the piezoelectric disc. - Force Mounting 9N
396 MARIAN AND ADRIAN

III. PROCESSING THE ACQUIRED VIBRATION SIGNAL As you can see, most of the vibration signal energy is at
frequencies above 1 kHz or even 10kHz.
Using data acquisition system [1] described above and The power spectrum for the “good” bearing is 20 times
generate vibration platform were obtained the following signals lower than the power spectrum for the “spent” bearing. As the
acquired for a bearing "good" and a bearing “spent" as in Fig.8. motor that rotates the mechanism of the 2 bearings that are
fitted vibration sensors have 1500rpm, i.e. around 25Hz, it can
be concluded that in this signal is included vibration and
effective response vibration sensor which has a important
weighting.

Fig.8.Vibration signals (acceleration) acquired

The acquisition was done for 1 second time and have
obtained 400,000 samples. Signals acquired for the 2 bearings Fig.10. Power spectrum of acquired signals
that made a zoom for 1000 samples is presented in Fig.9.
Basically, the response of the transducer xr(t) is obtained as
the convolution product (in case of a linear and time invariant
system) between signal x1(t) and weighting function h(t) of the
vibration transducer, where x1(t) is the real vibration signal.
xr (t ) = x1 (t ) * h (t ) (5)
∞
or: x r (t ) = ³ x1 (τ )h(t − τ )dτ (6)
−∞
for causal x1(t) and h(t), the convolution integral limits can
customize as follows:
t

³
x r (t ) = x1 (τ )h(t − τ )dτ (7)
0
Returning to equation (2), we have:
.. . ..
m x r (t ) + c x r (t ) + kx r (t ) = −m x1 (t ) (8)
Starting from the premise that mechanical vibration of the
structure is small displacement around a position of
Fig.9. Vibration signals for the 2 bearings consist equilibrium which is also a stationary point for the system,
1000 samples approximating the vibration transducer as a linear system is
Processing was realized using the development environment valid.
LabWindows/CVI 8.1 and processing was done for the So, we can apply the Laplace transform and we have zero
acquired signal from the charge amplifier, that is connected initial conditions:
accelerometer. The signal is a signal of acceleration (vibration) ms 2 X r ( s ) + csX r ( s ) + kX r ( s ) = −ms 2 X 1 ( s ) (9)
of the mechanical system.
Mathematical model (transfer function) of the vibration
We made for these 2 signals the power spectrum (using
transducer is:
Fourier transform) and obtain the processing of Fig.10.
 VIBRATION SIGNAL PROCESSING BY DECONVOLUTION METHOD 397

X r ( s) − ms 2
H ( s) = = 2 (10) After processing relationship (12), we have:
X 1 ( s ) ms + cs + k
X ( z)
Using the Tustin substitution: H ( z) = r = (13)
X1( z)
H ( z) = H ( s) 2 z −1 (11) − 4m + 8mz −1 − 4mz − 2
=
s=
T z +1 (4m + 2Tc + kT ) (
2
) (
+ − 8m + 2kT 2 z −1 + 4m − 2Tc + kT 2 z − 2)
4 (z − 1)2 Further we note this:
−m b1=-4m; b2=8m; b3=-4m;
T 2 (z + 1)2
Ÿ H ( z) = (12) a1=4m+2Tc+kT2; a2=-8m+2kT2; a3=4m-2Tc+kT2
4 (z − 1)2 2 (z − 1) We obtain discrete transfer function:
m +c +k
T 2 (z + 1)2 T (z + 1)
b + b z −1 + b z −2 X ( z)
H ( z ) = 1 2 −1 3 − 2 = r (14)
a1 + a 2 z + a3 z X 1( z)
Equation (14) can be implementated by considering the
delay operator z-1:
a1 x rk + a 2 x rk −1 + a3 x rk − 2 = b1 x1k + b2 x1k −1 + b3 x1k − 2 (15)
where:
xrk- is the signal of the output vibration transducer
(transducer response)
x1k- is the signal of the input vibration transducer
(real vibration of bearing)
Real vibration signal for the mechanical system after this
processing is:
b b a a a
x1k = − 2 x1k −1 − 3 x1k _ 2 + 1 x rk + 2 x rk −1 + 3 x rk − 2 (16)
b1 b1 b1 b1 b1
Because the transducer transfer function from relation (10) is
a proper function, can be implemented numerically as
relationship (16), we obtain the signal of input (real vibration
of mechanical structure) of the transducer, knowing output
signal of transducer and the mathematical model of the
transducer. In this way, is made a digital deconvolution, which
is valid given that the transfer function is proper.
In fig.11. first 2 graphs are the response of the transducer to
impulse mechanically generated and the next 2 graphs
represent the response of the mathematical model simulated.
Power spectrum obtained for the new signal x1 is presented
in fig.12. for the 2 bearings. As you can see, most of the signal
energy is in the low frequencies.
Although the 2 vibration signals are acquired on different
channels from acquisition board and from different bearings,
can be observed a correlation of spectral components
(highlighted by green dotted lines) in relation to frequency and
amplitude compared with the notes that the bearing used (2nd
graph) has some component of 200 times the bearing
something better (first graph).
This method, with digital deconvolution, introduces large
errors primarily due to initial conditions is not identical with
real conditions.
In the graphs presented, after the processing by the relation
Fig.11. Rectangular impulse response for the real system and (16) vibration signal, has eliminated a total of 100,000 samples
simulated system from 400,000 before taking a Fourier transform. By this was
removed from the calculated signal variation due to the initial
conditions.
398 MARIAN AND ADRIAN

Another error that affects processing is not known with structure "good", and this process taking as reference for
precision mathematical model of the vibration transducer and comparison to other similar structures in operation.
in particular the coefficients c and k which will have
REFERENCES
corresponding ‚n i ‡ . In this case it was considered
m=0.002; ‡=0.1; ‚n =45000;
[1] Grofu F., Popescu L. and Cercel C. (2008)
“Acquisition System for Monitoring Vibrations”,
International Joint Conferences on Computer,
Information, and Systems Sciences, and Engineering
(CISSE 2007) Conference, Conference Proceedings
book, December 3-12, 2007, University of Brigeport,
USA, publish in Novel Algorithms and Techniques In
Telecommunications, Automation and Industrial
Electronics, ISBN 978-1-4020-8736-3 (Print) 978-1-
4020-8737-0 (Online), pg. 89-92, DOI 10.1007/978-1-
4020-8737-0_16.
[2] Shin K., Hammond J., “Fundamentals of Signal
Processing for Sound and Vibration Engeneers”, John
Wiley&Soons Ltd., 2008, ISBN 978-0-470-51188-6
[3] R.Keyth Mobley,“An Introduction to Predictive
Maintenance”, 2nd Edition, 2002, Elsevier Ltd.,
ISBN: 978-0-7506-7531-4
[4] Mateescu A.,Dumitriu N., Stanciu L. “Semnale si
sisteme-Aplicatii in filtrarea semnalelor”, Ed.Teora,
2001, ISBN:973-20-0666-8
[5] Buzdugan G., Fetcu L., Rades M. “Vibratiile
sistemelor mecanice”, Ed. Academiei Romane
Bucuresti, 1975.
Fig.12. Power spectrum obtained after deconvolution [6] Moheimani, S.O. Reza, Fleming, Andrew J.,
“Piezoelectric Transducers for Vibration Control and
Damping”, Series: Advances in Industrial Control,
IV.CONCLUSIONS 2006, Springer-Verlag, ISBN: 978-1-84628-331-4
[7] Harris C.M.; Piersol, A.G. “Harris' Shock and
In an automation system, digital signal processing of Vibration Handbook”, (5th Edition), 2002, McGraw-
vibration is a very difficult problem which must take into Hill, ISBN 0-07-137081-1.
account the following aspects: [8] Douglas Thorby “Structural Dynamics and Vibration
> Acquisition system must meet the powerful features of real in Practice”, 2008, Elsevier Ltd., ISBN: 978-0-7506-
time, because the vibration signal is a very dynamic signal. 8002-8.
> Another problem is strongly nonlinear nature of a vibration [9] Dan Necsulescu, “Advanced mechatronics:
signal, which complicates its processing. Most mechanical monitoring and control of spatially distributed
systems that generate vibrations have imperfections in their systems”, 2008, World Scientific, ISBN 9812771816,
structures, to which adds that the vibration in a mechanical 9789812771810
structure have more degrees of freedom and transmission [10] www.leonovabyspm.com, Leonova Infinity-
through the structure. Instruction Manual, 2008
> Vibration transducers have limited responses and introduce [11] http://www.ni.com/soundandvibration/, Sound and
distortions of amplitude and phase signal provided. vibration systems, 2009.
> Any acquisition system (including transducer) and signal [12] Honeywell Sensotec, Accelerometers Frequently
processing of vibration require laborious calibration platforms Asked Questions, www.sensotec.com, 2003
specially designed for a specific type of application. [13] Maxim-IC, App Note 604: Fast Memory Transfers
> Structures analysis and on-line processing of vibration with the Ultra High-Speed Microcontroller -
signals are very difficult to achieve because in a mechanical DS89C420,
structure, the point where vibration measurements is made, is [14] I.J.Taylor. ”Identification of Bearing Defects by
far from processing system, and networks cable introduce Spectral Analysis”. Journal of Mechanical Design,
additional distortions Transaction of ASME, vol. 120, 1980, pp 199-204.
As a conclusion, the most applications for acquisition and [15] Vibration Measurement and Analysis
processing of vibration in purpose of predictive maintenance of Bruel & Kjaer Handbook BA 7676-12, 1998
a mechanical structures, processing the signal for a mechanical
 An IT Change Management Tool to Optimize
Scheduling and Planning
Denílson Cursino Oliveira, Raimir Holanda Filho
University of Fortaleza- UNIFOR
[email protected], [email protected]

Abstract – The IT governance has become strategic to dependencies and the rules to build an optimized schedule
business objectives. The main reason is the increase of of changes based in one change window. The main
importance that IT has in business decision. To achieve this feature of this model is the possibility of building a
importance, the IT management must ensure a Just-in-time parallel implementation of changes using a simpler
feedback, electing the change management as main point of IT
to help the business goal through the use of tools with optimized
heuristic with same complex set of changes if compared
procedures to guarantee a lower financial loss during the with the state-of-art.
implementation of changes. This work presents a tool to In [4] was added some resources to change
minimize the impact of the changes on the business goals, management tool: more dependency treatment as the
respecting the dependencies between the changes and featuring principle cross dependency; the allocation of changes in
innovations as dynamic change window determination. The tool different change windows; the automated determination
presented in the work builds an optimized determination of of time implementation of each family of changes,
change windows, treatment of dependencies and allocation of enabling the determination of the financial loss.
changes, helping the business to minimize the impact. In [5] is presented an optimization of IT changes,
showing mathematical concepts to determine the financial
I. INTRODUCTION AND MOTIVATION loss, cost of changes implementation over a set of
The actual IT governance aligns the real goals of changes attached to a change window. One change
the business with an automated response information management conceptual model of a decision support tool
technology to achieve strategic goals with more is presented. The implementation of changes is performed
efficiency and less time. To formalize the IT governance sequentially. On the other hand, our model does this in
globally, some set of procedures and good practice have parallel.
been developed. Control Objectives for Information and The CHAMPS [6] focuses on optimization of a
related Technology (COBIT) [1] is one of them that scheduling of changes to satisfy external RFC´s.
promotes a management and control of information with CHAMPS develops a complex dependence hierarchy that
use of measures, performance indicators and maturity is used to build a final scheduling, where it reflects a least
models. Other example is the IT information library financial impact during the changes. Our tool has a
(ITIL) [2], which consists of a set of best practices that simpler but powerful heuristic for the construction of the
determine how the IT infrastructure of a company is dependences between changes and their treatments. Also
organized. The library ITIL is an IT service management it builds an optimized scheduling based on financial or
framework that promotes an efficient management of IT timeline variable.
changes in order to minimize any impact upon IT The results obtained in previous work [3][4]
services. showed that formalize some treatments over the change
Some automated IT tools were developed, using dependence and a simpler heuristic to build an optimized
the good practice and procedures of ITIL and COBIT, allocation of changes, now we present more features to
aligning IT and business strategy, measuring the IT the tool and a conceptual model.
performance and building IT feedbacks to strategic
decisions. We propose a change management tool to II. THE CHANGE MANAGEMENT TOOL
select the needed changes, determine in an automated As commented in previous work, as important as
form the best time intervals to become change window doing a planned change management is the selection of
and build an optimized schedule of changes. changes that should be implemented. We present a model
The first results of this study were published in divided into three main steps: change filtering, refinement
[3], where we presented: the division of the changes in of changes and scheduling.
family of changes, where each one represents an IT
service; the concepts of direct and indirect dependencies
between the changes; the treatment realized over the

399
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_70, © Springer Science+Business Media B.V. 2010
400 OLIVEIRA AND FILHO

Figure 1: Change management Conceptual model

B. Refinement of Changes
A. Change Filtering In this step, all the changes have already been
The first step, we perform a filtering of all possible released. Our model requires treatments of these changes.
changes, represented by RFC documents. This document These changes are linked to one or more configuration
is a formal proposal to implement a change, referencing item (CI), which are part of a CMDB. The CMDB has a
details as implementation cost, implementation and not mapping of all CI, their dependencies and characteristics.
implementation risks, proposed date to implement, etc. Each change could be linked to new CIs (configuration
IT investments adds details to RFC documents and item), which are out of CMDB. If this occurs, is
the change advisor board (CAB) [2] analyze inherent to necessary to map it to determine the dependencies
the proposed changes, considering the financial reserve to between this new CI and the other that are part of CMDB.
IT investments, return of investments (ROI) and other After the mapping of dependencies between the
financial analysis. We want to propose a more detailed changes, formation of the family of changes, our model
analysis with respect to ROI linked to this tool. applies treatments over the dependencies. The model
The back-out plan is formed by all actions to return provides two cases: cross dependency and multiple
the proposed change to your previous state in case of indirect dependencies from one family of changes to
failure. another [3] [4].
The CAB is formed by stakeholders from each area The final object of phase is the set of changes with
affected by the proposed changes. This group is all CIs and the mapping and treatments of your
responsible for receiving, analyzing, allowing or denying dependencies.
the changes. The set of changes allowed is described in
forward schedule of changes (RFC). C. Scheduling
In [4], we described that each change is The schedule of changes centers the action
represented by C(i, j) , where i={1, 2, 3,…, n} and j={1, 2, implemented by the model to build an optimized
3, …, m}. The values of i and j will determine the allocation of the changes, considering a planning to get a
position of C(i, j) in the scheduling of changes. The set of minor implementation time and/or less financial loss.
changes is represented by SC={C(1, 1),…,C(n, m)}. The SC is The roll-back plan contains the actions to be
formed by one or more families of changes. A family of implemented and gets the previous state of one change in
changes is formed by changes C(i, j) with the same i and case of failure during your implementation. The roll-back
different value of j. In the Schedule, each family plan is different, bigger and more complex than back-out
represents a column. The indirect dependence (ID) plan, because it includes adding, modifying or removing
represents the dependence among changes with different some of them. It modifies the action, takes into
i. We have a direct dependence (DD) when the consideration two main points to do: The treatments of
implementation of C(i,j) depends on of C(i,j-1). dependencies in the previous step and new mapping of
 AN IT CHANGE MANAGEMENT TOOL TO OPTIMIZE SCHEDULING AND PLANNING 401

dependencies by the addition of new CIs. Our studies Analyzing the table 1, we have the change
showed that some modifications in the change actions are windows shown in table 2.
necessary to form the roll-back plan, although it inherits
some of them from back-out plan. Table 2. Change windows.

The change windows are formed by two types: IT Intervals ( hour/Day)

attributed and discovered change windows. Following the Service 0h to 3h/ - 3h to 6h/ -
actual state-of-art, the attributed change windows are 1 x
2 x
found negotiating with a client the time of this change
window that have less impact for business objectives. If
In example of table 2 the implementation time of
one or more changes affect more than one client, should
each family occurs in a sequential form, but it could be
be determined the change window considering these
done in a parallel model. It is possible because they don’t
clients. In this case, which the less financial impact is
have indirect dependency between them.
more important, the chosen interval to become a change
The action of getting amount of profits by interval
window is in periods of less intensity of work.
of time is one customizable point to tool. We could have
The discovered change windows are proposed by
an example where we need a customization to get the
our model, in which the determination of interval chosen
value in a just-in-time form.
to become a change window is carried through dynamic
The allocation of changes details all possible
form. If the stronger point to determine the interval is the
information about the chosen schedule. Here we have the
impact in financial loss, we propose some steps to do it.
final point of the model, in which the changes had been
In our model, each family of change represents an
allocated to change windows and their implementation
IT service, in which we can determine the amount of
sequences determined.
financial profits they generate by a time interval. The
interval used to generate the amount of financial profits is
III. A REAL-WORLD INSPIRED PROBLEM
determined by the planned implementation time of the
family of change linked to IT service. The choice of an Our model was implemented in many cases to
interval can consider the reading of just one day or validate it’s your performance. The real case of figure 3
measuring of several. The chosen interval will be the one was implemented in an e-commerce company. The
that generate minor profits (less financial impact). To company has offices and needs to perform the following
illustrate the determination of a discovered change changes: Change the core-routers and switches for models
windows, we will use the implementation of the set of belonging to the same manufacturer and capable of
changes below. handling 802.1x packages; make changes in servers to fit
the new authentication service. The company wants to
deploy authentication of all computers and users to use
network resources through 802.1X protocol.

Figure 2: Set of changes

The figure 2 present two families of changes

where each change has accurately one hour to be
implemented, totalizing three hours to finish each family
and consequentially the IT service linked to each one
comes up. The tool would evaluate the profits of each IT
service using a variable previously established, as showed
in table 1.
Table 1. Analysis of financial profits by interval time.

Intervals
IT Service 0h to 3h 3h to 6h ... 21h to 24h
1 $ 1.200 $ 1.400 ... $ 2.100
2 $ 1.400 $ 1.300 ... $ 1.700 Figure 3: scene to implement the changes.
402 OLIVEIRA AND FILHO

The branch office depends on the main one, existence of cross dependencies and multiple ID. After
because of this point, the exchange of core routers could that, the rules of allocation of changes on array [4] are
be performed in parallel form. When the change of core- applied. In figure 5 shows the last state of schedule of
router of main office begins, the IT services of branch changes.
office break at the same time. The changes of core-switch
were divided into three phases: change each unit;
configure the operational system; perform security
settings.
The change of switches could be performed just
after the first and second phases of the core-router
changes.
The changes of servers can start after the second
phase of core-routers and distribution-switches. This is
necessary because the change of servers to work with the
authentication service depends on the changes of the
routers and switches.
The model brings together the changes linked to an
IT service to a column: the family of changes. In this
case, to illustrate it, we consider the changes of each
office as two different IT services. The table 3 shows the
identification of IT service as family of changes.
Table 3: Identifying the IT services.

Family of Changes (columns)

Figure 5: Schedule of changes.
1 2 3 4 5 6
Server
X
(Branch office) It is important to note the fact that of family of
Distribution switch change 6 (servers of main office) can only begin when the
X
(Branch office)
Core-Router
distribution switches are finished because the changes of
X servers take effect only if the other succeeds.
IT (Branch office)
service Core-Router
X
In this case, the change window was attributed and
(main office) only one was necessary to implement all changes, with
Distribution switch the interval was of 12h and was not necessary to use the
X
(Main office)
Server resource of discovered change window of our model. The
X model has the function that calculates the total
(Main office)
implementation time in just-in-time form [4]. Considering
The identification of IT services to families does the implementation time necessary to each change of one
not follow a rule, therefore it is possible to identify them hour, the total implementation time of each family is
with any value of i. The representation of schedule of shown in table 4.
changes, implemented by our model, is in figure 4 below.
Table 4: Total implementation time of each family of changes.

Family of changes Implementation Time

1 1
2 3
3 3
4 3
5 2
6 3

Just the changes of servers located at the main

office generate financial loss. With an implementation
time of three hours and rate of $ 1.200/hour, the total
financial loss was $ 3.600. The other changes have an
Figure 4: Schedule of changes. administrative functions without financial loss rate.
The model allocates the change in an array to
The model takes the schedule of changes and
facilitate the timeline vision of the implementation of
applies the treatment of dependencies, verifying the
 AN IT CHANGE MANAGEMENT TOOL TO OPTIMIZE SCHEDULING AND PLANNING 403

changes as shown in table 5, following the last state of REFERENCES

schedule of change of figure 5.
[1] IT Governance Institute, “COBIT 4.1 Edition, 2006,
Table 5: Changes allocated in array. www.isaca.org/cobit.
[2] IT Infrastructure Library (ITIL 3.0), www.itil-fficialsite.com.
i
[3] Denilson Cursino de Oliveira, Raimir Holanda Filho, “A Highly parallel
1 2 3 4 5 6
scheduling model for IT change management”, in CISSE 2008.
1 C(3,1) C(4,1)
2 C(3,2) C(4,2) [4] Oliveira, D.C.; Filho, R.H., “A time and financial loss estimation
using a highly parallel scheduling model for IT change
3 C(2,3) C(3,3) C(4,3) C(5,3)
management”, in Integrated Network Management-Workshops,
j 4 C(2,4) C(5,4)
2009. IM apos; 09. IFIP/IEEE International Symposium on Volume,
5 C(1,5) C(2,5) C(6,1)
Issue, 1-5 June 2009 Page(s):1 – 9.
6 C(6,2)
[5] Rebouças, R., Sauvé J., Moura A., Bartolini C., Trastour D., “A
7 C(6,3)
Decision Support Tool for Optimizing Scheduling of IT Changes”,
10th IFIP/IEEE Symp. On Integrated Mgmt, 2007.
The implementation of changes follows two basic [6] A. keller, J. L. Hellerstein, J. L. Wolf, K. –L. Wu, V. Krishnan. “The
rules: all changes with the same value of i must to be CHAMPS System: Change Management with Planning and
implemented in a sequential mode and all changes with Scheduling” in 9th IEEE/ IFIP Network Operations and
Management Sysposium (NOMS 2004), Seoul, Korea, Abril 2004.
the same value of j could be implemented in a parallel [7] Sauvé, J., Moura, A., Sampaio, M., Jornada, J. and Radziuk, E., “An
mode [4]. The figure 6 shows the implementation of Introductory Overview and Survey of Business–Driven IT
changes into an attributed change window. Management”, in Proceedings of the 1st IEEE / IFIP International
Workshop On Business-Driven IT Management, in conjunction with
NOMS 2006, Vancouver, Canada, pp. 1-10.
1h 1h 1h 1h 1h 1h 1h Time
C(3,1) [8] Machiraju, V., Bartolini, C. and Casati, F., “Technologies for
C(4,1)
Business-Driven IT Management“, in “Extending Web Services
C(3,2) Technologies: the Use of Multi-Agent Approaches“, edited by
C(4,2) Cavedon, L., Maamar, Z., Martin, D. and Benatallah, B., Kluwer
C(2,3) Academic, 2005, pp.1-28.
C(3,3) [9] Sauvé, J. P., Marques, Filipe Teixeira, Moura, José Antão Beltrão,
C(4,3) Sampaio, Marcus Costa, Jornada, João, Radziuk, Eduardo Optimal
C(5,3) Design of E-Commerce Site Infrastructure from a Business
C(2,4) Perspective In: Hawaii International Conference on System
C(5,4) Sciences, 2006, Waikoloa, Hawaii. Proceedings of the 39th Annual
C(1,5) Hawaii International Conference on System Sciences (HICSS'06).
C(2,5) Washington DC: IEEE Computer Society, 2006. v.8. p.178.3 - 178.3
C(6,5) [10] Bartolini, C., Sauvé, J. and Sahai, A. (eds.), “Information
C(6,6) Technology Management from a Business Perspective. Proceedings
C(6,7) of the 1st IEEE / IFIP International Workshop on Business-Driven
IT Management (collocated with NOMS 2006), Vancouver,
Canada,April 2006
Figure 6: Schedule of changes.
[11] Abrahão, B., Almeida, V., and Almeida, J., “A Self-Adaptive
SLADriven Capacity Management Model for Utility Computing”,
Elsevier Service, 2005.
IV. CONCLUSION AND FUTURE WORK [12] Aib, I., Sallé, M., Bartolini, C. and Boulmakoul, A., “A Business
Driven Management Framework for Utility Computing
The results obtained through application of the Environments”, HP Labs Bristol Tech. Report, 2004-171.
[13] B. Abrahão, V. Almeida, and J. Almeida, “Self-adaptive SLA-
model in many cases show the ability to offer an
driven capacity management for internet services,” in 17th
automated response to change management. The main IFIP/IEEE International Workshop on Distributed Systems:
feature is a simpler heuristic, if compared to others, to Operations and Management,DSOM 2006, 2006.
build a best schedule of change and planning your [14] Brown, A.B., Keller, A., Hellerstein, J.L., "A model of
implementation for a smaller financial loss. configuration complexity and its application to a change
management system", in: Integrated Network Management, 2005.
This paper presents a conceptual model of the tool IM 2005, pp. 631- 644.
with new functions as: discovering change windows, which [15]Christian Mayerl, Jens-Uwe Gaspar, Christof Momm, Sebastian
can discover the time interval in an automated form; Abeck, “ Definition of Metric dependencies for Monitoring the
separating the concept of back-out plan and creating the impact of Quality of Services on Quality of Processes”, in: BDIM –
Business-Driven IT management 2007.
roll-back plan, which must be more complex and complete. [16] Gilad Barash, Claudio Bartolini, Liya Wu, “Measuring and
We plan to develop an automatic reading of the Improving the Performance of an IT Support Organization in
schedule of change to build a roll-back plan. We will Managing Service Incidents” in: BDIM – Business-Driven IT
develop a detailed construction of roll-back plan to ensure management 2007.
[17] J.A. Nilsson and A.U. Ranerup. Elaborate change management:
the return of each configuration item to its previous state Improvisational introduction of groupware in public sector. In
through an automated form. Proceedings of the 34th Annual Hawaii International Conference on
System Sciences, 2001.
 A Low-Power Content-Addressable Memory (CAM)
using Pipelined Search Scheme
Yibo Song, Zheng Yao, Xingguo Xiong
Department of Electrical and Computer Engineering,
University of Bridgeport, Bridgeport, CT 06604, USA

Abstract — A Content-Addressable Memory (CAM) is a order to choose the appropriate route. A CAM is a good
memory that implements the lookup-table function in a single choice for implementing this lookup operation due to its fast
clock cycle using dedicated comparison circuitry. CAMs are search capability.
especially popular in network routers for packet forwarding Fig. 1 [2] illustrates how a CAM accomplishes address
and packet classification, but they are also beneficial in a
variety of other applications that require high-speed table
lookup by implementing the routing table shown in the table.
lookup. The main CAM design challenge is to reduce power As shown in Fig. 1, the packet destination-address of 01101
consumption associated with the large amount of parallel is the input to the CAM. According to the table, two
active circuitry, without sacrificing speed or memory density. locations match, with the (priority) encoder choosing the
In this project, a low power content-addressable memory using upper entry and generating the match location 01, which
pipelined hierarchical search scheme is implemented. The corresponds to the most-direct route. This match location is
search operation is pipelined and the match-lines are broken to the input address to a RAM that contains a list of output
several segments. Whenever the stored words fail to match in ports, as depicted in Fig. 1. A RAM read operation outputs
their initial segments, the search operation is discontinued for the port designation, port B, to which the incoming packet is
subsequent segments, hence the power is saved. The proposed
scheme has been implemented in a 64×64 bit ternary CMOS
forwarded. We can treat the match location output of the
CAM. The schematics of both the pipelined and non-pipelined CAM as a pointer that retrieves the associated word from
CAMs are designed. PSPICE power simulation is used to the RAM. In the particular case of packet forwarding, the
extract the power consumption of both memory designs. The associated word is the designation of the output port. This
simulation results demonstrate an effective overall power CAM/RAM system is a complete implementation of an
reduction in the pipelined CAM compared to non-pipelined address-lookup engine for packet forwarding.
architecture for the given example memory pattern.

Keywords: Content-Addressable Memory (CAM), low power

VLSI, pipeline, power simulation

I. INTRODUCTION

A Content-Addressable memory (CAM) compares

input search data against a table of stored data, and returns
the address of the matching data [1]. CAMs have a single
clock cycle throughput, which make them faster than other
hardware- and software-based search systems. CAMs can be
used in a wide variety of applications which require high
search speed, such as parametric curve extraction, Hough Figure 1: CAM-based implementation of an example
transformation, Huffman coding/decoding, Lempel–Ziv routing table [2]
compression, and image coding. The primary commercial However, the speed of a CAM comes at the cost of
application of CAMs today is to classify and forward increased silicon area and power consumption, two design
Internet Protocol (IP) packets in network routers [1]. In parameters that designers strive to reduce. As CAM
networks like the internet, a message such an as e-mail or a applications grow, demanding larger CAM sizes, the power
web page is transferred by first breaking up the message problem is further exacerbated. Reducing power
into small data packets of a few hundred bytes, and then consumption without sacrificing speed or area is the main
sending each data packet individually through the network. thread of recent research in large-capacity CAMs. Various
These packets are routed from the source, through the low power CAM memory designs have been proposed [3]-
intermediate nodes of the network (called routers), and [9]. In this paper, we mainly implement the concepts in [3]
reassembled at the destination to reproduce the original for a 64×64 bit ternary CMOS pipelined CAM, and compare
message. The function of a router is to compare the its power consumption with the non-pipelined design.
destination address of a packet to all possible routes, in Simulation results demonstrate effective power savings in

405
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_71, © Springer Science+Business Media B.V. 2010
406 SONG ET AL.

the low power pipelined CAM memory design for the given are also becoming popular in solving other problems such as
example memory pattern. sorting and range searching.

II. CAM Basics and Matchline Structure

A small model of CAM architecture is shown in Fig. 2

[4]. The figure shows a CAM consisting of 4 words, with
each word containing 3 bits arranged horizontally
(corresponding to 3 CAM cells). There is a matchline
corresponding to each word feeding into matchline sense
amplifiers (MLSAs), and there is a differential searchline
pair corresponding to each bit of the search word. A CAM
Figure 3: Search operation in a TCAM with (a) global
search operation begins with loading the search-data word
masking and (b) local masking [5]
into the search-data registers followed by precharging all
matchlines high, putting them all temporarily in the match The Binary CAM Cell
state. Next, the searchline drivers broadcast the search word A CAM cell serves two basic functions: bit storage (as in
onto the differential searchlines, and each CAM core cell RAM) and bit comparison (unique to CAM). Fig. 4 shows a
compares its stored bit against the bit on its corresponding NOR-type CAM cell [Fig. 4(a)] and the NAND-type CAM
searchlines. Matchlines on which all bits match remain in cell [Fig. 4(b)]. The bit storage in both cases is an SRAM
the precharged-high state. Matchlines that have at least one cell where cross-coupled inverters implement the bit-storage
bit that misses, are discharged to ground. The MLSA then
nodes D and D . To simplify the schematic, we omit the
detects whether its matchline has a matching condition or
NMOS access transistors and bitlines which are used to read
miss condition. Finally, the encoder maps the matchline of
and write the SRAM storage bit. Although some CAM cell
the matching location to its encode address.
implementations use small area DRAM cells, typically,
CAM cells use SRAM storage. The bit comparison, which
is logically equivalent to an XOR of the stored bit and the
search bit is implemented in a somewhat different fashion in
the NOR and the NAND cells.

Figure 2: Simple schematic of a CAM model [4]

Binary Versus Ternary CAMs
CAMs can be divided into two categories [3]: binary Figure 4. CAM core cells for (a) 10-T NOR-type CAM and
CAMs and ternary CAMs (TCAMs). A binary CAM can (b) 9-T NAND-type CAM [9]
store and search binary words (consisting of ‘0’s and ‘1’s).
Thus, binary CAMs are suitable for applications that require III. A low-Power Content-Addressable Memory design
only exact-match searches. A more powerful and feature- This paper is mainly based on the concept in [3]. The
rich TCAM can store and search ternary states (‘1’, ‘0’, and technique is to pipeline the search operation by breaking the
‘X’). The state ‘X’, also called ‘mask’ or ‘don’t care’, can match-lines into several segments. Since most stored words
be used as a wild card entry to perform partial matching. fail to match in their first segments, the search operation is
Masking can be done both globally (in the search key) and discontinued for subsequent segments, hence reducing the
locally (in the table entries). Figure 3 shows examples of power.
global and local masking in TCAMs. In Figure 3(a), the Fig. 5 shows the non-pipelined as well as pipelined
search key 10110XXX matches with all the entries that fall matchline architectures. Both the non-pipelined and
in the following range: 10110000 to 10110111 (words pipelined architectures use a current-based MLSA. In this
located at addresses 1 and 4 in this case). It is called global ML sensing scheme, the match-line is pre-discharged to low
masking because the last three bits of all the table entries are and a current is forced into the ML. MLs in the miss state
ignored. In Figure 3(b), word 110-XX-010 (located at discharge the current to ground and thus there is little
address 2) will match with any of the following search keys: increase in the ML voltage. MLs in the match state collect
110-00-010, 110-01-010, 110-10-010 and 110-11-010. The charge and the ML voltage increases. The NMOS transistor
mask feature is particularly suitable for longest-prefix match turns on only for MLs in the match state, which in turn flips
searches in classless inter-domain routing (CIDR). TCAMs the state of the ensuing half-latch, indicating a match.
 A LOW-POWER CONTENT-ADDRESSABLE MEMORY 407

(a). Non-pipelined match-line architecture

Figure 7. Schematic design of one module in 64×64 CAM

The key benefit of the pipelined architecture is that it

exploits the same effect as the selective precharge scheme to
(b). Pipelined match-line architecture reduce match-line power consumption. In typical CAM
Figure 5. Non-piplelined and pipelined match-line applications, such as router address look-up, only one or
architecture of the CAM two words match and all other words miss. Furthermore,
Figure 6 shows the schematic design of the 64×64 most words will miss on the first few bits. This fact is
Content-Addressable Memory in PSPICE. Since the overall exploited in power reduction by allocating the first segment
memory is too large to fit in the design space, hierarchical so that the majority of words will miss in this segment. The
design strategy is utilized. As shown in Fig. 6, the overall subsequent segments are set to be relatively larger in size in
memory consists of 4×4 modules, with each module order to minimize the total amount of duplicated sensing
representing a 16×16 block. Figure 7 shows the detailed and pipeline circuitry required.
schematic design of one memory module. The ML is
divided into four ML segments, each evaluated sequentially In order to verify the power savings of the pipelined CAM
in a pipeline fashion. Each segment has 16 bits, hence the memeory design, PSIPICE simulation is used to etract the
ML has totally 64 bits. The MLSA current source that power consumption of both the non-pipelined and pipelined
provides the IML current is divided among the four segments CAM memory circuits for the given input patterns. In
in proportion to the number of bits in each segment. This is PSPICE power simulation, an auxiliary power measurement
to guarantee the identical speed in all ML segments and to circuit (including a current-controlled current source k·idd, a
allow a fair comparison with the non-pipelined architecture. capacitor C and a resistor R) is added, as shown in Fig. 8.
The pipelined ML operates from left to right, with each ML
segment acting as an enable signal for the MLSA of the
subsequent segment. Hence, only words that match a
segment proceed with the search in their subsequent
segments. Words that fail to match a segment do not search
for their subsequent segments and hence power is saved.
stage1 stage2 stage3 stage4

Figure 8. Power analysis using PSPICE

R1 L1 R1 L1 R1 L1 R1
R2 L2 R2 L2 R2 L2 R2
R3 L3 R3 L3 R3 L3 R3
R4 L4 R4 L4 R4 L4 R4
R5 L5 R5 L5 R5 L5 R5
R6 L6 R6 L6 R6 L6 R6
R7 L7 R7 L7 R7 L7 R7
R8 L8 R8 L8 R8 L8 R8
R9 L9 R9 L9 R9 L9 R9
R10 L10 R10 L10 R10 L10 R10
R11 L11 R11 L11 R11 L11 R11
R12 L12 R12 L12 R12 L12 R12
R13 L13 R13 L13 R13 L13 R13

The average power of circuit under test (CUT) in time

R14 L14 R14 L14 R14 L14 R14
D10
D11
D12
D13
D14
D15
D16
D17
D18
D19
D20
D21
D22
D23
D24
D25
D26
D27
D28
D29
D30
D31
D32

D10
D11
D12
D13
D14
D15
D16
D17
D18
D19
D20
D21
D22
D23
D24
D25
D26
D27
D28
D29
D30
D31
D32

D10
D11
D12
D13
D14
D15
D16
D17
D18
D19
D20
D21
D22
D23
D24
D25
D26
D27
D28
D29
D30
D31
D32

D10
D11
D12
D13
D14
D15
D16
D17
D18
D19
D20
D21
D22
D23
D24
D25
D26
D27
D28
D29
D30
D31
D32

R15 L15 R15 L15 R15 L15 R15

D1
D2
D3
D4
D5
D6
D7
D8
D9

D1
D2
D3
D4
D5
D6
D7
D8
D9

D1
D2
D3
D4
D5
D6
D7
D8
D9

D1
D2
D3
D4
D5
D6
D7
D8
D9

R16 L16 R16 L16 R16 L16 R16

stage1 stage2 stage3 stage4

period T is:
stage12 stage22 stage32 stage42

R1 L1 R1 L1 R1 L1 R1
u1
u2
u3
u4
u5
u6
u7
u8
u9

u1
u2
u3
u4
u5
u6
u7
u8
u9

u1
u2
u3
u4
u5
u6
u7
u8
u9

u1
u2
u3
u4
u5
u6
u7
u8
u9
u10
u11
u12
u13
u14
u15
u16
u17
u18
u19
u20
u21
u22
u23
u24
u25
u26
u27
u28
u29
u30
u31
u32

u10
u11
u12
u13
u14
u15
u16
u17
u18
u19
u20
u21
u22
u23
u24
u25
u26
u27
u28
u29
u30
u31
u32

u10
u11
u12
u13
u14
u15
u16
u17
u18
u19
u20
u21
u22
u23
u24
u25
u26
u27
u28
u29
u30
u31
u32

u10
u11
u12
u13
u14
u15
u16
u17
u18
u19
u20
u21
u22
u23
u24
u25
u26
u27
u28
u29
u30
u31
u32

R2 L2 R2 L2 R2 L2 R2
R3 L3 R3 L3 R3 L3 R3
R4 L4 R4 L4 R4 L4 R4
R5 L5 R5 L5 R5 L5 R5
R6 L6 R6 L6 R6 L6 R6
R7 L7 R7 L7 R7 L7 R7

E (t ) 1
R8 L8 R8 L8 R8 L8 R8
R9 L9 R9 L9 R9 L9 R9
R10 L10 R10 L10 R10 L10 R10
T Vdd T
Pavg (T ) = = ³ P (t )dt = ³ idd (t ) dt (1)
R11 L11 R11 L11 R11 L11 R11
R12 L12 R12 L12 R12 L12 R12
R13 L13 R13 L13 R13 L13 R13
R14 L14 R14 L14 R14 L14 R14
D10
D11
D12
D13
D14
D15
D16
D17
D18
D19
D20
D21
D22
D23
D24
D25
D26
D27
D28
D29
D30
D31
D32

D10
D11
D12
D13
D14
D15
D16
D17
D18
D19
D20
D21
D22
D23
D24
D25
D26
D27
D28
D29
D30
D31
D32

D10
D11
D12
D13
D14
D15
D16
D17
D18
D19
D20
D21
D22
D23
D24
D25
D26
D27
D28
D29
D30
D31
D32

D10
D11
D12
D13
D14
D15
D16
D17
D18
D19
D20
D21
D22
D23
D24
D25
D26
D27
D28
D29
D30
D31
D32

R15 L15 R15 L15 R15 L15 R15

D1
D2
D3
D4
D5
D6
D7
D8
D9

D1
D2
D3
D4
D5
D6
D7
D8
D9

D1
D2
D3
D4
D5
D6
D7
D8
D9

D1
D2
D3
D4
D5
D6
D7
D8
D9

0 0
R16 L16 R16 L16 R16 L16 R16

stage12

stage13
stage22

stage23
stage32

stage33
stage42

stage43
T T T
R1 L1 R1 L1 R1 L1 R1
u1
u2
u3
u4
u5
u6
u7
u8
u9

u1
u2
u3
u4
u5
u6
u7
u8
u9

u1
u2
u3
u4
u5
u6
u7
u8
u9

u1
u2
u3
u4
u5
u6
u7
u8
u9
u10
u11
u12
u13
u14
u15
u16
u17
u18
u19
u20
u21
u22
u23
u24
u25
u26
u27
u28
u29
u30
u31
u32

u10
u11
u12
u13
u14
u15
u16
u17
u18
u19
u20
u21
u22
u23
u24
u25
u26
u27
u28
u29
u30
u31
u32

u10
u11
u12
u13
u14
u15
u16
u17
u18
u19
u20
u21
u22
u23
u24
u25
u26
u27
u28
u29
u30
u31
u32

u10
u11
u12
u13
u14
u15
u16
u17
u18
u19
u20
u21
u22
u23
u24
u25
u26
u27
u28
u29
u30
u31
u32

R2 L2 R2 L2 R2 L2 R2

In power testing circuit:

R3 L3 R3 L3 R3 L3 R3
R4 L4 R4 L4 R4 L4 R4
R5 L5 R5 L5 R5 L5 R5
R6 L6 R6 L6 R6 L6 R6
R7 L7 R7 L7 R7 L7 R7
R8 L8 R8 L8 R8 L8 R8
R9 L9 R9 L9 R9 L9 R9
R10 L10 R10 L10 R10 L10 R10
R11 L11 R11 L11 R11 L11 R11

dQc d (CVC ) dV (T )
R12 L12 R12 L12 R12 L12 R12

k T
R13 L13 R13 L13 R13 L13 R13
R14 L14 R14 L14 R14 L14 R14

Ÿ VC (T ) = ³ idd (t )dt (2)

D10
D11
D12
D13
D14
D15
D16
D17
D18
D19
D20
D21
D22
D23
D24
D25
D26
D27
D28
D29
D30
D31
D32

D10
D11
D12
D13
D14
D15
D17
D16
D18
D19
D20
D21
D22
D23
D24
D25
D26
D27
D28
D29
D30
D31
D32

D10
D11
D12
D13
D14
D15
D16
D17
D18
D19
D20
D21
D22
D23
D24
D25
D26
D27
D28
D29
D30
D31
D32

D10
D11
D12
D13
D14
D15
D16
D17
D18
D19
D20
D21
D22
D23
D24
D25
D26
D27
D28
D29
D30
D31
D32

R15 L15 R15 L15 R15 L15 R15

k ⋅ idd (t ) = = =C C
D1
D2
D3
D4
D5
D6
D7
D8
D9

D1
D2
D3
D4
D5
D6
D7
D8
D9

D1
D2
D3
D4
D5
D6
D7
D8
D9

D1
D2
D3
D4
D5
D6
D7
D8
D9

R16 L16 R16 L16 R16 L16 R16

stage13 stage23 stage33 stage43

stage14

R1
stage24

L1 R1
stage34

L1 R1
stage44

L1 R1 dt dt dt C 0
u1
u2
u3
u4
u5
u6
u7
u8
u9

u1
u2
u3
u4
u5
u6
u7
u8
u9

u1
u2
u3
u4
u5
u6
u7
u8
u9

u1
u2
u3
u4
u5
u6
u7
u8
u9
u10
u11
u12
u13
u14
u15
u16
u17
u18
u19
u20
u21
u22
u23
u24
u25
u26
u27
u28
u29
u30
u31
u32

u10
u11
u12
u13
u14
u15
u16
u17
u18
u19
u20
u21
u22
u23
u24
u25
u26
u27
u28
u29
u30
u31
u32

L10
u11
u12
u13
u14
u15
u16
u17
u18
u19
u20
u21
u22
u23
u24
u25
u26
u27
u28
u29
u30
u31
u32

u10
u11
u12
u13
u14
u15
u16
u17
u18
u19
u20
u21
u22
u23
u24
u25
u26
u27
u28
u29
u30
u31
u32

R2 L2 R2 L2 R2 L2 R2
R3 L3 R3 L3 R3 L3 R3
R4 L4 R4 L4 R4 L4 R4
R5 L5 R5 L5 R5 L5 R5

Compare above two, if we select: k = ( V dd ⋅ C ) / T

R6 L6 R6 L6 R6 L6 R6
R7 L7 R7 L7 R7 L7 R7
R8 L8 R8 L8 R8 L8 R8
R9 L9 R9 L9 R9 L9 R9
R10 L10 R10 u10 R10 L10 R10
R11 L11 R11 L11 R11 L11 R11
R12 L12 R12 L12 R12 L12 R12
R13 L13 R13 L13 R13 L13 R13
R14
R15
L14
L15
R14
R15
L14
L15
R14
R15
L14
L15
R14
R15
V dd k
= Ÿ Pavg (T ) = V C (T )
R16 L16 R16 L16 R16 L16 R16

stage14 stage24 stage34 stage44

Figure 6. Schematic design of 64×64 CAM Then T C (3)

408 SONG ET AL.

That is, the voltage in capacitor C is numerically equal to line bits, the last stage is not turned on. As a result, power is
the average power of the CUT. Thus by measuring the saved.
voltage VC(T) in PSPICE simulation, we know the value of
Pavg(T). The power consumption of a circuit can be
extracted by PSPICE simulation.

IV. Results and Discussion

We performed transient simulation to verify the

function of the designed pipelined CAM memory. The
following input patterns are used for all inputs:
Pattern1:
0000000000000000000000000000000000000000000000
000000000000000000
Pattern2:
0000000000000000000000000000000000000000000000
000000000000000000 Figure 10. Simulated waveforms of CAM for last stage
Pattern3: mismatch
1111111111111111111111111111111111111111111111
111111111111111111 Fig 11 shows that the MLSA turns off the last two stages
Pattern4: due to a mismatch in the third stage. As pipelined ML
0000000000000000000000000000000000000000000000 operates from left to right, the first two stages all turn on
000000000000000000 during 10.5ns-14ns which means all the CAM memory cells
The precharge and evaluation phases for each pattern last in first two stages match their corresponding searching bits.
for 3.5ns and 3.5ns separately. That is, clock period Each ML segment acts as an enable signal for the MLSA of
Tclk=7ns. If all the CAM memory cells match the the subsequent segment. Since the third stage failed to
corresponding search line bits, the MLSA will detect the match the searching bits, the third and fourth stages are both
match condition and feed into next stage. Fig 9 shows that turned off to save the power.
all the memory bits are set as 1. As pipelined ML operates
from left to right, at 10.5ns-14ns all the MLs match. Each
ML segment then acts as an enable signal for the MLSA of
the subsequent segment. Hence, the MLSA detects the
matching condition for all stage.

Figure 11. Simulated waveforms of CAM for third stage

miss

Similarly, Fig. 12 shows that when the CAM cells mis-

Figure 9. Simulated waveforms of CAM for all stage match match the searching bits at second stage, then the MLSA
turns off the subsequent stages (the third and fourth stages)
Fig 10 shows that the last stage MLSA didn’t detect the and power can be saved. In Fig. 13, when the CAM cells fail
matching signal in last stage. As pipelined ML operates to match the searching bits at the very first stage, then all the
from left to right, the first three stages are all turned on MLSA will turn off all the four stages, and a maximum
during 10.5ns-14ns which means that all the CAM memory power saving can be achieved. While in the non-pipelined
cells in first three stages match their corresponding CAM design, all the bits will be enabled in order to decide a
searching bits. Each ML segment acts as an enable signal match or a miss, which leads to a larger power consumption
for the MLSA of the subsequent segment. Since the memory compared to the pipelined design.
cells in last stage didn’t match their corresponding searching
 A LOW-POWER CONTENT-ADDRESSABLE MEMORY 409

Fig 15 shows the simulated power curve of the 4-stage

pipelined CAM design with all the 4 stages in matching
condition. Based on the simulation result, the average power
of the pipelined CAM during time periods T1=21ns is found
to be Pavg=2.7671mW. Compared with the non-pipelined
CAM, the pipelined CAM does not save power in this case.
This is because all the four stages are enabled due to the
complete matching in all the four bits in this case. We can
see that actually it spends even a little more power
compared to non-pipelined CAM due to the extra pipeline
registers introduced.

Figure 12. Simulation of CAM for second stage mismatch

Figure 15. Power consumption for pipelined scheme (all

match)
Fig 16 shows the simulated power curve of the 4-stage
CAM design when the first stage is found to be a mismatch
with the searching bits. In this case, all the four stages will
be disabled and hence a maximum power saving is achieved.
Based on the simulation results, the average power
Figure 13. Simulation of CAM for all stage mismatch consumption of the pipelined CAM during time periods
T1=21ns is found to be Pavg=505.6μW, compared to the non-
In order to verify the power consumption of the pipelined pipelined scheme, a significant power saving of 77.1% is
CAM design, PSPICE power simulation is used to extract achieved in this case.
the power consumption of both the pipelined and non-
pipelined CAM memory for the same giving patterns. Fig
14 shows the simulated power curve V(Pavg) waveform of
the non-pipelined 64×64 CAM memory for the given input
pattern. Based on the cursor measurement results in V(Pavg)
waveform, the average power consumption of the non-
pipelined CAM during time periods T1=21ns is found to be
Pavg=2.2041mW.

Figure 16. Power consumption for pipelined scheme (all

miss)
The average power of the 4-stage pipelined CAM
memory with the mismatch in second stage is also simulated.
Figure 14. Power consumption for non-pipelined scheme According to the simulation results, the average power of
410 SONG ET AL.

the pipelined CAM with the second stage mismatch during amount of parallel active circuitry, without sacrificing speed
time period of T1=21ns is found to be Pavg=561.288μW. or memory density. In this paper, a low power 4-stage
Compared with the non-pipelined scheme, a power saving pipelined 64×64 CAM memory design [3] is implemented.
of 74% is observed. This power saving is slightly lower than The power savings of the pipelined match-lines is due to the
the case when the first stage is mismatched. This is fact that only a small portion of the ML segments will be
reasonable because the first stage is still turned on for activated in general searching operation, and the rest of the
comparison when the mismatch occurs in the second stage. segments will be deactivated to save power. The schematic
Similarly, the average power consumption of the 4-stage of the pipelined CAM memory is designed in PSPICE.
pipelined CAM during time period T1=21ns when the Simulation results demonstrate the correct function of the
mismatch occurs in the third stage is found to be design. Furthermore, PSPICE simulations are used to
Pavg=1.0417mW in PSPICE power simulation. This extract the average power consumptions of both pipelined
translates to a power saving of 52.65% compared to the non- and non-pipelined CAM memories. Simulation results show
pipelined CAM design with the same input pattern. that if all the four stages mismatch, a significant power
Further, the average power consumption of the 4-stage saving of 77.1% is observed for the pipelined CAM
pipelined CAM during time period T1=21ns when the compared to the non-pipelined design. The power savings
mismatch occurs in the 4th stage is found to be also depends on where the first mismatch occurs in the four
Pavg=1.45mW in PSPICE power simulation. Compared to stages. If the mismatch occurs in the early pipeline stage,
the original non-pipelined CAM design, a power saving of more power saving will be achieved due to the factor that
34% is observed. This leads to least power saving compared the rest of the stages can be turned off to save power.
to the previous cases because three out of four stages in the Similarly, power can also be saved if the hierarchical
pipelined CAM is activated for comparison in this case. search-lines activate only a small portion of the local
Compare the energy consumption of the propose searchlines (LSLs). In the future work, we will study how to
architecture with the non-pipelined architecture, we can see save power in search-lines by using hierarchical search
that the addition of pipelined match-lines reduces the energy scheme and combine it with the pipelined scheme.
consumption, consisting of a ML energy consumption and
no change in SL energy. Since most ML segments miss in REFERENCES
the first 16-bit stage (rarely activating the subsequent 48
bits), one expects the ML energy consumption to be reduced [1] A. S. Tanenbaum, “Computer Networks,” Prentice Hall,
by 74%, compared to the non-pipelined architecture. The Upper Saddle River, NJ, 2003.
simulation results measurement, however, the total 56% ML [2] J. M. Rabaey, A. Chandrakasan, and B. Nikoliæ, Digital
energy is reduction in the 64×64 pipelined content Integrated Circuits: A Design Perspective, 2nd ed.
addressable memory designs. Englewood Cliffs, NJ, Prentice-Hall, 2003.
From above simulations, we can see that generally the [3] K. Pagiamtzis and A. Sheikholeslami, “Pipelined match-
pipelined CAM design leads to significant power savings lines and hierarchical search-lines for low-power content-
compared to the non-pipelined CAM design. Especially if addressable memories,” Proc. IEEE Custom Integrated
the first stage is found to be a mismatch with the searching Circuits Conf., 2003, pp. 383–386.
bits, all the four stages will be turned off and a significant [4] R. Sangireddy, and A. K. Somani, “High-speed IP
power saving of 77.1% is observed for the given patterns. If routing with binary decision diagrams based hardware
the mismatch occurs in a later stage, less power saving will address lookup engine,” IEEE Journal on Selected Areas in
be resulted because more stages will be activated for Communications, vol. 21, no. 4, pp. 513-521, May 2003.
comparison. If no mismatch exists, i.e. all the four stages [5] K. Etzel, “Answering IPv6 lookup challenges,”
match the searching bits, then all the four stages will be Technical Article, Cypress Semiconductor Corporation, Oct.
activated and no power saving can be achieved. Generally, 27, 2004.
most of the patterns lead to mismatch in one of the four [6] A. J. McAuley, and P. Francis, “Fast routing table
stages. Hence effective power savings can be achieved for lookup using CAMs,” Proceedings INFOCOM’93, San
the pipelined CAM design. Francisco, CA, vol. 3, pp. 1382-1391, 1993.
[7] T.-B. Pei and C. Zukowski, “Putting routing tables in
V. Conclusions and Future Work silicon,” IEEE Network Mag., vol. 6, pp. 42–50, Jan. 1992.
[8] L. Chisvin and R. J. Duckworth, “Content-addressable
A CAM is a memory that implements the lookup-table and associative memory: Alternatives to the ubiquitous
function in a single clock cycle using dedicated comparison RAM,” IEEE Computer, vol. 22, pp. 51–64, July 1989.
circuitry. CAMs are especially popular in network routers [9] H. Miyatake, M. Tanaka, and Y. Mori, “A design for
for packet forwarding and packet classification, but they are high-speed lowpower CMOS fully parallel content-
also beneficial in a variety of other applications that require addressable memory macros,” IEEE J. Solid-State Circuits,
high-speed table lookup. The main CAM-design challenge vol. 36, pp. 956–968, June 2001.
is to reduce power consumption associated with the large
 Design and Optimization of Piezoelectric
Dual-Mode Micro-Mirror
Jichao Zhong, Xingguo Xiong, Zheng Yao, Junling Hu*, Prabir Patra*
Department of Electrical and Computer Engineering,
*Department of Mechanical Engineering
University of Bridgeport, Bridgeport, CT 06604, USA

Abstract—MEMS micro-mirrors have been widely used According to the working modes, MEMS micro-mirrors can
in optical communication, projection display, and be divided into two categories: piston mirrors and torsional
microscanner, etc. Various MEMS micro-mirror devices mirrors. Piston micro-mirrors move perpendicular to the
have been reported. In this paper, the design of a bulk- device plan, while torsional micro-mirrors rotate along a
micromachined MEMS piezoelectric dual-mode micro- certain axis. Various piston and torsional micro-mirrors
mirror is proposed. The working principle of the dual- have been reported [1]-[4]. The working principles and
mode micro-mirror device is analyzed. Based on the design considerations of MEMS micromirrors have also
analysis, a set of optimized design parameters of the been analyzed in details [5]-[6]. Furthermore, dual-mode
micro-mirror is suggested. Theoretical analysis shows micro-mirrors which can work in both piston and torsional
that the proposed micro-mirror can deflect for a modes have been reported [7]-[8]. Dual-mode micro-mirrors
maximum displacement of 2.5m in its piston mode, and offers more flexibility in light modulation. Users can
tilt for a maximum angle of 4.88º in its torsional mode. reconfigure them into either piston or torsional modes
The fabrication flow of the MEMS micro-mirror is also according to their individual needs.dd
suggested. The proposed micro-mirror has the MEMS micro-mirrors can be activated using various
advantages of large displacement and large tilt angle due mechanisms, such as electrostatic driving, electromagnetic
to piezoelectric activation. The PZT actuated micro- driving, thermal driving, piezoelectric driving, as well as
mirror (PAM) also leads to improved linearity in tilting shape-memory alloy driving, etc. Among them, electrostatic
angle and piston displacement. Further, fast response of driving is popular due to their easy in implementation. But
the PZT unimorph on the applied actuation voltage the maximum displacement is limited by snap-down effect
ensures wider operation bandwidth of the PAM in in perpendicular driving. Electromagnetic driving has larger
contrast to thermal activated micro-mirrors. activation force and displacement, but the fabrication is
more complex. Thermal driving is relatively slow in
Keywords: MEMS (Microelectromechanical System), dual- response, and the energy efficiency is not high due to
mode micro-mirror, piezoelectric technology (PZT), piston thermal dissipation. Piezoelectric driving utilizes
micro-mirror, torsional micro-mirror. piezoelectric effect to activate MEMS micro-mirrors. It can
achieve large displacement or rotation angle with high
I. Introduction energy efficiency. Various piezoelectric micro-mirrors have
been reported [9]-[11].
Microelectromechanical Systems (MEMS) deal with In this paper, a bulk-micromachined piezoelectric-driven,
devices or systems in the size range of 1μm~1mm, which dual-mode MEMS micro-mirror device is proposed. The
integrates both mechanical and electrical components into a dual-mode micro-mirror can be configured to work in either
single chip. MEMS merges with nanotechnology and piston or torsional mode according to the need. The working
creates the field of Nanoelectromechanical Systems principle of the dual-mode micro-mirror device is analyzed.
(NEMS). Optical MEMS is a unique application where Based on the analysis, a set of optimized design parameters
MEMS are used to direct, guide, filter, and, in some of the micro-mirror is suggested. Theoretical analysis shows
instances, amplify light. Well known applications of optical that the proposed micro-mirror can deflect for a maximum
MEMS include optical switching and digital light displacement of 2.5m in its piston mode, and tilt for a
projection. In optical switching applications, micro-mirrors maximum angle of 4.88º in its torsional mode. The
are used to steer light from an incoming fiber optic input to fabrication flow of the MEMS micro-mirror is also
a fiber optic output. In digital light projection applications, suggested. The proposed micro-mirror can achieve large
micro-mirrors, such as digital micro-mirror device (DMD) displacement and large tilt angle due to piezoelectric
by Texas Instruments Inc., incident light is reflected by an activation. The PZT actuated micro-mirror (PAM) also has
array of torsional micro-mirrors so that each mirror pixel improved linearity in tilting angle and piston displacement.
can control a corresponding spot on the screen to be bright
or dark. In this way, vivid pictures can be projected onto the
screen.

411
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_72, © Springer Science+Business Media B.V. 2010
412 ZHONG ET AL.

II. Piezoelectric Activation polarization is contributed by two parts – one stemming

from electrical biasing and one from mechanical loading.
Piezoelectric materials are used for both sensing and If no electric field is present (i.e., E=0), then the second
actuation purposes. It was observed that certain materials term on the right-hand side of the above equation can be
generate an electric charge (or voltage) when they are under eliminated.
mechanical stress. This is known as the direct effect of The general constitutive equation can be written in the
piezoelectricity. Alternately, the same materials would be full matrix form:
able to produce a mechanical deformation (or force) when
an electric field is applied to them. This is called the inverse
effect of piezoelectricity.
Piezoelectric materials are crystals. The microscopic
origin of piezoelectricity is the displacement of ionic
charges within a crystal, leading to the polarization and
electric field. A stress (tensile or compressive) applied to a
piezoelectric crystal will alter the spacing between centers
of positive and negative charge sites in each domain cell; the terms T1 through T3 are normal stresses along axes 1, 2
this leads to a net polarization manifested as open circuit and 3, whereas T4 through T6 are shear stresses. The units of
voltages measurable at the crystal surface. Compressive and electrical displacement (Di) , stress (Ti), permittivity (œi),
tensile stresses will generate electric fields and hence and electrical field (Ej) are C/m2, N/m2, F/m2, and V/m,
voltages of opposite polarity. respectively. The unit of the piezoelectric constant dij is the
Inversely, an external electric field will exert a force unit of electric displacement divided by the unit of the
between the centers of positive and negative charges, stress, namely:
leading to an elastic strain and changes of dimensions F V
depending on the field polarity. Not all naturally occurring [D ] [ε ][E ] m m Columb
[d ij ] = [T ] = [T ] = N = N
or synthesized crystals exhibit piezoelectricity. Crystals can
be classified into 32 groups according to crystal symmetry.
m2
Centrosymmetric crystal structures are crystals that are
symmetric along all axes through the center of the crystal. 3
These crystals occupy 11 out of 32 possible groups and are
non-piezoelectric materials because the positive and
negative charge sites will not be spatially separated under 6
stress. Out of 21 non-centrosymmetric groups, 20 are
piezoelectric crystals. Piezoelectric effects are strongly
orientation dependent. The notation conventions for crystal
orientations in the context of piezoelectric polarization are 4
discussed first. A piezoelectric material needs to be poled in 2
a particular direction to provide a strong piezoelectric effect,
although some materials exhibit natural or spontaneous 5
polarization. The direction of positive polarization is
customarily chosen to coincide with the Z-axis of a 1
Z
rectangular system of crystallographic axes X, Y, and Z.
Alternatively, the normal stress components along axes X,
Y, and Z are denoted by subscripts 1, 2 and 3, respectively,
as shown in Figure 1. As such, the poling axis always Y
coincides with axis-3. Shear stress and strain components X
about these axes are denoted by sub-scripts 4, 5 and 6
respectively. Figure 1: Schematic illustration of a piezoelectric crystal in
In a piezoelectric crystal, the constitutive equation that a Cartesian coordinate system [12]
relates electrical polarization (D) and applied mechanical
stress (T) is [12]: The inverse effect of piezoelectricity can be similarly
D = dT + εE (1) described by a matrix-form constitutive equation. In this
case, the total strain is related to both the applied electric
where d is the piezoelectric coefficient matrix, œ is the
field and any mechanical stress [12], according to
electrical permittivity matrix, and E is the electrical field.
Here, an electric field is applied in conjunction with the s = ST + dE
mechanical stress to provide more generality. The electrical Where s is the strain vector and S is the compliance matrix.
 DESIGN AND OPTIMIZATION OF PIEZOELECTRIC DUAL-MODE MICRO-MIRROR 413

This equation can be expanded to a full matrix form the PZT actuators will expand or shrink simultaneously,
[12]: which in turn activates the frame (and the mirror) to move
perpendicular to the device plane. Hence, the micro-mirror
works in piston mode. By applying electrical driving
voltages to the PZT actuators on either the piston or
torsional beams, the micro-mirror can work in either piston
or torsional mode. Hence this is a piezoelectric activated
dual-mode micro-mirror design.

if there is no mechanical stress present (Ti,i=1,6=0), the strain

is related to the electric field by [12]

tilting
axis
Note that, for any given piezoelectric material, the dij
components connecting the strain and the applied field in
the inverse effect are identical to the dij connecting the
polarization and the stress in the direct effect. Figure 2. Schematic diagram of the dual-mode piezoelectric
The electromechanical coupling coefficient k is a micro-mirror
measure of how much energy is transferred from electrical
to mechanical energy, or vice versa, during the actuation
process [12]:
energy _ converted
k2 =
input _ energy
this relation holds true for both mechanical-to-electrical and
electrical-to-mechanical energy conversion. The magnitude
of k is a function of not only the material, but also the
geometries of the sample and its oscillation mode.

III. Design and Optimization of the Micro-mirror

The schematic diagram of the proposed dual-mode

piezoelectric micro-mirror is shown in Figure 1. ANSYS
FEM simulation is used to verify the device function, as
shown in Figure 2. As shown in Figure 1 and 2, a MEMS
micro-mirror is connected to four torsional beams which Fig 3: ANSYS model of the dual-mode micro-mirror
extrudes from two torsional axes. On two of the four
torsional beams, PZT piezoelectric materials are deposited In this mirror design, PZT actuated micro-mirror (PAM)
to activate the torsional beams. When electrical driving can provide improved linearity tilting angle and piston
voltages are applied to the PZT actuators on torsional displacement. In addition, its displacement in piston mode is
beams, the PZT will expand or shrink, which generates not limited by pull-in phenomenon as in perpendicular
corresponding torque to activate the micromirror to electrostatic actuation. Fast response of the PZT unimorph
rotate/tilt along the torsional axes. In this way, the micro- membrane on the applied actuation voltage ensures wider
mirror works in torsional mode. The two torsional axes are operation bandwidth of the PAM in contrast to the thermal
again connected to the outer frame, which are anchored to activation. In this work, PZT-4 is chosen as piezoelectric
the substrate through four flexible piston beams. PZT material and polysilicon is used as substrate material for
actuators are also deposited on top of the four piston beams. PZT actuators.
When the electrical driving voltages are simultaneously In this piezoelectric actuated micro-mirror, cantilever
applied to the four PZT actuators on the four piston beams, piezoelectric actuator model is used to analyze the device
414 ZHONG ET AL.

behavior. Piezoelectric actuators are often used in δ ( L) = F / k

conjunction with cantilevers or membranes for sensing and the force can be expressed as
actuation purpose. General models for such piezoelectric
actuators are rather complex. Accurate analysis often
F = δ ( x = L) ⋅ k
involves finite element modeling. For limited cases, such as In this work, PZT-4 is chosen as the piezoelectric
a cantilever actuator with two layers, analytical solution has material. For PZT-4,
been successfully achieved. Based on the theoretical
analysis, a set of optimized design parameters of the micro-
mirror is achieved, as shown in Table 1.

Table 1. Optimized design parameters of the dual-mode

micro-mirror
Outer actuator beam Length 100 μm
Width 6 μm
Thickness 2 μm
Main frame (square) Length of side 90 μm
Width of side 10 μm
Thickness 4 μm
According to the structure we used for the actuator, we
Torsional axis Length 10/20 μm
should choose d31 as the piezoelectric parameter. So the
Width 2 μm
displacement can be calculated as below:
Thickness 2 μm
Inner actuator beam Length 24 μm x2 d31(t p + te )(Ap Ep Ae Ee )
δ (x) =
4(Ep I p + Ee I e )(Ap Ep + Ae Ee ) + (te + t p ) Ae Ee Ap Ep
2
Width 4 μm
Thickness 2 μm = 1.05 ×10 −6 × V (μm )
Gap between the outer 2.5 μm
Under small-deflection approximation, the dual-mode
actuator beam and the
micro-mirror can be treated as simplified spring-mass
main frame
model. In the piston mode, the four piston beams can be
Gap between the inner 2.5 μm
modeled as double-clamped beams. The spring constant of
actuator beam and the
one double-clamped beam can be calculated as
mirror
12 EI b
Gap between the mirror 3 μm Kb =
and the substrate L3b
Hence in our design, the displacement at the end of the
The deflection of a two-layer piezoelectric structure can piston beam can be calculated as
be derived based on above model. Consider a cantilever 1.05 × 10 −6 × V
with two layers, one elastic and one piezoelectric, joined δ (x ) = = 0.25V (μm )
4
along one side. Assume the two layers have the same length. The designed driving voltage of this micro-mirror is
The beam bends into an arc when the piezoelectric layer is 10V, so the maximum displacement of the micro-mirror in
subjected to a longitudinal strain, slong. The radius of
piston mode is 2.5 μm .
curvature can be found by [12]
2 slong (t p + te )(Ap Ep Ae Ee )
When the driving electrical voltage is applied to the PZT
1 actuators in the inner torsional beams, the PZT actuators
=
r 4(Ep I p + Ee I e )(Ap Ep + Ae Ee ) + (te + t p )2 Ae Ee Ap Ep will generate corresponding torque on the torsional beams,
hence the micro-mirror will tilt along the torsional axes. As
a result, the dual-mode micro-mirror works in torsional
In which Ap and Ae are the cross-sectional area of the mode. Assume the voltage is 50V, and the length of the
piezoelectric and the elastic layer; Ep and Ee are Young’s torsional axis is 2μm(h )× 2 μm(b )× 4 μm(l ) , based on the
modulus of the piezoelectric layer and Young’s modulus of above design parameters, the displacement of the actuator is
the elastic layer; and tp and te are the thickness of the calculated to be
piezoelectric layer and the elastic layer.
The amount of force achievable at the free end of a
δ ( x ) = 0.975( μm)
piezoelectric bimorph actuator equals the force required to The tilt angle θ is calculated to be
restore the tip of the actuator to its initial un-deformed state. θ = 2.32 $
Since the displacement is linearly related to force according we can optimize the mirror design on inner actuator, the
to actuator will counteract the torque of the torsional axis, and
the tilt angle of the torsional axis must follow this equation:
 DESIGN AND OPTIMIZATION OF PIEZOELECTRIC DUAL-MODE MICRO-MIRROR 415

Tl (d) Thin hard mask layer is deposited and patterned.

ϕ=
GBhb 3
where T is torque; l is the length of the torsional axis, G is
shear modulus (for polysilicon, G=69GPa), β is a constant,
h is the height of torsional beam, and b is the width of the
torsional beam. Now, the size of the torsional axis can be
redesigned as 2m(h)×2m(b)×4m(l). For β = 0.141 ,
(e) The silicon nitride at the backside of the wafer is
under the same torque, the displacement of the micro-mirror
patterned by double side aligning and subsequent RIE to
is found to be in linear relationship with the torsional
define the etch window for silicon bulk etching.
driving voltage:
δ ( x ) = 0.041V (μm )
For a torsional driving voltage of 50V, the corresponding
displacement and the torsional angle of the micromirror are
found to be
δ ( x ) = 2.05(μm )
θ = 4.88$ (f) The silicon under the suspended microstructures is
removed by anisotropic etching in aqueous KOH. The
IV. Fabrication Flow movable microstructure can be released to be free-standing.

A fabrication process of the above dual-mode micro-

mirror is illustrated in Figure 4. Starting from a bare silicon
wafer, we coat the top side of the silicon wafer with 2m
thickness LPCVD low stress silicon nitride. In order to (g) Suspended structures including the mirror plate are
pattern the PZT layer, a thin layer of 150 nm-thick platinum released by RIE from the front side of the wafer using the
layer is pre-deposited at the bottom electrode layer. Then a patterned metal film at the step (d) as the masking material.
0.2m-chick PZT-4 layer is deposited using sol-gel method.
On the top of the PZT-4 layer, we deposited a 150 nm-thick
ruthenium oxide layer. This forms the PZT piezoelectric
actuator structures for the micro-mirror.

(h) Finally, released mirror are diced into individual chips

after removing the hard mask metal
(a) Silicon wafer coated with Si3N4 and PZT film Figure 4. Fabrication flow of the piezoelectric dual-mode
micromirror

V. Conclusions and Future Work

In this paper, a piezoelectric actuated dual-mode micro-

mirror design is proposed. The working principle of the
(b) Patterning of the PZT layer is delineated by self-aligned micro-mirror is analyzed. The micro-mirror is actuated by
process using Cl/O-based plasma etching. four PZT actuator beams in piston mode. The inner set of
two PZT actuators is used to tilt the micro-mirror in
torsional mode. Each set of actuators can work
independently. Based on the analysis, a set of optimized
design parameters of the micro-mirror is achieved. The two
modes (piston and torsional) can be use for different
(c) The aluminum mirror surface is patterned by lift-off situations as needed. Theoretical analysis shows that the
process. proposed micro-mirror can deflect for a maximum
displacement of 2.5m in its piston mode, and tilt for a
maximum angle of 4.88º in its torsional mode. The
fabrication flow of the MEMS micro-mirror is also
suggested. The proposed micro-mirror can achieve large
displacement and large tilt angle due to piezoelectric
416 ZHONG ET AL.

activation. The PZT actuated micro-mirror (PAM) also has [6] X. M. Zhang, F. S. Chau, C. Quan, Y. L. Lam, A. Q.
improved linearity in tilting angle and piston displacement. Liu, “A study of the static characteristics of a torsional
In the future work, we will further improve the structure micro-mirror”, Sensors and Actuators A: Physical, Vol. 90,
design, so that the driving voltage can be reduced from 2001, pp. 73–81.
current 10V in piston mode to around 5V. In this way, it can [7] U. Krishnamoorth, K. Li, K. Yu, D. Lee, J. P. Heritage,
be compatible with the power supply of on-chip CMOS O. Solgaard, “Dual-mode micro-mirrors for optical phased
circuitry. By reducing the required driving voltage, the array applications”, Sensors and Actuators A: Physical, Vol.
power consumption of the micro-mirror can also be further 97-98, 2002, pp. 21-26.
reduced. [8] X. Xiong, T. Dallas, S. Gangopadhyay, J. Berg, T.
References Henryk, “Design and Simulation of Aluminum Bi-
functional Spatial Light Modulator", Proc. of the 48th IEEE
[1] G. Zhou, V.J. Logeeswaran, F.E.H. Tay, F.S. Chau, International Midwest Symposium on Circuits & Systems
"Digital deflection micro-mirrors for adaptive optics (MWSCAS’05), Cincinnati, Ohio, USA, Aug 7-10, 2005, pp.
applications", 12th International Conference on Solid-State 159-162.
Sensors, Actuators and Microsystems (Transducers'03), [9] Y. Yee, J. U Bu, M. Ha, J. Choi, H. Oh, S. Lee, and H.
Vol. 1, June 8-12, 2003, pp. a591-a594. Nm, “Fabrication and characterization of a PZT actuated
[2] K. Jia, S. Pal, H. Xie, "An electrothermal tip–tilt–piston micro-mirror with two-axis rotational motion for free space
micromirror based on folded dual S-shaped bimorphs", optics”, Proc. of IEEE MEMS Workshop 2001, pp. 317–320.
Journal of Microelectromechanical Systems, Vol. 18, Issue [10] H. Nam, Y. Kim, S. Cho, Y. Yee and J. Bu, "Low
5, Oct. 2009, pp. 1004-1015. voltage PZT actuated tilting micro-mirror with hinge
[3] G. Vdovin, S. Middelhoek, L. Sarro, "Deformable mirror structure", Conference Digest of 2002 IEEE/LEOS
display with continuous reflecting surface micromachined in International Conference on Optical MEMs, 2002, pp. 89-90.
silicon", IEEE Proceedings of Micro Electro Mechanical [11] S. J. Kim, Y. H. Cho, H. J. Nam, J. U. BU,
Systems (MEMS'95), Jan. 29 - Feb. 2, 1995, pp. 61–65. "Piezoelectrically pushed rotational micro-mirrors for wide-
[4] K.B. Lee, L. Lin, "A vertically-supported two-axial angle optical switch applications", IEEE The Sixteenth
torsional micro-mirror", Proceedings of 17th IEEE Annual International Conference on Micro Electro
International Conference on Micro Electro Mechanical Mechanical Systems (MEMS'03), Kyoto, Japan, Jan. 19-23,
Systems (MEMS'04), 2004, pp. 41-44. 2003, pp. 263-266.
[5] O. Degani, Y. Nemirovsky, “Design considerations of [12] C. Liu, "Foundation of MEMS", Chapter 7:
rectangular electrostatic torsion actuation based on new “Piezoelectric Sensing and Actuation”, Prentice Hall, Sept.
analytical pull-in expression”, Journal of 5, 2005, ISBN: 0131472860.
Microelectromechanical Systems, Vol. 11, pp. 20–26, 2002.
 Device for Measurement of Clamped Joints
Friction Torque
Ing. Eva Batešková, PhD., Ing. Jozef Maščenik,PhD., Ing. Jozef Haľko, PhD., Ing. Štefan Gašpár
Faculty of manufacturing Technologies, TU Košice with seat in Prešov, Štúrova 31,
08001 Prešov, Slovak republic
[email protected], [email protected], [email protected], [email protected]

Abstract - This contribution deals with a proposal of II. PROPOSAL OF 3D DEVICE FOR FRICTION JOINT
device for measurement of clamped joints. In the second MEASUREMENT
chapter, we propose a model of clamped joint. In the
experimental part we measured values of the friction torque Fig. 1 shows the proposed device solution. The
depending on twisting moment, and in the conclusion we device consists of frame 1, to which bearing shell sliding
measured shaft surface abrasivity. surface 14 is screwed. Holder 8 is welded to the frame and
there are pegs 11 screwed in it. Via screw 4, through
INTRODUCTION collar 13, dynamometer 5, peg holder and peg, the
clamping power is transferred on claws 3. Weight 12
One of the basic construction elements for slides axially along lever 6. Shaft 2 is from both sides
construction production either in engineering or building placed in bearings 9, which are in shell 10. Screw 7 serves
and construction industries are clamped joint claws. From to ensure the shaft position.
the point of view of working life and criteria on the
clamped joint jaws, the main importance belongs to Measuring procedure:
stating the correct clamping power, as well as stating the By the screw rotation, clamping of the shaft by means of
correct friction torque in the joint. If there is a small claws is generated. From the dynamometer, the clamping
clamping power on this joint, it can cause an insufficient power is read. By means of the weight, it slides axially in
friction torque and subsequently damage the clamped the direction from the shaft axis along the lever until the
joint and thus damage the whole mechanism. On the clamped joint is broken. From the weight mass and the
contrary, high clamping power can cause a big distance from the shaft, twisting moment of the clamped
deformation either on the hub or clamped joint claws, joint breach is calculated. The friction torque is
which can lead to cracks in the claws or disabling the determined from the condition for transfer of the twisting
mutual re-assembling and disassembling of the hub and moment. The disadvantage of this device is measurement
claws [1]. scope restriction by the lever length and relatively
complicated construction, which, however, eliminates the
I. BASIC REQUIREMENTS ON JOINTS shortcomings from the two previous proposals.

In appliances, there are two types of binding

between components and knots – movable and non-
movable. The type of movable binding results from the
required function and kinematic scheme. The non-
movable bindings ensure a constant position of
components and knots. The non-movable bindings are
obtained by linking the components and knots.
The binding elements and methods are to ensure
sufficient resistance and required constant position of the
components and knots. In precise mechanics, a higher
number of binding methods is used.
Fig. 1 3D device model
Reasons:
• Power and deformation ratios,
• Requirements on simple assembly and II.I PROCESS OF DEVICE SET-UP
disassembly,
Firstly, 8 holes were drilled into the construction
• High requirements on precision, frame, into which coils M8 x 1,25 were cut. Into these
• Simultaneous use of heterogeneous materials in holes, sliding surfaces of bearing shells were screwed by
one knot, means of screws with fixed head and internal hexagon.
• Requirements on outward form [2]. The bearings were inserted into shells and ensured by
snap rings 68.
The pegs were screwed into the peg holders either
into the external holes for big claws or into the internal

417
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_73, © Springer Science+Business Media B.V. 2010
418 BATEŠKOVÁ ET AL.

holes for small claws. The holder with the sliding surface
and collar were inserted into the tubes with the internal
diameter of φ 12 mm on the construction frame. The claws
were put on the pegs according to the peg distribution.
The shaft was forced into the bearings and subsequently
ensured against drop-out by snap rings 40. The lever was
screwed by a screw with a washer. Afterwards, a locking
element against lever fall was screwed into the frame. The
weight was put on the lever. The dynamometer was Fig. 3 Device for measurement of surface abrasiveness Mitutoyo SJ 301
screwed into the collar on the peg holder and on the
opposite side of the dynamometer the collar with bearing Measurement of equal pressure distribution on hub
were screwed. The frame was supplemented by screw perimeter:
M24 x 2, which presses on the dynamometer collar and For this measurement, we used big claws of the
serves for educing the axial power. Finally, the entire clamped joint, which were put on the pegs and
construction was fixed to the work table. subsequently tightened by the screw to 2000 N. The
measured values of the clamped joint breach, as well as
the calculated clamping power is in chart 1. Conversion of
the number of dynamometer divisions to power F was
done according to the equation (1)

F = 2000 / 552,1.ndivisions (1)

TABLE I
MEASUREMENT ON CLAWS WITH EXTERNAL DIAMETER OF
Ø 90 MM
Number of lever
divisions
33 31 29 27 25
Fig. 2 Complete clamped joint device
Number of
dynamometer 155 145 131 122 120
divisions
III. EXPERIMENTAL MEASUREMENTS
F [N] 561,5 525,3 474,6 441,9 434,7
The sample for this measurement is a shaft with the Number of lever
diametre of φ 40 mm and one type of claws : divisions
19 17 15 13 11
Number of
• For equal pressure distribution on the claw hub dynamometer 71 56 43 40 32
perimeter with external diametre of φ 90 mm divisions
Measuring method: F [N] 257,2 202,9 155,8 144,9 115,9
Every measurement began by clamping the Number of lever
divisions
13 15 17 19 21
clamped joint by power of 2 kN by a screw. On the lever,
Number of
the weight was set on 33 divisions, whereas every division dynamometer 55 55 78 80 81
corresponds with a different twisting moment and divisions
afterwards the screw was slowly loosened until the F [N] 199,2 199,2 282,6 289,8 293,4
clamped joint was broken. The breach occurs if the lever Number of lever
27 29 31 33
moves. When the lever moves, we read the number of divisions
divisions from the dynamometer. Gradually, the weight Number of
moves towards the rotation axis by two division up to 9 dynamometer 101 105 109 110
divisions
divisions. F [N] 365,9 380,4 394,9 398,5
Subsequently, the measurement was performed
from 9 divisions to 33 divisions. After these
measurements, we carried out measurement of the surface
abrasiveness by device Mitutoyo SJ 301, and The calculation of the friction torque for equal pressure
consequently soft regrinding of claws by abrasive-coated distribution was done according to equation (2)
paper and re-measurement.
π
M T = FN . f ´.d = FN . . f .d (2)
2

where f is chosen from 0,15 to 0,2 , for our measurement

we used the value .
 DEVICE FOR MEASUREMENT OF CLAMPED JOINTS FRICTION TORQUE 419

TABLE II
CHART OF CALCULATED FRICTION TORQUES FOR CLAWS OF
Ø 90 MM
F [N] 561,5 525,3 474,6 441,9 434,7
MT
5,2920 4,9508 4,4730 4,1648 4,0970
[N.m]
F [N] 257,2 202,9 155,8 144,9 115,9
MT
2,4241 1,9123 1,4684 1,3657 1,0923
[N.m]
F [N] 199,2 199,2 282,6 289,8 293,4
MT
1,8774 1,8774 2,6634 2,7313 2,7652
[N.m]
F [N] 365,9 380,4 394,9 398,5
MT
3,4485 3,5852 3,7218 3,7558
[N.m]

Fig. 4 shows the friction torque dependence on

twisting moment, which results in the fact that the
measurement on the device was with minor errors during
the first measurement, when the weight was moved from
the end of the lever towards the axis of rotation. During
the measurement from the axis, a bigger dispersion arose.

Fig. 5 Measured surface abrasivity

CONCLUSION
The calculated values of the friction torque are
distorted. They depend on many factors, among which
there are e.g.: precision of clamped joint make-out;
surface abrasivity; clamping power; conditions of
measurement; number of measurement etc. This causes
the fact that when more measurements are carried out, the
value of friction torque varies, which causes the
dispersion in measurements. Finally, for demanding
applications, it is suitable to recommend modern
Fig. 4 Friction torque dependence on twisting moment hydraulic tightening machines for their higher precision,
and precise digital torque wrenches for clamped joint
Surface abrasivity measurement: breach. This contribution forms a part of the solution of
Fig. 5 shows the recording of measuring abrasivity of the the grant task VEGA 1/4156/07 and KEGA 3/6279/08.
surface of shaft with diameter of 50 mm.

REFERENCES
[1] MEDVECKÝ, Š.: Základy konštruovania, EDIS, Źilina, 1999.
[2] PAVLENKO, S. – HAĽKO, J.: Časti strojov I. FVT Prešov, 2006,
ISBN 80-8073-100-4.
[3] ROVŇÁKOVÁ, Sylvia - LÍŠKA, Ondrej: Predchádzanie kolíziám
pri automatizovaných pracoviskách využitím modelovania a
simulačných programov, Prešov, TU Košice, FVT, 2008. s. 339-
342. ISBN 978-80-553-0044-3.
 Self-Optimization for Dynamic Scheduling in
Manufacturing Systems
Ana Madureira, Ivo Pereira
Institute of Engineering – Polytechnic of Porto, Portugal

Learning (ML), two subfields of Artificial Intelligence.

Abstract — Scheduling is a critical function that is present Traditional ML typically involves a single agent that is trying
throughout many industries and applications. A great need to maximize some utility function without any knowledge.
exists for developing scheduling approaches that can be applied
Traditional ML tasks include function approximation,
to a number of different scheduling problems with significant
impact on performance of business organizations. A challenge is classification, and problem-solving performance improvement
emerging in the design of scheduling support systems for given empirical data. MAS learning includes any situation in
manufacturing environments where dynamic adaptation and which an agent learns to interact with other agents [4].
optimization become increasingly important. In this paper, we This paper envisage the proposal of a Self-Optimizing
describe a Self-Optimizing Mechanism for Scheduling System mechanism for a Cooperative Scheduling System considering
through Nature Inspired Optimization Techniques (NIT).
that AutoDynAgents [8] must be able to perform scheduling
Index Terms— Self-Optimization, Multi-agent Learning,
in highly dynamic environments where there is incomplete
Autonomic Computing, Multi-Agent Systems, Nature Inspired information and changes often occur; modify previously
Optimization Techniques. formed schedules considering recent dynamic information,
minimizing the disruption of earlier schedules and still
I. INTRODUCTION aiming for the most effective possible use of resources and
Scheduling is an important aspect of automation in achievement of goals and provide flexibility to react robustly
manufacturing systems. Most of scheduling domains are to any disruption in an efficient and timely manner.
characterized by a great amount of uncertainty that leads to The remaining sections are organized as follows: Section 2
significant system dynamics [2]. Such dynamic scheduling is summarizes some related work on dynamic scheduling
receiving increased attention amongst both researchers and through MAS and Bio-Inspired Techniques (BIT). Some
practitioners [2][7-9][11]. However, scheduling is still having aspects of MAS and BIT are discussed on section 3 e 4. In
difficulties in real world situations and, hence, human section 5 the AutoDynAgents System is presented. Section 6
intervention is required to maintain real-time adaptation and describes the Self-Optimizing Mechanism for Scheduling
optimization. System through Nature Inspired Optimization Techniques.
A challenge is emerging in the design of scheduling Finally, the paper presents some conclusions and puts forward
support systems for manufacturing environments where some ideas for future work.
dynamic adaptation and optimization become increasingly
important incorporating expert’s knowledge. Despite the II. DYNAMIC SCHEDULING
numerous advances in scheduling and MAS [6][10], most Scheduling problems arise in a diverse set of domains,
development systems still require that users have substantial ranging from manufacturing to hospitals settings, transports,
knowledge of procedural-programming techniques as well as computer and space environments, amongst others. Most of
the specific computer system at hand. these domains are characterized by a great amount of
At this manufacturing scheduling scenario, self-optimizing uncertainty that leads to significant system dynamism. Such
arise as the ability of the agent to monitor its state and dynamic scheduling is receiving increased attention amongst
performance, and proactively tune itself to respond to both researchers and practitioners.
environmental stimuli. Dynamic changes of a problem could arise from new user
requirements and the evolution of the external environment.
Learning can be considered as the process of adapting
In a more general view, dynamic problem changes can be
behavior in response to events in the environment. Without
seen as a set of constraint insertions and cancellations.
learning, a system is limited by the ability of its designer to
For these dynamic optimization problems environments,
foresee all situations that might occur. Multi-agent Learning that are often impossible to avoid in practice, the objective of
is the intersection of Multi-agent Systems and Machine the optimization algorithm is no longer to simply locate the

421
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_74, © Springer Science+Business Media B.V. 2010
422 MADUREIRA AND PEREIRA

global optimum solution, but to continuously track the communities to look for inspiration in diverse fields, e.g.
optimum in dynamic environments, or to find a robust robotics, artificial intelligence or biology, to find new ways of
solution that operates optimally in the presence of designing and managing systems. Hybridization of different
perturbations [2][9]. In spite of all the previous trials, the approaches seems to be a promising research field of
scheduling problem is still known to be NP-complete, even computational intelligence focusing on the development of
for static environments. This fact poses serious challenges to the next generation of intelligent systems.
conventional algorithms and incites researchers to explore
new directions [9-11] and Multi-Agent technology has been V. AUTODYNAGENTS ARCHITECTURE
considered an important approach for developing industrial Distributed environment approaches are important in order
distributed systems. to improve scheduling systems flexibility and capacity to react
to unpredictable events. It is accepted that new generations of
III. MULTI-AGENT SYSTEMS
manufacturing facilities, with increasing specialization and
Multi-agent paradigm is emerging for the development of
integration, add more problematic challenges to scheduling
solutions to very hard distributed computational problems.
systems. For that reason, issues like robustness, regeneration
This paradigm is based either on the activity of "intelligent"
capacities and efficiency are currently critical elements in the
agents which perform complex functionalities or on the
design of manufacturing scheduling system and encouraged
exploitation of a large number of simple agents that can
the development of new architectures and solutions,
produce an overall intelligent behavior leading to the solution
leveraging the MAS research results.
of alleged almost intractable problems.
The work reported in this paper is concerned with the
Considering the complexity inherent to the manufacturing
resolution of realistic scheduling problems, called here
systems, dynamic scheduling is considered an excellent
Extended Job-Shop Scheduling Problems (EJSSP) [9].
candidate for the application of agent-based technology. In
Moreover, it is concerned with integrated scheduling of jobs
many implementations of MAS systems for manufacturing
which are products composed by several parts or components
scheduling, the agents model the resources of the system and
that may be submitted to a number of manufacturing and
the tasks scheduling are done in a distributed way by means
multi-level assembly operations.
of cooperation and coordination amongst agents [9-10]. When
AUTODYNAGENTS is an Autonomic Scheduling System
responding to disturbances, the distributed nature of multi-
in which communities of agents model a real manufacturing
agent systems can also be a benefit to the rescheduling
system subject to perturbations. Agents must be able to learn
algorithm by involving only the agents directly affected,
and manage their internal behavior and their relationships
without disturbing the rest of the community which can
with other autonomic agents, by cooperative negotiation in
continue with their work.
accordance with business policies defined by user manager.
IV. NATURE INSPIRED OPTIMIZATION TECHNIQUES It is a Multi-Agent system where each agent represents a
resource (Machine Agents) in a Manufacturing System. Each
Biological and natural processes have been a source of Machine Agent must be able: to find an optimal or near
inspiration for computer science and information technology. optimal local solution through Genetic Algorithms, Tabu
The interest of the Nature Inspired Optimization Search or other NIT; to deal with system dynamism (new jobs
Techniques, also named Meta-heuristics, is that they arriving, cancelled jobs, changing jobs attributes, etc.); to
converge, in general, to satisfactory solutions in an effective change/adapt the parameters of the basic algorithm according
and efficient way (computing time and implementation to the current situation; to switch from one Meta-Heuristic
effort). NIT have often been shown to be effective for algorithm to another and to cooperate with other agents.
difficult combinatorial optimization problems appearing in Scheduling approach followed by AUTODYNAGENTS
several industrial, economical, and scientific domains system is rather different from the ones found in the
[3][9][11]. Prominent examples are evolutionary algorithms, literature; as we try to implement a system where each agent
simulated annealing, tabu search, scatter search, memetic (Resource Agent) is responsible for optimizing the scheduling
algorithms, ant colony systems and particle swarm of operations for one machine through a NIT. This considers
optimization. a specific kind of social interaction that is cooperative
When considering and understanding solutions followed by problem solving (CPS), where the group of agents work
nature it is possible to use this acquired knowledge on the together to achieve a good solution for the problem.
resolution of complex problems on different domains. From The original Scheduling problem defined in [7][9], is
this knowledge application, on a creative way, arise new decomposed into a series of Single Machine Scheduling
computing science areas - Bionic Computing. Problems (SMSP)[9]. The Machine Agents (which has an
The complexity of current computer systems has led the NIT associated) obtain local solutions and later cooperate in
software engineering, distributed systems and management
 SELF-OPTIMIZATION FOR DYNAMIC SCHEDULING IN MANUFACTURING SYSTEMS 423

order to overcome inter-agent constraints and achieve a detect changes occurred in the schedule, allowing the system
global schedule. to a dynamic adaptation. With this agent, the system will be
Two possible approaches, to deal with this problem, could prepared to automatically handle dynamism by adapting the
be used. In the first, the AUTODYNAGENTS system waits solutions to external perturbations. While, on one hand,
for the solutions obtained by the machine agents and then partial events only require a redefinition of job’s attributes
apply a repair mechanism to shift some operations in the and re-evaluation of the objective function, on other hand,
generated schedules till a feasible solution is obtained (Repair total events require changes on the solution’s structure and
Approach). In the second, a coordination mechanism is size, carried out by insertion or deletion of operations, and
established between related agents in the process, in order to also re-evaluation of the objective function. Therefore, under
interact with each other to pursuit common objective through total events, the modification of the current solution is
cooperation. These coordination mechanism must be prepared imperative, through job arrival integration mechanisms
to accept agents subjected to dynamism (new jobs arriving, (when a new job arrives to be processed), job elimination
cancelled jobs, changing jobs attributes). mechanisms (when a job is cancelled) and regeneration
mechanisms in order to ensure a dynamic adaptation of
population/neighborhood.
The Self-Optimizing Agent is responsible for the
automatically tuning of the meta-heuristics’ parameters,
according to the problem. This agent receives the initial
problem, or the changes detected by Self-Configuring Agent,
and automatically choose the meta-heuristic to use, and
makes its self-parameterization. If some dynamism occurs,
parameters may change in run-time. This tuning of
parameters is made through learning and experience, since it
uses a Case-based Reasoning (CBR) module. Each time a new
problem (case) appears, the CBR uses past experience in
order to specify the meta-heuristic and respective parameters
for that case. When the new case is solved, it is stored for
later use.
Fig. 1. AutoDynAgents Architecture
Finally, the Self-Healing Agent gives the capacity to the
system for diagnosing deviations from normal conditions and
The AUTODYNAGENTS architecture (Figure 1) is based
proactively takes actions to normalize them and avoid service
on six different types of agents. In order to allow a seamless
disruptions. This agent monitors other agents in order to
communication with the user, a User Interface Agent is
provide overall self-healing capabilities. Since agents may
implemented. This agent, apart from being responsible for the
crash for some reason, self-healing provides one or more
user interface, will generate the necessary Task Agents
agents backup registries in order to grant storage for the
dynamically according to the number of tasks that comprise
reactivation of lost or stuck scheduling agents with
the scheduling problem and assign each task to the respective
meaningful results, thus enabling the system to restart from a
Task Agent.
previous checkpoint as opposed to a complete reset. With this
The Task Agent will process the necessary information
agent, the system becomes stable, even if some deadlocks or
about the job. That is to say that this agent will be responsible
crashes occur.
for the generation of the earliest and latest processing times,
Rescheduling is necessary due to two classes of events:
the verification of feasible schedules and identification of
Partial events imply variability in jobs/operations attributes
constraint conflicts on each job and the decision on which
such as processing times, due dates or release times; and
Machine Agent is responsible for solving a specific conflict.
Total events imply variability in neighborhood/population
The Machine Agent is responsible for the scheduling of the
structure, resulting from new job arrivals, job cancellations,
operations that require processing in the machine supervised
machines breakdown, etc.
by the agent. This agent will implement meta-heuristic and
local search procedures in order to find best possible VI. SELF-OPTIMIZING MODULE
operation schedules and will communicate those solutions to The objective is that each machine agent adopt and
the Task Agent for later feasibility check. provides self-parameterization of the solving method in
Respectively to the Self-* Agents, the Self-Configuring accordance with the problem being solved (parameters can
Agent is responsible for monitoring the system in order to change in run-time).
424 MADUREIRA AND PEREIRA

Each machine agent must be able to define which NIT will the use of a CBR system.
be used and define initial parameters of NIT according to the The CBR is an artificial intelligence methodology which
current situation or even to commutate from one algorithm to aims to solve new problems by using information about the
the other according to current state and previous learning. solutions to previous similar problems [5]. It operates under
Agents self-optimize through learning and experience. the premise that similar problems require similar solutions.
Nature Inspired Optimization Techniques like Tabu The CBR consists in a cycle (Figure 2), usually described as
Search, Genetic Algorithms, Simulated Annealing, etc., are the ‘4 Rs’ [1]:
very useful to obtain good solutions in feasible execution 1. Retrieve the most similar case or cases;
times. Sometimes they can even obtain the optimal solution. 2. Reuse the information and knowledge retrieved to
But to be possible to obtain optimal or near-optimal solutions solve the problem;
it is required the correctly tuning of the parameters from the 3. Revise the proposed solution;
different meta-heuristics. It was proved to be a very hard task 4. Retain the revised solution to be useful for future
since it needs some expertise knowledge from the meta- problem solving.
heuristic in use and from the concerned problem. Sometimes There are two approaches to use the CBR in our system.
it needs the trial-error method and the difficulties increase The first one consists in retrieving the most similar case or
when there are more than one meta-heuristic to be used, cases to the new problem, regardless the meta-heuristic to
because it requires an a priori choice of the meta-heuristic use. Therefore, it is retrieved the case(s) containing the meta-
and then the tuning of its parameters. heuristic and its parameters to use. The second approach
consists in choose the meta-heuristic a priori and then
 
retrieve the most similar cases, containing the parameters to
    use for that meta-heuristic. The first approach is more
appropriate to the problem in question since that is important
for the system to decide which meta-heuristic to use and the
     respective parameters, because not all meta-heuristics are
suitable to every types of problems. Consequently, it will be
  
possible to know, for example, which meta-heuristics are
       more suitable to a particular type of problem.

  
A. CBR Architecture
The architecture of the proposed CBR system is presented
 
in Figure 2. Initially, every new problem leads to a new case
for the system. In the first phase of the ‘4REs’ cycle, the
 
previous most similar cases are retrieved from the database.
 As it is possible to see in Figure 3, the retrieved cases are
compared with the new case and two things can happen. If
Fig. 2. CBR system architecture. there is a very low similarity between them, the retrieved case
will consist in a set of pre-defined parameters, tuned from
To solve this problem, we present a module, described in expertise knowledge. If not, the most similar case is retrieved.
this section, representing the third part from Autonomic
Computing Self-CHOP, named Self-Optimizing. Kephart and
Chess [4] defined it as the capacity of “monitor, experiment
with, and tune their own parameters and learn to make
appropriate choices about keeping functions or outsourcing
them”. With this definition we propose an embedded module
capable of monitoring the system and tuning the parameters
of the different meta-heuristics, with respect to each new
single problem emerging on the system.
This module needs to know how to tune up the different
parameters of every meta-heuristic. It is impossible to predict Fig. 3. CBR Retrieve phase
every single problem to treat, so the system must be capable
of learning about its experience during lifetime, e.g., as After being recovered, the case can be reused becoming in
humans do. To perform this learning mechanism, we propose a suggested solution. After this, and in the Revise phase, the
 SELF-OPTIMIZATION FOR DYNAMIC SCHEDULING IN MANUFACTURING SYSTEMS 425

CBR system executes the problem in the MAS, using the the parameters of the cases for tuning the respective NIT.
given solution. To escape from local optimal solutions and The CBR table, as referred before, saves the attributes of
stagnation, we propose the use of some controlled every single case in the CBR system. The fields are described
randomness in the parameters of the proposed solution. After in Table I.
the MAS execution end, the case are confirmed or not as a
good solution, and if so, it is retained on the database as a Table II – TS table fields description
new learned case, for future use. Field Data type Description
This represents the lifecycle of the proposed CBR system in NeighGen Double Neighborhood generation percentage
the Self-Optimizing module, as each new problem will be SubNeigh Double Sub-neighborhood percentage
TabuListLen Integer Tabu list length
evaluated and become a case of the casebase. We believe that StopCrit Integer Stopping criteria, number of iterations
the system will be capable of learning and evolving in the
parameterization of the meta-heuristics, adapting to every Table III – GA table fields description
new problem that may appear and improving the existing Field Data type Description
problems. InitPopGen Double Initial population generation
percentage
PopSize Double Population size percentage
NumGen Integer Number of generations
CrossRate Double Crossover rate
MutRate Double Mutation rate

Table IV – SA table fields description

Field Data type Description
InitTemp Double Initial temperature
Alpha Double Alpha factor of temperature
reduction

NumIteraK Integer Number of iterations at the same

Fig. 4. CBR casebase temperature

StopCrit Integer Stopping criteria, number of

iterations
B. Casebase
In Figure 4 it is presented the CBR casebase. It consists in
six tables, one for each meta-heuristic in use and one for As example we present in Table II, III, IV, V and VI the
saving the attributes of each case in the CBR. descriptions of the fields of each meta-heuristic, each field
corresponding to each parameter tuned by the CBR system.
Table I. CBR table fields description
Table V – ACO table fields description
Field Data type Description
Field Data type Description
ID Integer Primary key of the case, auto-incremented
MH String Name of the table of the meta-heuristic used NumCol Integer Number of colonies
NJobs Integer Number of jobs of the problem (e.g., 10) NumAnts Integer Number of ants per colony
Number of machines of the problem (e.g., EvapRate Double Pheromone evaporation rate
NMachines Integer Alpha Double Heuristic value importance
5)
The type of the problem (“single-machine”, Beta Double Pheromone importance
ProblemType String
“open-shop”, “flow-shop”, or “job-shop”) StopCrit Integer Stopping criteria, number of
Indicates if the problem has operations with iterations
MultiLevel Boolean
more than one precedent
Optimal makespan value known, if
CmaxOptimal Integer
available
Table VI – PSO table fields description
Heuristic used to calculate the initial Field Data type Description
InitSol String
solution NumPart Integer Number of particles
ObjFunc String Objective function used NumItera Integer Number of iterations
Cmax Integer Makespan value obtained from the case LowLimit Integer Lower limit
TimeExec Double Execution time, in seconds UpLimit Integer Upper limit
MinInertia Double Minimum inertia
The NIT tables are TS, GA, SA, ACO, and PSO, MaxInertia Double Maximum inertia
C1 Double Cognitive component
representing Tabu Search, Genetic Algorithms, Simulated Double Social component
C2
Annealing, Ant Colony Optimization, and Particle Swarm MinVeloc Double Minimum velocity
Optimization respectively. Each one of these tables contains MaxVeloc Double Maximum velocity
426 MADUREIRA AND PEREIRA

C. Similarity Measure ACKNOWLEDGMENT

The attributes to consider in the similarity measure are The authors would like to acknowledge FCT, FEDER,
some fields from the CBR table. The most important ones are POCTI, POCI for their support to R&D Projects and GECAD
NJobs, NMachines, ProblemType, MultiLevel, and - Knowledge Engineering and Decision Support Group Unit.
CmaxOptimal. These fields are correctly heavy in the
similarity measure. However, there are other important
attributes to consider for the selection of the most similar case
REFERENCES
to retrieve. These attributes are the Cmax value, because it is
important to know the makespan of the previous cases, in [1] Aamodt, A. and Plaza, E. Case-Based Reasoning:
order to be selected the best case from a set of similar cases, Foundational Issues, Methodological Variations, and System
Approaches, Artificial Intelligence Communications, vol. 7,
and the TimeExec value, which is important to know in those
pp. 39-52, 1994.
cases where exists very similar makespan values, in order to
[2] Aytug, H., Lawley, M. A, McKay, K., Mohan, S.&
select the more efficient case. So, with these two values we
Uzsoy, R. Executing production schedules in the face of
can correctly heavy the effectiveness-efficiency binominal. uncertainties: A review and some future directions. European
Journal of Operational Research, Volume 16 (1), 86-110,
2005.
VII. CONCLUSIONS AND FUTURE WORK [3] Gonzalez, T. Handbook of Approximation Algorithms
In the retrieving phase, when there could be more than one and Metaheuristics. Chapman&Hall/Crc Computer and
similar cases, it will be retrieved the case with the best Information Science Series, 2007.
makespan value obtained in the shortest execution time. [4] Kephart, J. and Chess, D. The Vision of Autonomic
It is expected a training phase of the CBR system, in which Computing, Computer, vol. 36, pp. 41-50, January 2003.
the system will be executed a number of times with some [5] Kolodner, J., Case-Based Reasoning, Morgan Kaufmann
knowing problems only with the objective of case generation. Publishers Inc, 1993.
Alternatively, there is the possibility to start without previous [6] Luck, M., McBurney, P., Shehory, O., Willmoth, S.
Agent Technology: Computing as Interaction. A Roadmap for
cases, using the pre-defined parameters.
Agent-Based Computing, AgentLink III, 2005.
We also predict that will be necessary to analyze and
[7] Madureira, A., Meta-Heuristics Application to
compare two different approaches of the CBR system
Scheduling in Dynamic Environments of Discrete
utilization. The first approach, consists in apply the CBR Manufacturing. PhD Dissertation. University of Minho,
system in a “global” level on MAS, optimizing the meta- Portugal, 2003 (in portuguese).
heuristics in the same way for all Resource Agents, i.e., the [8] Madureira, A., Santos, F. e Pereira, I. Self-Managing
CBR system specify the parameters before the start of MAS Agents for Dynamic Scheduling in Manufacturing,
execution. The second approach is to consider the CBR in a GECCO'2008 (Genetic and Evolut. Comput. Conference
“local” level of each Resource Agent, so each agent can 2008, Atlanta, Georgia (EUA), 2008.
optimize its parameters and/or apply a different meta- [9] Madureira, A., Santos, J., Fernandes, N. and Ramos, C.,
heuristic, according its single-machine problem. In this Proposal of a Cooperation Mechanism for Team-Work Based
second case, every problem will be single-machine, though Multi-Agent System in Dynamic Scheduling through Meta-
some Resource Agents can have different number of Heuristics, 2007 IEEE Intern. Symp. on Assembly and
operations to process, which could be beneficial to consider. Manufacturing (ISAM07), Ann Arbor (USA), 2007, pp. 233-
However, as a drawback, the makespan value to use will be 238, ISBN: 1-4244-0563-7.
the final time of processing in each machine, which can lead [10] Monostori, L., J. Váncza, Kumara, S. Agent based
the agents to fall in local optimal, minimizing their own systems for manufacturing, CIRP Annals - Manufacturing
Technology, vol. 55, no. 2, pp. 697–20, 2006.
processing time, and not the global makespan of the MAS.
[11] Xhafa, F. and Abraham A. Metaheuristics for scheduling
First, we will implement the first approach, but with the
in industrial and manufacturing applications series: Studies
intent to implement the second one, to be possible to compare
in computational intelligence, volume 128, Springer, 2008.
and analyze each approach.
We believe that the resulting system will be able to evolve
and improve the performance of the scheduling problems.
 A Decomposition Method in Modeling Queuing
Systems
Eimutis Valakevicius, Vaidotas Valiukas, Mindaugas Snipas
Department of Mathematical Research in Systems
Kaunas University of Technology
Kaunas, LT - 51368, Lithuania

Abstract-The analysis of stochastic systems is most difficult stages are obtaining the set of all the possible
notoriously hard, especially when their performance is states of a system and transition matrix between them. In
described by multidimensional Markov chains. the paper is used a method for automatic construction of
Approximation of general distribution functions by phase-type numerical models for systems described by Markov chains
distributions rise to large Markov chains. If the model is too with a countable space of states and continuous time.
large to analyze in its entirety, it is divided into To construct a model we need to describe the
subsystems. Each subsystem is analyzed separately and performance of a stochastic system in the event language
global solution is constructed from the partial solutions. [3]. It allows automating some stages of the model. The
Numerical algorithm to solve for the steady-state probabilities created software in C++ generates the set of possible states,
of these Markov chains from a system of linear equations by
decomposition method is presented in the paper. Based on
the matrix of transitions among states, constructs and
these probabilities, we can compute a wide range of relevant solves equilibrium equations to find steady state
performance characteristics, such as average number of probabilities.
customers of a certain type in the system and expected
postponement time for each customer class.
Keywords: Queuing system, Markov chain, decomposition II. APPROXIMATION OF QUEUING SYSTEM MODELS
method, steady state probabilities.
Continuous-time Markovian processes with countable
space of states are related with exponential distributions. It
I. INTRODUCTION means that transition time from one state to another must be
exponentially distributed. However, the mentioned
Complex queuing systems can arise in various practical condition often does not hold in the real life systems, so
applications, for example, telecommunication and computer additional difficulties are expected to arise in the
systems, production, logistics etc. Queuing models are investigation and modeling of such systems. To solve this
important tools for studying the performance of complex problem, an approximation of the stochastic systems by
systems, but despite the substantial queuing theory Markovian processes is used, when fictitious phase-type
literature, it is often necessary to use approximations in the method is applied [4-6]. The diagram of approximation of a
case the system is non-markovian. Use of phase-type (PH) general distribution G by phase-type distribution is shown
distributions is a common means of obtaining tractable in Fig. 1.
queuing models [1-2], but it rises to large Markov chains. The examples of approximation schemes of queuing
Markov modelling is a common approach to analyzing the systems are presented in figures 2 and 3.
performance of various stochastic systems. However, the
performance analysis of complex queuing systems has a
common source of difficulty: The Markov chain that
models the system behaviour has a state space that grows
infinitely in multiple dimensions. A method of
decomposition to analyze Markov chains of large size is
given in next sections.
It is known that creation of analytical models requires
large efforts. Use of numerical methods permits to create
models for a wider class of systems. The process of creating Fig.1. Approximation of general distribution G using phase-type
numerical models for systems described by Markov chains distribution.
consists of the following stages: 1) definition of the state of
a system; 2) creating equations describing Markov chain; 3)
computation of stationary probabilities of Markov chain; 4)
computation characteristics of the system performance. The

427
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_75, © Springer Science+Business Media B.V. 2010
428 VALAKEVICIUS ET AL.

medium priority customer is being served; 3 – if low

priority customer is being served);
n5 – indicates the state of the second server (0 – if server is
empty; 1 – if high priority customer is being served; 2 - if
medium priority customer is being served; 3 – if low
Fig. 2. Approximation of M/G/1 by mixture of Erlang distributions priority customer is being served);
The following events can occur in the system:
E = {e1 , e 2 , e3 , e 4 , e5 , e6 , e7 , e8 , e9 } ,
where
e1 – a high priority customer arrived to the system with
intensity λ ;
h

e2 – a medium priority customer arrived to the system

Fig. 3. Approximation of M/G/1 by
Coxian distributions with intensity λ m ;
e3 – a low priority customer arrived to the system with
III. CONCEPTUAL MODEL OF MULTICLASS QUEUING
intensity λ ;
l
SYSTEM
e4 –a high priority customer was served in the first
Consider a multi-class, multi-server queuing system
shared by N customer classes, numbered 1,…, N . A server with intensity μ ; h

number of class indicates the priority rank (class 1 has e5 –a medium priority customer was served in the first
highest priority and class N has the lowest priority).
Priority rule is nonpreemptive. Class i customers arrives in server with intensity μ m ;
the system according to a Poisson process with rate λi . The e6 – a low priority customer was served in the first server
service times of class i customers are exponentially
with intensity μ l ;
distributed with mean 1 / μ i . Within each class, service
discipline of customers is a First Come First Served e7 – a high priority customer was served in the second
(FCFS). Queue length of each class customers can have server with intensity μ h ;
various limitations.
e8 – a medium priority customer was served in the second
IV. NUMERICAL MODEL OF THE SYSTEM server with intensity μ m ;

Consider a queuing system with 3 priority classes – high, e9 – a low priority customer was served in the second
medium and low, and 2 servers, which we call first and server with intensity μ l ;
second. In addition, we consider limitation L ( L ∈ N ) for
summary waiting space of each customer class. The The set of transition rates between the states is the
system’s performance is described in the event language, following:
using methods from [1].
The set of system states is the following: {(
INTENS = λh , λm , λl , μ h , μ m , μ l )}
N = {(n1 , n2 , n3 , n4 , n5 )}, n1 + n2 + n3 ≤ L; n4 ≤ 3; n5 ≤ 3 , The created software generates the possible set of states
and matrix of transition probabilities.
n1 , n 2 , n3 , n 4 , n5 ∈ Z + , Calculation of steady state probabilities can require a
where large amount of calculations and computer resources. For
n1 – the number of high priority customers in the queue; example, if the set of states is described as
N = {(n1 , n 2 , n3 , n 4 , n5 )}, n1 + n 2 + n3 ≤ L; n 4 ≤ 3; n5 ≤ 3 ,
n 2 – the number of medium priority customers in the it is easy to prove that the total number of states equal to
queue; ( L + 1)(L + 2 )(L + 3)
⋅ 4 ⋅ 4 = ( L + 1)( L + 2)( L + 3) .
8
N =
n 3 – the number of low priority customers in the queue; 3! 3
For example, if L = 10 then system has 4576² states!
n 4 – indicates the state of the first server (0 – if server is
empty; 1 – if high priority customer is being served; 2 - if
 A DECOMPOSITION METHOD IN MODELING QUEUING SYSTEMS 429

V. DECOMPOSITION OF THE MATRIX OF TRANSITION Stochastic matrix

PROBABILITIES of transition
probabilities P
Transition probability matrix P of Markov chain,
which gives the conditional probability p ij of making a
transition from the state i to the state , j has the form P is partitioned Each block is Global solution
into block analyzed π is constructed
structure separately
§ p11  p1 j  p1n ·
¨ ¸
¨      ¸ Fig.4. Scheme of the decomposition approach
P = ¨ pi1  pij  pin ¸
¨ ¸ The following three main steps are implemented during
¨      ¸ the partition of the matrix:
¨p  p nj  p nn ¸¹ 1. The decomposability factor γ is introduced,
© n1 −10 −1
which varies from 10 to 10 .
Let us assume that the matrix is of big dimension. So 2. All elements, less than or equal to γ are
the calculation of the stationary probability vector π from replaced by zero.
π ⋅P =π 3. The resulting matrix is treated as the
becomes difficult, because classic methods and algorithms representation of a directed graph and a search is
is not working properly due to the amount of computation initiated for strongly connected components.
and residuals. Therefore we will to apply the decomposition After all the result is the matrix P having the block
technique [9]. structure.
The algorithm of decomposition is presented in Fig.4. Let π is partitioned respectively with P:
In Markov modeling it is frequently the case that the π = (π 1 , π 2 , , π N )
state space of the model can be partitioned into disjoint
subsets, with strong interactions among the states of the
and πi is a vector of length ni . If the off-diagonal blocks
subset, but with weak interactions among the subsets are all zero, so the matrix P is completely decomposable
themselves. Such problems are sometimes referred to as and have the form
nearly completely decomposable (NCD) [7,8].
Strong interactions among the states of a group and § P11 0  0 0 ·
weak interactions among the groups themselves imply that ¨ ¸
the states of a nearly completely decomposable Markov ¨ 0 P22  0 0 ¸
chain can be ordered so that the stochastic matrix of (π1,π2,,πN)¨      ¸=(π1,π2,,πN)
transition probabilities has the form ¨ ¸
¨ 0 0  PN−1N−1 0 ¸
¨ 0 0  0 PNN ¸¹
§ P11 P12  P1N · ©
¨ ¸
¨P P22  P2 N ¸ Each πi can be found directly from
P = ¨ 21
    ¸ π i ⋅ Pii = π i .
¨ ¸
¨P PN 2  PNN ¸¹ In general case of nonzero off-diagonal blocks we
© N1
assume that the system is completely decomposable and the
stationary probabilities for each block are calculated in
in which the nonzero elements of the off-diagonal blocks
same way.
are small compared to those of the diagonal blocks. The sub
Another problem now arises. We have calculated the
blocks Pii are square and of order ni : stationary probability vector from for each block. But
N simply connecting them together will not give a probability
i = 1,2,  , N , n = ¦ ni vector for all system. Why? The elements of each vector
i =1
π i sum to 1, while sum of the vector π this condition
does not hold. We still need to weight each of the
probability sub vectors by a quantity that is equal to the
430 VALAKEVICIUS ET AL.

probability of being in that sub block of states. These

weights are given by 1. Initial 2. Calculation
approximation ofϕ
π
(π 1 , π 2 ,, π N ) , (7)

where ⋅ denotes the spectral form. 5. Calculation 3.

of the new Construction
In purpose to compute the probability of leaving block approximation of the

i and entering block j the coupling matrix A is formed: 4. Finding of

§ a11  a1 j  a1N · ξ
¨ ¸
¨      ¸
¨
A = a i1  aij  aiN ¸
Fig. 5. The KMS algorithm
¨ ¸
¨      ¸ The Takahashi algorithm holds the same scheme of
¨a finding the solution as it is shown in the Fig. 5. The only
© N1  a Nj  a NN ¸¹ difference between this algorithm and the KMS algorithm is
πi in step 5, where the solution vector π is constructed (more
where aij = Pij e = ϕ i Pij e , e = (1,1, ,1) T . on that in).
πi The software was created in order to realize this
Thus, the element a1N gives the probability that the system technique above. So let’s look how decomposition
approach works in G/M/1 and M/G/1 models. In purpose to
will enter one of the states of block N when it leaves one model these systems let’s say G is the Weibull distribution
of the states of block 1. with the distribution function
Similarly in case of the matrix P, the stochastic matrix k
­ t½ λ >0.
A possesses a unique stationary probability vector p(t ) = 1 − exp®− ¾ , t ≥ 0, k > 0,
¯ λ¿
ξ = (ξ1 , ξ 2 , , ξ N ) : For k = 0,7 , λ = 0,7 parameters of the mixture of Erlang
ξ ⋅A=ξ . distribution and Coxian distribution accordingly are:
Only this time the element ξ i is the stationary probability n = 1 , p = 0,3106 , μ1 = 0,5313 , μ 2 = 2,2869 ;
of being in block i (one of the states of block i). p = 0,2385 , μ1 = 2,2869 , μ 2 = 0,5313 .
Furthermore, after some operations it becomes apparent In Fig. 6 we see the transition probability matrix P
that ( n = 12 ) of the model G/M/1 when the distribution
function G is approximated by the mixture of Erlang
ξ = ( π 1 , π 2 ,, π N ) distribution.

Now we have solutions πi of each block and weights ξi

of being in the block i from. So the global solution of the
system can be constructed by
π = (ξ1π 1 , ξ 2π 2 ,, ξ N π N ) .
The noted steps are the basis of iterative KMS (Koury,
McAllister and Stewart) and Takahashi algorithms [3].
They rapidly converge onto the exact solution when the
Markov chain is nearly completely decomposable. Fig. 5
visually shows how the KMS algorithm is performed.

Fig. 6. Transition probability matrix of G/M/1 (G – mixture of

Erlang distribution)
 A DECOMPOSITION METHOD IN MODELING QUEUING SYSTEMS 431

There are only few elements less than 10 −1 in the

matrix (bolded in the picture), so the decomposability factor
γ has no influence any more. If these probabilities (less
than 10 −1 ) are replaced by zero as the theory expects, strong
interactions don’t exist in the model. Due to the similar
structure of the transition probability matrix in the model
G/M/1 (now G is approximated by the Coxian distribution),
there are no strongly connected components (Fig. 7).
Therefore, the first part of the decomposition method fails,
because no block structure was found during this step.
Some kind of the complement to the procedure of the
partition is expected to be done.
During the analysis was noticed that both type of
matrices have zero elements in the main diagonal. Fig. 8. Block structure in G/M/1 model (G – mixture of Erlang
Moreover, this condition remains for sub matrices which distribution)

were partitioned randomly. Elements p ii in the transition

As was mentioned above, we didn’t pay attention to the
probability matrix mean the possibility to stay at the same size of probabilities in sub matrices. But the theory requires
state. Since they all are equal to zero, every transition takes that elements of the diagonal blocks are close to 1 as
to another state. possible. Besides, the analysis of larger models showed
Considering the remarks above, we can modify the some elements in the transition probability matrix repeat
procedure of the partition. Now we will be looking for sub row by row (let’s say in Fig. 7 elements p 4,5 , p5, 6 ,
matrices where all rows and columns have at least one
nonzero element. The size of probabilities doesn’t matter p 6,7 ). These two conditions imply the fact that the
this time.
partition in small matrices can be repeated in the large ones.
After applying the modification to G/M/1 model (G is
So, the resulting strategy of the partition in the G/M/1
Coxian distribution), we have got 3 sub matrices with the
model is:
size of 4 × 4 . All three blocks hold the condition about
nonzero elements in every row and column. In G/M/1 ƒ P11 (size of 6× 6 ), P22 ( 8× 8 ), P33 ( 8 × 8 ),...,
model with the mixture of Erlang distribution the situation PNN (smaller or equal to sub matrix of size
is different. The last sub matrix has only zero elements
(Fig. 8), because it left as the residual of the procedure. In 8 × 8 ), when G is the mixture of Erlang
this case we can aggregate two last blocks into one. Now distribution.
we have got 2 sub matrices with the size of 5× 5 and ƒ P11 (size of 5× 5 ), P22 ( 6 × 6 ), P33 ( 6 × 6 ),...,
7 × 7 , where at least one element in each row and column
PNN (smaller or equal to sub matrix of size
differs from zero.
6 × 6 ), when G is Coxian distribution.
These strategies of the partition are optimal due to they give
faster convergence speed in comparison with previously
mentioned ones. In Fig. 9 the example with the transition
probability matrix of size n = 104 shows that the number
of iterations is less of the optimal strategy (line “Blocks2”)
than the first approach of the partition (line “Blocks1”). The
significant difference between the accuracy wasn’t noticed.

Fig. 7. Transition probability matrix of G/M/1 (G – Coxian

distribution)
432 VALAKEVICIUS ET AL.

TABLE 1.
1
0,1
RESIDUALS OFLARGE G/M/1 MODELS

11

13

15

17

19

21

23

25

27
1

9
0,01
0,001 Matrix size Average residual
0,0001 n KMS Takahashi
Residual

1E-05 Blocks1
1E-06 Blocks2 123 1,4055E-12 1,6567E-12
1E-07
1E-08
243 3,5846E-011 9,6571E-012
1E-09 483 6,2866E-011 7,9955E-011
1E-10
1E-11
Iteration

REFERENCES
Fig. 9. Comparison of convergence in G/M/1 model
[1] G. Mickeviius, E. Valakeviius , “Modelling of non-
Markovian queuing systems”, Technological and economic
Applying the decomposition method with the optimal development of economy, vol . XII, No 4, pp. 295-300, 2006.
strategy of the partition into the block structure, the [2] T. Osogami , M. Harchol-Balter., “Necessary and Sufficient
approximate stationary probabilities of large G/M/1 models Conditions for Representing General Distributions by Coxians”,
School of Computer Science, Carnegie Mellon University,
were computed. The results require a lot space in the paper, 2003.
so Table 1 shows average residuals. The accuracy (from [3] E. Valakevicius, H. Pranevicius, “An algorithm for creating
Markovian Models of Complex System”, In Proceedings of the
10 −12 to 10 −10 ) remained as in models, where the size of 12th World Multi-Conference on Systemics, Cybernetics and
the transition probability matrix was around 15. The Informatics, Orlando, USA, June-July 2008, pp. 258-262.
number of iterations taken to reach this accuracy varies [4] M.A. Johnson, “An empirical study of queuing
depending on the size of the matrix, but there were no cases approximations based on phase-type distributions”, Commun.
Statist.-Stochastic Models, vol. 9, No. 4, pp. 531-561, 1993.
when algorithms exceeded the number of 30 iterations. [5] J.A. Luhman, M.J. Johnson, “Behaviour of queuing
approximations based on sample moments”, Applied Stochastic
Models and Data analysis, vol. 10, pp. 233-246, 1994.
VI. SUMMARY [6] M. J. Faddy, “Examples of fitting structured phase-type
distribution”, Applied Stochastic Models and Data Analysis,
vol. 10, pp. 247-255, 1994.
At the beginning of this paper we mentioned not only [7] S.W. Fuhrmann, R.B. Cooper, “ Stochastic decompositions in
G/M/1, but also M/G/1 models (accordingly G is Coxian the M/G/1 queue with generalized vacations”, Operations
distribution or the mixture of Erlang distribution). But the Research, vol. 33(5), pp.1117-1129,1985.
[8] A. Horváth,, G. Horváth, and M. Telek., “A traffic based
analysis of these models showed that the different structure decomposition of two-class queueing networks with priority
of the transition probability matrix doesn’t let partitioning service”, Computer Networks, vol. 53, pp.1235-1248, 2009.
into sub matrices. After applying the procedure of the [9] W. J. Stewart, Introduction to the Numerical Solution of
partition (even with simple condition – each row and Markov Chains”,.Princeton University Press, Princeton, New
Jersey, 1994.
column must have at least one nonzero element), resulting
matrix is the same as primary transition probability matrix.
Queuing systems requires more investigation in the
future.
Modular Design and Structure for a Mobile Sensory
Platform
Ayssam Elkady, Vipul Babariya, Jovin Joy and Tarek Sobh
School of Engineering
University of Bridgeport

Abstract—A mobile manipulator is a manipulator mounted

on a mobile platform with no support from the ground. We
are already in the process of building a platform (RISCbot II)
which consists of a comprehensive sensor suite and significant
end-effector capabilities for manipulation. In order to reduce
the uncertainty in localization, sensor fusion is used to create
an efficient and effective user interface to facilitate teleoperation
by enhancing the quality of information that is provided to the
teleoperator.
This paper presents the modular design process of the RISCbot
II mobile manipulator. In the design process, the overall design of
the system is discussed and then the control process of the robot
is presented. Furthermore, the tasks that the RISCbot II can
perform such as teleoperation, navigation, obstacle avoidance,
manipulation, face detection and recognition, and map building
are described.

I. I NTRODUCTION
A mobile manipulator offers the dual advantage of mo-
bility offered by the platform, and dexterity offered by the
manipulator. The mobile platform extends the workspace of
the manipulator. We are developing and constructing a mobile
manipulation platform called RISCbot II. The RISCbot II mo-
bile manipulator platform is shown in figure 1. Mobile manip-
ulators are potentially useful in dangerous and unpredictable
environments such as construction sites, space, underwater,
service environments, and in nuclear power stations.
Sensor fusion has been an active area of research in the field
of computer vision and mobile robotics. Sensor fusion is the Fig. 1. The RISCbot II mobile manipulator platform.
combination of sensory data from different sources, resulting
in better and more reliable information of the environment than
data derived from any individual sensor. Sensor fusion algo-
rithms are useful in low-cost mobile robot applications where
results can be merged by fusing the data from multiple sensors,
acceptable performance and reliability is desired, given a lim-
and presenting the result in a way that enables the tele-operator
ited set of inexpensive sensors like ultrasonic and infrared sen-
to perceive the current situation quickly. Sensor fusion is used
sors. Depending on the modalities of the sensors, sensor fusion
to reduce the workload for the operator to enable him or her
can be categorized into two classes (as described in [1]): sensor
to concentrate on the task itself. Sensor fusion is commonly
fusion using complimentary sensors, and sensor fusion using
used to reduce uncertainty in localization, obstacle avoidance,
competing sensors. Complementary sensors consist of sensors
and map building. Furthermore, sensor fusion may be used
with different modalities, such as a combination of a laser
to improve teleoperation by creating user interfaces which
sensor and a digital camera. In contrast, competing sensors are
efficiently facilitate understanding of remote environments and
composed of sensors suit which have the same modality, such
improve situational awareness.
as two digital cameras which provide photographic images of
the same building from two different viewpoints. Sensor fusion In this paper we describe our mobile manipulation platform,
has some critical problems, including the synchronization of and introduce the tasks that the RISCbot II is performing them
sensors. Different sensors have different resolutions and frame in, specifically teleoperation, navigation, obstacle avoidance,
rates, so the sensors need to be synchronized before their manipulation, face detection, and face recognition.

433
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_76, © Springer Science+Business Media B.V. 2010
434 ELKADY ET AL.

Fig. 2. The Prototype Structure.

II. D ESIGN STRUCTURE OF THE RISC BOT II

In the final design as shown in figure 2, the three layer
design was maintained. The base comprises of plywood hold-
ing a square core iron structure to provide support for the
manipulator. Cylindrical structure to mount sensors is fixed
on the central layer. The topmost layer is used to mount a
stereo camera and a manipulator. The manipulator is mounted
in such a way as to transfer its weight to the square core iron
from the base.
Furthermore, there is a 2” diameter hole being drilled in
the bottom layer to pass the wirings from the battery. The
joystick of the wheelchair is mounted on the core iron structure
Fig. 3. Different views of the prototype structure.
vertically. There are seven sensors mounted on the back face
of the bottom plywood and the center of all sensors is kept in
one line. Different views of the structure are shown in figure
3. Plywood was chosen in the final design for building the
layered structure due to its cost effectiveness and availability.

A. Individual Parts Description

A summary of individual parts used in the final prototype
is described in table I.
TABLE I
I NDIVIDUAL PARTS USED IN THE FINAL PROTOTYPE .

Part Material Type Quantity Fig. 4. Bottom plywood sheet.

Top Plywood Sheet Plywood 1
Middle Plywood Sheet Plywood 1
Bottom Plywood Sheet Plywood 1
Central Support Structure Core Iron 1 shown in figure 5. This plywood sheet is directly mounted on
Support Part Core Iron 4
Cylindrical Part Wood 1 six bosses extending from the wheelchair frame. Two I-support
Backward Sensor Part Wood 3 and Central support structure are mounted on top of the sheet.
2) Middle Plywood sheet: This plywood sheet shown in
1) Bottom Plywood sheet: The bottom plywood sheet used figure 6 is mounted on the square brackets of I-support and
has a dimension of 32” × 22” × 0.75” as shown in figure 4. In central support structure extending out of the bottom. Its
order to mount all components, thickness of the plywood sheet dimensions are 16”×22”×0.75”. This sheet is used to support
was set at 0.75”. Front pocket is used to mount the laptop is the cylindrical sensor mount.
 MODULAR DESIGN AND STRUCTURE FOR A MOBILE SENSORY PLATFORM 435

Fig. 5. The prototype showing structural support.

Fig. 7. I-Support Part.

Fig. 6. Middle Plywood sheet.

3) I-Support Part: It is manufactured from core iron of 0.5”

Fig. 8. The prototype showing structural support without Cylindrical Sensor
diameter with square brackets welded on each end as shown Mount.
in figure 7. Four of these were used in the design; two of
them are used for supporting the middle plywood sheet and
the others are mounted on the middle plywood and is located 6) Top Plywood sheet: The top plywood sheet, shown in
in the Cylindrical Sensor mount to support the top plywood figure 11 is mounted on top of the central support structure. It
as shown in figure 8. is also supported by two I-support extended from the middle
4) Cylindrical Sensor Mount: This structure is assembled plywood, cylindrical sensor mount, and the rear sensor rack.
from 9 rectangular wooden plates arranged around a incom- This sheet supports the manipulator and stereo camera.
plete circle (1.25π) of radius 9.5”. Each rectangular plate is
2.75” thick and mounts 2 type of sensors, namely infrared and B. Analysis Of the final Structure
ultrasonic, as shown in figure 9. The analysis of the structure was done on ANSYS Work-
5) Central Support Structure: This structure is a rectan- bench 11.0. Static structural analysis was done to check stress,
gular hollow frame made from core iron of 1” thickness. In strain, and deformation of the structure. The procedure used
the middle, an angle bracket is welded to support the middle is described below:
plywood sheet. This support structure rests on the bottom 1) Import the final prototype model designed using Pro-
plywood sheet and it provides support for the top plywood Engineer software into ANSYS.
and manipulator. 2) Defining the material properties of each individual part
436 ELKADY ET AL.

Fig. 9. Cylindrical Sensor Mount with rectangular plates on perimeter. Fig. 12. Structure Meshing

Fig. 13. Equivalent stress distribution throughout structure.

3) Assign meshes to the structure (figure 12)

4) Choose various support conditions and load application
on the structure
5) Assign all the forces that affect the structure
The stress distribution in the structure is shown figure 13
Fig. 10. Central Support Structure.
and the maximum stress of 2.0086 kPa is measured on the
edges of plywood. Strain generated in all the assembled com-
ponents can be observed in figure 14. Total deformation found
throughout the structure from the simulation was uniform as
shown in figure 15. The maximum equivalent Elastic Strain
and Total deformation found is negligible.

III. A PPLICATIONS
A. Face Detection
Efficiency of any detection process can be improved dra-
matically if the detection feature encodes some information
Fig. 11. Top Plywood sheet.
about the class that is to be detected. Our primary method
 MODULAR DESIGN AND STRUCTURE FOR A MOBILE SENSORY PLATFORM 437

Fig. 16. Overview of Face Recognition Module

B. Face Recognition
Face Recognition is a task that identifies or verifies a
Fig. 14. Equivalent elastic Stain simulation result.
person from a digital image or a video source. The primary
method of face detection uses a overlapping energy histogram
as described in Face Recognition by Overlapping Energy
Histograms [3]. For the sake of efficiency, OpenCV library
was used extensively while developing the code.
1) General Approach: Face Recognition algorithm is a
two step process with a training stage and search stage. The
general approach is shown in figure 16. The training stage
is done initially to create a database which is not computed
again as it is a time consuming process. The search phase
uses the database created from the previous stage to look up
unknown images. Overlapping energy Histogram comprises of
taking DCT coefficients of every 8x8 pixel values with a 75
percent overlap as shown in figure 17 (i.e Taking DCT for
every 8x8 block after every 2 pixels). A sample of which is
shown in Figure 18. Once the image has been transformed
into a histogram Euclidean Distances as show in Eqn 1
is computed, where Ω is the feature vectors calculated via
the histogram. Further Euclidean Distance has been used in
Fig. 15. Total deformation.
many face recognition techniques [3]. This distance is used
to calcualte the Database Threshold(θ), which is used for
identifying and rejecting images at the search stage
of detection is based on Haar-Link Features (as described in
[2]) and OpenCV computer vision library. Haar encodes the 2
n = Ω − Ωn  (1)
existence of oriented contrasts between regions in images. A
set of features can be used to encode the contrast exhibited by
human faces and their special relationships. C. Feature Extraction
Our detection program is derived primarily from OpenCV Once the image with a face is transformed using overlapping
and uses classifiers (cascade of boosted classifiers working DCT procedure values of Feature Set F2,
with haar-like features) that are trained with a few hundred
sample views of human faces. Classifiers used in our project F2 = [DC; AC01 ; AC10 ; AC11 ]
were obtained from OpenCV library. The classifier is built
using positive and negative examples of the same size. After a ,is put into a histogram with a bin size of 60. The real
classifier is trained it can be used to search across an image at challenge has been in selecting the appropriate appropriate
locations. The output of the classifier is 1 if it finds an object bin sizes for calculating threshold values. A general deviation
(face) in a region. The classifier is resized, rather than the for the algorithm was using a predetermined bin size of 60
image, to detect objects (faces) of different sizes. Therefore, which proved to be optimally good, both computationally and
to find faces of unknown sizes, the scan procedure is done for detection. This was determined by running a series of test
several times at different scales. on known images at the time of development.
438 ELKADY ET AL.

Fig. 19. Face Detection Module Detecting Multiple People in a Single Frame

Fig. 17. Energy Histogram With 75 Percent Overlap

Fig. 20. Face Recognition Module Output

with the closest threshold value is usually the correct person.

3) Face Detection/Recognition Testing: Face Detection and
Fig. 18. Image and Corresponding Overlapping DCT Representation Recognition modules has to be tested simultaneously with the
input of Detection Program fed into the recognition program.
The Face Detection program on an average detects up to 5
1) Threshold Selection: Threshold for the image database faces real time(30 Frames/Second) running on a Dual Core In-
is calculated by intra and inter class information gained from tel Processor, therefor bringing the total to 150 Images/Second.
the training dataset. Intra class(D) information is obtained This is a 10 fold increase from the original Viola-Jones
via computing the Euclidean distance between the images of implementation. Figure 19 shows a sample output from the
an individual and inter class(P) information is obtained via detection program. The main concern regarding the number
computing the distance between images of an individual with of images that can be detected per frame is the computational
others in the database. Once this is done for all the images in requirement and the need to maintain real time performance.
the database the Threshold(θ) is defined by Recognition module on the other hand can take each of
the detected faces and search the database to find possible
Dmax + Pmin
θ= (2) matches. The initialization process of the Face Recognition
2 database was found to be a processor hog, hence plans to re-
2) Image Search: Once the required feature from the image compute the database values at run time had to be abandoned.
to be recognized is extracted, it’s features are cross computed Another bottleneck noticed was the total memory requirements
with each individual in the database to see if it falls withing the for the database, which increased due to storing the feature
range of the database threshold. If more than one individual vectors in uncompressed formats in system memory. Output
matches with the image that is searched for, the individual of face recognition module can be seen in Figure 20.
 MODULAR DESIGN AND STRUCTURE FOR A MOBILE SENSORY PLATFORM 439

to the design and analysis of a robot manipulator. The ma-

nipulability is the ability to move in arbitrary directions.
The manipulability index is considered as a quantitative and
performance measure of the ability for realizing some tasks.
This measure should be taken into consideration in the design
phase of a serial robot and also in the design of control
algorithms.
In [4], we presented a new method for measuring the
manipulability index, and then justified this concept by visu-
alizing the bands of this index resulting from our experiments
implemented on different manipulators, such as the Puma
560 manipulator, a six DOF manipulator, and the Mitsubishi
Movemaster manipulator. In fixed-base serial manipulators,
manipulability depends on link lengths, joint types, joint
motion limits and the structure of the manipulator. In mobile
manipulators, the manipulability depends on the kinematics,
geometric design, the payload, and the mass and mass distribu-
tion of the mobile platform. Thus, the manipulability measure
in mobile manipulators is very complicated due to the coupling
between the kinematic parameters and the dynamics effect.
Fig. 21. RISCbot II website Architecture.
Furthermore, we use the proposed method for measuring the
manipulability index in serial manipulators to generalize the
standard definition of the manipulability index in the case of
mobile manipulators.

F. Navigation and Obstacle Avoidance

A prerequisite task for an autonomous mobile robot is the
ability to detect and avoid obstacles given real-time sensor
readings. Given partial knowledge about the robot’s environ-
ment and a goal position or a series of positions, navigation
encompasses the ability of the robot to act based on its
knowledge and sensor values so as to reach its goal positions
as efficiently and reliably as possible. The techniques used in
the detection of obstacles may vary according to the nature
of the obstacle. The resulting robot motion is a function of
Fig. 22. Web-based manual control.
both the robot’s sensor readings and its goal position. The
obstacle avoidance application focuses on changing the robot’s
trajectory as informed by sensors during robot motion. The
D. Teleoperation
obstacle avoidance algorithms that are commonly used can be
We have implemented a web-based application in order summarized as the following [5]: the bug algorithm, tangent
to control the RISCbot II. The framework is implemented Bug, Artificial Potential Fields, and Vector Field Histogram.
with ASP .NET 2.0 in C#. The Web Form communicates
with the physical robot using Web Services which are also G. Path planning and map building
implemented in C#. In this application, a graphical user Given a map and a goal location, path planning involves
interface is provided to RISCbot’s teleoperator to allow him identifying a trajectory that will bring the robot from the
to interact with the robot. There are two modes to control the initial location to the goal location. During execution, the robot
robot; autonomous mode and manual mode. Since the RISCbot must react to unforeseen obstacles and still reach its goal.
II could carry out a number of tasks, this web-based console In the navigation problem, the requirement is to know the
allows the user to link the video/sensor data from the robot positions of the mobile robot and a map of the environment
with various computer vision applications. The architecture of (or an estimated map). The related problem is when both the
the web-based application for RISCbot II is shown in figure position of the mobile robot and the map are not known. In
21 and the manual control mode is shown in figure 22. this scenario, The robot starts in an unknown location in an
unknown environment and proceeds to gradually build the
E. Manipulability map of the existing environment. In this case, the position
Studying the performance characteristics of the robot such of the robot and the map estimation are highly correlated.
as dexterity, manipulability, and accuracy is very important This problem is known as Simultaneous Localization and
440 ELKADY ET AL.

Map Building (SLAM) ([6] and [7]). SLAM is the process of

concurrently building a feature based map of the environment
and using this map to get an estimation of the location of the
mobile platform.
In [8], the recent Radio Frequency Identification (RFID)
was used to improve the localization of mobile robots. This
research studied the problem of localizing RFID tags with a
mobile robot that is equipped with a pair of RFID antennas.
Furthermore, a probabilistic measurement model for RFID
readers was presented in order to accurately localize RFID
tags in the environment.

IV. I MPLEMENTATION AND R ESULTS

We are developing and constructing the mobile manipulator
platform called RISCbot II( as shown in figure 1). The Fig. 23. Working Envelope of EZ0 sonar Sensors.
RISCbot II mobile manipulator has been designed to support
our research in algorithms and control for an autonomous
mobile manipulator. The objective is to build a hardware
platform with redundant kinematic degrees of freedom, a
comprehensive sensor suite, and significant end-effector capa-
bilities for manipulation. The RISCbot II platform differs from
any related robotic platforms because its mobile platform is a
wheelchair base. Thus, the RISCbot II has the advantages of
the wheelchair such as high payload, high speed motor pack-
age (the top speed of the wheelchair is 6 mph), Active-Trac and
rear caster suspension for outstanding outdoor performance,
and adjustable front anti-tips to meet terrain challenges.
In order to use the wheelchair as a mobile platform , a
reverse engineering process has been used to understand the
communication between the joystick of the wheelchair and the
motor controller. This process was done by intercepting the Fig. 24. A close-up view of the sonar and infrared sensors array.
continuous stream of voltages generated by the joystick after
opening the joystick module and reading the signals within
joystick wires that send the signals to the wheelchair controller. A main drawback of these sensors is that they can only
We used different types of sensors so that the RISCbot II accurately measure obstacle distances within a range of 0.1m
can perceive its environment with better accuracy. Our robot to 0.8 m. Another drawback of these sensors is that they are
hosts an array of 13 LV −M axSonar R
−EZ0T M ultrasonic susceptible to inaccuracies due to outdoor light interference as
sensors. The working envelope of the 13 sonars is shown well as an obstacle’s color or reflectivity characteristics which
in figure 23 The sensors are suitable for obstacle avoidance can be seriously affected by windows and metallic surfaces.
applications but their wide beams are unable to distinguish Note that since our sonar and infrared sensors are in fixed
features within the beam angle, making sonars a poor choice of positions, our experiments concentrated on performing data
sensor for fine feature extraction within indoor environments. fusion on data obtained from a particular fixed height in
This resolution problem is magnified for objects further away the environment. In this project, sonar and infrared sensors
from the robot (i.e., objects appearing at the wide end of the are used together in a complementary fashion, where the
beam). Lastly, our robot is also equipped with an array of 11 advantages of one compensate for the disadvantages of the
Sharp GP20A21YK infrared proximity sensors above the sonar other.
ring. The sonar and infrared sensors were mounted together As shown in figure 25, the RISCbot II software which is
so that their beams are oriented in the same direction. The written in Visual C# and runs on a laptop reads the values of
configuration of sonar and infrared sensors is shown in figure all sensors at a rate of 10 HZ gathered in the data acquisition.
24. These sensors allow the RISCbot II to obtain a set of The RISCbot II software maps the sensory inputs to a series
observations and provide these observations to the controller of actions which are used to achieve the required task. Based
and higher decision making mechanisms. The controller acts on the used algorithm, the RISCbot II software responds to the
upon the received observations to cause the robot to turn in sensor data by generating streams of voltages corresponding to
the correct direction. The Integration of these modules together the joystick signals to the wheelchair controller. These voltages
constitutes an intelligent mobile robot. control the direction and the speed of the wheelchair causing
 MODULAR DESIGN AND STRUCTURE FOR A MOBILE SENSORY PLATFORM 441

tion, and face recognition. There are great benefits in using

a mobile manipulator in dangerous, inaccessible and toxic
environments. In teleoperation, a human operator controls the
RISCbot II from a distance. The teleoperator has some type
of display and control mechanisms, and the RISCbot II has
Fig. 25. The components of the RISCbot II system. sensors which gather all the information about the remote
environment, an end-effector, and mobility.
In our anticipated future work, there will be an ongoing
the RISCbot II to turn in the desired direction. effort for the development of multiple mobile manipulation
The experimental result indicates that the RISCbot II can systems and platforms which interact with each other to
detect unknown obstacles, and avoid collisions while simul- perform more complex tasks exhibiting intelligent behaviors
taneously steering from the initial position towards the target utilizing the proposed manipulability measure.
position.
R EFERENCES
V. C ONCLUSIONS AND FUTURE WORK [1] A. Yilmaz, “Sensor fusion in computer vision,” in Urban Remote Sensing
In this paper, the mobile manipulation platform RISCbot Joint Event, April 2007, pp. 1–5.
[2] P. Viola and M. Jones, “Rapid object detection using a boosted cascade
II has been presented. The RISCbot II platform differs from of simple features,” in Proc. IEEE Computer Society Conference on
any other robotic platform because its mo-bile platform is Computer Vision and Pattern Recognition CVPR 2001, vol. 1, 2001, pp.
a wheelchair base. Thus, the RISCbot II has the advantages I–511–I–518.
[3] R. Tjahyadi, W. Liu, S. An, and S. Venkatesh, “Face recognition via
of the wheelchair. Furthermore, the RISCbot II consists of a the overlapping energy histogram,” in IJCAI, 2007. [Online]. Available:
comprehensive sensor suite, and significant end-effector capa- http://www.ijcai.org/papers07/Papers/IJCAI07-465.pdf
bilities for manipulation. In addition, we have used infrared [4] T. S. M. Mohammed, A. ElKady, “New concept in optimizing manip-
ulability index of serial manipulators, using svd method,” in Industrial
and sonar sensors to monitor if any type of obstruction is in Electronics, Technology and Automation (IETA 07), Dec 2007.
the path of the robot. This research aspires to find real-time [5] R. Siegwart and I. R. Nourbakhsh, Introduction to Autonomous Mobile
collision-free trajectories for mobile manipulation platforms in Robots, 1st ed., ser. Intelligent Robotics and Autonomous Agents series.
The MIT Press, April 2004.
an unknown static or dynamic environment containing some [6] S. C. H. D.-W. M.W.M.G. Dissanayake, P.Newman and M. Csorba, “A so-
obstacles, between a start and a goal configuration. lution to the simultaneous localization and map building (slam)problem,”
Path planning for mobile robots is one of the key issues in IEEE Transactions on Robotics and Automation, vol. 17, no. 3, pp. 229
– 241, Jun 2001.
robotics research that helps a mobile robot find a collision-free [7] J. E. Guivant and E. M. Nebot, “Optimization of the simultaneous
path from the beginning to the target position in the presence localization and map-building algorithm for real-time implementation,”
of obstacles. Furthermore, it deals with the uncertainties in IEEE Transactions on Robotics and Automation, vol. 17, no. 3, pp. 242
– 257, Jun 2001.
sensor data. [8] D. F. K. F. Dirk Hähnel, Wolfram Burgard and M. Philipose, “Mapping
The objective for this project is to implement a general and localization with rfid technology,” in IEEE International Conference
purpose mobile manipulator that can be used for various appli- on Robotics and Automation (ICRA 2004), vol. 1, New Orleans. LA, April
2004, pp. 1015 – 1020.
cations such as teleoperation, navigation, obstacle avoidance,
manipulation, 3-D reconstruction, map building, face detec-
 wradvs: A Didactic Server for
IPv6 Stateless Autoconfiguration
Gabriel Abadi, Francisco Jimenez, and Eric Gamess
School of Computer Science
Central University of Venezuela
Caracas, Venezuela
[email protected], [email protected], [email protected]

Abstract — In this paper we present wradvs, a free open source graphical user interface, assists users during the configuration
application developed under the GNU General Public License process with tooltips, and validates all the submitted
that provides a comprehensive solution for IPv6 stateless information. Additionally, wradvs provides an Event Log
autoconfiguration. The application works as a service and Viewer with detailed information of the messages that are sent
implements RFCs 2460, 4861, 5006, 4191, 3775, and 3963. It also and received in the autoconfiguration process. Even if the main
has an event log viewer that records detailed information of all goal of the application is to support the teaching and learning
events in real time, allowing users to troubleshoot network process, it can be very useful for a network administrator that
problems without the need of additional tools. The main goal of uses IPv6 servers based on Windows. The rest of the paper is
wradvs is to be used as a didactic application in network
organized as follows: the IPv6 Router Advertisement message
advanced courses at Central University of Venezuela. Thus, it has
a friendly graphical user interface and presents a lot of valuable
is presented in Section II; IPv6 stateless autoconfiguration is
information to users while they configure all the parameters and discussed in Section III; related works are viewed in Section
options of the Router Advertisement messages. IV; wradvs is presented and justified in Section V; and finally
Section VI concludes the paper.
Keywords: ipv6; stateless autoconfiguration; router
advertisement; windows; service; daemon; didactic application. II. ROUTER ADVERTISEMENT
A Router Advertisement (RA) message is an Internet
I. INTRODUCTION Control Message Protocol version 6 (ICMPv6) message
Due to the imminent IPv4 address space depletion, defined by the Neighbor Discovery (ND) protocol [2].
LACNIC (Latin American and Caribbean Internet Addresses IPv6 routers send unsolicited RA messages pseudo-
Registry) is encouraging the adoption of IPv6 [5][13] as soon periodically and solicited RA messages in response to the
as possible. However, this adoption has been slow in receipt of a Router Solicitation (RS) message. The interval
Venezuela and in many other countries. One of the reasons for between unsolicited advertisements is randomized to reduce
the delay is the lack of IPv6 network specialists. Therefore, the synchronization issues when there are multiple advertising
training of IPv6 specialists has become an important issue. In routers on a link. RA messages contain the information
the undergraduate program of Computer Science at Central required by hosts to determine default gateways, the link
University of Venezuela (in Spanish: Universidad Central de prefixes, the link MTU, specific routes, home agent
Venezuela), some courses have been upgraded or added to the information, recursive DNS servers, whether or not to use
curriculum to face the problem. For example, Advanced stateful autoconfiguration, and the duration for which addresses
Network Protocols (in Spanish: Protocolos Avanzados de created through stateless address autoconfiguration are valid
Redes) is a new course that was introduced to the curriculum of and preferred [5].
the undergraduate program of Computer Science in 2005. Its
objectives include the understanding of IPv6 standards such as Fig. 1 shows the structure of the RA message which
the stateless autoconfiguration process described in [1]. contains the following fields:
Some daemons or servers have been developed, by the • Type: the value of this field must be 134 and identify
community or manufacturers, to support the stateless the type of ICMPv6 message (in this case, a RA).
autoconfiguration process. However, most of these daemons do
not support all the parameters and options of the stateless • Code: the value of this field must be 0.
autoconfiguration process, and sometimes, configuration can • Checksum: stores the checksum of the ICMPv6
be very difficult and the debugging system can be very poor. message.
These deficiencies can make the teaching and learning activity
process very laborious. For these reasons, we developed • Current Hop Limit: indicates the default value of the
wradvs (Windows Router Advertisement Server). wradvs is not Hop Limit field to be put in the IPv6 header for packets
only one of the most complete daemon created for IPv6 sent by hosts that receive this RA message. A value of
stateless autoconfiguration, but it also provides a friendly 0 indicates that the default value of the Hop Limit field
is not specified by this router.

443
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_77, © Springer Science+Business Media B.V. 2010
444 ABADI ET AL.

Figure 1. Structure of the Router Advertisement message.

• Managed Address Configuration Flag (M): when set, The options that can be present in a RA message are the
it indicates that hosts receiving this RA message following:
must use a stateful configuration protocol, such as
DHCPv6 (Dynamic Host Configuration Protocol for • Source Link-Layer Address option: contains the
IPv6), to obtain addresses in addition to the link-layer address of the interface from which the
addresses that might be generated from stateless RA message was sent.
address autoconfiguration.
• MTU option: contains the recommended MTU for
• Other Stateful Configuration Flag (O): when set, it the link. Should be sent on links that have a variable
indicates that hosts receiving the RA message must MTU.
use a stateful configuration protocol, such as
DHCPv6, to obtain other configuration information. • Prefix Information option: contains the prefixes that
are on-link and/or used for stateless address
• Home Agent Flag (H): when set, it indicates that the autoconfiguration.
router sending the RA message is also functioning as
a Mobile IPv6 home agent on this link [3]. • Route Information option: contains more-specific
routes that improve the ability of hosts to pick an
• Default Router Preference (Prf): indicates whether to appropriate next hop for an off-link destination [4].
prefer this router over other default routers according
to the level of preference. Valid values in binary are: • Recursive DNS Server option: contains one or more
01 (high), 00 (medium) and 11 (low) [4]. IPv6 addresses of recursive DNS servers [6].

• Reserved: this field is reserved for future use and • Home Agent Information option: contains
must be set to 0. information specific to this router’s functionality as a
home agent [3].
• Router Lifetime: indicates the lifetime (in seconds)
of this router as a default router. A value of 0 • Advertisement Interval option: contains the interval
indicates that this router is not a default router; at which this router sends unsolicited multicast RA
however, all other information contained in the RA messages [3].
message is still valid.
III. IPV6 STATELESS ADDRESS AUTOCONFIGURATION
• Reachable Time: indicates the amount of time (in
milliseconds) that a host should assume a neighbor is The IPv6 stateless address autoconfiguration defines the
reachable after receiving a reachability confirmation. mechanism that allows a host to generate its own addresses
A value of 0 indicates that this router does not make and others network parameters using local information and
any recommendation about the reachable time. information advertised by routers in RA messages. This
mechanism requires no manual configuration of hosts. The
• Retransmission Timer: indicates the amount of time address autoconfiguration process [1] for a physical interface
(in milliseconds) that a host should wait before the of an IPv6 node is as follows:
retransmission of a Neighbor Solicitation message.
A value of 0 indicates that this router does not When an interface becomes enabled, a tentative link-local
specify the retransmission timer. address is generated based on the link-local prefix
(FE80::/64) and an interface identifier, which is commonly
derived from the link-layer address. The uniqueness of the
tentative link-local address is verified using the duplicate
address detection (DAD) algorithm. If the address is in use
by another host, the autoconfiguration stops and manual
configuration must be performed on the host.
 WRADVS: A DIDACTIC SERVER FOR IPV6 STATELESS AUTOCONFIGURATION 445

Otherwise, the host sends a RS message to request an • Event log: refers to the existence of event registry
immediate RA message, rather than waiting for the next and how clear the events registered are.
scheduled RA message.
The studied daemons were: radvd, rtadvd, in.ndpd, the
The information advertised in the RA message (Current Cisco IOS, and netsh.
Hop Limit, Reachable Time, Retransmission Timer) is set.
For each Prefix Information option sent in the RA message, Table I shows a summary of the study for all the Router
the following actions occur: Advertisement daemons. For each criterion, a grade was
given between 0 and 5 stars, where 0 stars indicate a very
1. If the On-Link flag is set to 1, the prefix is added to poor or a lack of support, and 5 stars represent an optimum
the prefix list. or efficient implementation.
2. If the Autonomous flag is set to 1, the prefix and an radvd is one of the most popular Router Advertisement
appropriate interface identifier are used to generate daemons. Most of the stateless autoconfiguration parameters
a tentative address. and options can be specified with radvd. Configuration is
done through a text file; there is no graphical user interface.
3. Duplicate address detection is used to verify the
The syntax of the configuration file is easy to read and learn.
uniqueness of the tentative address.
radvd provides an efficient and informative error handler; the
4. If the tentative address is in use, it is not assigned to depuration process shows information about the daemon’s
the interface. events. However, this information is not specific enough.
The depuration information can be saved in a log file for
A RA message could also include other options, such as: posterior analysis. We tested radvd on Debian GNU/Linux 4.
Source Link-Layer Address, MTU, Route Information,
Recursive DNS Servers, Home Agent Information, and rtadvd is another Router Advertisement daemon mostly
Advertisement Interval. used on FreeBSD and Mac OS. It supports most of the
stateless autoconfiguration parameters and options. The
IV. RELATED WORKS syntax of the configuration file (text file) is hard to read and
learn, the depuration information does not provides valuable
A Router Advertisement daemon listens to RS messages information (e.g., value of the Router Advertisement’s fields)
and sends RA messages on demand and periodically as and it does not allow a custom path for the log event. We
described in [2]. These advertisements allow any listening tested rtadvd on FreeBSD 7.0 and Mac OS X Leopard.
host to configure its addresses and some other parameters
automatically through IPv6 stateless autoconfiguration. in.ndpd was developed by SUN Microsystems for
Solaris. It is compliant with RFC 2461 which was deprecated
We studied several Router Advertisement daemons of by RFC 4861. in.ndpd just supports an incomplete set of
different operating systems to make a comparative study. stateless autoconfiguration parameters and options. The
The criteria for evaluating these daemons were: debugging process displays important and well formatted
• Installation: refers to the easiness of the installation information. It was tested on Solaris 10.
process. Some versions of the Cisco IOS (Internetwork Operating
• Configuration: refers to the easiness of the System) include a Router Advertisement daemon. This
configuration of parameters and options. daemon supports just an incomplete set of stateless
autoconfiguration parameters and options; however, it offers
• Supported parameters: takes in consideration the a very good debugging system. Since the IOS is a
amount of fully supported parameters and options. commercial product, no source code is available.
• Graphical user interface: refers to the existence of a netsh is a command-line scripting utility that allows
graphical user interface and its friendliness. (among other things) sending RA messages in a Windows
system. It does not have a graphical user interface. It only
• Error handling: refers to how specific and clear the permits to specify a few parameters and options of the
error messages are when there are troubles in the stateless autoconfiguration process. The error messages are
configuration. very poor. The source code is not available.

TABLE I. COMPARATIVE ANALYSIS RESULTS.

radvd rtadvd in.ndpd Cisco IOS netsh

Installation
Configuration
Supported parameters
Graphical user interface
Error handling
Events log
Total 3.3 3.2 3.0 3.2 2.3
446 ABADI ET AL.

For teaching and learning purposes, support of all software development [7]. Scrum method is divided in
parameters and options of the stateless autoconfiguration Sprints. A Sprint is a period of time, where every phase of
process is an important issue. Another important issue is the the development occurs. The four main modules were
availability of the source code so students can improve it and developed in the first four Sprints, starting with the wradvs
adapt it to specific needs. Additionally, a friendly graphical Server that is installed as a service, and finishing with the
user interface can be very useful for non-expert students. construction of the detail of the events that occur in real time,
Since none of the existing daemons offers all the required showing a hexadecimal dump of the messages that are sent
features, we decided to develop our own server. and received. Then, there was a Sprint for the creation of the
Tooltips that gives all the information of the parameters,
V. WINDOWS ROUTER ADVERTISEMENT SERVER options and protocols that can be configured, making the
application an excellent teaching tool. The following Sprint
wradvs (Windows Router Advertisement Server) is a was introduced to improve the feedback to users and make
Router Advertisement server for Windows developed under the application more intuitive. Finally, there was a Sprint for
the GNU General Public License. It supports RFCs 2460, testing and debugging all the software.
4861, 5006, 4191, 3775, and 3963. It has a friendly graphical
user interface, checks entered values for consistency, offers a The application uses two external projects that were
good debugging system for troubleshooting purposes, and necessary to accomplish some functionality:
shows tooltips to guide users during the configuration
• WinPcap: an industry-standard tool for link-layer
process. At the present time, wradvs has an interface for
network access in Windows environments. It allows
English and Spanish users. Switching from one language to
applications to capture and transmit network packets
another can be done with just a few clicks.
bypassing the protocol stack, and it has additional
useful features; including kernel-level packet
A. Methodology filtering and a network statistics engine [9].
We wrote the application with Microsoft Visual C#
Express Edition [8]. wradvs was divided in four main • SharpPcap: a packet capture framework for the .NET
modules: the wradvs Server, the Server Configuration Tool, environment based on the famous WinPcap libraries.
the Log Viewer and the IPv6 Address Configuration Tool. It provides an API for capturing, injecting, analyzing
Fig. 2 shows a simplified class diagram of the application. and building packets using any .NET language such
as C# and VB.NET [10].
To develop wradvs, we used Scrum, an iterative
incremental framework for managing complex work (such as
new product development) commonly used with agile

Figure 2. Diagram of the main classes.

 WRADVS: A DIDACTIC SERVER FOR IPV6 STATELESS AUTOCONFIGURATION 447

intended for educational purposes, a lot of valuable

B. Modules information is shown on labels, tooltips, and messages. Also,
As stated previously, wradvs is divided in four main this module checks all the data entered by users for
modules (the wradvs Server, the Server Configuration Tool, consistency. Fig. 3 shows the main window of the server
the Log Viewer and the IPv6 Address Configuration Tool) configuration tool.
that we describe below.
3) Log Viewer
1) wradvs Server The log viewer provides visualization in real time of the
The server (wradvs) is in charge of listening to the RS server’s events. For every event, the log viewer displays the
messages and sending RA messages on demand and following information: date, time, involved network
periodically as described in [2], allowing any listening host interface, and description message. By double clicking on a
on the link to configure its addresses and some other network specific event, the log viewer shows a hexadecimal dump
parameters automatically through IPv6 stateless (including every fields’ values of the headers) of sent RA
autoconfiguration. The server is installed on Windows as a messages and received RS messages. Fig. 4 shows the main
service that will run on every system start. window of the log viewer.
2) Server Configuration Tool 4) IPv6 Address Configuration Tool
The server configuration tool provides a friendly The IPv6 address configuration tool provides a graphical
graphical user interface that allows users to configure all the user interface that allows the execution of netsh commands
parameters and options of the RA messages sent by the related to the configuration of IPv6 addresses, default
server. This configuration is saved as an XML document routers, routing table, DNS servers, and forwarding. Fig. 5
which is read by the server when it is started. The server shows the main window of the IPv6 address configuration
configuration tool also provides a way to start, stop, and tool.
restart the server’s execution. Since the application is

Figure 3. Main window of the server configuration tool.

448 ABADI ET AL.

Figure 4. Main window of the log viewer.

We will continue to support wradvs. Our goal is to offer

it to other national and international universities to be used in
their curriculum for teaching networking advanced courses.
Thus, we plan to give support to many other languages (at
the moment only English and Spanish are supported), and we
also want to port wradvs to Linux. Since the stateless
autoconfiguration process is vulnerable to various attacks
[12] when it is not secured, the original specifications of ND
called for the use of IPsec as a security mechanism to protect
ND messages. However, the use of IPsec is impractical due
to the very large number of manually configured security
associations needed for protecting ND. For this reason, the
Secure Neighbor Discovery Protocol [11] (SEND) was
proposed. We also plan to develop an open source
implementation of a SEND server that can be used in
production environment or as a didactic application.
REFERENCES
[1] S. Thomson, T. Narten, and T. Jinmei, “IPv6 Stateless Address
Autoconfiguration”, RFC 4862. September, 2007.
[2] T. Narten, E. Nordmark, W. Simpson, and H. Soliman, “Neighbor
Discovery for IP version 6 (IPv6)”, RFC 4861. September, 2007.
Figure 5. Main window of the IPv6 address configuration tool. [3] D. Johnson, C. Perkins, and J. Arkko, “Mobility Support in IPv6”,
RFC 3775. June, 2004.
[4] R. Draves and D. Thaler, “Default Router Preferences and More-
VI. CONCLUSION AND FUTURE WORKS Specific Routes”, RFC 4191. November, 2005.
In this paper we presented wradvs, a full solution for the [5] J. Davies, “Understanding IPv6”, 2nd ed. Microsoft Press. February,
IPv6 stateless address autoconfiguration process. Thanks to 2008.
the amount of information shown in the Server [6] J. Jeong, S. Park, L. Beloeil, and S. Madanapalli, “IPv6 Router
Configuration Tool and the amount of information stored in Advertisement Option for DNS Configuration”, RFC 5006.
September, 2007.
the Log Viewer for each event, wradvs is an excellent tool for
teaching. Currently, it is used to teach the Advanced Network [7] K. Schwaber, “Agile Project Management with Scrum”, Microsoft
Press. March, 2004.
Protocols course at Central University of Venezuela. The
[8] P. Pelland. “Microsoft Visual C# 2008 Express Edition: Build a
feedback that we received from students and professors is Program Now”, Microsoft Press. March, 2008.
very helpful and positive. [9] http://www.winpcap.org
Although wradvs was primarily developed as a didactic [10] http://sourceforge.net/projects/sharppcap
application, it is a complete and powerful tool for network [11] J. Arkko, J. Kempf, B. Zill, and B. Nikander. “Secure Neighbor
administrators. Its source code is freely available and can be Discovery (SEND)”. RFC 3971. March, 2005.
easily adapted to specific requirements by students. It has [12] P. Nikander, J. Kempf, and E. Nordmark. “IPv6 Neighbor Discovery
been released recently and can be downloaded from (ND) Trust Models and Threats”. RFC 3756. May, 2004.
Sourceforge at http://wradvs.sourceforge.net. [13] S. Deering and R. Hinden. “Internet Protocol, Version 6 (IPv6)
Specification”, RFC 2460. December, 1998.
 Securing P2P Wireless Communications by
Deploying Honeytokens in a Cooperative
Maritime Network
Jonathan White, Jamelle Srini Ramaswamy, Sean Mhamed Itmi
Brown Geoghan INSA de Rouen
University of Arkansas Univ. of AR. at Little Rock Rouen, France
Fayetteville, AR USA Little Rock, AR USA [email protected]
{jlw09, jab17}@uark.edu {sxramaswamy,
sjgeoghnan}@ualr.edu

Abstract - As wireless devices become ubiquitous in society, involves the movement of goods from large centralized ports to
proper security methodologies must be deployed that are other ports by utilizing cargo ships that traverse the oceans and
cognizant of the physical limitations inherent in the device design other waterways. The maritime shipping industry relies on
in order to ensure that the information produced in these systems shipping freight that is containerized, meaning that the cargo is
is not abused. One such area of abuse involves an attacker
sealed by the owner in a standardized metal container and is
locating and disabling a wireless device that is transmitting a
relatively large amount of data compared to the other devices in not palletized or packaged by the maritime shipping provider.
the same physical area; this large amount of transmitted data This containerization has allowed the shipping industry to
may indicate that the wireless device is important or valuable to achieve great efficiencies as the whole system has become
the owner. standardized.
In this work, we propose and test a novel design that mitigates There are only a few standard sizes of containers that are
this threat as related to P2P wireless communication systems that used throughout the industry (e.g., 8’x8’x20’, 8’x8’x40’, etc).
have battery life as a limiting constraint. Our solution uses The container shipping industry measures freight in terms
specialized honeytoken devices that transmit realistic noise in twenty foot equivalent units (TEUs), which represents the
order to mask the presence or absence of communication from the
capacity of a single 20’x8’x8’6” container. Estimates for 2008
actual wireless devices. This will allow the wireless devices to
only transmit when required, conserving battery life and adding indicate that over 18 million containers make over 200 million
to the security of the system by making it impossible to locate a trips per year, which account for approximately 90% of non-
highly active device. The honeytoken noise is easily removed by bulk cargo shipped worldwide [6]. Currently, the largest
trusted insiders; however it can not easily be removed by container transport ship is the Emma Maersk which can
attackers. The idea of deploying honeytoken devices to protect transport over 11,000 20-foot containers [5]. The ports are
P2P devices with battery life constraints is a new approach, and impressive locations, for example, the Port of Los Angeles
our key contribution is the proposal, design, and testing of this encompasses 7500 acres, 43 miles of waterfront and features
system. 27 cargo terminals.
Container shipping via large cargo ships is an industry that
I. INTRODUCTION transports millions of tons of cargo on hundreds of cargo ships
In the following sections we identify the maritime system in to destinations all over the world. In a typical journey, a
which we are deploying wireless peer to peer sensors in order shipper loads goods into a container and then seals the
to better track and secure intermodal shipping containers. We container. Next, the container is transported via truck to a port
also define several key properties that all honeytokens possess, where it is loaded onto a container ship for overseas transport.
regardless of the type of object they are protecting. We also Once the ship arrives at its destination port, the container is
present the motivation for using honeytokens to better protect unloaded and again loaded onto a truck for transport to its final
our wireless P2P system. destination where it is unsealed and unloaded. Due to the value
of the cargo and the potential for terroristic actions such as
A. Maritime Shipping Industry smuggling contraband, poisoning food cargo or stealing
The shipping industry is a complex system, composed of contents, the container must remain secure throughout its entire
many subsidiaries that allow and facilitate the transport of journey. Also, the container must remain at a safe temperature
goods from manufacturers to retail outlets or other entities in for refrigerated goods and the containers should not experience
the supply chain. In this work, we are focused on one a high level of acceleration for fragile goods. A system that
particular subsystem, the maritime shipping industry, which can monitor the position of the container at a port or on a ship

449
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_78, © Springer Science+Business Media B.V. 2010
450 WHITE ET AL.

and whether or not the container has been breached will battery constraints, the devices will only transmit when they
provide shipping personnel with additional information to have information to communicate or are passing on
ensure the secure transport of the containers and the goods information received from another node. This allows for fault
within the container. Our system provides part of this security tolerance and long battery life, which are two requirements of
and is described in section 3. the system.
However, the proposed system exposes a security risk.
B. Honeytokens While an attacker may not be able to understand what the
Honeytokens are an information security tool that operates devices are communicating, they can easily locate the devices
on the theory of deception. Honeytokens are related to that are communicating the most as most devices will be in a
honeypots, which are systems whose value lies in attracting, sleep state the majority of the time. For example, when a
distracting, and potentially tracing attackers by presenting a container is breached it will raise a signal that will propagate
seemingly worthwhile, but unprotected, computer. through the system. As this is a rare event, if an attacker has
Honeytokens are different from honeypots in that they aren’t been monitoring the wireless communications with a high
an actual device. Instead, they are seemingly worthwhile and powered receiver and suddenly notices a spike in the received
valid pieces of digital information. While organizations have traffic, they can be assured that something important has
sophisticated firewalls and intrusion detection systems in place happened. Also, if a device is consistently being queried for
to detect physical attacks on their networks or data systems, the information by the owner, this device will also communicate
loss of information posses a much greater threat, especially in relatively more data than the other containers in the P2P
wireless systems like ours [1]. network, making it stand out from the rest. An intelligent and
The term ‘honeytoken’ was coined by Augusto Paes de motivated attacker could potentially target this container for
Barros in 2003 in a discussion on honeypots. Honeytokens are malicious actions, such as stealing the contents, as it appears to
manufactured by security engineers to look attractive to an be valuable to its owners.
attacker [1], hence the name. Examples would be a customer We are going to use honeytokens to mitigate this risk. The
database of personal information, a list of passwords, honeytokens will be deployed throughout the ship and they
proprietary design documents, a list of credit card numbers, or will produce realistic looking noise; the communication from
a seemingly worthwhile wireless signal passing through a P2P the honeytokens will be very similar to the actual
network. communications produced by the sensors. Each honeytoken
Honeytokens have common properties, regardless of the device will be given a unique identification value so that they
digital object they mimic. First, the honeytoken must appear can be separated from the actual data produced by the P2P
valuable to an adversary [2]. If the honeytoken does not sensors. The honeytoken identification values will be known
appear worthwhile, it will never be interacted with, stolen, or by the controlling system, and the controlling system will
examined. As wireless systems become ubiquitous, the notify the sensors to not pass the noise produced by the
information that they produce will become valuable as it will honeytokens through a learning phase when the ship is loaded.
be used in many different areas to communicate important The noise will appear at random times and at random
data. durations, effectively masking the actual signal
Second, the honeytoken must be designed in such a manner communications that the sensor nodes produce. Because there
so as to not interfere with any real data. The implementers of will always be a high level of realistic noise in the RF
the honeytokens must be able to tell what is real and what is environment, an attacker will no longer have the ability to
not, and thus the honeytoken must have unique, identifiable single out an individual container, adding to the security of the
properties. If the honeytokens can not be separated from the system. The use of honeytokens also allows the actual P2P
real data, the whole system suffers and the accuracy of the sensors to transmit only when they have actual information to
system is compromised. send, conserving battery life.
Third, honeytokens are implemented so that they appear The rest of this paper is organized as follows. Section 2
real to an attacker. If the honeytokens are obviously an details some previous work in the area of P2P security and
imitation, the attacker would have no reason to use or examine honeytokens. Section 3 describes the current P2P wireless
them. If the honeytoken is easily detected, it might also be system and our proposal for using honeytokens. Section 4
easily removed by the attacker while still accessing the actual shows how we simulated the system and took comparison
data that the honeytoken was designed to protect. Secrecy is of measurements on a system that uses honeytokens and with a
the utmost importance because if an attacker is able to find out system that does not, and section 5 presents these results.
what objects are honeytokens, they can easily ignore them [3]. Section 6 concludes the work.

C. Motivation II. BACKGROUND

The system in which we are operating consists of many Honeytokens are used in several diverse industries [4].
distributed sensors that communicate in a peer to peer to They have been used with great success, and this is an
manner. The wireless system encrypts all communications, indication that P2P honeytokens can potentially be used in the
making it unintelligible to attackers. Also, due to the inherent maritime shipping domain. We show what industries use
 SECURING P2P WIRELESS COMMUNICATIONS BY DEPLOYING HONEYTOKENS 451

honeytokens and also other honeypot methods that have been equipment that requires frequent maintenance will not perform
used to secure P2P systems. well within the economic constraints of the shipping industry.
Honeytokens have been used in personally identifiable Solid state electronic components are generally rugged enough
information (PII) databases to detect unauthorized access [1]. to withstand the application as long as they are mounted to the
In many industries, privacy laws prohibit individuals from container in a position that is not vulnerable to damage by the
accessing personal information that they don’t have a valid use equipment used for loading and unloading. However, the
for. Detecting when unauthorized access has occurred is power requirements of the instrumentation must be addressed.
difficult, as the data requirements for applications and users Typically, containers do not have access to a direct, always on,
vary greatly. power source. Therefore, each wireless node must rely on
Honeytokens are known as ‘copyright traps’ when used in battery power, which must last several months to prevent
the cartographic and dictionary industries [3]. Mapmakers are excessive maintenance requirements.
known to insert fake streets, misspelled objects, and non- Each container node periodically broadcasts its status
existent bends in rivers to catch illicit copying. Dictionaries (breached or secure) to all the container nodes that are
have also inserted fake words into their dictionaries in order to listening. For example, the sensor nodes may broadcast a
catch copyright violators. Both of these uses of honeytokens system secure message every hour on the hour; also the sensors
have been shown to work [3]. will wake every ten minutes to listen for breaches or other
Honeypot systems have been used to analyze P2P botnets messages from the system. If a breach message is received, the
[7]. The honeypot attracts the bots and the actions that are message is passed on throughout the P2P network until it
performed are analyzed. If the bot interacts with the honeypot reaches the intended recipient. This is a basic overview of the
for an extensive period of time, many of its abilities can be P2P environment in which we are operating in. While the
identified and analyzed. Honeytoken email addresses have focus of this work is not the implementation of the network
been used to monitor distributed P2P spam networks [8]. The itself, it is necessary to have an appreciation of the physical
addresses appear valid, but they are present just to attract and constraints on the devices in order to understand the security
waste attacker’s time and resources. Many other approaches risks inherent in this design.
have been used to secure P2P communications [9], [11], but
little work has been done on using honeytokens to add noise to B. Honeytoken Deployment
the system and to distract attackers from the actual The honeytokens will be deployed throughout the RF
communications. environment on the boat. They must be deployed in sufficient
amounts in order to effectively cover the actual
III. PROPOSED SYSTEM communications that the sensors are transmitting. Also, as the
In the following sections, we describe the P2P wireless sensors are synchronized to transmit and/or receive only at
maritime system. As the sensors are going to be deployed onto certain predetermined discrete intervals, the honeytokens must
shipping containers, there are several physical constraints that also be attuned to this. The honeytokens must produce noise
must be accounted for. Also, we describe in detail our that appears realistic to an outsider when compared to the
proposal for deploying the honeytoken devices throughout the actual transmissions, however since the communications are all
RF environment on the ship. encrypted, this will be a straightforward process.
The honeytoken devices themselves will be simpler than the
A. P2P Wireless Maritime System P2P devices. The honeytokens will have a transmission only
The container network is created by instrumenting each radio, a battery, and a small processor that can store the
container with intrusion detection hardware and a wireless device’s assigned identification number and perform the
network device. The device will have a battery, radio, necessary calculations in order to produce random noise at a
temperature sensor, light sensor and accelerometer in order to specified time slot. The honeytokens will run on a separate
perform the necessary physical security requirements. Due to battery based power supply from the P2P devices so as to not
the limited range of the wireless devices, many containers will have a negative effect on the battery of the actual devices.
not be able to communicate directly to the ship or port There will be no sensors or other physical interfaces as the
network. However, the containers are packed tightly together honeytokens are present to only produce noise.
and each container will be in range of several neighbor The honeytoken devices can be placed on physical
containers. Typically, containers are stacked several units high containers or throughout other areas of the ship. When the
in rows on the ship or port. honeytokens are placed throughout the ship, the controlling
Additionally, some of the containers will be in range of the computer records the identification number of the honeytoken
ship or port network. Therefore, all co-located containers, and notifies collocated devices about their presence. If an
either on a ship or a port, will form an ad-hoc peer-to-peer actual P2P device retransmits the noise, it will eventually reach
(P2P) network. Part of the cost-effectiveness of the the ship’s computer. The computer will have access to a list of
containerization shipping process is the fact that individual the honeytoken identification values and it will be able to
containers are designed to be rugged and require very little determine that this message is in fact noise. The shipboard
maintenance. Therefore, the addition of sensitive electronic computer will then send a message to the device that
452 WHITE ET AL.

retransmitted the noise alerting it to the fact that this device is communications. We compared these four scenarios against
in fact a honeytoken. In this way, all trusted devices can learn the system with no honeytokens.
what devices are actual sensors and which devices are present The attacker only has a limited range in which they can
just to produce noise. This method of learning what devices receive signals from the sensors successfully. For our tests, we
are honeytokens will require extra communication during the assumed that the attacker was at a point in the middle of the
setup phase, but once the honeytokens are learned, the P2P ship and that the attacker’s receive antenna had a range of 50
devices can ignore the noise while still being able to retransmit meters [10]. In reality, an attacker could also be scanning from
valid messages. a port at the side of the boat, and we will take this into account
As mentioned previously, honeytokens must appear in a future work. In order for the attacker to receive a
valuable, realistic, and must not interfere with the valid data. transmission from a sensor or honeytoken, the receive antenna
The honeytokens on our system are valuable and realistic has to be less than 50 meters from the transmission source, or:
because they produce the same type of signals that the actual
P2P devices transmit. An attacker does not have the ability to
remove this noise, and they can no longer locate a signal (S x − I x )2 + (S y − I y )2 + (S z − I z )2 < 50 m
device that is transmitting a large amount of data. As each of
the honeytokens has a unique identification value that is only where (Sx, Sy, Sz) represents the spatial coordinates of the
known to the trusted devices, each device can correctly attacker’s antenna, and (Ix, Iy, Iz) represents the spatial
identify what is and what is not noise, keeping the devices coordinates of a particular sensor. The above equation gives
from interfering with each other. the distance the object is from the antenna if a line is drawn
from the sensor to the antenna in three dimensional space.
IV. SYSTEM SIMULATION For all of the simulations, we ran the same series of
In the following section, we detail the steps that were taken simulated messages. For the simulations, the P2P sensors
to test our hypothesis that honeytokens could be used to mask broadcast a system secure message every 60 minutes, unless
the presence of devices that were transmitting a lot of data the container had been breached, in which case a message is
relative to other devices in the P2P network. First, we detail sent stating that fact. Also, the sensors wake every 10 minutes
the physical setup of the simulation with several thousand for 30 seconds in order to listen for breach signals or other
shipping containers, each with an affixed P2P device. We then transmissions from the system. We simulated the probability
describe the placement and amount of deployed honeytokens. of each container needing to transmit a message at a given time
We perform several experimental simulations using various slot as 5%, which is what we expected the probability to be in
amounts of honeytokens. We conclude by describing the actual usage on the ship as the breach and other events are rare
message set that was used to compare the systems. occurrences. We then simulated 2.5% of the sensors as “hot”
sensors. These sensors communicated at a much higher rate,
A. System Setup even though it was still a rare event for the system overall; for
For our experiments, we simulated one ship with several our experiments we used a Poisson distribution with a lambda
thousand containers, each with a P2P sensor attached. As value of one. Again, this relatively small amount of “hot”
mentioned previously, the Emma Maersk is the largest sensors is what was expected in actual usage. As the
container ship currently sailing [5]. It is capable of holding honeytokens only produced noise, we allowed them to all be
approximately 11,264 twenty foot equivalent containers. We considered “hot” sensors, with a Poisson distribution that was
located the dimensions of the ship and placed each of the skewed towards a higher transmission probability. If we
11,264 containers at a specific place on the ship. The ship is allowed the honeytokens to transmit all the time, this would
397 meters in length, and 56 meters wide. Based upon the easily be removed by the attacker as it would simply raise the
dimensions of the ship and other technical data that was noise floor threshold; instead, we allowed the honeytokens to
obtained, the simulated containers were stacked 8 tall by 22 act like the devices that were transmitting the most, which
wide by 64 containers in length. We then simulated placing a made the noise harder to remove as it looked like the real
sensor on the outside of each container, each having a specific thing. We ran each simulation for a total of 3 hours, so as to
location in three dimensional space. collect a sufficient amount of data. At each ten minute interval
when the sensors woke up, measurements were taken on the
B. Test Setup amount of simulated transmissions that the attacker would
Next, we simulated various amounts of honeytokens intercept given the constraints identified above. As we get
distributed throughout the ship. The honeytokens were placed more real world usage data, we will adjust these simulation
randomly throughout the ship on the exterior of the containers properties to accurately reflect the environment.
and at the side of the ship’s deck. We ran simulations with 50,
500, 1000, and 5000 honeytokens. We anticipated that if only V. RESULTS
a small amount of honeytokens were used then the produced We first ran the simulation with no honeytokens present to
noise would not be enough to effectively mask the actual get a baseline comparison of how the system operates. As
expected, there were certain peaks in the transmissions that
 SECURING P2P WIRELESS COMMUNICATIONS BY DEPLOYING HONEYTOKENS 453

were received which allowed the attacker to singulate out

individuals or small groups of containers that were transmitting
500 Honeytokens
at a much higher relative rate compared to the rest of the
containers. Figure 1 on the opposite column shows this 600
weakness; in time frames four, twelve, and sixteen, the amount

Number of Transmissions
500
of communication is over twice what is normally transmitted.
400
Time frames one, seven, thirteen, and nineteen are the highest
300
communications where all sensors transmit a signal to indicate
that they are still operational. As this is a system wide event, it 200

does not reveal secure information. 100

0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19
Transm issions w ith no Honeytokens
Time

600
Figure 3: Simulation results with 500 honeytokens.
500

400
The results with the usage of 500 honeytokens are
presented above in Figure 3. The communications are
300 beginning to appear much more random; however the largest
200 amounts of communications can still be seen at the fourth,
twelfth, and sixteenth time frame. However, the other time
100
frames have a much more similar distribution, and the peaks
0 are beginning to even out.
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19
T ime ( 10 minut e int er val s)
1000 Honeytokens
Figure 1: Intercepted transmission with no honeytokens.
600
Number of Transmissions

500
As mentioned previously, our goal of using the
honeytokens is to obfuscate the actual communications. Our 400

first attempt at doing this used 50 honeytokens randomly 300

distributed throughout the ship. While the honeytokens were 200

still producing noise at the same rate as the most active 100

sensors, they were unable to effectively mask the true signals. 0

The results are presented in Figure 2 below, and it can be seen 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19

that the same peaks of communications that appeared in our Tim e

baseline comparison in Figure 1 are still present.

We then proceeded to test the system with 500, 1000, and Figure 4: Results with 1000 honeytokens.
5000 honeytokens. We expected to find that the intercepted
communications appeared much more random as more Figure 4 above shows the results when one thousand
honeytokens were added. honeytokens were deployed randomly throughout the ship. At
this point, the communications have become obfuscated. The
honeytokens have masked the presence of the active devices by
50 Honeytokens transmitting a sufficient amount of noise. While the clock
signals that are transmitted once every hour are still present,
600
this does not present much of a security risk as every device in
500 the network transmits this.
400 Figure 5 shows the scenario when five thousand
honeytokens were used; even the clock signals were masked
300
by the honeytoken noise. The communications appear pseudo
200 random, and an attacker can develop very little information
100
about the system. In the case where five thousand
honeytokens were used throughout the ship, the attacker is
0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19
actually receiving significantly more noise than actual sensor
T ime
communications, which makes actually deciphering any useful
information difficult.
Figure 2: Intercepted signals with 50 honeytokens.
454 WHITE ET AL.

the communications were made the most random when the

5000 Honeytokens
largest amount of honeytokens were used. In our tests, this
1000 occurred when 5000 honeytokens were deployed among the
Number of Transmissions 900 11,264 containers. However, we also found that even a
800
700 smaller amount of honeytokens are sufficient to remove the
600 ability for an attacker to singulate out a sensor or group of
500
400
sensors that are communicating consistently. When 1000
300 honeytokens were used, the communications appeared random
200 to an attacker, effectively mitigating this risk. Our experiments
100
0 showed how effective honeytokens are in this environment and
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 our work was successful in identifying how and in what
Time amount honeytokens should be used.
Figure 5: Results with 5000 honeytokens.
ACKNOWLEDGEMENT
This work is supported, in part, by research supported by the
The greatest amount of obfuscation was observed when the
National Science Foundation (under Grant Nos. CNS-
largest amounts of honeytokens were employed, which is what
0619069, EPS-0701890, and OISE 0729792); NASA EPSCoR
was expected. However, it is likely not economically feasible
Arkansas Space Grant Consortium (#UALR 16804); and
to deploy such a great amount of honeytokens throughout the
Acxiom Corporation (#281539).
ship. When one thousand honeytokens were employed,
sufficient noise was present to hide the actual communications
REFERENCES
from an attacker that might be listening in on and the one 1. I. Mokube,, M. Adams, “Honeypots: concepts, approaches, challenges”,
thousand honeytokens represent slightly less than ten percent Proceedings of the 45th annual southeast regional conference, ACM
of the number of sensors that are present on the ship. Also, in Press, New York, NY, (2007), pp. 321–326.
some circumstances, even less honeytokens might be 2. C. McRae., R. Vaughn, “Phighting the Phisher: Using Web Bugs and
Honeytokens to Investigate the Source of Phishing Attacks”, 40th Annual
appropriate. Hawaii International Conference on System Sciences (HICSS'07), (2007),
pp. 270.
VI. CONCLUSIONS AND FUTURE WORK 3. J. Yuill, M. Zappe, D. Denning, F. Feer, “Honeyfiles: Deceptive Files for
As pervasive computing become more common in our Intrusion Detection”, Proceedings of the 2004 IEEE Workshop on
Information Assurance, IEEE publishers, West Point, NY, (2004), pp.
society, proper security measures must be in place that are 116-122.
aware of the physical limitations of the electronic devices [10]. 4. S. Weis, “Security parallels between people and pervasive devices”,
In this work, we have focused on a maritime environment Proceeding of the 2nd IEEE International Workshop on Pervasive
where wireless sensors are deployed onto shipping containers. Computing and Communication Security - PerSec 2005, IEEE publishers,
West Point, NY, (2005) , pp. 105–109.
In order to conserve battery, the sensors only transmit when 5. Maersk Line Facts and figures. Retrieved July 4, 2008 from the Maersk
required, saving battery. However, this opens up a risk in that Line web site.
an attacker can easily locate and disable a sensor that is http://www.maerskline.com/link/?page=brochure&path=/about_us/facts_
transmitting a relatively large amount of data when compared and_figures.
6. Intermodal freight transport, Retrieved February 11, 2009,
to the other sensors in the collocated RF environment. http://www.nationmaster.com/encyclopedia/Intermodalfreighttransport.
We have designed and tested a novel methodology that 7. B. McCarty, "Botnets: Big and bigger," IEEE Security & Privacy
enables the mitigation of this risk. We have proposed Magazine, vol. 1, no. 4, July 2003.
deploying honeytokens throughout the ship. Honeytokens are 8. R. Vogt, J. Aycock, and M. Jacobson, "Army of botnets," in Proceedings
of 14th Annual Network and Distributed System Security Symposium
devices that transmit realistic, but synthetic noise meant to
(NDSS), February, 2007.
distract the attacker and hide the actual valuable 9. E. K. Lua, J. Crowcroft, M. Pias, R. Sharma, and S. Lim, "A survey and
communications. The honeytokens transmit noise that appears comparison of peer-to-peer overlay network schemes," IEEE
realistic to an outsider; however it is easily separated and Communications Surveys and Tutorials, vol. 7, no. 2, 2005.
10. S. Weis, “Security parallels between people and pervasive devices”,
removed from the actual communications by trusted insiders.
Proceeding of the 2nd IEEE International Workshop on Pervasive
The honeytoken devices themselves are simpler than the Computing and Communication Security - PerSec 2005, IEEE publishers,
sensors that are deployed on the shipping containers on the West Point, NY, (2005) , pp. 105–109.
ship and they run on a separate power supply, conserving the 11. J. White, “Radio Frequency Identification (RFID) Honeytokens as a
battery life of the actual sensors. Mitigation Tool against Illicit Inventorying”, in Proc. International
Conference on Information Security and Privacy (ISP-2009), Orlando,
In terms of future work, we would like to begin field trials Florida, July 13 – 16, 2009, pp. 67 – 73
of our work. We intend to measure more accurately the 12. S. Geoghegan, G. McCorkle, C. Robinson, G. Fundyler, S. Ramaswamy,
probabilities at which the sensors transmit data. We plan to M. Tudoreanu, R. Seker, J. Brown, and M. Itmi, “A Multi-Agent System
test with physical devices and measure how far the range is of Architecture for Cooperative Maritime Networks”, in Proceedings of ISC
2009, Toronto, Canada, 2009.
the devices.
We have simulated several scenarios with honeytokens
deployed randomly throughout a ship. Our results showed that
 Integrating Information Security into
Quality Management Systems
Dr. PhD Margareth Stoll [email protected]
EURAC Research, Drususallee, 1
39100 Bozen, South Tyrol, Italy

Abstract-Due to globalization, stronger competition, in- sharpened, the IT governance requirements and thereby data
creased complexity, information explosion, interconnection and integrity requirements are increased and based on the main
extensive use of technology, information management is a main role of data, information and knowledge the requirements for
performance driver and key differentiator for sustainable or- confidentiality and availability are increased.
ganization success. Data, information, knowledge and asset are
exposed to most different threats. Therefore more than 5800
In this respect information and system availability, confiden-
organizations worldwide are implementing an information se- tiality and data integrity, as well as data protection and ful-
curity management system in accordance with ISO/IEC 27001. fillment of legal and regulatory requirements is central for
Many organizations of different sizes are implementing also organizations’ success.
standard based management systems, such as ISO9001 or oth- One million organizations of different sizes and scopes are
ers. On the basis of several practical applications in most dif- implementing, already since several years, quality manage-
ferent organizations are described applied methods and expe- ment systems, such as ISO9001, or other management sys-
riences by implementing suitably holistic management systems tems based on international standards (e.g. environment
for risk prevention and quality improvement. This promotes ef- ISO14001, IT service management ISO22000). All these
fectiveness, efficiency, legal conformity, information security
awareness and organization development including improved
management systems require common principles: organiza-
information security for sustainable organization success. tion objectives and strategies, business processes, resource
management and continuously optimization of the organiza-
tion [4]. The established management system must be docu-
Keywords: information security, quality management, holis-
mented, communicated, implemented and continuously im-
tic enterprise management, data protection, risk management
proved. These systems are implemented more frequently ho-
listic, whereby are integrated according with the organiza-
tional purpose and objectives different aspects, like quality,
environment, hygiene, occupational health and safety, as
I. INTRODUCTION well as personnel development, resource management, IT -
management, communication management, controlling and
A Starting Situation also knowledge management.
Due to globalization, ever stronger competition, information
explosion, increasing interconnection, extensive use of IT- B Purpose of the article
systems and the growing complexity data and information
management for knowledge generation and continual im- The increased requirements for information security includ-
provement are main performance driver and key differentia- ing availability, confidentiality and integrity, international
tors for competitive advantages and sustainable organization standards such as IEC/ISO 27001 [5], IEC/ISO 27002 [6],
success. Thus data, information, knowledge, IT and produc- ISO/IEC 20000-1, Sarbanes-Oxley or other legal or regulato-
tion systems become a completely new role. But data, in- ry requirements, the improved risk awareness promote a ho-
formation and knowledge are exposed to most different listic information security approach. There are just now more
threats, such as physical and environmental threats, technical than 5800 IEC/ISO 27001 information security systems im-
threats, organizational and human related threats. The Com- plemented and certificated [7]. The information security
puter Security Institute (CSI) and the FBI describe that virus management systems are often implemented parallel to other
attacks are the source of the greatest financial losses, fol- standard based management systems. The trend to holistic
lowed by unauthorized access to networks, lost or stolen lap- management systems and the need of developing an inte-
tops or mobile hardware, and theft of proprietary information grated solution to support optimally all requirements of an
are the next and all together was reported a lost account for organization and to promote a consistent and integrated im-
more than $39,370,717 [1]. The kes Microsoft study and the plementation and operation influenced us to integrate an in-
report about the IT security situation in Germany explain formation security management system with an ISO 9001
that collaborators error and negligence are the main cause for quality management system. This will enrich management
damage followed by software and hardware problems [2], systems by information security aspects, use common syner-
[3]. In the last years the data protection law requirements are gies and resources, and it will contribute efficiently, re-

455
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_79, © Springer Science+Business Media B.V. 2010
456 STOLL

sources carefully, fast and successfully to support informa- Thus information security and thereby availability of data,
tion security for sustainable organizational development and documents, asset and infrastructures, confidentiality of in-
success. formation, documents and data, as well as data integrity
should be optimized and the organization will be promoted
C Research Approach in achieving its objectives including information security ob-
jectives and reducing risks.
We developed a holistic information security management
model by integrating the ISO 9001 standard for quality man-
agement systems or other international standards for man-
agement systems and the information security management III. MAIN STANDARD REQUIREMENTS
standard in accordance to IEC/ISO 27001 and IEC/ISO
27002. This model was implemented after in different small A Main Requirements of Standard Based Management Systems
and medium sized organizations. The case study results were
collected by established process and system measurement The ISO 9001 quality management standard and other stan-
methods and interviewing the managers and collaborators. dards for management systems require common principles
[4]:
D Structure of the Article • Organizations’ objectives and strategies must be estab-
lished regarding stakeholder requirements [see top of
Firstly we describe based on the starting situation the project Fig.1].
objectives [II]. After we explain the requirements for our ap-
• All business processes including the management process,
proach based on the main requirements of quality manage-
support processes, resource processes and optimization
ment systems (ISO 9001) and other international standards
processes must be defined and promote the optimized ful-
for management systems [III A]. Also the main requirements
filment of the organizations’ objectives under the focus of
for information security management in accordance to
the respective standard [see the graphic under the top of
IEC/ISO 27001 and IEC/ISO 27002 [III B] are illustrated.
Fig. 1].
After we present our holistic information security manage-
ment model [IV] and the implementation of that model [V] • Process oriented resource management must be promoted
with the development of policy and objectives [V A], the including human resource development, IT – management
risk assessment, risk treatment and business continuity plan and other infrastructures, tools and instruments [see bot-
[V B], the process analysis and process improvement [V C], tom of Fig. 1].
the resource management [V D], the continually improve- • The organization, their objectives and strategies, services/
ment of the whole system and the system documentation [E]. products and processes must be continually optimized ac-
Finally we report about our project experiences and results cording to established processes in sense of a PDCA cycle
of the implementation in different small and medium sized (plan, do, check, act) [see the circle around in Fig.1].
organizations [VI] with the achievement of the project prin-
ciples [VI A] and we reflect about success factors [VI B]. At The established management system must be structured and
the end we give an outlook [VII] and our conclusions [VIII]. systematically documented and communicated within the
organization and the collaborators must be continually moti-
vated for implementing the system and for recommending
improvements.
II. PROJECT OBJECTIVES These standard based management systems are implemented
more frequently in a holistic way. In accordance with the or-
How can be integrated the requirements of information secu- ganizational purpose and objectives are integrated into a
rity management in accordance to IEC/ISO 27001 and management system eventually environmental, hygienic, oc-
IEC/ISO 27002 and the requirements ISO 9001 quality man- cupational health and safety aspects, as well as human re-
agement systems or other standards based management sys- source development, resource management, knowledge
tems in order to contribute to organizations’ success and to management, IT - management or controlling.
be considered, used and improved as base for organization
development?
By integrating information security management in accor- interested parties
dance to IEC/ISO 27001 and IEC/ISO 27002 into a, the or- vision
policy
ganization best adapted, holistic information security man- objective, priority, strategy
agement model we expect to foster:
service/product realization
• effectiveness, resource management
• efficiency and cost reduction,
• shorter initial training periods for new collaborators, and
• improved information security. Fig. 1. Main requirements of standard based management systems
 INTEGRATING INFORMATION SECURITY INTO QUALITY MANAGEMENT SYSTEMS 457

B Main Requirements of Information Security Management Systems

§
We present the requirements of ISO/IEC 27001 [5] and Interested Parties
Vision
ISO/IEC 27002 [6]: Policy + IS
Objectives, Strategies + IS
• An information security policy must be defined and ap- P o te
Im pac
o f n ti a l A m
Le v D a m a o un t
e ls g ,
o f R ise
ks
ts
P o te n t i a l
T h re a ts
L i k e l ih o o d

A sset

proved by the management. C o n f i d e n t ia l i ty A v a ib il it y

P r e v e n t iv e
A c t io n
I n te g r it y

R i s ik s
A c c e p ta b le ?
Re
R iss id u
ks a l

Services,
• A risk assessment must be conducted to establish a risk Production + IS
treatment plan to reduce risks to acceptable levels of risk. Ressource Management
+ IS
For the identified remaining risks a business continuity
plan must be developed, implemented, maintained, tested
and updated regularly.
• The needed resources must be determined and provided. Fig. 2. The Holistic Information Security Management Model

All collaborators must be competent to perform their tasks

and be aware of the relevance and importance of their in- integrated into the operational processes and thereby im-
formation security activities and how they can contribute plemented, maintained, tested and updated regularly to en-
to the achievement of the information security manage- sure there effectiveness. Afterwards the business processes
ment system objectives. are improved according to the established objectives.
Thereby all stakeholder requirements inclusive the infor-
• The effectiveness and adequacy of the information security mation security aspects are improved. The process descrip-
management system must be continually improved using tion establishes for all process steps the associated respon-
controls, objectives, audit results, by monitoring, applying sible and accountable function and the relevant informa-
corrective and preventive actions and management review tion security requirements (corporate, regulatory or legal
in sense of a PDCA cycle (plan, do, check / study, act). requirements) [see middle of Fig. 2: the main business
The established information security management system processes start from the first contact with the customer to
must be structured and systematically documented, commu- the delivery of the product/service and the smile of the
nicated, implemented and continual improved. customer].
• The resource management specifies competence objectives
aligned to the corporate objectives for all functions. Ap-
propriate trainings or other actions must be taken to
IV. HOLISTIC INFORMATION SECURITY MODEL achieve and maintain these objectives according to defined
processes. Their effectiveness must be evaluated. Thereby
Due to the great diffusion of ISO 9001 quality management
also all required information security competences and the
systems or other standard based process oriented manage-
collaborators’ awareness is constantly identified and im-
ment systems, the role of knowledge and information as
proved. The organization must also define, plan and pro-
main performance driver and key differentiator and the vari-
vide the necessary resources, tools and instruments to ob-
ous most different threats for information, data and asset, as
tain and continually improve its objectives including in-
well as the alignment of different standard based manage-
formation security objectives [see bottom of Fig. 2].
ment systems we integrate an information security manage-
ment system and standard based management systems to a • The effectiveness and adequacy of the established holistic
holistic information security management model. management system and objectives must be evaluated pe-
riodically by suitable methods for monitoring and, where
Based on the main requirements of standard based manage-
applicable, measurement. It must be continually improved
ment systems [III A] and the information security manage-
according to established processes in sense of a PDCA cy-
ment system requirements [III B] we developed a holistic in-
cle (plan, do, check / study, act) [see the circle in Fig.2].
formation security management model:
• The corporate vision, policy, objectives and strategies are
extended by information security aspects [see top of Fig.
2]. V. IMPLEMENTATION
• A risk assessment must be conducted to establish a risk
treatment plan to reduce risks to an acceptable level of A Policy and Objectives
risk. For the identified remaining risks a business continu- Starting from the requirements of interested parties, legal or
ity plan must be developed, implemented, maintained, regulatory requirements, contractual security obligations, the
tested and updated regularly [see graphic under the top of characteristics of the business, the organization, its location,
Fig. 2]. environment, assets and technology we elaborate using
• All business processes are analyzed regarding also infor- Quality Function Deployment [8] the corporate policy in-
mation security aspects and all in the risk assessment iden- cluding information security elements and integrate it to a
tified risk treatments and the business continuity plan are holistic corporate policy. Thereby we define the information
458 STOLL

security management system scope and boundaries. We de- ing, generating, changing, handling and the accountable for
duce after from the policy the objectives and strategies inclu- it, as well as the functions, which will collaborate and/or be
sive information security objectives, such as availability, informed. Also external information sources or receivers
confidentiality and integrity. Also all legal, regulatory and (document logistic), the data protection class in accordance
corporate information security requirements are analyzed to the data protection law and further the required archiving
and adequate information security strategies deduced. Thus methods (for current, intermediate and historical archives)
the entire organization is focused on the fulfillment of stake- are identified. Thereby we establish necessary and licit
holder requirements including information security objec- access rights, archiving procedures, information security re-
tives and requirements. quirements regarding confidentiality, availability and integri-
ty, necessary data encryption with encryption procedures, as
B Risk Assessment, Risk Treatment and Business Continuity Plan well as signature rights and signature procedures as basis for
the implementation of adequate signature regulations. All
A risk assessment is conducted to establish a risk treatment treated data are examined for there necessity and lawfulness.
plan to reduce risks to an acceptable level of risk. Therefore Thus all information flows, documents, data and document
we identify the assets, define the required levels of confiden- handling and all processes are analyzed regarding informa-
tiality, integrity and availability for the assets, identify the tion security and optimized.
potential threats to those assets, define the realistic likelih- Function profiles and required competences are deduced
ood of security failures, the impacts that losses of confiden- from the role of the different functions at the single process
tiality, integrity and availability may have on the assets, the steps.
controls currently implemented, estimate the levels of risks
regarding the implemented controls and elaborate for the
higher risks adequate risk treatment plans. The management D Resource Management
must approve whether the risks are acceptable or further risk
treatments must be implemented. For the risk treatments ap- The organization must determine and provide due to stan-
propriate risk control objectives and controls are selected dard requirements necessary resources, tools and instruments
and approved. The risk treatment plan is integrated into the to obtain the established objectives and to continually im-
operational processes. Thereby the risk treatment and the ex- prove the organization. Thereby an optimal infrastructure for
isting standard based management system are integrated. For information security is promoted. Training and competence
the identified remaining risks a business continuity plan is objectives are planed and implemented according to defined
developed, integrated into the operational processes. Thus it processes and their effectiveness are evaluated. Thus also the
is implemented, maintained, tested and updated regularly to competences and awareness of all collaborators and persons
ensure that it is effective to maintain or restore operations to whom are assigned information security responsibilities
and ensure availability at the required level and in the re- are promoted systematically and structured, their effective-
quired time scale following interruption to, or failure of, crit- ness evaluated and possibly necessary corrective actions are
ical business processes. taken.

C Process Analysis and Process Improvement E Continually Improvement

The management process, all business processes including Due to standard requirements we deduce from the estab-
supporting processes, resources processes and optimization lished corporate objectives including information security
processes are analyzed bottom up by interviewing the con- objectives the business process objectives with correspond-
cerned collaborators and defining necessary information, da- ing measurement methods to demonstrate the process ability
ta, knowledge and documents [9]. Thereby also all relevant to achieve planned results. When planned results are not
standards, guidelines, legal and regulatory requirements are achieved, corrective and eventually preventive actions must
defined and integrated. The processes are optimized regard- be taken. All corrective actions, improvements or changes
ing the established organizational objectives, as well effi- must be discussed, approved, documented, communicated,
ciency, effectiveness, information security requirements, da- implemented and their effectiveness evaluated. Corrective
ta protection and other corporate, legal and regulatory re- actions, improvements or preventive actions are introduced
quirements. Thereby procedures, regulations, legal and regu- also by means of collaborator ideas, the results of periodical-
latory interpretations, checklist, forms and IT systems are ly internal and external audits, feedbacks from stakeholders
harmonized and also knowledge and information manage- or external sources (e.g. supplier, technical journals). There-
ment improved [9], [10]. by also information security is constantly evaluated and if
By process modeling for all process steps the responsible necessary changed and optimized. Organizational knowledge
and accountable functions, as well as necessary documents, is changed and therefore new individual learning becomes
information, tools, IT - applications including the observing possible. Thus the knowledge and learning spiral is constant-
information security laws and regulations are defined. For all ly pushed again and the organizational knowledge base ex-
documents we deduce the responsible function for establish- tended [12].
 INTEGRATING INFORMATION SECURITY INTO QUALITY MANAGEMENT SYSTEMS 459

F System Documentation could be abbreviated around a half. New collaborators

execute their tasks faster well regarding information secu-
The established management system with policy, objectives,
rity, whereby the productivity increase and possible errors
strategies, processes, risk assessment and risk treatment plan,
are avoided.
business continuity plan, function profiles, templates, check-
lists and others must be documented. Every collaborator • Information security improvement by employee involve-
must be trained on it. All changes of the documentation must ment: Due to the standard all ideas, optimizations and sug-
be distributed traceably and trained. The collaborators and gestions of the collaborators must be evaluated, eventually
particularly the managers must implement constantly the es- implemented systematically and structured and the effec-
tablished processes and procedures. tiveness of the implementation must be controlled. Thus
information security is promoted. Thereby we received av-
To promote need-oriented, workplace integrated access to
eraged monthly 0.1 until 0.6 ideas from each collaborator.
system documentation we prepare the documentation regard-
This continuously organizational improvement by ideas
ing didactical principles [12], [13] and distribute it electroni-
and suggestions of the collaborators and the systematic
cally through web-based intranets, pdf or organizational
und structured process for determining, implementing and
learning and knowledge systems based on constructivist
evaluating the effectiveness improve information security
theory [14].
and the organization itself.
Data protection and information security legal and regulatory
requirements are analyzed, implemented and maintained by
defined improvement processes. Thereby the legal and regu-
VI. PROJECT EXPERIENCES AND RESULTS latory conformity is sustainable implemented. This is spe-
cially appreciated by managers, controllers, risk managers
This holistic information security model was implemented and/or responsible persons for data protection and/or infor-
with distinct process oriented standard based management mation security.
systems in different small and medium sized organizations. All the discussions about information security during the
Thereby implementing process thinking, harmonizing and system introduction and the trainings increased severely the
simplifying processes, process controlling and overall main- information security awareness of the collaborators.
tain sustainable security awareness were great challenges.
By integrating information security completely into a stan-
dard based management system the information security
A Achieving the Project Objectives management system uses synergies, such as the existing
The following case study results were collected by estab- process and system thinking, the systematic process oriented
lished process and system measurement methods and inter- approach, the existing policy, strategy, objectives, the train-
viewing managers and collaborators: ing and improvement processes, and the management re-
sponsibility. In many technical organizations the collabora-
• Effectiveness: The fulfilment of the established objectives
tors feel the information security management system more
is periodically controlled and studied by defined measure-
helpfully than the quality management system or others and
ment methods and if necessary there are due to standard
thus also the existing standard based management system
requirements elaborated and implemented appropriate cor-
benefits from the integration. The collaborators feel the inte-
rective or prevention actions. Their effectiveness is con-
gration more than an extension and enriching.
trolled. Thereby the enterprises achieve their planned ob-
jectives in average more than 92%. In a service organiza-
tion e.g., where confidentiality was a main factor for cus-
tomer satisfaction it was improved constantly from 85% at B Success Factors
the end of 2002 to 92% at the end of 2008. Corporate culture processes and technology must be inte-
• Efficiency and cost reduction: Integrating information se- grated optimally according to the organizations’ objectives
curity into an existing standard based management system and to collaborators needs and requirements, as well as con-
reduces in our case studies the effort for the development sidering didactical principles. Standards, methods and tech-
about 10-20% and thereby also the costs. The advantages nology are thereby only tools, which support organizational
are still higher during the implementation, where only the development and information security so far as this is admit-
audit is extended and the risk assessment must be main- ted by the culture. Therefore we need an open, confident
tained. based, participative corporate learning culture with criticism
• Shorter initial training periods for new collaborators: New and change readiness [11].
collaborators are quickly introduced into the management The collaborators should be interested in new knowledge,
system at their first working day. Thereby they focus on able for self-driven learning, have personal employment,
the principle information. Afterwards they access the team ability and change willingness [11]. All managers must
documentation work-integrated and need-oriented, when promote constantly information security and motivate their
they have questions for fulfilling their job. The lead time collaborators in following these principles.
460 STOLL

Introducing information security into holistic standard based organizations’ success, information security management
management systems requires a, the organization best should be enhanced by integrating it into all existing man-
adapted, system. This certainly can not be achieved by using agement systems.
a purchased general manual, which does not correspond with Implementing the holistic information security management
the lived processes and corporate objectives. Further the model promotes information security and thereby in accor-
management system must be constantly and effectively im- dance to organizations’ objectives availability, confidentiali-
plemented. ty and integrity, effectiveness, efficiency, legal and regulato-
This holistic information security model requires from the ry conformity, information security awareness and organiza-
system or information security manager, a part from standard tional development including information security improve-
management system skills, also information security skills, ment to guarantee stakeholder orientation and sustainable
the acquisition of necessary change management, organiza- organizations’ success.
tional learning, and controlling knowledge, the knowledge
about all relevant information security and data protection
laws, standards and regulations, as well as experience in in-
formation technology, knowledge management and informa- REFERENCES
tion management.
[1] Gordon, M. Loeb, W. Lucyshyn: CSI/FBI Computer Crime and Secu-
Sufficient IT-infrastructure and IT-support are also very im- rity Survey, i.cmpnet.com/gocsi/db_area/pdfs/fbi/FBI2006, 2006
portant for the project success. Only an appropriate IT sys- [2] <kes>-Microsoft® Sicherheitsstudie 2006, kes, vol. 4-6, Sonderdruck,
tem, technical equipment and infrastructure in accordance 2006
with organizational objectives promote continuously optimi-
[3] Federal Office for Information Security: Report about the IT security
zation of information security and sustainable organizations’ situation in Germany, bsi.de/english/publications/securitysituation/-
success. Lagebericht_2007_englisch.pdf
[4] ISO, ISO 9001:2008, Quality Managemnt Systems – requirements,
ISO, 1.12.2008.
[5] ISO, ISO/IEC 27001:2005, Information technology – Security tech-
VII. OUTLOOK niques – Information security management systems – Requirements,
5.10.2005.
Due to these excellent project experiences in several organi-
[6] ISO, ISO/IEC 27002:2005, Information technology – Security tech-
zations with different management systems there should be niques – Code of practice for information security management,
introduced enhanced information security management into 15.6.2005.
standard based management systems regarding all success [7] http://www.iso27001certificates.com/
factors [VI B].
[8] Y. Akao, Quality Function Deployment, integrating customer require-
Standard management system trainings should inform about ments into product design, Productivity Press, Portland, 1990.
the principles of information security management systems [9] M. Stoll Managementsysteme und Prozessorientiertes Wissensma-
and vice versa. nagement in Proc. 4th Conference on Professional Knowledge Man-
The management systems standards should emphasize the agement – Experiences and Visions, Gito Verlag, Berlin (2007), vo-
lume 1, pp. 433-434.
importance of information security and require the integra-
tion of information security into management systems. [10] T. Davenport: Thinking for a living, how to get better performance and
results from knowledge workers, Harvard Business School Press, Bos-
ton, 2005.
[11] P. Senge: Taking personal change seriously, The impact of Organiza-
tional Learning on management practice, Academy of Management
VIII. CONCLUSION Executive, vol. 17, p 47-50, May2003.

Integrating information security management into a process [12] M. Rosenberg: E-Learning, Strategies for Delivering Knowledge in the
Digital Age, McGraw-Hill, New York, 2001
oriented standard based management system enriches it.
Thereby information security can be introduced efficiently, [13] G. Reinmann-Rothmeier Mediendidaktik und Wissensmanagement. in
MedienPädagogik 10, www.medienpaed.com/02-2/reinmann1.pdf,
resources carefully, fast and successfully using many com- 2002.
mon synergies.
[14] M. Stoll, Workplace Process Integrated Learning and Knowledge Or-
Especially in organizations, where data, information and ganization, in H. Maurer and K. Tochtermann Eds. Proc. I-Know 07,
knowledge or availability of services, asset and/or infrastruc- 7th International Conference on Knowledge Management, J.UCS
ture or confidentiality or integrity has a high importance for Journal of Universal Computer Science, Graz, 2007.
 Evaluation of Asynchronous Event Mechanisms for
Browser-based Real-time Communication Integration
Keith Griffin Colin Flanagan
Cisco University of Limerick
Galway, Ireland Limerick, Ireland
[email protected] [email protected]

Abstract — Desktop based real-time mark on a green icon is used to display availability while
communication applications are commonly an icon representing the hands of a clock is overlaid on a
used for presence based instant messaging blue icon to represent an inactive state.
and telephony applications. Such applications
use installed desktop components to handle
real-time asynchronous events on the client
originating from the communication system.
When the client is based in a web browser
these installed components are not available
however browser-based communication
applications are required to handle the same
type of asynchronous events. Moreover,
browser-based clients which typically run
over HTTP are challenged by HTTP itself
which is designed to be a synchronous
request-response protocol. We contend that a
suitable mechanism can be found to deliver
asynchronous real-time events to browser-
based applications

Keywords: Browser, HTTP, Communication Integration,

Telephony, Presence, Web.

I. INTRODUCTION
In real-time communication integration systems such
as those providing telephony, instant messaging and
presence integration asynchronous events are a core and
mandatory feature. Basic events like a presence state
change from busy to available or lifting a telephone
handset resulting in an offhook event being delivered to an Fig. 1 Presence enabled buddy list in a desktop
application are fundamental to traditional desktop based communication client
communication integration applications. An example of
such an application displaying presence state is shown in A user’s presence state can be affected by multiple
Fig. 1 [1]. Presence is defined as a person’s availability sources as described in Fig. 2. These sources result in a
and propensity to communicate [2]. A person’s availability mix of user and system driven presence state changes.
can be reflected by a communication state. This state is
often presented as a colour, icon or colour coded icon. A • User driven presence: A user manually sets their
red phone icon for a contact may represent that the contact presence or availability state using a user interface
is busy and unavailable for voice communication, with such as a drop down menu. Presence states are
amber representing that they have been idle for some time often restricted to include “available”,
and green representing that they are available for voice “unavailable” or “busy” and “idle”.
communication. In this case the red, amber and green • System driven presence: A user’s state is
scheme which is commonly used in road traffic signals is automatically set by a system event e.g. logging
being use to represent availability and a phone icon is off from an instant messaging client would
being used to represent telephony or voice communication.
automatically set the user to “unavailable”.
In Fig. 1 the scheme that is used is to overlay a graphical
indication along with a colour scheme. A positive tick

461
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_80, © Springer Science+Business Media B.V. 2010
462 GRIFFIN AND FLANAGAN

Presence Source Sample Presence States first place. It has been found that several alternatives exist
User Busy, Available, Idle, In a to continuous polling which will be explored in the
meeting following sections.
System Logged in, Logged out, III. OVERVIEW OF ALTERNATIVE APPROACHES
Idle
Finding a solution for asynchronous events to browser-
Calendar Available, Busy based applications is not a trivial task. Browser-based
Online Meeting System / In a meeting, Meeting applications typically run over HTTP which despite having
Web Share state persistent connections is designed as a synchronous,
Telephony One the phone, call state stateless and connectionless protocol. Furthermore
Instant Messaging Message State communication related services such as presence generate
Fig. 2 Sample input sources for presence asynchronous events which require real-time handling.
Take the example of a user who has a presence state of
Regardless of how a presence state change occurs the “BUSY - ON THE PHONE” as a result of an integration
change must be reflected in applications subscribing to that between a telephony and presence service. This busy state
user’s presence in real-time. Some applications can may be displayed on a number of other clients who are
integrate directly with a protocol while others use subscribers or followers of this user’s presence state. The
middleware based integration [3]. While it is clear that number of clients depends on the number of others who
presence functionality is available to middleware and subscribe to the user’s presence and whether those users
installed desktop applications, it is proposed that suitable have a running client capable of displaying the presence
mechanisms can be found to offer presence functionality to state. When the user finishes the phone call the telephony
browser-based applications. We will use presence as a system will update the presence system to reflect a state of
sample real-time communication integration service to “AVAILABLE”. Once this state change happens, all other
discuss various mechanisms to extend real-time clients subscribed to that user’s presence must be updated
asynchronous presence events to a browser-based in real-time. Without this real-time update the value of the
application. system would be significantly diminished. In the case of an
installed desktop client which typically includes a local
II. ASYNCHRONOUS EVENTS FOR BROWSER-BASED protocol handler, receiving an asynchronous update could
APPLICATIONS OVERVIEW be described as straightforward although not trivial as it is
still a complex task. It is straightforward as the locally
Browser-based communication applications typically
installed desktop component can include a protocol
rely on HTTP as the protocol used to communicate with a
handler acting as a listener and ready to receive and handle
web server. HTTP 1.0 [4] improved on previously defined
incoming asynchronous events over a persistent
versions of the protocol which provided for raw data
connection. In the case of a browser-based client it is not
transfer over the internet. Improvements included allowing
so straightforward. The application running in the browser
messages to be of MIME like format containing meta
can make a request for a particular user’s presence from a
information about the data being transferred. Amongst
presence web service and it will get a response with the
other things HTTP 1.0 did not sufficiently take into
presence state. If that user’s presence state changes at an
consideration persistent connections. However persistent
unknown time in the future how will the browser-based
connections are considered in HTTP 1.1 [5] in order to
application become aware of this change? In the following
reduce the number of TCP connections and reduce
sections we will explore mechanisms to extend
congestion. However this does not address the ability for a
asynchronous real-time events to browser-based presence
client to receive real-time asynchronous events from a
applications.
server to a browser-based application. An asynchronous
event effectively amounts to a response without an explicit A. Continuous Polling and Streaming
request. The browser-based application will have One straightforward way to keep the client aware of
implicitly solicited an asynchronous event as part of an presence updates on the server is to not use an event
earlier request but the timing for an availability of that mechanism at all but instead to continuously ask what the
event will be unknown and driven by other events in the state is. Using this approach, rather than waiting for the
communication system. A mechanism is required which server to push events to the browser-based application the
will allow the delivery of asynchronous events which application implements a continuous cycle of HTTP
occur in the communication integration system to browser- requests for a given user or list of user’s presence. For
based clients. However any such mechanism must respect example request a presence state every second, receive and
the behavior of web-based systems and not allow immediate response, close the connection and repeat.
unsolicited events from the server to the client which could Various approaches to this technique have been suggested
create a security vulnerability. and implemented from the brute force technique suggested
One way to handle asynchronous events would be to above to more elegant standards submissions such as WS-
use a continuous polling mechanism but polling would Polling [6]. An obvious disadvantage of continuous polling
impact both the performance and scalability of the system is that it results in the server handling all of the polling
which are among the benefits that HTTP presents in the requests and as a result can be very inefficient on the
 EVALUATION OF ASYNCHRONOUS EVENT MECHANISMS 463

server and also on the client which has to handle Bayeux suggests that all requests originate from the
continuous page refreshes. An improvement to continuous client i.e. a server may not initiate a connection to a client
polling would be to use a technique commonly referred to unless a client has previously sent a request to a server
as streaming [7]. This involves making a request from the which supports the browser security model. However
browser-based application to the presence web service and asynchronous events intended for a given client are
then not closing the connection at the server after the supported once a client has made a request to the server.
response has been sent. Subsequent asynchronous presence
events for the initially requested user would then be The transport for server to client event delivery can
streamed over this connection. Unlike continuous polling terminate the HTTP response after all messages are sent as
the server is not required to handle continuous requests in the case of a polling transport or use a streaming
however it is required to keep a connection continuously technique that allows multiple event messages to be sent in
open which means that this connection cannot be used for the same HTTP response. Two transport types are
other purposes. This approach takes the connectionless, supported by Bayeux: Long-Polling and Callback-Polling.
synchronous request-response nature of HTTP and turns it Callback-polling is a recommended but not mandatory
into a connection oriented approach like an installed transport type as defined by Bayeux. Callback polling uses
desktop client might use with a persistent TCP socket HTTP GET to pass the name of the callback to be
connection. For a web based application providing real- triggered to the server whereas long polling uses a POST
time asynchronous events this is not thought to be a request and response.
scalable solution, as it locks up too many connections. In Connections between the client and server are
particular, browsers often place a limit on the number of negotiated with handshake messages that allow the content
simultaneous connections they will hold open to a server. type, authentication, protocol version and other parameters
A technique that can be combined with polling for a to be agreed. Comet is a term often used to describe the
more efficient approach is Asynchronous Javascript And behavior of the Bayeux protocol. In fact the Bayeux
Xml (AJAX). Ajax consists of a number of technologies protocol definition was led by the same person that coined
that combine as a mechanism which can be used to poll for the term “Comet”. Cometd is an implementation of the
a specific update without updating the entire content of the Bayeux protocol by the Dojo Foundation [10].
page. Two key technologies used by Ajax are JavaScript Bi-directional Asynchronous Events with LongPolling
and XMLHttpRequest (XHR) [8]. XHR can be used to In order to support bi-directional communications as
request or poll a presence web service for updates. required for asynchronous presence updates, Bayeux
Following such an update the XHR object can use a call clients use two connections to a server to allow client to
back function to update the browser-based application server and server to client messaging occur simultaneously
using JavaScript. An advantage of Ajax is that it can be as shown in Fig. 3. Bayeux offers a transport option called
used to update a specific data set such a user’s presence long-polling which attempts to hold open a request until
state without requiring an entire page refresh. This offers a there are related events to deliver to the client. The intent
more efficient approach however it still relies on a brute is to always have a pending request to deliver events as
force polling technique to realize presence state changes they occur. The reconnect and interval advice fields are
on the server, as described in this example. used to maintain requests over time and address resource
B. Long Polling starvation and increased server load. In the worst case this
behavior can degenerate to that of traditional polling but
Long polling is a technique whereby a browser-based used efficiently it can minimize both latency in server to
client sends a request to a server where the server holds the client message delivery and the network resources required
request open until a response is available. This is similar to for the connection.
the streaming technique in that the request is held open but
different in that the connection can be closed either due to Bayeux Bayeux
a defined timeout or when a response is received. Once a Client Server
Req 0
connection is closed a new connection is initiated. As a
result a given client typically has a request open when an Req 1
asynchronous event such as a presence event becomes .
available. .
Wait
Resp 1 .
The Bayeux Protocol and Comet
Bayeux is a protocol used to transport messages Resp 0
between a web server and web browser using named
channels [9]. Messages can be delivered as follows:
Fig. 3 Bi-directional Asynchronous Operation
• Client to server
Bidirectional Streams Over Synchronous HTTP
• Server to client The XMPP Standards Foundation describes
• Client to client (via server) Bidirectional streams Over Synchronous HTTP (BOSH)
[11] as “a transport protocol that emulates the semantics
of a long-lived, bidirectional TCP connection between two
464 GRIFFIN AND FLANAGAN

entities (such as a client and a server) by efficiently using IV. ASYNCHRONOUS EVENTS FOR BROWSER-BASED
multiple synchronous HTTP request/response pairs COMMUNICATION INTEGRATION
without requiring the use of frequent polling or chunked
Two of the previously outlined mechanisms were used
responses”. As can be seen from this definition, BOSH is
to generate results for presence updates in a browser-based
a long polling technique and claims to be more efficient in
application, namely continuous polling and BOSH. Ajax
terms of bandwidth and responsiveness than Ajax. The
was used as a common technique to update only the
BOSH standard calls for the use of a HTTP Connection
presence data-set in the browser without requiring a full
Manager which is effectively a HTTP web server capable
page refresh.
of implementing the long polling technique as described in
the standard. BOSH works by having the HTTP A. Results for Continuous Polling
connection manager respond to a browser side request only Using Cisco’s Presence Web Service [18] a browser-
when a server side event is available. Once the client has based client was used to get and set presence states. In this
received this event it immediately makes another request case presence updates were polled using Ajax at the client.
to provide a vehicle for the next asynchronous event. The The resulting presence state change was observed in the
differences between the BOSH and Bayeux/Comet browser client as shown in Fig. 4. The web page shows a
techniques appear to be minimal and are related to client for user Colin Flanagan who is subscribed to
implementation differences when run over HTTP 1.0 and presence state changes for user Keith Griffin. The presence
1.1. state for Keith Griffin has just changed to Do Not Disturb
Interestingly, a standard has been defined for the use of (DND) at this point of the test.
the commonly used Extensible Messaging and Presence
Protocol (XMPP) [12] standard with BOSH. The XMPP
over BOSH standard [13] defines how XMPP stanzas can
be transported over HTTP. As defined in XMPP an XML
stanza is a discrete unit of structured information which
can be used to represent a user’s presence state. Clearly
this proven technique provides a mechanism for real-time
asynchronous presence events to be delivered to browser-
based applications for XMPP based presence and instant
messaging systems.
C. Web Sockets
Two key components of the HTML5 [14]
specification, as it relates to real-time asynchronous events
for browser-based applications, are the web sockets API
[15] and server sent events [16]. Web sockets attempt “to
enable Web applications to maintain bidirectional
communications with server-side processes”. The web
socket protocol [17] claims to provide an alternative to
HTTP polling for bi-directional communications between
a web page and server by using a single TCP connection to
manage traffic in both directions. It uses HTTP to achieve
a handshake between client and server by requesting a
HTTP “upgrade: websocket”. Following this the Fig. 4 Presence enabled client with continuous polling
underlying TCP connection is used for bi-directional
client-server communications. If the protocol delivers on While performance related timing information was
this claim it will offer a single socket connection capable available in the traces the goal of the test was not to
of bi-directional messaging between client and server. This measure performance but to determine if asynchronous
sounds like the definition that one might expect for an presence state changes could be successfully delivered to
installed desktop client running natively over TCP. the browser. Trace output in Fig. 5 shows a basic presence
update delivered to the user Colin Flanagan as a result of
Once a websocket is established server sent events can user Keith Griffin changing presence to a “Do Not
be used to push real-time asynchronous events to a Disturb” (DND) state.
browser-based application. This is achieved by using the
event source interface which allows for the creation of an "<presenceList type="basic">[\n]"
event source object in the client and registering an event httpclient.wire.content - <<
listener. "<contact contactURI="[email protected]"
presenceStatus="DND"/>[\n]"
"</presenceList>[\n]"

Fig. 5 Trace displaying XML Presence Representation for DND

 EVALUATION OF ASYNCHRONOUS EVENT MECHANISMS 465

B. Results for Long Polling using BOSH V. EVALUATION

Using the sample JavaScript libraries with the Jabber Considering the nature of real-time communication
Extensible Communications Platform (XCP) XMPP based events, in particular presence, it is clear that a mechanism
implementation [19], BOSH long polling was observed in is needed to allow browser-based communication clients to
the client shown in Fig. 1. A long-poll interval set to 30 receive asynchronous real-time events in a manner similar
seconds was used. When an event was received to desktop clients which can use a dedicated protocol
asynchronously on the server it was sent to the browser as handler. Our goal is not to attempt to select the best web
a response to an open HTTP POST request. Once the event based asynchronous event mechanism for general purpose
was received in the browser a new HTTP POST was use as it is expected that research in this area will continue
initiated according to the BOSH long polling technique. for quite some time. However it is necessary to evaluate
available options that are capable of providing
asynchronous events to browser-based communication
applications.
All of the approaches outlined in this paper share the
goal of enabling the delivery of asynchronous events to
browser-based client applications. Continuous polling can
be wasteful of network and server resources by polling
continuously even when no data is available. Ajax is a
useful technique when used to update a particular data set
such as presence state while avoiding the need for a full
page refresh. Long polling techniques such as the Bayeux
Protocol and BOSH respect the fundamentals of HTTP
such as only maintaining two connections with the server.
Long polling also offers a more efficient use of network
and server resources by handling asynchronous events
over time and event controlled request-response pairs. It
should be noted that for high asynchronous event rates the
behavior of a given long polling technique approaches that
of continuous polling as a high rate of asynchronous
events result in an equivalent high rate of client driven
requests. This is perhaps a useful reminder that while
useful, long polling is itself a work-around technique as
opposed to a true solution for the delivery of asynchronous
Fig. 6 Presence enabled client with BOSH Long Polling events to browser-based applications. For the purposes of
this research BOSH is noteworthy as not only is it a long
It can be seen from the debug output in Fig. 6 that a polling technique that has been proven to work for instant
presence state change to DND for user “griffin” has been messaging and presence based communication systems but
received. Furthermore, it can be seen that the update was it is also designed to be fully compliant with HTTP 1.0 and
received in response to a HTTP POST request sent 16.08 therefore suitable for use with constrained clients. While
seconds earlier. It can also be seen in the debug output that the long polling technique can clearly offer a suitable
the following long poll expired without receiving an event solution for asynchronous presence event delivery to
before the 30 second timeout and the subsequent long poll browser-based applications the promise of bi-directional
is held open awaiting an asynchronous event from the messaging via HTTP using the websocket technique to
server. utilize the underlying TCP connection is promising.
However, at time of writing the websocket standard is still
In both examples presence state changes were observed being developed and it is too early to draw significant
on browser-based clients and installed desktop clients conclusions for its use with browser-based presence
based on an underlying protocol stack. In the case of applications. It is suggested that, at the time of writing,
continuous polling the installed client was SIP based and long polling is the most suitable mechanism that can be
in the case of long-polling the installed client was XMPP used to deliver asynchronous real-time communication
based. What has been shown in the results for the sake of events to browser-based applications.
brevity is an example of a single state change. While
intentionally not quantified by measurement, no difference VI. CONCLUSION
could be observed to the human eye for presence state
changes in the browser-based client. However it is We set out to identify a mechanism that would be
expected that the use of a continuous polling technique suitable for the delivery of asynchronous real-time
would result in degraded performance and scalability of communication events, namely presence events to
such a system when compared with a system using a long browser-based applications. After evaluating several
polling technique. techniques, including some recent and still under
development, it is clear that it is certainly possible to
466 GRIFFIN AND FLANAGAN

deliver asynchronous presence events to browser-based [4] Berners-Lee, T., R. Fielding, and H. Frystyk. HTTP – Hypertext
applications. While basic continuous polling techniques Transport Protocol (HTTP 1.0). RFC1945 1996 [cited; Available
can be used, they cannot scale. Web sockets offers the from: ftp://ftp.isi.edu/in-notes/rfc1945.txt.
promise of a “more native” solution in time, however [5] Fielding, R., et al. HTTP – Hypertext Transport Protocol (HTTP
currently it is suggested that long polling techniques
1.1). RFC2616 1999, ftp://ftp.isi.edu/in-notes/rfc2616.txt
(the Bayeux protocol or BOSH) are a suitable mechanism
[6] Brown, K., et al. Web Services Polling (WS-Polling). 2005,
for the delivery of asynchronous real-time events to
browser-based applications. http://www.w3.org/Submission/ws-polling/.
[7] Franklin, M. and S. Zdonik, Data in your face: push technology in
VII. FUTURE WORK perspective, in Proceedings of the 1998 ACM SIGMOD
While a suitable mechanism has been found for international conference on Management of data. 1998, ACM:
asynchronous real-time event delivery to browser-based Seattle, Washington, United States.
applications this mechanism has been identified as a work [8] W3C. XMLHttpRequest. W3C Working Draft 20 August 2009,
around technique rather than a solution. An obvious area http://www.w3.org/TR/XMLHttpRequest/.
for future research would be the pursuit of a true solution [9] Russell, A., et al. Bayeux Protocol Draft 1.0. 2007,
for real bi-directional messaging between web client and [10] Dojo. Cometd Bayeux Ajax Push. 2009,
server which could be used by real-time communication http://alex.dojotoolkit.org/2006/03/comet-low-latency-data-for-
systems.
the-browser/.
ACKNOWLEDGMENTS [11] Paterson, I., et al. XEP-0124: Bidirectional-streams Over
Synchronous HTTP (BOSH). 2005,
With thanks to David Mathieson and Seamus Kerrigan for
http://xmpp.org/extensions/xep-0124.html.
their tutorials on the implementation of JavaScript with
[12] Saint-Andre, P. XMPP Extension Protocols. 2001. XEP-0001.
Comet and XCP with BOSH respectively.
Available from: http://xmpp.org/extensions/xep-0001.html
REFERENCES [13] Paterson, I. and P. Saint-Andre. XEP-0206: XMPP Over BOSH.
[1] Jabber. The Power of Presence. 2008, 2007, http://xmpp.org/extensions/xep-0206.html.
http://www.jabber.com/CE/VideoVoiceCommunications. [14] Hickson, I. and D. Hyatt. A vocabulary and associated APIs for
[2] Rosenberg, J. A Presence Event Package for the Session Initiation HTML and XHTML. 2009, http://www.w3.org/TR/html5/.
Protocol (SIP). The Internet Society 2004, [15] Hickson, I. The Web Sockets API. 2009,
http://www.ietf.org/rfc/rfc3856.txt. http://dev.w3.org/html5/websockets/.
[3] Beltran, V. and J. Paradells, Middleware-based solution to offer [16] Hickson, I. Server-Sent Events. 2009,
mobile presence services, in Proceedings of the 1st international http://dev.w3.org/html5/eventsource/.
conference on MOBILe Wireless MiddleWARE, Operating [17] Hickson, I. The Web Socket protocol, 2009,
Systems, and Applications. 2007, ICST (Institute for Computer http://tools.ietf.org/html/draft-hixie-thewebsocketprotocol-44.
Sciences, Social-Informatics and Telecommunications [18] Cisco, Developer Guide for Cisco Unified Presence v7.0. 2008.
Engineering): Innsbruck, Austria. [19] Jabber. Jabber Extensible Communications Platform. 2008,
http://www.jabber.com/CE/JabberXCP.
 An Ontology-based Intrusion Detection for RFID
Systems
M. Esposito1,2, G. Della Vecchia1
1
ICAR-CNR, Via Castellino 111, 80131 Naples, Italy
{massimo.esposito, gennaro.dellavecchia}@na.icar.cnr.it
2
University of Naples “Parthenope”, Centro Direzionale, Isola C4, 80143, Naples, Italy
[email protected]

Although incremental improvements to the aforementioned

Abstract- In the last decade, RFID systems have gained
increasing popularity, but they also pose many critical security trade-off, many reasons to assume that the existing solutions
concerns. The most challenging security threat in several RFID are not be proficient to completely protect tags from cloning
applications is tag cloning. In this paper, we do not focus on can be pointed out.
preventing tag cloning, but propose the application of the Firstly, solutions instanced on the tag are intrinsically not
intrusion detection model to identify when tag cloning has secure, because, in an RFID system, tags are the weakest link
occurred. In particular, we present an ontology-based misuse in the whole chain due to their limited functional capabilities:
detection system (IDS) that integrates information coming from an attacker, even with poor resources, can violate their
RFID middleware layer to detect tag cloning. Ontologies and rules security quite easily. Moreover, it is disputable whether it will
are applied to formalize the declarative and procedural
be possible to produce a truly secure RFID tag, able to address
knowledge required to implement a “track & trace” technique,
with the final aim of modelling two levels of detection against tag all known vulnerabilities without improving the overall cost.
cloning. An inference engine is used to provide the proposed IDS Secondly, existing solutions mainly focus on preventing tag
with the advanced capacity of automatically reasoning about cloning, but a secure RFID system should go beyond
ontologies and rules in order to actually apply the formalized prevention measures, by providing detection capabilities to
model of detection and infer when a tagged object is cloned or identify when tag cloning has occurred [6].
victim of a cloning attack. In this work, we present an ontology-based intrusion
detection system (IDS) that integrates information coming
I. INTRODUCTION from RFID middleware layer to detect tag cloning. More
In the last decade, RFID (Radio Frequency IDentification) specifically, we propose a misuse detection approach that
systems have gained increasing attention, due to the growing implements a “track & trace” technique, essentially based on
number of investigations into different scenarios, varying from the reasoning that “when you know where a genuine tagged
personnel tracking and localization to healthcare monitoring object is, a cloned ones can be detected”. Such a technique has
[1-3]. Nevertheless, there are still a vast number of problems been implemented by applying ontologies in conjunction with
that need to be solved before their massive deployment. As a rules, respectively to i) develop a formal description of a
matter of fact, an intense public concern with security issues “track & trace” model and ii) formalize the declarative and
has gone largely unaddressed and poses critical challenges to procedural knowledge modelling two levels of defence against
several RFID applications. tag cloning. Ontologies and rules are utilized in conjunction
From the point of view of RFID technology, the most with an inference engine to provide the IDS with the advanced
challenging security threat in several RFID applications is tag capacity of automatically reasoning about the defined “track &
cloning. The research community addresses this threat trace” model, in order to infer if a tagged object is authentic,
primarily by applying variety of countermeasures, such as tag cloned or currently under attack.
cryptography, authentication and access control. The rest of the paper is organized as follows. Section II
The fundamental difficulties of such approaches revolve discusses related work. Section III describes our proposal of
around the trade-off between tag cost, level of security, and ontology-based intrusion detection for RFID systems. Section
hardware functionalities. As a matter of fact, RFID tags are IV overviews our IDS architecture and outlines the
typically deployed in numbers of several millions and the end- implementation details. Finally, section V concludes the
user companies have a strong financial incentive to minimize paper.
the tag cost and, thus, the features the tags provide [4]. II. RELATED WORK
As a result, it is extremely challenging to use cryptography for
Despite the benefits of RFID, it is essentially an insecure
protecting low cost tags from cloning, due to their limited
technology. In particular, tag cloning is one of the most
power, storage and processing resources.
serious threats to the security of RFID systems. Tag cloning,
Moreover, in order to supply cryptographic components with
i.e. when an attacker makes an exact copy of an RFID tag,
sufficient power, tags would need to be read from a shorter
enables an attacker to simply obtain the tag unique identifier,
distance, which would degrade the read-rate of readers [5].
typically not kept secret on the majority of tags, so as to make

467
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_81, © Springer Science+Business Media B.V. 2010
468 ESPOSITO AND VECCHIA

their cloned tags indistinguishable from the originals. Once Our approach closely resembles such an idea of applying
legitimate tag data has been obtained, attackers can reproduce the intrusion detection to identify tag cloning. But, differently,
their cloned tags on a wide scale. our research efforts have been primarily focused on
In high cost RFID tags, where resources are not very developing an intelligent misuse detection system based on
restricted, several countermeasures have been devised to ontologies and rules, that, applied in conjunction with an
combat tag cloning, such as deactivation of tags, encryption, inference engine, provide the advanced capacity of
authentication and hash codes [7]. In [8], Juels has automatically reasoning to detect tag cloning. To the best of
demonstrated some techniques for strengthening the resistance our knowledge, the RFID security literature has not yet
of EPC tags against cloning attacks, using PIN-based access to addressed applications of ontologies and rules to implement
achieve challenge response authentication. In [9], Weis et al. intrusion detection in RFID systems, neither system-oriented
have proposed a cryptographic approach to lock the tag researches appear to have been developed in that direction.
without storing the access key, but only a hash of the key on
III. THE ONTOLOGY-BASED INTRUSION DETECTION APPROACH
the tag. The key is stored in a back-end server and can be
found using the tag’s meta-ID. A. Misuse Detection
Avoine et al. [10] have proposed another hash-based RFID Traditionally, there are two basic approaches to intrusion
protocol that provides modified identifiers for improved detection: anomaly detection and misuse detection. In
privacy and that can be applied for authentication. In addition, anomaly detection, the goal is to define and characterize
hash-based RFID protocols for mutual authentication have legitimate behaviours of the users, and, then, detect anomalous
been proposed in [11-12]. All these protocols rely on behaviours by quantifying deviations from the former by
synchronized secrets residing on the tag and back-end server means of established metrics. Differently, misuse detection
and require a one-way hash function from the tag. involves the comparison of a user’s activities with known
In contrast, in low cost tags, due to their small size and behaviours of attackers, formalized in different models.
constrained resources, complex cryptographic solutions like The approach proposed in such a work relies on the
hash functions can not be executed. In [13] the authors application of a misuse detection based on ontologies and
mention scarcity of tag resources in low-cost RFID systems as rules to identify tag cloning in RFID systems. Such a choice is
a primary challenge in providing security and privacy due to a several of issues.
mechanisms, and, also, in combating cloning. To address this Firstly, a misuse detection based on ontologies and rules
issue, few lightweight authentication protocols that do not enables to have a clear understanding of what constitutes a
require cryptographic hash/keys in the tag have been proposed scenario of tag cloning, and, as a result, it is characterized by a
[14-15]. Another approach to combat cloning is the use of a low false positive rate. Secondly, it is more intuitive,
Physical Unclonable Function (PUF) [16]. PUFs significantly particularly in the case when users need to formalize and
increase physical security by generating volatile secrets that understand their knowledge about specific scenarios. This
only exist in a digital form when a chip is powered on and involves that a misuse detection system based on ontologies
running. Its main property is that it is easy to generate but hard and rules knows how tag cloning should manifest itself. This
to characterize. leads to a simple and efficient processing of the audit data.
All the cited solutions aim at preventing tag cloning. The obvious disadvantages of such a model rely on the time-
Additionally, tag cloning can be also detected by correlating consuming task of specifying ontologies and rules and the
information in the middleware layer. In such a sense, a very scarce ability to identify novel situations of tag cloning.
interesting approach has been proposed by Mirowski et al. [6]. Nevertheless, we believe that, in the context of our scenario,
More precisely, they have proposed an intrusion detection such issues do not constitute drawbacks and the proposed
system for RFID systems, named Deckard, based on a misuse detection can be proficiently applied, due to the very
statistical classifier and focused on the detection of change of limited amount of available data that, thus, significantly
tag ownership. This is one of the preliminary research to facilitates the task of specifying and keeping ontologies and
address the need for intrusion detection systems in RFID. rules up to date.
Another remarkable approach, similar to Deckard in its
intrusion detection architecture, has been proposed in [17]. B. The “track & trace” technique
Such an approach focuses beyond change in tag ownership The misuse detection has been actually realized by
and provides a more generic security framework to detect applying the “track & trace” technique. This technique relies
various RFID attacks. on the reasoning that “when you know where the genuine
The approaches outlined in [6, 17] are very remarkable for a tagged object is, the fake/cloned ones can be detected”.
several of ways. Firstly, they go beyond preventing tag In detail, “track & trace” refers to generate and store
cloning by actually determining whether the tag cloning has inherently dynamic profiles of RFID tagged object’s
occurred. Secondly, they avoid the difficulties onboard the movements through the supply chain. Such profiles are built
tags, by working with the middleware. As the reader and exploiting the set of location events related to one tagged
middleware components are typically accepted as the object that can be retrieved from a tracing system, such as the
expensive components of RFID, such proposals are very EPC network [18]. In addition to such dynamic profiles,
practical [6]. tagged objects are also characterized by static profiles, i.e.
fixed paths composed of point of interests (POI) through the
 AN ONTOLOGY-BASED INTRUSION DETECTION FOR RFID SYSTEMS 469

supply chain. Moreover, POIs specify normal usage conditions B

for tagged objects.

According to such profiles, it is possible to apply two A
C

levels of detection against tag cloning. More precisely, a first

level of defence refers to detect all the tagged objects that have C

an invalid tag serial number. The second level of defence is to D

detect whether a tagged object introduced into the supply

chain is cloned or currently under a cloning attack.
The first situation, i.e. a cloned tag is detected, occurs when: D

• a genuine tagged object is in multiple locations at the

same time, that means it is expected to be at a
particular point, and, at the same time, a tagged Fig. 1. An example of RFID tagged object path
object with the same ID is scanned at a different
point in the supply chain; Such a fine granularity of formalization is required because it
• a genuine tagged object moves in the supply chain is worth specifying, for example, which room inside a clothing
too fast, that means it is scanned at a particular point, shop at “Piazza del Plebiscito” in Naples the tagged object is
and, then, scanned again at a different point after a placed in. As a matter of fact, if the room inside the clothing
too short time interval. shop is different from the expected one, it is possible that the
tagged object is currently victim of a cloning attack.
The second situation, i.e. a genuine tag is recognized as victim In contrast, the dynamic profile of a tagged object
of a cloning attack, occurs when: represents an enlarging path of POIs, structured in terms of
• a genuine tagged object moves in the supply chain in physical locations transited through the supply chain. Physical
an abnormal way, that means it does not move locations can be inferred by means of the audit records
according to the fixed path defined in the static outcome by RFID read/write operations. As a matter of fact, a
profile, but goes, for example, from the stocking typical audit record can be logically structured as follows:
room of a clothing shop to its bathroom; <tag number, reader number, RFID operation, timestamp>.
• a genuine tagged object is utilized in the supply chain According to such data, it is possible to infer the physical
in an abnormal way, that means it is scanned at a location of an RFID tagged object for proximity to the reader
particular point too many times in a fixed time that has performed the operation.
window with respect to the expected usage condition. In order to determine which semantic location corresponds
to a physical one, semantic locations have been subdivided in
In order to formalize both the static and dynamic profiles many physical locations, that represent the minimal area in
to be associated to each tagged object, we have devised a which a tag can be localized, as shown in Fig.2.
“track & trace” model, based on the ideas of physical and
semantic locations. These notions are not new in literature [2],
but we have partially re-elaborated them.
Generally, a physical location specifies a precise region
identified by means of the proximity to well-known points.
Differently, a semantic location specifies the meaning of a
location and usually covers more physical locations. More
specifically, physical locations are the areas covered by RFID
readers, whereas a semantic location can be a country, a city, a
building, a room inside a building, a railway station and so on.
According to the typical use of RFID technology in indoor
situations, we have defined semantic locations for a generic
indoor environment, i.e. building, room, corridor, and so on. Fig. 2. Representation of the physical and semantic location perspectives
Each of these semantic locations is related to the city and
As a result, according to such a formalization and to the
country where it is placed in.
physical locations obtained from the audit records, it is easily
The static profile to be associated to each tagged object
to identify, for a tagged object, the corresponding semantic
represents a fixed path of POIs, structured in terms of
location, and, thus, verify the consistency of its movements
semantic locations to be transited through the supply chain.
with respect to its static profile. Moreover, for a tagged object,
An example of fixed path is shown in Fig.1. Such a path is
an expected usage condition is associated to each POI of its
composed of four notable POI: the first one is located in
static profile: if, for example, in the warehouse of the clothing
France, the second one in Switzerland, the third and the last
shop at “Piazza del Plebiscito”in Naples, the number of
ones in Italy. Each of these POIs has to be described in detail
readings related to an RFID tagged object exceeds a threshold
in terms of country, city, address, building and room inside the
indicating a normal behaviour, it is possible that the tagged
building, such as reported, as an example, in Fig.1.
object is currently victim of fraudulent readings.
470 ESPOSITO AND VECCHIA

Country
is As s oc iatedTo is As s oc iatedTo
RFID Tagged RFID Tagged Dynamic s ubClass Of is PartOf
Static Path Object Path
Object
has Dy namic POI Semantic
has Static POI City
Location
Static Point Dynamic Point maps s ubClass Of is PartOf
Of Interest Of Interest s ubClas s Of

has Us ageCondition has Semantic Loc ation

has Us ageModality has Phy s ic alLoc ation Physical
Location Building

Usage Semantic Usage Physical

is PartOf is PartOf
Condition Location Modality Location c ov ers

s ubClas s Of subClas s Of s ubClas s Of s ubClass Of RFID

Reader Room Corridor
RFID Tag RFID Tag RFID Tag RFID Tag
Writing Reading Writing Reading
subClas s Of
a) Static profile b) Dynamic profile c) Location intormation

Audit
Record
has Operation has Tag
hasReader

Usage RFID RFID Tagged

Modality Reader Object

d) Audit Record

Fig. 3. “Track & Trace” Ontology

C. Ontologies and rules for the “track & trace” model The second part of the ontology models a dynamic profile.
The “track & trace” model specifies information regarding The main concepts and properties are outlined in Fig. 3 (b),
static and dynamic profiles, locations and audit records in a whereas a more detailed list of properties is reported in Tab. 2.
unique and sharable manner. Nevertheless, this information
Role Domain Range Inverse Trans.
must be represented in an unambiguous and well-defined hasDynamicPOI DynamicPath DynamicPoint isDynamic no
formalism and expressed in a machine-readable format in hasDynamicPathID DynamicPath
OfInterest
Datatype: String
POIOf
- -
order to facilitate the application of reasoning techniques to isAssociatedTo DynamicPath RFIDTaggedObject has no
DynamicPath
compare static and dynamic profiles of tagged objects. hasUsageModality DynamicPoint UsageModality isUsage no
As a result, we have defined an ontology to formally OfInterest ModalityOf
hasPhysicalLocation DynamicPoint PhysicalLocation isPhysical no
describe the “track & trace” model, in terms of relevant OfInterest LocationOf
concepts, properties and interrelations. More precisely, Fig.3 hasOperation
Counter
UsageModality Datatype: Integer - -

presents a high level view of the defined ontology, named hasTimestamp UsageModality Datatype: Time - -
hasTagID RFIDTaggedObject Datatype: String - -
“Track and Trace” Ontology, and formalized in standard OWL hasTagData RFIDTaggedObject Datatype: String - -
(Web Ontology Language) [19].This ontology is subdivided in hasValidID RFIDTaggedObject Datatype: Boolean - -
isCloned RFIDTaggedObject Datatype: Boolean - -
four logical parts for sake of clarity. isUnderAttack RFIDTaggedObject Datatype: Boolean - -
The first part of the ontology is intended to model a static hasPhysical
LocationID
PhysicalLocation Datatype: String - -

profile of a tagged object. In Fig. 3 (a), the main concepts and

properties are outlined. A complete and detailed list of Tab. 2 Properties for modelling a dynamic profile
properties of each concept and sub-concept (denoted by The third part of the ontology is devised to specify location
ellipses) is not shown in figure, because it would make the information by means of concepts, as outlined in Fig. 3 (c),
representation unwieldy, but it is reported in Tab. 1. and properties, as diffusely reported in Tab. 3.
Role Domain Range Inverse Trans. Role Domain Range Inverse Trans.
hasStaticPOI StaticPath StaticPoint isStatic no isPartOf SemanticLocation SemanticLocation hasPart yes
OfInterest POIOf
hasLocationName SemanticLocation Datatype: String - -
hasStaticPathID StaticPath Datatype: String - -
hasPhysical PhysicalLocation Datatype: String - -
isAssociatedTo StaticPath RFIDTaggedObject has no LocationID
StaticPath
maps PhysicalLocation SemanticLocation isMappedWith -
hasUsageCondition StaticPointOf UsageCondition isUsage no
covers RFIDReader PhysicalLocation isCoveredBy no
Interest ConditionOf
hasReaderID RFIDReader Datatype: String - -
hasSemanticLocation StaticPointOf SemanticLocation isSemantic no
hasAddress Building Datatype: String - -
Interest LocationOf
hasFloors Building Datatype: Integer - -
hasStaticPOIID StaticPointOf Datatype: String - -
Interest hasRooms Building Datatype: Integer - -
hasTagID RFIDTaggedObject Datatype: String - - isOnFloor Room Datatype: Integer - -
hasTagData RFIDTaggedObject Datatype: String - - isBorderingOn Room Room - -
hasSemantic SemanticLocation Datatype: String - -
LocationName Tab. 3 Properties for modelling location information
hasConditionName UsageCondition Datatype: String - -
hasMaxValue UsageCondition Datatype: Integer - - Finally. the last part of the ontology specifies information
Tab. 1 Properties for modelling a static profile contained in an audit record by means of concepts, as outlined
in Fig. 3 (d), and properties, as diffusely reported in Tab. 4.
Each property is defined in terms of domain (the set of
possible subject concepts) and range (the set of possible object Role Domain Range Inverse Trans.
hasOperation AuditRecord RFIDOperation isOperationOf no
concepts or datatypes). Besides, the inverse property is hasTag AuditRecord RFIDTaggedObject isTagOf no
reported, if it is admissible, and the transitiveness is specified, hasReader AuditRecord RFIDReader isReaderOf no
hasTimestamp AuditRecord Datatype: Time - -
if it exists. It is worth noting that each sub-concept inherits
super-concept properties and adds new specialized ones. Tab. 4 Properties for modelling an audit record
 AN ONTOLOGY-BASED INTRUSION DETECTION FOR RFID SYSTEMS 471

Moreover, according to such an ontology, we have formalized, usage condition of the corresponding static
in a set of SWRL (Semantic Web Rule Language) [20] rules, POI, in terms of number of operations
the declarative and procedural knowledge modelling the two allowable in a pre-defined time window.
levels of detection previously described. For sake of brevity
and clarity, we have grouped the implemented rules in logical An inference engine is utilized in conjunction with such
sets and reported their description only in natural language: formalized ontologies and rules to automatically reason about
• Analysis and processing of data contained in audit static and dynamic profiles of tagged objects and, thus, to
records. Depending on data contained in an audit implement the two reported levels of detection against tag
record, diverse sets of rules have been formalized for: cloning.
o identifying the dynamic profile to be updated
IV. THE ONTOLOGY-BASED IDS ARCHITECTURE
with new information;
o determining the physical location of the The proposed IDS has been devised to be transparently and
sensed tag for proximity to the reader that seamlessly integrated into a given RFID system by means of
has performed the operation; being positioned at the middleware level in order to effectively
o adding a new dynamic POI in the case when shield back-end systems from cloning attacks. Thus, even if a
the detected physical location belong to no cloned tag enters the reader field, its data will not propagate
POI. Such a POI will be also characterized beyond the IDS, the backend will never be aware of its
by the usage modality and the relative presence and normal operations will continue as if nothing
timestamp, both extracted by the audit happened. The architecture of the proposed IDS consists of the
record; set of components described below:
o updating an existing dynamic POI in the • Target System. It is a typical RFID system capable of
case when the detected physical location producing data that summarizes activity of tags in its
belong to an existing POI, but a new coverage area. It is responsible for recording the
operation has been performed. The existing outcome of a RFID read/write operation into an audit
POI will be characterized by the new usage record.
modality and the relative timestamp, both • Knowledge Base. It contains the formalization of the
extracted by the audit record; domain knowledge regarding the possible scenarios
o updating the usage modality of an existing of tag cloning. It consists of three entities: i) a
dynamic POI in the case when the detected Terminological Box, populated with the previously
physical location belong to an existing POI, described OWL ontologies, ii) the Rule Box,
and the operation has been already populated with the previously described SWRL rules,
performed in the past. The existing POI will and iii) the Assertional Box, populated with the audit
be updated with the timestamp and the records, appropriately formalized in RDF (Resource
occurrences of the last performed operation. Description Framework) [21].
• Execution of the first level of detection. A set of rules • Auditing Module. It is the component responsible for
have been formalized for detecting all the tagged collecting and logging the audit records received
objects that have an invalid tag serial number. from the target system into the Assertional Box of the
• Execution of the second level of detection. Depending Knowledge Base.
on data stored in both the static and dynamic profiles, • Decision Module. It is the smart component that
sets of rules have been formalized for: implements the “track & trace” technique by
o determining the correspondence existing applying ontologies and rules, as described in the
between semantic and physical locations; previous section. It relies on the logic inference
o inferring that a tagged object is cloned in the engine presented in [22], able to integrate and reason
case when it is detected simultaneously in at about ontologies and rules to perform inferential
least two different and not bordering reasoning on the domain knowledge stored into the
dynamic POIs in a too short time window; Knowledge Base.
o inferring that a tagged object is cloned in the
case when a dynamic POI is not expected in With such components in mind, the proposed IDS operates in
the static profile and it is not bordering with the following manner. Firstly, the Target System records the
static POIs; details of a new operation into an audit record. This record is,
o inferring that a tagged object is under attack then, stored by the Audit Module inside the Assertional Box of
in the case when a dynamic POI is not the Knowledge Base. Secondly, according to the data
expected in the static profile, but it is contained in the new audit record, the Decision Module
bordering with static POIs; updates the dynamic profile of the appropriate tag and
o inferring that a tagged object is under attack activates the inference engine in order to compare the
in the case when the usage modality of a appropriate static and dynamic profiles stored into the
dynamic POI is not in accordance with the
472 ESPOSITO AND VECCHIA

Knowledge Base, and, thus, infer if the sensed tag is authentic, [4] M. Lehtonen, D. Ostojic, A. Ilic, and F. Michahelles, “Securing RFID
Systems by Detecting Tag Cloning”, in Proceedings of the 7th
cloned or currently under attack. International Conference on Pervasive Computing, Lecture Notes In
An experimental prototype for testing purpose has been Computer Science, Springer-Verlag, 2009, vol. 5538,pp. 291-308. DOI
implemented at the Institute for High Performance Computing 10.1007/978-3-642-01516-8_20.
and Networking (ICAR) of Italian National Research Council [5] D.C. Ranasinghe, D.W. Engels, and P.H. Cole, “Low-Cost RFID
(CNR). All the described components have been implemented Systems: Confronting Security and Privacy”, in Auto-ID Labs Research
as fully portable Java entities, with the exception of the Audit Workshop, Zurich, Switzerland, 2005.
Module, which also integrate, as native code, Microsoft Visual [6] L. Mirowski, and J. Hartnett, “Deckard: A system to detect change of
C++ 6.0 control drivers provided by the RFID technology RFID tag ownership”, International Journal of Computer Science and
Network Security, 2007, vol. 7, pp. 89-98, ISSN 1738-7906 (2007).
manufacturers. Specifically, the Target System consists of
RFID readers belonging to FEIG’s Passive RFID Long Range [7] T. Karygiannis, B. Eydt, G. Barber, L.Bunn, and T. Phillips, “Guidelines
for securing radio frequency identification (RFID) systems,” April 2007,
Reader series, model ISC.LRU2000-A. These readers are able NIST Special Publication 800-98.
to interact with EPC Class 1 Gen 2 transponders for operating [8] A. Juels, “Strengthening EPC tags against cloning”, in Proceedings of
frequencies in the UHF range 865-870 MHz and reading range WiSe '05 (Cologne, Germany, September 2005), ACM, New York, NY,
of up to 10 meters (approx. 33 feet). FEIG’s UHF antennas 2005, pp. 67-76, DOI 10.1145/1080793.1080805.
model ID ISC.ANT.U250/250-EU have been installed, too. [9] S. Weis, S. Sarma, R. Rivest, and D. Engels, “Security and Privacy
These antennas are circular polarized for operating frequencies Aspects of Low-cost Radio Frequency Identification Systems”. in
in the UHF range 865-870 MHz. Proceedings of SPC 2003, Lecture Notes in Computer Science,
Springer-Verlag, 2004, vol. 2802, pp. 50-59, DOI 10.1007/b95124.
Preliminary tests performed on the experimental prototype
aimed at detecting synthesized cloning attacks gave a proof of [10] G. Avoine, and P. Oechslin, "A scalable and provably secure hash based
RFID protocol”, in Proceedings of PerSec 2005, IEEE Computer Society
the feasibility of our approach, suggesting that an actual Press, 2005, pp. 110–114, DOI 10.1109/PERCOMW.2005.12.
deployment of the proposed IDS can effectively support the
[11] E.Y. Choi, S.M. Lee, and D.H. Lee, “Efficient RFID authentication
detection of cloned tags in RFID systems. protocol for ubiquitous computing environment”, in Proceedings of
SecUbiq 2005, Lecture Notes in Computer Science, Springer-Verlag,
V. CONCLUSION
2005, vol. 3823, pp. 945-954, DOI 10.1007/11596042.
Detection may be seen as the first step in defending against [12] S. Lee, T. Asano, and K. Kim, “RFID Mutual Authentication Scheme
tag cloning and preventing RFID-enabled crime from based on Synchronized Secret Information”, in Proceedings of In
occurring. In this perspective, the present paper has described Symposium on Cryptography and Information Security, Hiroshima,
Japan, January 2006.
a misuse detection approach, based on the “track & trace”
technique, for RFID systems. Ontologies and rules have been [13] S. Sarma, S. Weis, and D. Engels, “Radio-frequency identification:
Security risks and challenges”, RSA Laboratories Cryptobytes, 2003,
used in order to implement such a technique, by enabling the vol. 6, no 1, pp. 2-9.
formalization of the declarative and procedural knowledge
[14] S. Karthikeyan, and M. Nesterenko, “RFID security without extensive
required to model two levels of defence against tag cloning. cryptography”, in Proceedings of SASN'05, ACM, New York, NY,
Moreover, an inference engine has been used to provide the 2005, pp. 63-67, DOI 10.1145/1102219.1102229.
proposed IDS with the advanced capacity of automatically [15] H.-Y. Chien, “SASI: A new ultralightweight RFID authentication
reasoning about ontologies and rules in order to actually apply protocol providing strong authentication and strong integrity”, IEEE
the formalized model of detection, and infer when a tagged Transactions on Dependable and Secure Computing, 2007, vol. 4, no 4,
pp. 337-340, DOI 10.1109/TDSC.2007.70226.
object is cloned or victim of a cloning attack.
As concluding remarks, it is relevant to observe that the [16] S. Devadas, E. Suh, S. Paral, R. Sowell, T. Ziola, and V. Khandelwal,
“Design and implementation of PUFbased “unclonable” RFID ICs for
approach described in this paper represents only a preliminary anti-counterfeiting and security applications”, in Proceedings of the
step of an ongoing research, because, at the moment, an 2008 IEEE international conference on RFID, IEEE Computer Society
experimental prototype has been tested only over proof Press, 2008, pp. 58–64, DOI 10.1109/RFID.2008.4519377.
cloning attacks. Next step of our research will be to minutely [17] G. Thamilarasu, R. Sridhar, "Intrusion detection in RFID systems", in
investigate the feasibility of the approach in real cases, in Proceedings of MILCOM 2008, IEEE Computer Society Press, 2008,
terms of performance evaluation and experimental assessment. pp. 16-19, DOI 10.1109/MILCOM.2008.4753218.
[18] EPCglobal Inc. “EPCglobal Architecture Framework Version 1.3”,
REFERENCES
http://www.epcglobalinc.org/standards/architecture/architecture_1_3-
[1] M. Ciampi, A. Coronato, G. De Pietro, M. Esposito, “A Location framework-20090319.pdf.
Service for Pervasive Grids”, in Proceedings of CISSE 2005, Book
[19] Web Ontology Language, available at www.w3.org/TR/owl-features/ .
“Advances in Systems, Computing Sciences and Software Engineering”,
Springer, 2006, pp. 119-123, DOI 10.1007/1-4020-5263-4_19. [20] Semantic Web Rule Language, available at
www.w3.org/Submission/SWRL/ .
[2] A. Coronato, M. Esposito, G. De Pietro, "A Multimodal Semantic
Location Service for Intelligent Environments: An Application for Smart [21] Resource Description Framework, available at www.w3.org/RDF/ .
Hospitals", Journal of Personal and Ubiquitous Computing (JPUC), [22] M.Esposito, “An Ontological and Non-monotonic Rule-based Approach to
Springer-Verlag, 2009, vol. 13, no 7, pp. 527-538, DOI 10.1007/s00779- Label Medical Images”, in Proceedings of SITIS 2007, IEEE Computer
009-0223-x. Society Press, 2007, pp. 603-611, DOI 10.1109/SITIS.2007.77.
[3] G. Della Vecchia, M. Esposito, "A Pervasive System for Nuclear
Medicine Departments", Journal of Wireless Personal Communications
(WPC), Springer, 2009, DOI 10.1007/s11277-009-9789-x.
 A New Algorithm and its Implementations for RC Filters
GUOLIANG ZENG
Phone: (480) 727-1905; Fax: (480) 727-1773; E-mail: [email protected]
Motorola DSP Laboratory
Department Engineering
Arizona State University Polytech
7231 Sonoran Arroyo mall, Santan Hall, Mesa, AZ 85212

ABSTRACT: In digital communication, KEYWORDS: RC filter, Convolution,

it has been an important topic to Correlation, Digital communication.
implement an RC (Raised Cosine) filter
digitally. Based on the characteristics of INTRODUCTION: In a digital
the digital communication, a new communication system, information bits
symbol-based algorithm is proposed and are carried by symbol pulses. Due to the
implemented with MATLAB. It is also finite bandwidth of a practical channel, a
realized on Motorola’s DSP56303 in time-limited pulse, like square pulse will
assembly language. The new algorithm be distorted by the channel, causing
features the minimum data buffer and Inter-Symbol-Interference (ISI). This is
execution time, hence the minimum out- because a time-limited pulse has an
put time delay. Therefore, it is very infinite bandwidth of frequency. To
practically useful for high-speed data achieve zero ISI in a practical system,
transmission. This new method can be each symbol pulse received must be a
applied to any convolution and/or band-limited Nyquist pulse. The widely
correlation of two digital signals with used band-limited Nyquist pulse is the
different sampling periods. It should be family called Raised Cosine (RC) pulse.
applied in all up- and down-sampling A raised cosine (RC) pulse is
signal processing systems. given by the following equation [1]:

sin(πt / Ts ) cos(παt / Ts ) cos(παt / Ts ) , − ∞ < t < ∞ (1)

rc(t ) = = sin c(t / Ts )
πt / Ts 1 − (2αt / Ts ) 2 1 − (2αt / Ts ) 2

Here, 0 ¡ α ¡ 1 is called roll-off factor. The Ts is the symbol period. The spectrum of the
RC pulse is the Fourier transform of equation (1):

­Ts , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , ω ≤ π (1 − α ) / Ts
°
°Ts {1 + cos[ 1 (T ω − π (1 − α ))]} =
°2 2α s
(2)
RC (ω ) = ®
°Ts cos 2 [ 41α (Ts ω − π (1 − α ))], , , , , , , , , , , , , π (1 − α ) < ω ≤ π (1 + α )
° Ts Ts
°
¯0, , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , ω > π (1 + α ) / Ts

Therefore the frequency bandwidth of an RC pulse is finite. It can pass some practical
channels without distortion.
473
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_82, © Springer Science+Business Media B.V. 2010
474 ZENG

A filter is called an RC filter if its The output sequence is the convolution

impulse response h(t) is an RC pulse: of input sequence and impulse response
h(t) = rc(t). Clearly, an RC filter is a sequence:
low-pass non-causal filter.
A well equalized (flat) digital y[n] = x[n] ∗ h(n) = ¦ j =− DN h[ j ]x[n − j ] ,
DN −1

transmission system can be modeled as

an RC filter with the following input: n = 0,±1,±2, (7)

x(t ) = ¦ k = −∞ d [k ]δ (t − kTs ) ,
∞ CURRENT ALGORITHM: Consider
the input sequence x[n] obtained by (6).
−∞<t <∞ (3)
From equation (3), every N elements of
x[n] , there is only one nonzero element,
Here, k = 0,±1,±2, is the index for
which is the information data d[k]. Other
symbols. Ts is the symbol period. And, N – 1 elements are zeros. In fact, the
d[k] is the information data sequence. It sequence of d[k] is the only physical
consists of 1s and 0s for binary unipolar input. In practical implementations, for
signaling, or 1s and –1s for binary example, in the function “rcosflt” of
bipolar signaling. The output is the MATLAB [2], x[n] is formed by
convolution of input and impulse “upsampling” which pads N – 1 zeros
response: for each d[k]. Since “rcosflt” uses
built-in functions, it is rewritten for
y (t ) = x(t ) ∗ h(t ) = ¦k = −∞ d [k ]h(t − kTs ) (4)
∞
comparison as follows:

function [yo,to] = rcosflt1(x,

Note that the duration of h(t ) is from - Fd, Fs, R, Delay)
to  as shown in (1). For a practical %x = the input symbol sequence.
digital filter, it must be truncated, %The symbol rate is Fd(sym/sec).
shifted, and sampled. Assume that h(t ) %The sample rate is Fs(sam/sec).
%Fs must be an integer multiple.
is truncated from -D Ts to D Ts , for some %of Fd. R = the roll-off factor,
integer D. And N samples are taken in % R must be in the range [0, 1].
%DELAY is time delay.
each symbol period Ts . That is the %Y = the output sample sequence.
sampling period is Ts /N. Therefore the %
N=Fs/Fd;
impulse response sequence has 2DN % Design filter.
samples: th=linspace(-Delay, Delay,
2*Delay*N+1);
for i=1:length(th)
h[n] := rc (n Ns ) ,
T
if th(i)==0.0
n = -DN, -DN + 1, ¢¢¢, DN – 1. (5) yf(i)=1.0;
elseif abs(th(i)) == 1/
(2*R*Fd)
To get the output by convolution, the yf(i)=(sin(pi/(2*R)))*
input x(t) must be sampled with the same (R/2);
sampling period Ts /N: else
yf(i)=((sin(pi*th(i)*Fd))
*(cos(pi*R*th(i)*Fd)))
x[n] := x(n TNs ) , n = 0,±1,±2, . (6) / ((pi*th(i)*Fd).*(1–
 A NEW ALGORITHM AND ITS IMPLEMENTATIONS FOR RC FILTERS 475

(2*R*th(i)*Fd).^2)); k = -D, -D+1, ¢¢, D–1; i = 0, 1, ¢¢, N–1 (8a)

end Here, integer k is the symbol index,
end
% Upsample input x from Fd to Fs integer i is the sample index in each
%by padding N-1 0s per symbol. symbol period. Therefore,
lx=length(x); § h−D 0 , h− D1 ,, h−D ( N −1) ·
xx = zeros(1,(lx+Delay*2)* N); ¨ ¸
for i = 1:lx ¨ h( − D+1) 0 , h( − D+1)1 ,, h( − D+1)( N −1) ¸
xx((i-1)*N + 1) = x(i); ¨ ¸
end; h = ¨ ¸ (8b)
% Convolve xx and yf for output. ¨ ¸
lxx=length(xx); lyf=length(yf); ¨ ¸
lo=lxx+lyf-1; ¨ h( D−1) 0 , h( D−1)1 ,, h( D−1)( N −1) ¸
if lyf > lxx
© ¹ 2 D×N
tmpt=yf; yf=xx; xx=tmpt; At time t = k Ts , let the data vector be
end
d=(d[k+D],d[k+D-1],…,d[k-D+1])1×2D
yo=[];
for n=0:lo-1 k = 0,±1,±2, (9)
temp=0.0;
if n<lyf
for m=0:n
Hence, the output is the sequence of the
temp=temp+xx(m+1) * following vectors:
yf(n-m+1);
end yk = (d * h)1× N , k = 0,±1,±2, (10)
elseif n<lxx
for m=(n-lyf+1):n
temp=temp+xx(m+1)* Here, y k is a vector consisting of N
yf(n-m+1);
end
elements. It is not difficult to check that
else equation (10) gives the same out put as
for m=(n-yf+1):(lxx-1) y[n] in (7). Equations (8), (9) and (10)
temp=temp+xx(m+1)* form a new, symbol-based algorithm.
yf(n-m+1); An algorithm specifies the arithmetic
end
end operations to be performed but does not
yo=[yo temp]; specify how that arithmetic is to be
end implemented. Therefore, the implements
to = [0:lo-1]/Fs; of an algorithm are also important
especially for engineers. Now, we
The above current algorithm is sample- implement the new algorithm with
based, that is it processes signal sample MATLAB:
by sample. This requires a lot of zeros
padded in x[n] . They waste a lot of function [yo,to] = rcosflt2(x,
memory space and calculation time, Fd, Fs, R, Delay)
%x = the input symbol sequence.
causing bigger delay for the output. This %The symbol rate is Fd(sym/sec).
limits the speed of data transmission. To %The sample rate is Fs(sam/sec).
overcome this shortcoming, a new %Fs must be an integer multiple.
algorithm is proposed as follows. %of Fd. R = the roll-off factor,
% R must be in the range [0, 1].
%DELAY is time delay.
NEW ALGORITHM: Let us rewrite %Y = the output sample sequence.
equation (5) as a 2D×N matrix h: %
h = (hki)2D×N , with hki := rc((k + Ni )Ts ) , N=Fs/Fd;
476 ZENG

%Design impulse response matrix:

for k=1:2*Delay Furthermore, since the decisions in
for i=1:N
th=((k-Delay-1)+ digital communication are made symbol
(i-1)/N)/Fd; by symbol, the new algorithm can easily
if th==0.0 be realized in hardware for real-time
h(k,i)=1.0; processing as shown below.
elseif abs(th)==1/
(2*R*Fd)
h(k,i)=(sin(pi/ HARDWARE REALIZATION: The
(2*R)))*(R/2); architecture and addressing modes of a
else DSP chip are designed for vector
h(k,i)=((sin(pi*th* (parallel) processing. Hence, the
Fd)).*(cos(pi*R*th*
proposed new algorithm can be
Fd)))./((pi*th*Fd).*
(1-(2*R*th*Fd).^2)); implemented with the DSP chips
end efficiently. The following assembly
end language program is tested with
end Motorola’s DSP56303 in a real-time
y=[x' zeros(1,2*Delay)];
d=zeros(1,2*Delay);
processing system [3].
yo=[];
for j=1:length(y) ;M = 2D = # of symbols for the
d=circshift(d,[0 1]); ;impulse response h(t).
d(1)=y(j); ;N=# of samples in each symbol.
yj=d*h; ;x0 = input symbol.
yo=[yo yj]; ;a = output sample.
end ;x:$ffef=SSI RX/TX data registr.
to = [0:length(yo)-1]/Fs; ;Xbase = base location of filter
;states d[k] in X memory.
;Ybase=base location of impulse
Obviously, the new algorithm is symbol- ;response h[k,i] in Y memory.
based, that is it processes signal symbol
by symbol. In contrast with the sample- ;Assume M=4, N=6:
based one, it only requires about one Nth M equ 4
N equ 6
memory space and processing time ;X memmmory:
(number of multiplications). Usually, N Org x:Xbase
= 8 or 16. List1 dc d[-2], d[1],
The following plot shows the same result d[0], d[-1]
(output) of the above two algorithms ;new data d[2] in x:$ffef
;Y memmmory:
with Fd=1, Fs=8, R=0.5, Delay=3. Org y:Xbase
List2 dc h-2,0,h-2,1,h-2,2,
1.5
h-2,3,h-2,4,h-2,5,
h-1,0,h-1,1,h-1,2,
1
h-1,3,h-1,4,h-1,5,
0.5 h0,0, h0,1, h0,2,
h0,3, h0,4, h0,5,
Amplitude

0
h1,0, h1,1, h1,2,
-0.5 h1,3, h1,4, h1,5
-1
move #Xbase,r0
-1.5 ;point to states d[k].
0 5 10 15
Time
20 25 30 move #M-1,m0 ;mod(M).
 A NEW ALGORITHM AND ITS IMPLEMENTATIONS FOR RC FILTERS 477

move #Ybase,r4 CONCLUSION: This paper proposes

;point to impulse rspns. a new, symbol-based algorithm and its
move #MN-1,m4
;mod(MN). implementations in MATLAB and on
move #N,n4 Motorola’s DSP56303 for RC filters.
;offset(N). Contrasting the traditional sample-based
algorithm, the amount of reduction in
loop
movepx:$ffef, x0
memory space and execution time is
;readA/D. directly proportional to N. Generally, N
do #N,doend is greater than or equals 8. Therefore, the
clr a x0,x:(r0)+ improvement is very notable.
y:(r4)+n4,y0
rep #M-1
mac x0,y0,a REFERENCES:
x:(r0)+,x0 [1] A. Bateman, Digital
y:(r4)+n4,y0 Communications, Addison-Wesley,
macr x0,y0,a x(r0),x0 1999.
(r4)+
doend movep a,x:$ffef
[2] MATLAB R2008b, The Math Works
;write D/A Inc., 2008.
operation __, __, _ [3] Guoliang Zeng, A New
(r0)- (r4)-n4 Implementation of Root Raised
jmp loop Cosine Filter with Motorola’s
DSP56K, Proceedings of the
International Embedded Solutions
Event, September, 2004.
 A Computationally Dynamic Resource
Management for Multicarrier Communications
in Downlink for OFDMA Systems
Kabaou Mohamed Ouwais1, Rhaimi Belgcem Chibani1, Abdelkrim Med Naceur1, Marzoug Mongi2
1
National School of Engineering (ENIG), R.U: Modeling, Analysis and Control Systems (MACS), 6029-Gabès-
Tunisia
2
Orange-Groupe France Telecom, Département Réseaux fixes et Mobiles, 41-45 Bd Romain Rolland, 75672
Paris, Cedex 14, Paris-France
Email: [email protected], [email protected], [email protected],
[email protected]

Abstract subcarrier and power allocation strategy for

downlink communication to multiple users in an
Multicarrier transmission schemes have been widely OFDM based wireless system. We formulate the
adopted in wireless communication systems. Multiple
problem of minimizing total power consumption
users can share a multicarrier channel using multiple
access mechanisms such as OFDMA. This paper
with constraints on BER (Bit Error Rate).
studies some unique features of multicarrier Orthogonal Frequency Division Multiple Access
transmission and how these effect the performance of (OFDMA) is a promising multiple access technique
a multiple access system. Algorithms are derived for for next wireless broadband networks. By dividing
resource allocation in static and dynamic OFDMA the bandwidth into multiple subsets of subcarriers
systems. Numerical results demonstrate that the (named subchannels), OFDMA exploits the multi-
proposed approach (algorithm) offer comparable user diversity and thus increases the system
performance with existing algorithms. In this paper, a throughput.
new algorithm for subcarrier allocation (Hybrid
BABS) for the downlink of multiuser OFDM II. RADIO RESOURCE MANAGEMENT FOR
transmission is presented. The proposed algorithm is
more stable and it offers a lower complexity and
ORTHOGONAL FREQUENCY DIVISION
better performance than previous existing algorithms. MULTIPLE ACCESS (OFDMA)

Keywords One of the biggest advantages of OFDM systems is

the ability to allocate power and rate optimally
OFDMA, Resource Management, Bandwidth across frequency, using multiple methods
allocation, Greedy algorithm, Multiuser, Subcarrier, (algorithms) for scheduling, management and
Frequency allocation, Dynamic allocation. allocation of radio resource. Computationally
efficient algorithms exist to perform management
I. INTRODUCTION for single or multiple users communications.
Kivanc et.al. [1] was one of the researchers to show
In recent years, wireless communications has
the problem of finding an optimal sub-carrier and
grown to permeate all facets of life. From cellular
power allocation strategy for downlink
access to wireless data networks, there is a demand
communication to multiple users in an OFDM
for clear, fast transmission of multimedia
based wireless system. Lengoumbi et.al [2] shows
information. In an orthogonal frequency division
the Rate Adaptive optimization (RA) problem,
multiple access (OFDMA) system users are
which maximizes the sum of user data rates subject
assigned different carriers instead of sharing them.
to total power constraint and individual guaranteed
Optimal algorithms for frequency allocation are
rates. The first algorithm for resource allocation
found to be computationally demanding. In this
was introduced by Rohling and Grunheid [3]. They
paper a new class of algorithms is proposed which
present a simple heuristic greedy algorithm, and
achieves the users QoS objectives, compared to
show that it performs better than simple banded
existing algorithms and at lower computational
OFDMA. Several problems have been studied,
cost. We study the problem of finding an optimal

479
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_83, © Springer Science+Business Media B.V. 2010
480 OUWAIS ET AL.

including maximizing the number of users with IV. OPTIMIZATION TECHNIQUES FOR RADIO
equal rate requirements [4], maximizing the RESOURCES MANAGEMENT
minimum transmission rate for users with limited
power [5]. The objective is to find a subcarrier allocation
which allows each user to satisfy its rate
The main result of this paper is a class of requirements. Most famous problems can in fact be
algorithms which have been proposed for subcarrier described by: [4], [6].
assignment based on BABS, modified BABS and
our Hybrid BABS algorithm. These algorithms - RA (Rate Adaptive optimization
achieve comparable performance to the BABS problem),
algorithm but do not require intensive computation. - MA (Margin Adaptive optimization
A single cell with one base station and many problem),
mobile stations is considered. The algorithms - Equity optimization problem,
assume perfect information about the channel state - Outage optimization problem.
due to multipath fading as well as path loss and
shadowing effects, and the presence of a medium For more details, [2], [7], [9] could be named as
access protocol to convey information about several researches having been working about the
channel state and subcarrier allocation between the Rate Adaptive optimization problem (RA). We can
base station and the mobile stations. also refer to [3], [10], [11] for the Margin Adaptive
optimization problem (MA). Doufexi and Armour
III. SYSTEM MODEL AND PROBLEM [14] investigate the problem of dynamic multiuser
DESCRIPTION subcarrier allocation in a coded OFDMA system
based on the Equity problem optimization. Finally,
This section introduces the system model that is for Outage problem optimization we name to [4],
used, and formally introduces the problem being [6].
addressed in this paper. The criteria used for resource
allocation are introduced, and the subcarrier allocation A. Rate Adaptive problem formulation (RA)
method is formulated. We consider the downlink
transmissions of a multi-cell OFDMA system. Our objective is to optimize bandwidth allocation in
the system such that a maximum number of users
The system under consideration is an OFDM can achieve their data rate requirements. We aim to
system with frequency division multiple access find a bandwidth allocation that allows each user to
(FDMA). Perfect channel state information is satisfy its rate requirements while maximizing the
assumed at both the receiver and the transmitter, i.e. sum of user data rates subject to total power
the channel gain on each subcarrier due to path loss, constraint. Individual guaranteed rates, is
shadowing, and multipath fading is assumed to be considered [8], [10].
known. Channel parameters are assumed to be
perfectly estimated by a channel estimation The optimization problem can be expressed by:
algorithm for downlink OFDMA [3]. The design
and performance of this algorithm is outside the ­ max ¦ ru with constraint
scope of these studies. The system does not employ ° u=1..U
spreading in either time or frequency, and each
® ¦ p ≤P for b=1...B
°n=1..N b ,n Maxb
subcarrier can only be used by one user at any ¯ b
given time. Subcarrier allocation is performed at the and
base station and the users are notified of the carriers
chosen for them. After the allocation, each user ru ≥ ru0
performs power allocation and bit loading across Where U is the total number of users, N is the
the subcarriers allocated to it to find the number of OFDMA subcarriers, Ptr,max the total
transmission power. power constraint, r and r0 are respectively the data
rate and data rate constraint (minimum data rate)
We consider a system with K users, and N for user u.
subcarriers. Each user k must transmit at least Rk_min
bits per unit time. Let Hk(n) be the channel gain, V. SENSIBLE GREEDY APPROACH
pk(n) the transmission power and rk(n) the
transmission rate for user k on subcarrier n. The problem posed by Wong.et.al [5] is
computationally intractable, and as described
 A COMPUTATIONALLY DYNAMIC RESOURCE MANAGEMENT 481

above, a direct approach to solving it does not yield ª Rk º

a good algorithm. This paper examines two mk ← « m in »
«¬ R m ax »¼
different algorithms (resource allocation and
subcarrier assignment) and which use information w hile
K
¦ m > N do
about users channel and rate requirements to find a k =1 k
*
close approximation to the solution. k ← arg m in m
1≤ k ≤ K k
m * ← 0
Intuitively, the problem is separated into two k
stages: end w hile
K
w hile ¦ m < N do
1. Resource Allocation: Decide the number k =1 k
of subcarriers each user gets its bandwidth mk + 1 § Rk · m § k
R m in ·
based on rate requirements and the users Gk ← * f ¨ m in ¸ − k * f ¨ ¸ , k = 1.. K
Hk ¨ mk + 1 ¸ H k ¨ mk ¸
average channel gain. © ¹ © ¹
2. Subcarrier Allocation: Use the result of l ← arg m in G
1≤ k ≤ K k
the resource allocation stage and channel ml ← ml + 1
information to allocate the subcarriers to
end w hile
the users.
Where Hk, is a channel gain on every subcarrier,
Rmax is the maximum transmission rate per carrier,
A. Resource Allocation Algorithms and Rkmin is a minimum transmission rate requested
for user k. Let user k be allocated mk subcarriers.
In a wireless environment, some users will see a
lower overall SNR than other users. These users Example
tend to require the most power. Studying the
- 64 subcarriers,
subcarrier allocations from the LR algorithm shows
that once users have enough subcarriers to satisfy - 8 users [u1 u2... u8] and [m1 m2 ... m8] is the
their minimum rate requirements, giving more vector of number of subcarriers allocated
subcarriers to users with lower average SNR helps for each user.
to reduce the total transmission power. This section
Before BABS algorithm application, the vector
describes the bandwidth assignment based on SNR
number of subcarriers is: [m1 m2 m3 m4 m5 m6 m7
(BABS) algorithm which uses the average SNR for
m8] = [14 20 11 16 19 21 23 26].
each user to decide the number of subcarriers that
user will be assigned. In the second part of this The new vector of number of subcarriers after
paper, it finds our contribution for the best BABS algorithm application is: [m1 m2 m3 m4 m5
assignment of subcarriers to users. Two different m6 m7 m8] = [0 0 0 0 0 0 23 26].
approaches are presented, the rate-craving greedy
algorithm (RCG) and the amplitude-craving greedy B. Subcarrier Assignment Algorithms
algorithm (ACG).
Once the number of subcarriers is determined, we
1) Bandwidth Allocation Based on SNR (BABS) move on to assigning specific subcarriers to users.
In this section, we propose two suboptimal
To find the optimal distribution of subcarriers algorithms to allocate subcarriers to users:
among users given the flat channel assumption, we
propose using a greedy descent algorithm similar to - Amplitude Craving Greedy Algorithm
that used for discrete water filling. (ACG),
- Rate Craving Greedy Algorithm (RCG).
482 OUWAIS ET AL.

To minimize the transmission power and allocate Algorithm 2: RCG

the most appropriate subcarriers to the users
according to the channel quality information, the for each sub carrier, n = 1.. N
amplitude craving greedy (ACG) algorithm is k * ← arg m ax rk ( n )
formulated in a mathematical model. The amplitude 1≤ k ≤ K

craving greedy algorithm allocates subcarriers in a Ak * ← Ak * ∪ { n }

random order [2]. The aim of the ACG algorithm is, end for
if the individual users did not have rate constraints
to satisfy we would allocate each subcarrier to the for all user, k such that # Ak > m k , do
user who has the highest gain on that subcarrier. w hile # Ak > m k , do
The RCG algorithm begins with an estimate of the *
l ← arg m in m in − rk ( n ) + rl ( n )
users transmission rate on each carrier and aims to {l :# Ak < m l } {1≤ n ≤ N }
maximize the total transmission rate. The ACG n * ← arg m in − rk ( n ) + rl ( n )
algorithm is a modification of RCG which achieves {1 ≤ n ≤ N }
comparable performance at reduced computational Ak ← Ak /{ n * }, Al * ← Al * ∪ { n * }
complexity. ACG reduces complexity but has more
power transmitted than RCG [3]. end w hile
end for
The following algorithms, represents respectively
the ACG and RCG algorithms for subcarrier where mk, is the number of subcarriers allocated to
assignment (Equation.2): each user. rk(n), is the estimated transmission rate
of user k on subcarrier n.
Algorithm 1: ACG
The aim is to find the suboptimum subcarrier
fo r each su b carrier, n = 1 .. N
allocation algorithm that consumes the minimum
2
k
*
← arg m ax H k ( n ) transmission power and satisfies each user’s rate
1≤ k ≤ K requirements. This is accomplished in:
L et # Ak d en o te th e card in ality o f set A k
­ B U

w h ile (
#A * = m * ) do ° max ¦ ¦ ru, b
k k ® b =1 u =1
2 ° P( b ) < P
H k (n) ← 0 ¯ tr tr ,max (1 < u < U ),(1 < b < B)

2 and
k ← arg m ax H k ( n )
1≤ k ≤ K
en d w h ile
r ≥ r0
A
k
*
← A
k
*
∪ {n} VI. SIMULATION SETUP
en d fo r
In simulations Bandwidth Assignment Based on
SNR (BABS), Modified BABS (M-BABS), and
Where mk, is the number of subcarriers allocated to Hybrid BABS (H-BABS) are described. The
each user, and Ak ← { } initially for both channel gain g(u,n), of user u on subcarrier n in the
algorithms. serving cell, is summarized as:
-α
g (u ; n ) = K * d (u ) * ash (u ) * a f (u , n ) where
af has a Rayleigh distribution and represents the
small scale fading. We consider additive white
Gaussian noise (AWGN) which is characterized on
each subcarrier by a Gaussian random variable
with σ 2 = N 0 .W / N . The system under
consideration has parameters given in the following
Table. The channel model and traffic source models
used in simulations are described in more detail
below.
 A COMPUTATIONALLY DYNAMIC RESOURCE MANAGEMENT 483

Figure 2 shows a comparison of the transmitted

power, when bandwidth allocation is performed
using BABS and modified BABS algorithm. The
transmitted power of the BABS algorithm is more
important than that of the modified BABS. The
requested power is also comparable with the two
previous parameters. We conclude that the
advantages of the two algorithms are to optimize
the transmission power.

A similar result is developed with the same

parameters mentioned in Table I. From our
simulations, we obtained the average channel gain
versus the time (cf. figure 1).

Fig. 2. Transmission power versus Number of Iterations

A. A proposed Hybrid BABS Algorithm for

Bandwidth Allocation (H-BABS)

To allocate the bandwidth, we propose an

enhancement of BABS. H-BABS is a modified
Fig. 1. Channel Gain (dB) versus Time (s) version of BABS and M-BABS, that takes into
account the instantaneous algorithms (BABS and
VII. SIMULATION RESULTS H-BABS) during the scheduling. The aim of the
proposed algorithm is presented by the following
The algorithms are compared based on three graph:
criteria: the efficiency that a BABS algorithm is
able to allocate the bandwidth less than Modified
BABS (M-BABS) and Hybrid BABS (H-BABS),
the optimality, and the data rate per subcarrier
offered from each one.

This studies presents performance analysis of

resource allocation algorithms for a multicarrier
wireless system. The main contributions of this
paper are:

- One resource allocation algorithm is

proposed: the Hybrid Bandwidth
Allocation Based on SNR Algorithm (H-
BABS) for decide the number of subcarriers
mk needed for each user. The algorithms are
compared using simulations.
- A comparison studies are presented
between the two subcarrier allocation
algorithms: Amplitude Craving Greedy
(ACG) and Rate Craving Greedy (RCG).
Fig. 3. Hybrid Bandwidth Allocation Based on SNR
The main simulation results underlined
Algorithm
efficiency and the importance of phase
transitions in the study of search
suboptimal algorithms.
484 OUWAIS ET AL.

Figure 4 show the variation in data rate with VIII. Conclusion

number of iterations for the BABS, Modified
BABS and Hybrid BABS. There are 32 carriers and In this paper, we have designed a radio resource
8 users in the system. In figure 4, the total management scheme that integrates dynamic
bandwidth is W= 1MHz, and every subcarrier is subcarrier assignment in OFDMA wireless
used for data transmission. Figures also show plots networks. We have formulated the optimal
of an algorithm labeled as Modified BABS and subcarrier assignment scheme, and proposed a new
Hybrid BABS. We observe, that the modified suboptimal algorithm for resource allocation. Our
BABS algorithms allow an important gain of data simulations have demonstrated that RRM can more
rate than the based BABS algorithm. We can see efficiently utilize the available subcarriers. In this
also in the same figure a clear gap between the two paper, a computationally efficient class of
data rates offers by the algorithms. That's why, we algorithms for allocating subcarriers among users in
can affirmed that the performance obtained by the a multicarrier system has been described. Dividing
proposed Hybrid BABS (H-BABS) algorithm is the problem into two stages: resource allocation and
validated. subcarrier assignment, we present the main
objective of each sub-problem, the simulation
results underlined the opportunity of the approach.
This approach allows efficient use of system
resources in terms of bandwidth efficiency.

REFERENCES
[1] Didem. Kivanc and Hui. Liu, Subcarrier
Allocation and Power Control for OFDMG, IEEE
Transactions on Wireless Communications,
November 2003, vol. 2, no. 6, pp. 1150-1158.

[2] Carle. Lengoumbi, Philippe. Godlewski,

Philippe. Martins, An Efficient Subcarrier
Assignment Algorithm for Downlink OFDMA, The
64th IEEE Semiannual Vehicular Technology
Fig. 4. Comparison between Modified BABS and Hybrid Conference, Montreal, Canada, 25-28 september
BABS 2006, pp. 1010-1020.
B. Subcarrier Assignment Algorithms Results
[3] Didem Kivanc. Tureli, Resource Allocation for
Figure 5 gives a closer look at the BABS-ACG, Multicarrier Communications, Doctorat of
BABS-RCG, algorithms for the typical channel Philosophy, University of Washington, 2005.
with the gain mentioned by figure1. The plot for
BABS-ACG and BABS-RCG are nearly adjacent [4] Carle. Lengoumbi, Accès multiple OFDMA
and is difficult to see the difference between both. pour les systémes cellulaires post 3G: Allocation de
The BABS-ACG algorithm searches all possible ressources et ordonnancement, Thèse Doctorat,
allocations, and yields the lowest transmission Ecole Nationale Supérieur des Télécommunications
power per bit. (ENST Paris), Décembre 2008.

[5] C.Y.Wong, R.S. Cheng, K.B. Letaief, and R.D.

Much., Multiuser OFDM with adaptive subcarrier,
bit and power allocation, IEEE Journal. on Sel,
Areas in Communmmunication, pp. 17(10):1747-
1757, October 1999.

[6] Hammouda. Soumaya, Analyse et amélioration

de la gestion des ressources radio du système radio
mobile de troisième génération UMTS et au-delà,
Thèse doctorat, Ecole Supérieure des
communications de Tunis (SUP'COM), Technologie
de l'information et de la communication, Juin 2007.

[7] Carle. Lengoumbi, Philippe. Godlewski,

Philippe. Martins, Dynamic subcarrier reuse with
rate guaranty in a downlink multicell OFDMA
Fig. 5. Power per bit Transmitted (PT =Rtot)
system, The 17th Annual IEEE International
 A COMPUTATIONALLY DYNAMIC RESOURCE MANAGEMENT 485

Symposium on Personal, Indoor and Mobile Radio Allocation in Multiuser OFDM, In Proc. IEEE
Communications (PIMRC'06), 2006. Global Telecommunications Conf., volume 6,
pp. 3648 -3652,2001.
[8] Jiho. Jang, Kwang Bok. Lee, Transmit Power
Adaptation for Multiuser OFDMA Systems, IEEE [12] Lei. Yan, An introduction to Bandwidth
Journal on selected areas in communications, 21. Allocation in OFDM, March 2008.
NO 2, pp. 171- 178, February 2003.
[13] Carle. Lengoumbi, Philippe. Godlewski,
[9] Tommy K. Chee, Cheng-Chew. Lim, Jinho. Philippe. Martins, Comparison of different
Choi, Suboptimal Power Allocation for Downlink subchanalization modes for OFDMA, The 18th
OFDMA Systems, IEEE 60th VTC 2004, 26-29 Annual IEEE International Symposium on Personal,
September 2004. Indoor and Mobile Radio Communications
(PIMRC'07), 2007.
[10] D. Kivanc, H. Liu, Subcarrier Allocation and
Power Control for OFDMA, Conference Record of [14] A. Doufexi, S. Armour., Design considerations
the Thirty-Fourth Asilomar Conference on Signals, and Physical Layer Performance Results for a 4G
Systems, and Computers, pp. 147 -151, 2006. OFDMA System Employing Dynamic Subcarrier
Allocation, The 16th IEEE International
[11] Inhyoung. Kim, Hae Leem. Lee, Beomsup. Symposium on Personal, Indoor and Mobile Radio
Kim, and Yong H. Lee, On the Use of Linear Communications, PIMRC 2005.
Programming for Dynamic Sub channel and Bit
 A General Method for Synthesis of Families
of Orthogonal Complementary Codes
B. Y. Bedzhev, Member, IEEE
University of Shumen “Bishop Konstantin Preslavsky”
115 Universitetska Str.
Shumen, Bulgaria 9712, EU
M. P. Iliev
University of Ruse “Angel Kanchev”
8 Studentska Str.
Ruse, Bulgaria 7017, EU

Abstract-Families of orthogonal complementary codes (FOCCs) With regard to above described situation, our paper aims to
are called families of sets of radio signals possessing both zero develop a general method for synthesis of families of orthogonal
autocorrelation functions (ACFs) (except for the zero time shift) and complementary codes (FOCCs), which will allow the most part
zero cross-correlation functions (CCFs) among all pairs of the of techniques for synthesis of FOCCs, known today, to be viewed
members of a family. They are viewed as promising tool for a
radical enhancement of the data transmission rate of the future
in a common theoretical frame. The importance of the studied
wireless communication systems, because they allow the users to problem results from the following facts. First, today the FOCCs
communicate practically without multiple access interferences in a are the only known radio signals that have perfect (both periodic
multipath environment. Due to this reason in the paper a general and aperiodic) correlation properties [1], [3], [4], [5], [8], [9],
method for synthesis of FOCCs is proposed. It allows the [10], [11], [12], [13], [14], [15]. This means that every member
parameters of the FOCCs to be independently managed and the of a FOCC has an ideal ACF, resembling the Dirac delta –
most part of techniques for synthesis of FOCCs, known today, to be function, and the CCFs among all pairs of members of a FOCC
viewed in a common theoretical frame. are zero for every time shift. Second, despite of taken efforts
Index Terms – Synthesis of signals, Families of orthogonal many problems in the field of the synthesis of the FOCCs are
complementary codes. open still.
Paper is organized as follows. First, the basics of the FOCCs
I. INTRODUCTION
are recalled. Second, our method for synthesis of FOCCs is
The wireless communication systems are in very rapid suggested. At the end some important conclusions are given.
progress today. Despite of this the users and industry expect an
II. BASICS OF THE SYNTHESIS OF THE FAMILIES OF
even more significant enhancement of the quality of the services
and the transmission rate, offered by these systems. In order to ORTHOGONAL COMPLEMENTARY CODES
meet these expectations, the experience, obtained during the past Let K ≥ 2, M ≥ 2, N ≥ 1, be integers and let us consider a
twenty years, has been analyzed by the theorist and developers. family V(K, M, N), comprising K sets as every set consists of M
The main conclusion is that the critical role for the whole sequences with (code) length N:
performance of the wireless communication systems has the
system of employed radio signals. Due to this reason great efforts
have been directed to finding families of radio signals possessing
{ak (m, n)}nN=−01 = (1)
both autocorrelation functions (ACFs) with a small level of the = {ak (m, 0), ak (m,1), ..., ak (m, N − 1)}, m = 0,1, ..., M − 1
side-lobes and small cross-correlation functions (CCFs) among
all pairs of members of a family. The focusing of the research Then V(K, M, N) is called family of orthogonal complementary
activities on this problem can be explained by the following facts codes (FOCCs) if the following conditions are satisfied [4], [7],
[1], [3], [4], [5], [8], [9], [10], [11], [12], [13], [14], [15]. First, [9]. First, the aggregated ACFs of the sequences of every set are
the pointed out families of radio signals allow the so-named self- zero except for the zero time shift. Second, the aggregated CCFs
interference (SI), caused by multipath spreading of of the sequences of all pairs of sets are zero for every time shift.
electromagnetic waves, to be reduced by a separate processing of These conditions can be mathematically described as follow:
the direct and reflected signals. Second, it is possible the negative
effect of simultaneous transmission of numerous users, named ­M .N , k = l ∩ r = 0;
multi user interference (MAI), to be minimized. R Ak , Al (r ) = ® (2)
¯ 0, k ≠ l ∪ r ≠ 0,

487
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_84, © Springer Science+Business Media B.V. 2010
488 BEDZHEV AND ILIEV

N −1− r M −1
where: R Ak , Al (r ) = ¦ ¦ ak (m, n).al* (m, n + r ) =
- r is the time shift, r = −( N − 1), ( N − 2), ..., − 1, 0,1, ..., N − 1 ; n=0 m=0
(6)
N −1− r
- Ak , k = 0,1, ..., K − 1 is the matrix, denoting the k-th set of = ¦ CAk (n) ⊗ CAl* (n + r ).
the family V(K, M, N), i.e.: n=0

Here CAk (n) is the n-th vector – column of the matrix Ak and
ª ak (0, 0) ak (0,1) ... ak (0, N − 1) º
« a (1, 0) the symbol “ ⊗ ” denotes the inner product of the n-th vector –
ak (1,1) ... ak (1, N − 1) »»
Ak = « k
(3) column of the matrix Ak and the complex conjugated (n+r)-th
« ... ... ... ... »
« » vector – column of the matrix Al . Now it should be recalled that
a
¬ k ( M − 1, 0 ) a k ( M − 1, 1) ... ak ( M − 1, N − 1)¼ the evaluation of the CCFs can be simplified by means of
multiplication of polynomials [2]. More specifically, let us
- R Ak , Al (r ) is the aggregated CCF (ACF if k = l) of the rows introduce the following polynomials
of the k-th and l-th sets (matrices), k , l = 0,1, ..., K − 1 .
N −1
The aggregated R Ak , Al (r ) CCF (ACF if k = l) is evaluated by Fk ( x) = ¦ CAk (n).x n , k = 0,1, ..., K − 1,
the well known formula [1], [2], [3], [4], [13], [14], [15]: n =0
, (7)
N −1

R Ak , Al (r ) =
Fl* ( x −1 ) = ¦ CAl* ( j ).x − j , l = 0,1, ..., K − 1
j =0

­M −1 N −1− r
° ¦ ¦ ak* (m, n).al (m, n + r ), − ( N − 1) ≤ r ≤ 0, (4) then the coefficients of their polynomial product Fk ( x).Fl* ( x −1 )
°
= ® m = 0M −n1=N0 −1− r will be the aggregated cross-correlations R Ak , Al (r ) for all time
°
°¯ m¦ =0 n =0
¦ ak (m, n).al* (m, n + r ), 0 ≤ r ≤ N − 1. shifts r = −( N − 1), − ( N − 2), ..., − 1, 0,1, ..., N − 1 according to (6).
With regard to (7), the conditions (2) can be rewritten as
Here the symbol “*” means “complex conjugation”. follows
It should be mentioned that FOCCs are mathematical models
of the radio signals for the multicarrier code-division multiple ­M .N , k = l ;
access (MC-CDMA) wireless communication systems. In such a Fk ( x).Fl* ( x −1 ) = ® (8)
¯ 0, k ≠ l.
system K user can simultaneously communicate practically
without SI and MAI. The k-th set (i.e. the matrix
It should be stressed that multiplication of the coefficients in
Ak , k = 0,1, ..., K − 1 ) of the exploited FOCC is assigned to the k- (8) is performed as multiplication of vectors, i.e.
th user as its signature code. The entry ak (m, n) ,
m = 0,1, ..., M − 1 , n = 0,1, ..., N − 1 of the matrix M −1
CAk (n) ⊗ CAl* (n + r ) = ¦ ak (m, n).al* (m, n + r ) (9)
Ak , k = 0,1, ..., K − 1 is the complex envelope of the n-th m=0
elementary pulse of the m-th sequence. In fact the m-th sequence
describes mathematically the law of the amplitude and phase and due to this reason it is possible CAk (n) ⊗ CAl* (n + r ) = 0 .
modulation of the m-th subcarrier frequency during the every
clock interval in the process of the spreading of the spectrum of III. A GENERAL METHOD FOR SYNTHESIS OF FAMILIES
the radio signals. Most often only phase manipulation (Q-phase OF ORTHOGONAL COMPLEMENTARY CODES
shift keying) is used, i.e. The complementary codes were invented by M. Golay in 1949.
In his origin works [6], [7] K = 1, M = 2 and
∀ak (m, n) ∈ {exp[(2πil ) / Q; l = 0, 1, ..., Q − 1}, i = − 1 , (5) ∀a (m, n) ∈ {− 1, + 1}, m = 0,1, n = 0,1, ..., N − 1 , which gave the
reason these codes to be named complementary pairs. Later, in
but this restriction will not be used in the sequel. 1972, Tseng and Liu [13] extended the Golay’s conception and
For brevity but without loss of generality we shall suppose that introduced the so - named mutually orthogonal complementary
0 ≤ r ≤ N − 1 in (4). Then, after changing the order of sets, consisting of M ≥ 2 complementary sequences.
summations, (4) can be transformed as follows: Due to their unique correlation properties the complementary
codes have been intensively studied. For instance the study in [4]
 A GENERAL METHOD FOR SYNTHESIS OF FAMILIES 489

is based on over 300 journal papers and conference reports, are a FOCC with parameters: family size – K, number of
concerning this theme. The complementary codes have also sequences in every set – M’= K.M and length of the sequences –
exploited in some types radars and specialized communication N.
systems and obtained experience has proved the practical Proof: According to (7) let F (x) and F * ( x −1 ) be the
effectiveness of these signals. As a result at the beginning of the polynomials of (N-1)-th degree which coefficients are the vector
new century it was suggested the FOCCs, described in the – columns of the matrix A. As A is a set of M complementary
previous section, to be the base tool for multiple access to the sequences with length N:
communication channel in the next generations wireless
communication systems [1], [3], [4], [5], [8], [9], [12], [14], [15]. F ( x).F * ( x −1 ) = M .N . (13)
At the moment despite of the taken efforts in the study of
complementary codes, it is not known if or not the parameters K,
M, N of a FOCC can be independently managed during the Let CE (n) be the n-th column – submatrix of the matrix A:
developing of a new wireless communication system. For
example in [3], [4], [9], [13], [14] methods for synthesis of CE (n) = [d (0, n). A d (1, n). A ... d ( K − 1, n). A]T , (14)
FOCCs are described, according to which an augmentation of the
number of users (i.e. K) is possible only by enlarging of the where the symbol “T” denotes “transposition”.
length N of the sequences. Unfortunately, this approach leads to After taking into account (11) and (13), the CCF of the n-th
decreasing of the data transmission rate, which is very and l-th column - submatrices CE (n) and CE ( j ) of the tensor
undesirable. Due to this reason in this part of the paper a general
method for synthesis of FOCCs is suggested. In allows the product (12) can be evaluated as follows
parameters K, M, N of a FOCC to be chosen independently
during the process of system optimization. Besides our method ª d (0, n).F ( x) º ª d * (0, l ).F * ( x −1 ) º
provides the opportunity a significant part of known today « d (1, n).F ( x) » « −1 »
techniques for synthesis of FOCCs to be viewed from a common « » ⊗ « d * (1, l ).F * ( x ) » =
theoretical base. « ... » « ... »
« » « −1
»
The suggested general method for synthesis of GOCCs is ¬ d ( K − 1, n ).F ( x ) ¼ ¬ « d * ( K − 1, l ).F * ( x ) »¼
based on the following propositions.
¦ [d (k , n).F ( x)].[d * (k , l ).F * ( x −1 )] =
K −1
Proposition 3.1: Let A be a set of M complementary sequences = (15)
with length N and let D be a matrix k =0
K −1 ­(K .M ).N , n = l
ª d (0, 0) d (0,1) ... d (0, K − 1) º = M .N ¦ d (k , n).d * (k , l ) = ®
« d (1, 0) k =0 ¯ 0, n ≠ l.
d (1,1) ... d (1, K − 1) »»
D=« (10)
« ... ... ... ... » Equation (15) proves the Proposition 3.1.
« » It should be emphasized on the following facts.
¬ d ( K − 1, 0) d ( K − 1,1) ... d ( K − 1, K − 1)¼
First, the rows of D are mutually orthogonal also, because D is
a square matrix [1].
which columns are mutually orthogonal, i.e. Second, the Proposition 3.1 remains true if we reduce some of
columns of D and use only K’ (K’<K) columns for creating the
K −1 ­K , n = l; tensor product (12). Anyway in this case the size of the
¦ d (k , n).d * (k , l ) = ® 0, n ≠ l , (11)
synthesized FOCC will be smaller – namely, it will be K’.
k =0 ¯
Third, on the base of a known initial set of M complementary
for 0 ≤ n, l ≤ K − 1 . Then the column - submatrices of the tensor sequences with length N the Proposition 3.1 proves a technique
product for synthesis of a FOCC with parameters: family size – K,
number of sequences in every set – M’= K.M and length of the
sequences – N.
ª d (0, 0). A d (0,1). A ... d (0, K − 1). A º Proposition 3.2: Let V(K, M, N) be a FOCC consisting of K
« d (1, 0). A d (1,1). A ... d (1, K − 1). A »» sets A0 , A1 , ..., AK −1 as every set comprises M complementary
E=« (12)
« ... ... ... ... » sequences with length N. Besides, let D be a matrix, which
« » columns are orthogonal, according to (10) and (11). Then the
¬d ( K − 1, 0). A d ( K − 1,1). A ... d ( K − 1, K − 1). A¼
column - submatrices of the all tensor products
490 BEDZHEV AND ILIEV

Ek = ª d (0, n).Fk ,0 ( x) º ª d * (0, j ).Fl*,0 ( x −1 ) º

« d (1, n).F ( x) » « * −1 »
ª d (0, 0). Au ( k ,0) ... ... d (0, K − 1). Au ( k ,0) º « k ,1 » ⊗ « d * (1, j ).Fl ,1 ( x ) » =
« d (1, 0). A ... ... d (1, K − 1). Au ( k ,1) » « » « »
... ...
=« », (16)
u ( k ,1)
« » « »
« ... ... ... ... » ¬« d ( K − 1, n ).F k , K −1 ( x ) * −1
¼» ¬«d * ( K − 1, j ).Fl , K −1 ( x )¼»
« »
¬«d ( K − 1, 0). Au ( k , K −1) ... ... d ( K − 1, K − 1). Au ( k , K −1) ¼»
¦ [d (k , n).Fk , m ( x)].[d * (k , j ).Fl*,m ( x −1 )] =
K −1
= (20)
k = 0,1, ..., K − 1 m=0

are a FOCC with parameters: family size – K’ = K2, number of ­(K .M ).N , k = l ∩ n = j
=®
sequences in every set – M’= K.M and length of the sequences – ¯ 0, k ≠ l ∪ n ≠ j.
N.
Here the all rows of the matrix U
Equation (20) proves the Proposition 3.2.
It should be pointed out that the Proposition 3.2 suggests a
ª u (0, 0) u (0,1) ... u (0, K − 1) º
« u (1, 0) technique for expanding K times the size of an initial FOCC
« u (1,1) ... u (1, K − 1) »» without changing the length of the sequences. This is a very
U= (17)
« ... ... ... ... » useful opportunity, because the known today methods [3], [4],
« » [9], [13], [14] expand the size of an initial FOCC only by a
¬u ( K − 1, 0) u ( K − 1,1) ... u ( K − 1, K − 1)¼
simultaneous enlargement of the length of the sequences.
Anyway, in order to show that parameters of a FOCC can be
are permutations of the set {0,1, ..., K − 1} , i.e. independently managed it is necessary to prove the following
{u (k , 0), u (k ,1), ..., u (k , K − 1)} are the numbers {0,1, ..., K − 1} in propositions.
Proposition 3.3: Let V(K, M, N) be a FOCC, consisting of K
a possibly changed order.
sets A0 , A1 , ..., AK −1 as every set comprises M complementary
Proof: According to (7) let Fk , m ( x) and Fl*,m ( x −1 ) be the
sequences with length N. Besides, let D be a matrix which
polynomials of (N-1)-th degree which coefficients are the vector columns are orthogonal, according to (10) and (11). Then the row
– columns of the matrices Au ( k , m) and Au*(l , m) respectively. As - submatrices of the tensor product
all matrices Au ( k , m) are sets of M complementary sequences with
H=
length N:
ª d (0, 0). A0 d (0,1). A1 ... d (0, K − 1). AK −1 º
« d (1, 0). A d (1,1). A1 ... d (1, K − 1). AK −1 »» (21)
­M .N , k = l ,
Fk , m ( x).Fl*,m ( x −1 ) = ® (18) =« 0
« ... ... ... ... »
¯ 0, k ≠ l , « »
¬d ( K − 1, 0). A0 d ( K − 1,1). A1 ... d ( K − 1, K − 1). AK −1 ¼
for 0 ≤ k , l ≤ K − 1 .
Let CEk (n) be the n-th column – submatrix of the matrix Ek, are a FOCC with parameters: family size – K, number of
i.e.: sequences in a set –M and length of the sequences – N’ = K.N.
Proof: According to (7) let Fk (x) and Fk* ( x −1 ) be the
CEk (n) = polynomials of (N-1)-th degree which coefficients are the vector
,(19) – columns of the matrix Ak. As A0 , A1 , ..., AK −1 are a FOCC with
[
= d (0, n). Au ( k ,0 ) ... ... d ( K − 1, n). Au ( k , K −1) ] T
parameters K, M, N:

where the symbol “T” denotes “transposition”. ­ M .N , k = l ,

After taking into account (11), (18) and (19), the CCF of the n- Fk ( x).Fl* ( x −1 ) = ® (22)
th column – submatrix CEk (n) of the k-th tensor product and j-th ¯ 0, k ≠ l ,
column - submatrix CEl ( j ) of the l-th tensor product (16) can be for 0 ≤ k , l ≤ K − 1 .
evaluated as follows
Let RH k (m) be the m-th row – submatrix of the tensor product
H:
 A GENERAL METHOD FOR SYNTHESIS OF FAMILIES 491

CH k (m) = [d (m, 0). A0 d (m,1). A1 ... d (m, K − 1). AK −1 ] .(23) with parameters K’ = 2, M’ = K.M = 2 and N’ = 1.
Now, using the Proposition 3.3, FOCC from (27) and D,
After taking into account (22) and (23), the CCF of the m-th defined by (26), we create the following FOCC
and l-th row - submatrices CH k (m) and CH l (m) of the tensor
ª+ 1 + 1º ª+ 1 − 1º
product (21) can be evaluated as follows A' '0 = « », A' '1 = «+ 1 + 1» (28)
¬ + 1 − 1¼ ¬ ¼

ª K −1 k .N º
ª K −1 º with parameters K’’ = 2, M’’ = 2 and N’’ = K.N’ = 2.
« ¦ d (m, k ).Fk ( x ).x » « ¦ d * (l , j ).F j ( x ) x
* −1 − j . N
»= At the end, employing the simplest matrix U, which rows are
¬k =0 ¼ ¬« j = 0 ¼»
the numbers {0,1, ..., K − 1} in their natural order (here K = 2)
K −1
= ¦ d (m, k ).d * (l , k ).Fk ( x).Fk* ( x −1 ) +
k =0 ª0 1º
K −1K −1 (24) U =« » (29)
+ ¦ ¦ d (m, k ).d * (l , j ).Fk ( x).F j* ( x −1 ).x ( k − j ).N = ¬0 1¼
k = 0 j = 0,
j≠k and applying the Proposition 3.2, FOCC from (28), again D,
K −1 ­M (K .N ), m = l defined by (26), we create the following FOCC
= M .N ¦ d (m, k ).d * (l , k ) + 0 = ®
k =0 ¯ 0, n ≠ l.
ª+ 1 + 1º ª+ 1 + 1º
Equation (24) proves the Proposition 3.3. «+ 1 − 1»» «+ 1 − 1»»
Proposition 3.4: After interleaving of the columns of the row – A' ' '0 = « , A' ' '1 = « ,
«+ 1 + 1» «− 1 − 1»
submatrices CH k (m) , defined by (23), a FOCC with parameters: «+ 1
¬ − 1»¼ «− 1
¬ + 1»¼
family size – K, number of sequences in a set –M and length of (30)
the sequences – N’ = K.N will be obtained. ª+ 1 − 1º ª+ 1 − 1º
The Proposition 3.4 can be proven in a truly similar way as «+ 1 + 1»» «+ 1 + 1»»
above and due to this reason the proof is left to the reader. A' ' '2 = « , A' ' '3 = «
«+ 1 − 1» «− 1 + 1»
IV. RESULTS «+ 1 + 1»¼ «− 1 − 1»¼
¬ ¬
It should be pointed out that the Proposition 3.3 and 3.4
suggest techniques for expanding K times the length of the The parameters of the FOCC, presented by its sets of
sequences of an initial FOCC without changing the number of the complementary sequences (30), are: K’’’ = K.K’’ = 4,
sequences. Consequently, Propositions 3.2, 3.3 and 3.4 together M’’’ = K.M’’ = 4 and N’’’ = N’’ = 2.
prove that the parameters of the FOCCs can be independently It should be stressed that the analogous FOCC, synthesized by
managed. the methods, known from [3], [4], [13], [14] is
The Propositions 3.1, 3.2 and 3.3 will be explained by the
following example.
ª+ 1 + 1 + 1 + 1º ª− 1 − 1 + 1 + 1º
Example 4.1: Let us make use of the simplest set of «+ 1
complementary sequences + 1 − 1 − 1»» «− 1 − 1 − 1 − 1»»
B0 = « , B1 = « ,
«− 1 + 1 − 1 + 1» «+ 1 − 1 − 1 + 1»
A = [+ 1] (25) «− 1
¬ + 1 + 1 − 1»¼ «+ 1
¬ − 1 + 1 + 1¼»
(31)
ª− 1 + 1 − 1 + 1º ª+ 1 − 1 − 1 + 1º
with parameters K = M = N = 1 and the simplest orthogonal «− 1
matrix + 1 + 1 − 1»» «+ 1 − 1 + 1 − 1»»
B2 = « , B3 = «
«+ 1 + 1 + 1 + 1» «− 1 − 1 + 1 + 1»
«+ 1 + 1 − 1 − 1»¼ «− 1 − 1 − 1 − 1»¼
ª+ 1 + 1º ¬ ¬
D=« ». (26)
¬+ 1 − 1¼
As seen from (30) and (31), the classical methods for synthesis
After applying the Proposition 3.1 we obtain the FOCC of FOCCs extend the size of a FOCC only by the cost of a
directly proportional enlargement of the length of the sequences,
ª+ 1º ª+ 1º used for spreading the spectrum of the radio signals. This
A'0 = « », A'1 = « » (27) peculiarity often has negative effect over the performance of the
¬+ 1¼ ¬− 1¼ developed wireless communication system as it leads to an
492 BEDZHEV AND ILIEV

undesirable decrease of the data transmission rate. In contrast, the ACKNOWLEDGMENT

general method for synthesis of FOCCs, suggested in this part of The authors wish to thank Prof. Dr Kunchev, Prof. Dr E.
the paper, allows the size of the family and the length of the Ferdinandov, Prof. Dr H. Hristov and Prof. Dr B. Tsankov.
sequences of a FOCC to be independently managed. Finally, the authors appreciate the thoughtful critique by the
V. CONCLUSION referees.
In the paper a general method for synthesis of FOCCs is
suggested. It is based on the Propositions 3.1, 3.2, 3.3 and 3.4, REFERENCES
proved in paper. [1] B. Y. Bedzhev, “Generalized orthogonal complementary codes for CDMA
It should be emphasized on the following facts. wireless communications”, in Proc. The 2004 European Conference on
First, the Proposition 3.1 provides a technique for synthesis of Wireless Technology, pp. 329–332, Amsterdam, Kingdom of Netherlands,
Oct. 2004.
an initial FOCC with parameters: family size – K, number of [2] R .E. Blahut, Theory and Practice of Error Control Coding, Addison-Wesley,
sequences in a set – M’= K.M and length of the sequences – N, if Reading, Mass., USA, 1987
an initial set of M complementary sequences with length N is [3] H.-H. Chen, J.-F. Yeh and N. Suehiro, “A multicarrier CDMA architecture
known. Methods for synthesis of initial sets of complementary based on orthogonal complementary codes for new generations of wideband
wireless comunicatios,” IEEE Communications Magazine, pp. 126–135,
sequences are presented in [1], [4], [7], [8], [13]. Oct. 2001.
Second, the Propositions 3.3 and 3.4 allow the length of the [4] H.-H. Chen, Next Generation CDMA Technologies, John Wiley&Sons,
sequences of a FOCC to be enlarged if it is necessary in order to Chichester, West Sussex, England, 2007
expand the spectrum of the employed signals. Here it ought to be [5] A. T. Fam and I. Sarkar, “A new class of interlaced complementary codes
based on components with unity peak sidelobes,” Signal Processing, vol. 88
recalled that the spreading of the spectrum of the signals provides (2008), pp. 307–314
high processing gain (i.e. increasing the signal-to-noise ratio) in [6] M. J. E. Golay, “Multislit spectroscopy,” J. Opt. Soc. Amer., vol. 39, pp. 437–
receivers of the developed wireless communication system. 444, 1949
Third, the Proposition 3.2 substantiates a technique for [7] M. J. E. Golay, “Complementary series,” IRE Trans. Inform. Theory, vol. 7,
pp. 82–87, £2, 1961.
expanding K times the size of a FOCC without changing the [8] V. V. Ignatov, S. A. Dobrovolsky and A. Y. Guzhva, “Matrix systems of
length of the sequences. Here it ought to be stressed that family signals for the CDMA networks”, Elektrosvyaz, pp. 41-42, Sept. 2003 (in
size of a FOCC is very important parameter as it determines the Russian).
number of users, who can simultaneously exploit the resources of [9] Y. Jin and H. Koga, “Basic properties of the complete complementary codes
using the DFT matrices and the Kronecker products”, in Proc. International
the wireless communication system without MAI. In contrast the Symposium on Information Theory and its Applications, ISITA2008, pp. 1-6,
classical methods for synthesis of FOCCs extend the size of the Auckland, New Zealand, 7-10 December, 2008
FOCC only by the cost of a directly proportional enlargement of [10] S. P. Maity, M. Mukherjee, “A high capacity CI/MC-CDMA system for
the length of the sequences. This “hard” connection between the reduction in PAPR,” in Proc. Seventh Annual Communication Networks and
Services Research Conference, pp.297-304, Moncton, New Brunswick,
family size and the length of the FOCC could negatively Canada, 11-13 May, 2009
influence the process of development of the new wireless [11] A. Milev, V. Mutkov and B. Najdenov, “An algorithm for construction of
communication systems, because often leads to an undesirable optical orthogonal codes,” unpublished
diminishing of the data transmission rate. [12] T. R. Qureshi, M. D. Zoltowski and R. Calderbank, “A MIMO-OFDM
channel estimation scheme utilizing complementary sequences,” in Proc.
Forth, the Example 3.5 shows that if we flexibly apply the The 2009 IEEE International Conference on Acoustics, Speech and Signal
techniques, provided by Propositions 3.1, 3.2, 3.3 and 3.4, we can Processing, pp.2677-2680, Taipei, Taiwan, 19-24 April, 2009
obtain the most part of the FOCCs, known today. For instance, it [13] C.-C. Tseng and C. L. Liu, “Complementary set of sequences,” IEEE Trans.
is obvious that the columns of the orthogonal matrix D, defined Inf. Theory, vol. 18, pp. 206–209, Sept. 1972.
[14] S.-M. Tseng and M. R. Bell, “Asynchronous multicarrier DS-CDMA using
by (10) and (11), is a FOCC with parameters: K = M and N = 1. mutually orthogonal complementary sets of sequences,” IEEE Trans. on
Then using this FOCC, Propositions 3.3 and the same matrix D, Communications, vol. 48, pp. 53–59, Jan. 2000.
we can easily obtain the constructions, presented in [1], [4], [8], [15] M. D. Zoltowski, T. R. Qureshi and R. Calderbank, “Complementary codes
[13], [14]. Moreover, if we use the Proposition 3.4 instead of the based channel estimation for MIMO-OFDM systems,” in Proc. Forty-Sixth
Annual Allerton Conference, pp. 133-138, Allerton House, UIUC, Illinois,
Proposition 3.3, we shall obtain a new construction, which is not USA, 23-26 Sept., 2008
covered by [1], [4], [5], [8], [12], [13], [14], [15].
With regard to all above stated we hope the general method for
synthesis of FOCCs, suggested in the paper, to be useful in the
process of development of the new generation wireless
communication system, which aim to provide a radical
enhancement of the data transmission rate.
 E-Speed Start- A Window Based Additive
Increase, Multiplicative Decrease Network
Congestion Control Technique
Oyeyinka I. K1, Akinwale A.T2, Oluwatope A O3. Folorunso O2, Aderounmu G. A3, Abiona O. O4
1
Computer Technology Department, Yaba College of Technology, Yaba, Lagos-Nigeria.
[email protected]
2
Computer Science Department, University of Agriculture, Abeokuta, Nigeria.
{atakinwale, folorunsolusegun}@yahoo.com
3
Comnet Lab., Department of Computer Science & Engineering, Obafemi Awolowo University, Ile-Ife. Nigeria
{aoluwato, gaderoun}@oauife.edu.ng
4
Department of Computer Information System. Indiana University, NorthWest, USA
[email protected]
Abstract: The two major algorithms in the Additive Increase, product. It uses loss-delay to detect congestion. Wei in [16]
Multiplicative Decrease Network Congestion Control Algorithm (AIMD) proposed the FAST TCP which uses delay instead of loss to
are the slow start and the congestion avoidance. Much works has signal congestion. Other protocols in this category include
been done on improving the congestion avoidance stage. Not until HCTP [8], BIC TCP [17], STCP [6], CUBIC TCP [17] etc.
recently has attention shifted to slow start. The proposals for
improving slow start include swift start, quick start etc. These
These protocols modify the window growth function of TCP
modifications are deficient in one way or the other. Hence we to match large bandwidth delay product. It appears easy to
propose e-speed start for AIMD. We attempt to incorporate accomplish this modification but the issue of fairness with
environmental factors into slow start speed determination. these protocols has remained a challenge. Fairness in this case
Furthermore, we propose multiple selectable startups for AIMD. involves both intra and inter-protocol fairness. In addition,
Hence a connection may use the conventional slow start, e-speed none of these protocols modified the startup behaviour –slow
start or any other startup algorithm depending on the size of the start of AIMD, which has come under investigation lately.
available bandwidth. In fact, we are of the view that AIMD slow There have been several modifications of the slow start
start may be selectable from n-ary set of algorithms. The e-speed stage to overcome the problem of performance associated with
start uses the available bandwidth to calculate a factor  which is
used to determine the congestion window (cwnd) suitable for
it. According to Liu in [9], there are three lines of studies to
transmission at that point. The result obtained shows that e- slow start mechanism modification. The first approach uses
speed start performs far better than the slow start where the capacity estimation techniques to estimate available
available bandwidth is high and the environmental factors are bandwidth and sets the congestion window size using the
favourable. It was observed that e-speed start is able to increase estimated bandwidth. Partridge C et al in [12] proposed Swift
throughput between 700-51100% and network link utilisation by Start for TCP. Swift start employs an initial window (cwnd)
37.5% as against the convectional slow-start technique in TCP. of 4 packets thereafter estimates the available bandwidth in the
first round trip time using packet pair. Swift start uses the
I. INTRODUCTION estimated bandwidth to compute the bandwidth delay product
TCP New Reno was upgraded from experimental status to (BDP) of the network and set cwnd to a percentage of the
full protocol status in 2004. Several proposals and researches estimated BDP. Lawas-Grodek and Tran in [7] carried out a
had been put forward to improve its performance. Many of performance evaluation of Swift Start and submitted that Swift
these proposals tend to improve TCP in a high speed network Start improves network performance when the network is not
where it has been shown that TCP mechanism may lead to congested. However, when the network becomes overflowed
network resources underutilization [14]. TCP Westwood the estimation of the cwnd drifts away from accuracy. This
proposed bandwidth estimation as congestion measure [10], it drifting is due to retransmission of delayed or lost ACKs and
specified that a TCP sender continuously computes the Retransmission Time Out (RTO) timeouts.
connection bandwidth estimate by properly averaging the A second approach employs sharing of congestion
returning ACKs and the rate at which the ACKs are received. information among connections in the network. An example of
After a loss has occurred, the sender uses the estimated congestion control algorithm that used this approach is the
bandwidth to properly set the sending rate and the congestion Congestion manager proposed by [2]. Congestion manager
window. This is an improvement on standard TCP which half collects congestion status information and feedback from
its window on loss detection [10]. However, TCP Westwood receivers, shares the information with endpoints and
have not proved better in term of stability and fairness when it connections in the network which enables connections
co-habit with the standard TCP and its suitability for general determine the congestion status of the network and thereby
deployment has not been ascertained. determine an initial sending rate. The congestion manager has
Other proposed protocols in this category include XCP [5] a weakness of being beneficial to only connections that are
which requires modification to router algorithm. However, it initiated almost at the same time. Secondly, it is only those
is not visible to modify all existing routers’ algorithms hence connections and endpoints that supplied feedback that can
XCP will remain experimental protocol for a long time. benefit from this scheme. And lastly, there is an approach that
HSTCP [3] was also designed for high bandwidth delay

493
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_85, © Springer Science+Business Media B.V. 2010
494 OYEYINKA ET AL.

requires explicit network assistance to determine a practicable The remaining of this paper is organized as follows:
starting sending rate called Quick Start [4]. section II presents the existing TCP slow start mechanism.
According to [4], the experimental Quick start TCP Section III introduces our proposed approach called e-speed
extension is currently the only specified TCP extension that start, the design of e-speed start and its mechanism. Section IV
realizes a fast startup. A large amount of work has already discusses initial results obtained. Section V concludes the
been done to address the issue of choosing the initial paper.
congestion window for TCP. RFC 3390 [1] allows an initial
window larger than one packet. Quick start is based on the fact II THE EXISTING MODEL
that explicit feedback from all routers along the path is Window based congestion control has four stages; this is
required to be able to use an initial window larger than those to say that the life time of a connection passes through
specified by RFC 3390. possibly four stages before connection termination. These are
In quick start proposals, a sender (TCP host) would the slow start, congestion avoidance, fast retransmit and fast
indicate its intention to send at a particular rate in bytes per recovery. Most research efforts in the past have been
second. Each router along the path could approve, reduce, concentrated on the congestion avoidance stage and several
disapprove or ignore the request. Approval of the request by advances have been made until recently when research efforts
the router indicates that it is being underutilized currently and began to focus on the slow start stage due to advances in
it can accommodate the sender’s requested sending rate. The network speed technologies which made it possible to have
quick start mechanism can detect if there are routers in the gigabit networks operating at a speed of 1 gigabit/second (gbs)
path that disapproved or do not understand the quick start and above, compared to the maximum speed of 100 megabits
request. The receiver communicates the response to the sender per second (mbps) that existing networks transmit. The
in an answering TCP packet. If the quick start request is implication of this higher speed of transmission is that the
approved by all routers along the path, then the sender begins current congestion control techniques being used on the
transmission at the approved rate. Subsequently, transmissions internet may not be suitable for the best effort network in the
are governed by the default TCP congestion control future. The slow start stage is modeled as described below.
mechanism. If the request is not approved, then the sender The algorithm below applies:
transmits at the normal TCP startup speed [15].
According to [15], TCP is effective for both flow control The Slow Start Algorithm
and congestion control. TCP flow control is a receiver-driven 1. Start with a window size of 1 i.e cwnd=1
mechanism that informs the sender about the available 2. Increase the window size by 1 i.e. cwnd=cwnd+1; for
receive-buffer space and limits the maximum amount of every ACK1 received. Repeat step 2 until the ssthresh is
outstanding data. Flow control and congestion control are reached OR Packet loss is detected
independent while size of receive buffer space depends on the 4. If ssthresh is reached, go to congestion avoidance phase.
capacity of the receiver network. However, if the TCP is used 5. Else If a packet loss is detected;
with a link with large bandwidth-delay product, both set to ssthresh2 = 0.5*cwnd3
congestion window and flow control window need to be large set to cwnd =1
in other for TCP to perform well [4]. This makes both flow go to step 2
control and congestion control to overlap. The quick start TCP 6. Endif
extension assumes independence of flow and congestion Note: Set the initial value of ssthresh to a fraction of the
control which is not true and makes it not appropriate for the maximum window size. This is determined at the beginning of
global internet. transmission. Table I shows the simulation of the slow start
Another issue with the quick start approach is that the algorithm stated above. From table I the window size is
starting speed of a connection using quick start may send expressed as
thousands of packets in one RTT. The implication of this is W(nT + m/c) = 2n – 1 + m + 1, 0 ¡ m ¡ 2n –1 (1)
that, thousands of packets may be dropped in one RTT, And the maximum window size in a cycle is derived as
thereby necessitating a retransmission of these packets which W(m) = 2n (2)
may in turn be dropped and retransmitted over and over again, There will be two slow start phase. The first phase, buffer
thereby saturating the network which can lead to congestion. overflows and there is a packet loss which reduces the window
In addition, most routers and other network components on the size to 1 and slow start is re-entered. Equations 3 and 4
global internet are not built with capabilities to be quick start describe the ssthresh in both phases respectively.
aware. The implication is that they need either to be replaced ssthresh = (cT + B)/2 (3)
or modified before deployment. Hence, we propose a milder ssthresh = Min (2B-1, (cT + B)/4) (4)
approach to fast startup which called e-speed start. This is a where B is buffer size of link, c is the link capacity, T is sum
form of fast startup that does not require routers’ response or of propagation and queuing delay, W(m) is maximum window
modification. Rather, it employs end-to-end principles to size in a cycle, n is the number of cycles and m is number of
determine a suitable startup speed. This proposal is adaptable
to the current internet as well as the future gigabit internet. 1
Acknowledgment
2
Slow start threshold
3
Congestion window
 E-SPEED START- A WINDOW BASED ADDITIVE INCREASE, MULTIPLICATIVE DECREASE NETWORK 495

TABLE I of end-to-end control becomes useful. End-to-end control

MODELLING THE SLOW START builds congestion control complexity at the end nodes while
Cycle1 Time Acked Window Max packet the link is assumed unintelligent and less complex.
packet size released
A. Design of the E-Speed Start
Cycle 0 O - 1 1
The current design of the slow start state of the internet
Cycle 1 T 1 2 3 as expressed in New Reno [19] and other proposed
Cycle 2 2T 2 3 5 modification specified a single start speed for the TCP.
2T + 1/c 3 4 7 Although, slow start can work fairly well with networks with
Cycle 3 3T 4 5 9 low bandwidth delay product, its performance suffers from
3T + 1/c 5 6 11 severe inefficiency when it is applied to high bandwidth delay
3T + 2/c 6 7 13 product (BDP) networks. This is so because it takes too long
3T + 3/c 7 8 15
for the sender to reach an available high rate.
The slow start had been the traditional determinant of
Cycle 4 4T 8 9 17 start up speed for all proposed window based congestion
4T + 1/c 9 10 19
4T + 2/c 10 11 21 control algorithm until recently when [4] proposed the quick
4T + 3/c 11 12 23 start modification to TCP slow start algorithm. With the quick
4T + 4/c 12 13 25 start proposal, large data could be sent using an initial large
4T + 5/c 13 14 27 window that sends thousands of packets at the startup stage.
4T + 6/c 14 15 29
4T + 7/c 15 16 31 However, this has a requirement that the network presents
Cycle 5 5T 16 17 33
high bandwidth capacity to be able to transmit at the high rate
5T + 1/c 17 18 35 of quick start. This also implies that thousands of packets may
5T + 2/c 18 19 37 be lost in a single round trip time thereby overburdening the
… network.
Our proposal is to build an algorithm that is intelligent
RTT within a cycle. ((4) follows from the fact that ssthresh is
enough to be aware of the availability or non availability of
half of window size.) (1) to (4) was modified from [18].
certain environmental factors that determine an initial speed
III. WORK DONE for the startup stage of the window based congestion control
In our approach presented here, we attempt to find a algorithm. In this work, it is proposed that initial startup speed
balance between slow start low startup speed transmission and may be chosen from two options (e-speed start and slow start).
the large number of segment transmission of quick start. The However it is possible to make startup speed selectable from
approach introduces two network environmental variables an n-arry set of algorithms instead of binary option as
hose, periodic values are determined and assigned. These proposed in this work.
values are based on two operating conditions prevalent in the
network, peak or off-peak network utilization periods. B. Environmental Factor
Network utilization periods are representations of network The environmental factor referred to in this work
users’ behaviour, users are usually not using the network at performs two functions along with a bandwidth estimation
certain times e.g. night session, holiday period etc. Network function that determines the available bandwidth. These are
utilization is used to determine whether or not to employ the time function and user input function. The time function
conventional slow-start mechanism for transmission. At a obtains the system’s time of the day and expresses it as ‘peak’
particular point in time, if the network utilization is off peak or ‘off peak’ time. An initial definition of peak and off-peak
the available bandwidth[20] module is used to estimate the period may be set and may be refined or redefined by using
available bandwidth of the network. This value is used to network usage statistics. User inputs specify whether the day
compute a factor  which in turn is used to determine an is a night period, workday or holiday. Some networks are not
appropriate initial congestion window size and hence the heavily used in some day of the week, for instance, over the
initial start-up speed. On the other hand, if the available weekends or in the night. These periods may afford a higher
bandwidth is too low, the connection defaults to slow start. At startup speed without necessarily congesting the network and
any sign of congestion indication within the network, the e- thereby optimizing the use of network resources – bandwidth.
speed start mechanism is dropped and transmission proceeds The e-speed start is designed to be an end-to-end control
via the conventional TCP. not requiring any input or modification from the network
Approaches to congestion control are of two types; rate unlike the quick start that require routers input/modification.
based control and window based control. A window based The design of the E-speed start decouples speed determination
congestion control approach is presented in this work. A from actual data transfer at this stage. Hence the startup stage
window based congestion control uses a congestion window to is divided into two: speed determination stage and the data
determine the sending rate of the end nodes. Here the concept transfer stage.
496 OYEYINKA ET AL.

C. Speed Determination Stage The system transits into bandwidth estimation mode if
At this stage, two main environmental variables are the value of Time-of-day variable is off-peak or if user-input
introduced and their values determined in the following order: is yes in response to the holiday prompt. The bandwidth
• Time-of-day: a binary value of peak or off-peak estimation module estimates the available unused bandwidth.
• User-input: a binary value of workday or holiday. If bandwidth usage is low and the estimated available
In standard TCP, after handshake, the connection transits bandwidth is high, the system enters e-speed start otherwise
from idle to slow start state where data transmission proceeds slow start is activated.
immediately. In e-speed start, successful handshake transits From figure 3 below, the time function captures the time
into the speed determination state. At the speed determination of connection, determines if it can be categorized as off-peak
state, the system time is captured into the variable Time-of- or peak period, and passes control to bandwidth estimation
day and the value is compared with a table that contains data function or user input as the case may be. The bandwidth
collected from network usage statistics. If bandwidth usage is estimation function uses the packet pair algorithm to
high at this time (as contained in the network usage statistics determine the available bandwidth and thereby passes control
table), the value of Time-of-day is set to peak and if to either the slow start or e-speed start. If control is passed to
bandwidth usage is low, it is set to off-peak. It is assumed that the slow start, it proceeds as in conventional TCP. If the user-
if network usage is low for a period of time at a particular time input module is activated, the user is prompted with an
of the day, it is likely to be low always at such period of time optional message e.g. “press spacebar if it is an holiday”. If
and vice versa. However, the determination of the period of user responds to this message the bandwidth estimation
the day to be categorized as low or high bandwidth usage time module is activated else the slow start state is entered into and
is not trivial. We intend to use fuzzy logic to determine the data transmission starts. If the bandwidth estimation function
probability of low or high bandwidth usage time. If the value returns a high value for available bandwidth, e-speed start can
of Time-of-day is peak the connection activates user input be used for startup for data transmission.
module and if it is off-peak it proceeds to bandwidth
estimation. The e-speed Start Algorithm
The second environmental variable is user-input, the user 0. Estimate the value of  using available bandwidth. If =1
optionally responds to the prompt, “is it an holiday?” for go to 1else if >1 go to 7
example, during connection attempt. If yes, the speed 1. Start with a window size of 1 i.e cwnd=1
determination module default to bandwidth estimation to 2. Increase the window size by 1 i.e. cwnd=cwnd+1; for
determine available bandwidth, otherwise it proceeds to slow every ACK received. Repeat step 2 until the ssthresh is
start. Time-of-day and User-input are two variables that reached OR Packet loss is detected
determine if a connection proceeds to bandwidth estimation 4. If ssthresh is reached, go to congestion avoidance phase.
state or directly to slow start state. 5. Else If a packet loss is detected;
set to ssthresh = 0.5*cwnd
Congestion set to cwnd =1
Avoidance
go to step 2
Fast 6. Endif
Recovery
7. If =n, Increase the window size by 2 i.e.
Idle Slow Start
cwnd=cwnd+2; for every ACK received. Repeat step 7
until ssthresh is reached OR Packet loss is detected
ack of 8. If ssthresh is reached, go to congestion avoidance phase.
Exponential 9. Else if a packet loss is detected;
b
set to ssthresh = 0.5*cwnd
set to cwnd =1
go to step 2
Fig 1: TCP state transition using Slow Start 10. Endif
Using decision table to describe these scenarios, we
, Congestion
Avoidance obtain table II below: From table II, it is observed that E-
Speed
speed is used for startup at 37.5% of the time while slow start
start Fast
Retransmit
still dominate most of the connections at 62.5% of the time.
This implies that the link/network would have been
Idle Environmen
tal factor
underutilized 37.5% of the time. It is clear that e-speed start
optimizes the available bandwidth utilisation at these times for
faster connection and data transfer.
Slow Start

Fast Recovery
IV DISCUSSION OF E-SPEED START MECHANISM.
In e-speed start mechanism, new connection starts by
TCP sending an initial packet of 1. Thereafter, the congestion
Fig 2: TCP state transition using e-speed start
window is grown by (5);
 E-SPEED START- A WINDOW BASED ADDITIVE INCREASE, MULTIPLICATIVE DECREASE NETWORK 497

cwnd = 2n  (5) bandwidth delay product (BDP) as well as and forward

This is a modification of maximum window size per cycle compatible with networks with higher BDP
from (2) above where n is the number of RTT and  is a
function of the available bandwidth. For instance, if  is set to V. CONCLUSION
two, then four packets are sent per each ACK received and it We have presented here, a modification of the slow start
continues until there is a notification of congestion. The stage of the window based congestion control algorithm. This
resulting table as compared with the conventional slow start is modification is different from other modifications in literature
presented table III and IV. From table III, it is observed that as it achieved the following. It clearly decoupled speed
when =1, the throughput at RTT=7 is 128 packets while =2 determination from data transfer mode at the slow start. This
yielded over 16000 packets for the same number of RTT. This gives room for a differential start up speed depending on the
represents 12400% increase at slow start stage as long as other environmental factors like peak or off-peak period, a weekend,
operating conditions remain stable. night session etc. In addition, available bandwidth is major
For case of =2, the new table for packet transmission is determination of the startup speed. It is used to determine a
as presented in table V: The model describing e-speed start, factor  which is used as a multiplier of n, the power in 2n. If
window size is given by (6) the calculated value of  is 1, then TCP proceed with the
W(nT + m/c) = 2 ( n – 1) + 3m + 3, 0 ¡ m ¡ 2 ( n – 1) – 1 (6) normal slow start. If  is 2, TCP starts with a transmission of
and maximum window size in a cycle one packet and transmit 4 packets for each ACK received, If 
W(m) = 2 n (7) is 3, TCP starts with a transmission of 1 packet and transmit 8
Equations 6 and 7 form the bases for the window growth packets for each ACK received.  can take any integer value
function in the e-speed start.. provided the available bandwidth can support the resulting
From table III, when =2, four packets are transmitted in transmission speed. In addition, we propose that TCP start-up
response to one ACK received. This yielded a higher may be selected from n-arry possible algorithms instead of a
throughput. For instance a total of 4,096 packets are sent by e- single slow start as in the New Reno. We have proposed a
speed start as against 64 packets in the slow start at RTT=6. binary selection for start up stage and suggested a further
For =3, the same RTT yields 262,144 packets. Figure IV investigation to using an n-arry selection mechanism for
shows that by the second cycle, e-speed start speed can AIMD. This work is ongoing and these functions will be
increase TCP slow-start throughput between 700 and 51100%. analyzed for stability, e-speed start will be benchmarked
It is important to emphasise here that network capacity against major window based congestion control algorithms to
utilisation is the main trust of the e-speed start algorithm. The determine its fairness in terms of intra protocol and inter
algorithm can adapt its speed from network to network hence protocol fairness.
it is both backward compatible with networks with low

TABLE II : DECISION TABLE FOR E-SPEED START

Conditions ` Rules
Time of Day is off-peak Y Y Y Y Y Y Y Y N N N N N N N N
Available bandwidth is high Y Y Y Y N N N N Y Y Y Y N N N N
User Input is yes Y Y N N Y Y N N Y Y N N Y Y N N
It is an holiday Y N Y N Y N Y N Y N Y N Y N Y N
Action Result
Go to E-speed start X X X X X X
Go to slow start X X X X X X X X X X

Fig 3: Schematic diagram for E-Speed Start

 498 OYEYINKA ET AL.

TABLE III
E-SPEED START TRANSITION CYCLES WITH  =2 and 3
70
Slow start with B = 1
E-Speed start with B = 2
E-Speed start with B = 3
60
Congestion Congestion Congestion
Window for Window for E- Window for E- 50
Slow-Start speed Start speed Start

NUMBER OF PACKETS
Cycle (=1) (=2) (=3)
40

0 1 1 1
1 2 4 8 30

2 4 16 64
3 8 64 512 20

4 16 256 4,096
10
5 32 1,024 32,768
6 64 4,096 262,144 0
7 128 16,384 2,097,152 0 0.2 0.4 0.6 0.8 1
CYCLE
1.2 1.4 1.6 1.8 2

8 256 65,536 16,777,216

9 512 262,144 134,217,728 Figure. 4: Graph of slow start and e-speed start with =2 and =3
10 1,024 1,048,576 1,073,741,824
11 2,048 4,194,304 8,589,934,592 [4] S Floyd., M. Allman, A. Jain, P. Sarolahti. Quick-Start for TCP and IP.
12 4,096 16,777,216 68,719,476,736 RFC 4782. 2007.
[5] D. Katabi, , M .Handley and C. Rohrs. Congestion control for high
bandwidth-delay product networks. In proceedings on ACM Sigcomm.
TABLE V 2002.
MODELING THE E-SPEED START WITH =2 [6] T. Kelly, Scalable TCP improving performance in high speed wide area
networks. Computer communications Review 2003. 32(2).
Cycle Time Acked Window Max packet [7] F. Lawas-Grodek and D. Tran. Evaluation of Swift Start TCP in Long
packet size released Delay Environment. Technical report, Glenn research Center, Cleveland,
Cycle 0 O - 1 1 OH. 2004.
Cycle 1 T 1 4 5 [8] D. Leith, R. Shorten and Y. Li. H-TCP: A framework for congestion
Cycle 2 2T 2 7 9 control in high speed and long distance networks. Hi tech report. 2005.
2T + 1/c 3 10 13 http://www.hamilton.ie/net/htcp.
2T + 2/c 4 13 17 [9] D. Liu., M. Allman., S..Jin, and L. Wang. Congestion control without a
2T + 3/c 5 16 21 startup phase. In 5th International Workshop on Protocols for fast long
Cycle 3 3T 6 19 25 distance networks, Marina Del Rey, CA, 2007.
3T + 1/c 7 22 29 [10] S. Mascolo, C. Casetti., M. Gerla., M. Sanadidi., R. Wang. TCP
3T + 2/c 8 25 33 Westwood: end-to-end bandwidth estimation of efficient transport over
3T + 3/c 9 28 37 wired and wireless networks, in: Proceedings of the ACM Mobicom,
3T + 4/c 10 31 41 Rome, Italy. 2001.
3T + 5/c 11 34 45 [11] A. .Oluwatope,, A. Obadire,, G. Aderoumu, and M. Adigun, End-to-
3T + 6/c 12 37 49 End evaluation of selected TCP variants across a hybrid wireless
3T + 7/c 13 40 53 network, Issues in Information Science and Technology, 2006.Vol. 3.
3T + 8/c 14 43 57 [12] C. Partridge, D. Rockwell, M. Allman, R. Krishnan, and J. Sterbenz, A
3T + 9/c 15 46 61 swifter start for TCP. In Technical Report TR-8339, BBN Technologies.
3T + 10/c 16 49 65 2002.
3T + 11/c 17 52 69 [13] I. Rhee, and L. Xu. CUBIC: A new TCP – friendly high speed CTP
3T + 12/c 18 55 73 variant. In International Workshop on protocols for fast Long – distance
3T + 13/c 19 58 77 Networks, Lyon. 2005.
3T + 14/c 20 61 81 [14] P. Sarolahti, M. Allman, and S. Floyd. Determining an Appropriate
3T + 15/c 21 64 85 Sending Rate Over an Underutilized Network Path. February. 2006.
Cycle 4 4T 22 67 89 http://www.icir.org/floyd/quickstart.html
4T + 1/c 23 68 93 [15] M. Scharf, and S. Floyd, TCP Flow Control for Fast Startup Scheme
4T + 2/c 24 71 97 draft-scharf-tcpm-flow-control-quick-start-00.txt
4T + 3/c 25 74 101 [16] D. Wei, C. Jin, S. Low, and S. Hedges, Fast TCP: motivation,
4T + 4/c 26 77 105 architecture, algorithms, performance IEEE-ACM Transaction on
4T + 5/c 27 80 109 networking, 2006. 14(6): 1246 – 1259.
4T + 6/c 28 83 113 [17] L. Xu, K. Harfoush, and I. Rhee,. Binary Increase Congestion control for
… fast long distance networks. In proceedings of IEEE INFOCOMM ’04.
2004
[18] R. Srikant. "The Mathematics of Internet Congestion Control."
REFERENCES Birkhauser, 2004. Amazon Link. Notes and Errata
[1] M. Allman, Increasing TCP’s initial Window, RFC3390, 2002 [19] S Floyd., T. Henderson and A. Gurtov. The New Reno modification to
www.faqs.org/rfcs/rfc3390.html. TCP’s fast recovery algorithm RFC 3782 (2004).
[2] H. Balakrishman., H. Rahul. and Seshan, An integrated .congestion [20] A. O. Oluwatope, G. A. Aderounmu, E. R. Adagundo, O.O. Abiona and
management architecture for Internet hosts. In Proceedings of ACM F.J. Ogwu. Available Bandwidth Based Congestion Avoidance Scheme
SIGCOMM ’99, Cambridge, MA. 1999. for TCP: Modelling and Simulation. Advances in Computer and
[3] S. Floyd High-speed TCP for large congestion windows RFC 3649, information Sciences and Engineering, Springer Science +Business
December 2003. Media B.V. 2008. Pp 85-89
 Common-Wavelengths Design Scheme for Efficient
Management of Spare Resources in WDM Networks
Ardian Greca
Department of Computer Sciences, Georgia Southern University
P.O. Box 7997 Statesboro, GA, [email protected]

Abstract-The author addresses the problem of self-healing

capability to recover from network failures for WDM optical
networks. One problem in realizing such survivable WDM
networks is the inefficient utilization of large spare resources
(wavelengths) designed for restoration of the failed traffic, which
might not happen frequently. We study “common-wavelengths”
design scheme to increase the utilization efficiency of these
resources. The “common-wavelength” is an approach that allows
some working paths under the failure to efficiently use the
available common-wavelengths in the network. The effectiveness
of this proposed approach and the effect of physical and traffic
connectivity are investigated via simulations. Thus, the survivable (a) (b) (c)
WDM networks can be cost effectively designed according to the Fig. 1. Failure restoration designs.
reliability levels of all working paths in the network.
the WDM layer, some others on the IP/GMPLS layer, and recently
many others have addressed the multilayer survivability between these
I. INTRODUCTION two layers. Although there is a significant progress done so far, there
The survivability of a network against unexpected failures is one of is still a need for research focused on the WDM layer survivability
the important characteristics for optical WDM networks, due to a mechanisms. In order to provide a comprehensive and efficient
huge traffic loss. Network survivability can be carried out by using survivable network service, let us compare the pros and cons for each
self-healing techniques, which can restore the failed traffic of these layers. The survivability mechanisms in WDM layer are
automatically by rerouting the failed paths to other unaffected faster and more scalable than those of IP/GMPLS layer; however,
alternate routes when the failure occurs. In order for the self-healing they cannot handle faults such as router fault and service degradation
technique to succeed a sufficient amount of spare resources that usually are covered by IP/GMPLS survivability techniques. On
(wavelengths) should be pre-designed to complete the restoration the other hand, survivability mechanisms at the IP/GMPLS layer are
under the failure scenario. While the amount of spare resources usually slower and less scalable. One such example is a link failure
depends on several factors such as type of failure considered, network which can result in a failure of thousands IP layer traffic. Thus,
topology, self-healing scheme, still a large amount, exceeding 40% of providing a better survivability in WDM layer can improve the over
working resources, might be required for even simple failure all performance of the network and might even lead to efficient ways
scenarios. In order to realize a cost-effective survivable network the for multilayer survivability.
amount of spare resources in the network should be decreased. Many Researchers at [16] have described the concept of common pool
researchers have studied the cost-effective ways to reach the desired survivability, which allows some spare capacity of the lower SDH
level of availability and enhance the survivability performance of the layer to be used for recovery of the upper ATM layer traffic. The
network [1-3 and the references therein]. In [1] authors studied how approach is studied in the framework of multilayer survivability and
the optical layer survivability can be introduced in SDH over WDM can perform poorly if it is not carefully designed
transport networks. Researchers in [2] studied the survivable traffic In this paper we propose and study the “common-wavelengths”
grooming in WDM mesh networks. They proposed the protection-at- design scheme at the WDM layer to increase the efficiency of
lightpath (PAL) level and the protection-at-connection (PAC) level, resource utilization in the network under failure. The approach
and concluded that PAL outperforms PAC for the dynamic considers “multi-reliability class” concept to design the “common-
provisioning shared-mesh-protected sub-wavelength connections wavelengths” resources needed for protection of the affected traffic
against single-fiber failures. In [3] authors proposed a cost-effective and to reduce the number of spare wavelengths in the network. The
provisioning approach to provide differentiated per-connection-based method uses integer linear programming (ILP) model. In our
services to carry connections by appropriate routes and to protect simulations we show that the scheme reduces spare resources in the
them according to their availability requirements. network. To the best of our knowledge, we are not aware of such an
Furthermore, considerable research efforts have been dedicated to approach being studied so far for this layer.
the study of survivability mechanism in WDM and IP/GMPLS The rest of the paper is organized as follows. Section 2 explains the
network respectively [5-15]. Some of these efforts have been put on concept of “common wavelengths” and the policies used to realize

499
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_86, © Springer Science+Business Media B.V. 2010
500 GRECA

such a design. Section 3 discusses the formulation of the problem

using ILP. Simulated network topology and simulation results are
discussed in Section 4. Finally, conclusions are given in Section 5.

II. THE CONCEPT OF COMMON-WAVELENGTHS

Figure 1a, shows the straight forward approach where there is no
reliability class among traffic demands. This leads to the most
expensive approach since spare resources have to be designed for all
demands. A more efficient approach is shown in Fig. 1b, where
demands are grouped in two classes, protected class and unprotected
one. In this approach the total spare resources can be decreased since
they will be designed only for the demands of the protected class. The Fig. 2. Original and expanded graph of a network.
best approach appears to be if working resources of both the
be the set of optical paths demanded, and let a(d) and b(d) denote the
unprotected class and spare resources are combined together to assure
source and the destination of the path d ∈ D respectively. Suppose
the recovery under the failure. These resources are referred to as
that a flow of the commodity associated with the path d is directed
“common-wavelengths”. The policy of using “common-wavelengths”
from sa(d) to tb(d) in the converted graph, and the flow volume of each
is as follows:
commodity is 1. If the capacity of every edge is 1 and if the flow
1. During the normal operation of the network, “common-
volume on an edge is restricted to 0 or 1, the route of the flow will
wavelengths” will be used by demand of low priority class.
have the following characteristics:
2. Under the failure scenario, alternative paths for protected paths will
• The flow of a commodity passes a single route.
have the first priority to use these “common-wavelengths”. If the
• The flow of a commodity goes through only one of the subgraphs
available resources are not enough some of the demands of the
G(1), ..., G(k).
unprotected class have to release their wavelengths for the
• At most one commodity can go through each edge of G.
alternative paths.
Among these characteristics, the first and the third one follow
3. Low priority class paths can either be randomly dropped or their
straightforwardly from the edge capacity and the integer restriction.
bandwidth requirement can be decreased by average amount.
The second characteristic comes from the fact that in order to satisfy
This approach reduces the amount of spare resources in the network
the wavelength continuity constraint of the WP there are no routes
and leads to a cost-effective solution for WDM networks. The
between subgraphs.
drawback of this scheme is that it will degrade the QoS for some of
In summary we have the following notation:
the demands in the unprotected class. But this degradation, if it
• V the set of nodes in the network.
happens, will be only during the failure time which will not occur
• E the set of links in the network.
frequently.
• Λ the set of wavelengths in the network (subgraphs).
• G(V,E) network graph.
III. PROBLEM FORMULATION • G(i)(V,E) network graph for each wavelength.
This study uses the same approach as in [4] and considers static • Ggen(V1,E1) modified network graph with network topology,
optical path assignment; that is, a request for setting up a set of optical links in each subgraph, wavelengths and added nodes.
paths is given first. These optical paths are not released once they are • lin link that enters a node.
set up. The optical paths are assumed to be wavelength paths, (WP), • lout link that exits a node.
(lightpaths). Thus, the wavelength of a path is not converted at any We consider variable ymax to denote the maximum number of
transit node in the network. We consider an optical network that offers demands going through each link of the general graph. Also let xdl be
k wavelengths for setting up WPs. Let G (V , E ) denote the direc- a variable that denotes a demand d going through a link l of the
ted graph that models the network. This model is equivalent to the general graph. xdl =1 if demand d goes through the link l and 0
model with k separate networks, each of which has the same topology otherwise. The formulation of the problem is as follows:
as G (V , E ) and provides only one wavelength (Fig. 2). Let the
graphs of these k single networks be G (1) ,..., G (k ) . Also, let v (i ) Minimize ymax
denote the vertex in G (i ) corresponding to the vertex v ∈ V in the Subject to:
original graph. It is possible to combine k G (i ) graphs into one 1. the sum of all demands on each link should be less or equal than
ymax
extended graph in the following way. Add vertices s1,…s|v| and create
D
an edge from sv to v(i) for V every v ∈ V and every 1<i<k.
Similarly, add |V| more vertices t1, … t|v| and create an edge from v(i)
¦x
d =1
dl ≤ ymax ∀l ∈ E1 , λ ∈ Λ
to tv for every v ∈ V and every 1<i<k. With this graph conversion,
2. all paths that share a link should have different wavelengths
the routing and wavelength assignment problem becomes a multi-
commodity flow problem for each sub-graph G (i ) . Let D = {d1, … dj}
 COMMON-WAVELENGTHS DESIGN SCHEME 501

¦x
d =1
dl ≤1 ∀l ∈ E , λ ∈ Λ
3. each demand uses the same wavelength through its path
E1

¦x
l =1
dl ≤1 ∀d ∈ D, λ ∈ Λ
4. There are no loops
E1 D E1 D

¦¦ x
l =1 d =1
dl in + ¦¦ xdl out ≤ 2
l =1 d =1
∀n ∈ V , λ ∈ Λ
5. Conservation law
­ 1 if n is the origin
E1 D E1 D
°
¦¦ x dlin − ¦¦ xdlout = ®− 1 if n is the destination
l =1 d =1 l =1 d =1 ° 0 otherwise
¯
6. All variables can be either 1 or 0
x dl = 1 or 0 ∀d ∈ D, λ ∈ Λ

IV. SIMULATIONS AND RESULTS

The network models used for our simulations are shown in Fig. 3.
The parameters of these networks are the number of network nodes
and the physical connectivity. The number of fiber links between all
network nodes defines the physical connectivity in our simulations.
The total number of fibers per link is considered 1. Then we generate
4 different traffic patterns (demands) randomly chosen for the
network to be fully loaded. Each of these traffic patterns is a
combination of three classes of service. The first class consists of Fig. 3. Topologies of the photonic networks under investigation: a) regular
networks and b) network with random topology.
20%, the second class consists of 30% and the third class consists of
remaining 50% of the total demand. Each is chosen randomly. In case
of a failure the first and second class will have priority for restoration
and the third one would have to release some of its paths to make the
“common-wavelengths” available to the traffic demand. Based on the
requirements described above the ILP model is build to solve the
problem as follows:
1. Calculate the number of wavelengths needed to satisfy the
demand in the network.
2. Generate a random link failure and calculate the number of
wavelength needed to satisfy all traffic in the network.
3. Calculate the spare wavelengths needed to satisfy the
affected wavelength paths.
4. Release some or all traffic from the third class and calculate
Fig. 4. Simulation results for 15 nodes α=0.238 network.
the number of wavelengths needed to satisfy the affected
traffic and released traffic.
5. Repeat the above steps for different failure scenarios and traffic matrix, like physical connectivity itself, is represented by a
different traffic patterns. graph. The difference between the traffic matrix and the physical
Results of simulations for the three schemes described in Fig. 1 network is that the edges in the graph are represent the end-to-end
considering 15 nodes =0.238 network are shown in Fig. 4. We can connections. Then we studied the impact of the traffic connectivity on
see that on average scheme (a) requires about 40%, scheme (b) the number of wavelengths required to fulfil the given demand. Two
requires about 29% and our “common-wavelength” scheme (c) regular networks, consisting of 9 nodes =0.5 and 15 nodes =0.238
requires only 10% of total resources. The same trend is observed for respectively, were considered. Results are shown in Fig 5. As we
the other networks presented in Fig. 3a (not presented here). observe from this figure the increase in the traffic matrix will result on
We determine the connectivity of the traffic matrix from the the increase of the number of required wavelengths. Furthermore, a
number of communication paths in it and the number of nodes. The decrease on the physical connectivity will result on the increase of the
502 GRECA

REFERENCES
[1] P. Demeester, M. Gryseels, K. van Doorselaere, A. Autenrieth, C.
Brianza, G. Signorelli, et al. “Network resilience strategies in
SDH/WDM multilayer networks”, 24th European Conference on Optical
communication, Vol. 1, pp. 579-580, September 1998.
[2] C. Ou, K.Y. Zhu, B. Mukherjee, “Survivable traffic grooming in WDM
mesh networks”, IEEE OFC, Vol. 2, pp. 624-625, March 2003.
[3] J. Zhang, et al IEEE OFC, Vol. 2, pp. 622- 623, March 2003.
[4] A. Greca, et al IEEE Workshop on HPRS, pp. 146-149, Dallas 2001.
[5] S. Ramamurthy, B. Mukherjee, “Survivable WDM mesh networks, part I
– protection”, Proc. IEEE Infocom 99.
[6] S. Ramamurthy, B. Mukherjee, “Survivable WDM mesh networks, part I
– restoration”, Proc. ICC 99.
[7] O. Gerstel, R. Ramaswami, “Optical layer survivability – an
implementation perspective”, IEEE Journal of Selected Areas in
Communications, Vol. 18, No. 10, pp. 1885 – 1889, October 2000.
[8] J. Manchester, P. Bonenfant, C. Newton, “The evolution of transport
network survivability”, IEEE Communication Magazine, August 1999.
[9] O. Gerstel, R. Ramaswami, “Optical layer survivability: a service
Fig. 5. Wavelength requirement vs. traffic connectivity in regular network
perspective”, IEEE Communication Magazine, pp. 104 –113, March
topologies. 2000.
[10] Ch. Li, R. Ramaswami, “Automatic fault detection, isolation, and
number of wavelengths. This increase is almost linear. These results recovery in transparent all-optical networks”, Journal of Lightwave
Technology, Vol. 15, No. 10, pp.1784 –1793, October 1997.
are in line with other observations given in literature. [11] T.M. Chen, T.H. Oh, “Reliable services on MPLS“, IEEE
Communication Magazine, pp. 58 – 62, December 1999.
[12] K. Owens, S. Makam, V. Sharma, B. Mack-Krane, “A path
V. CONCLUSIONS protection/restoration mechanism for MPLS networks”, IETF draft, draft-
chang-mpls-path-protection-02.txt, July 2000.
In this paper we studied the design of “common-wavelengths” [13] V. Sharma, B.M. Crane, S. Makam, et al., “Framework for MPLS-based
scheme which can be used to build cost-effective survivable recovery”, IETF draft, draft-ietf-mpls-recovery-frmwrk-01.txt, November
2000.
WDM networks. We used ILP to solve the problem. Simulation [14] L. Mo, “General considerations for bandwidth reservation in protection”,
results confirmed that the proposed scheme reduces the number IETF draft, draft-mo-mpls-protection-00.txt, work in progress, July 2000.
of required wavelengths when a failure happens in the network. [15] D. Haskin, R. Krishman, “A method for setting an alternative label
switched path to handle fast reroute”, IETF draft, draft-haskin-mpls-fast-
reroute-05.txt, November 2000.
[16] P. Demeester, M. Gryseels, A. Autenrieth, et.al., “Resilience in
Multilayer Networks“, IEEE Communication Magazine, August 1999.
 ICMP Covert Channel Resiliency
Kristian Stokes, Bo Yuan, Daryl Johnson, and Peter Lutz
Department of Networking, Security, and Systems Administration
B. Thomas Golisano College of Computing and Information Sciences
Rochester Institute of Technology
Rochester, New York 14623

Abstract—The ICMP protocol has been widely used and II. C URRENT C OUNTERMEASURES
accepted as a covert channel. While the ICMP protocol is very
simple to use, modern security approaches such as firewalls, deep-
Countermeasures exist within the ICMP covert channel
packet inspection and intrusion detection systems threaten the use realm, however they come at a cost. As with all aspects of
of ICMP for a reliable means for a covert channel. This study security, tradeoffs exist.
explores the modern usefulness of ICMP with typical security Blocking all ICMP traffic from entering the network pre-
measures in place. Existing ICMP covert channel solutions are vents all aspects of ICMP communication, including covert
examined for compliance with standard RFCs and resiliency with
channels. This methodology may not be acceptable due to
modern security approaches.
the loss of network troubleshooting abilities. Blocking ICMP
communications at a central firewall through the use of ACL
I. I NTRODUCTION mechanisms may not solve the problem. The capability to send
ICMP messages in an intranet environment may still exist.
Blocking specific ICMP messages from entering the net-
The Internet Control Message Protocol (ICMP) [1] is de- work. This methodology also incorporates the use of segment-
signed to provide feedback about problems in the communi- ing incoming and outgoing connections. If this countermeasure
cation environment. ICMP relies on the basic support of IP is enabled, both parties involved in the covert channel can
as part of a higher level protocol. Due to this dependency, develop fuzzing techniques to modify ICMP messages used
both ICMPv4 and ICMPv6 exist for both versions of IP. for communication or party initiation.
Many message codes exist within ICMP to properly diagnose Restricting the size of ICMP packets. By blocking large
network problems and traffic flow. ICMP messages are sent ICMP packets, an ICMP message with an extensive data field
under many different circumstances such as an unreachable will be dropped and perceived as a crafted packet with a
destination or general congestion control on the network. malicious or unknown payload. Large ICMP packets can be
Simple network troubleshooting utilities such as ping and used to test a network for proper handling of large packets.
traceroute utilize explicit ICMP messages to gather informa- An adversary can also overcome this limitation by fragmenting
tion about a network. their ICMP covert channel to smaller packets.
Covert channels often refer to a hidden information stream Traffic normalization. Traffic normalization techniques such
and more specifically, hidden streams embedded in IEEE as state preservation will reply and generate new packets on the
802 networks. Lampson [2] originally defined covert channels senders behalf. The normalizer, typically a firewall, will serve
under a number of categories such as storage, timing, termi- as a proxy, rebuild messages and construct new payloads. This
nation, resource exhaustion and power. Most covert channels activity essentially disrupts ICMP covert channel communica-
that involve the use of ICMP are largely storage channels tion by stripping out the message payload. State preservation
where unused fields are utilized for covert communication. requires significant computational power and may not scale to
ICMP as a covert channel provides many benefits due to particular environments with high traffic loads.
the overall simplicity. Only several fields exist within most
ICMP messages which enables quick implementation and III. R ELATED W ORK
simple channel setup/teardown. The idea of using ICMP as Research in this area focuses on existing solutions currently
a covert channel is to use an atypical communication protocol available for ICMP covert channel communication. Loki, an
rather than ordinary TCP or UDP. This will have a smaller ICMP tunneling back door application, tunnels remote shell
footprint across the network and may go unnoticed by network commands in ICMP echo reply / requests and DNS query /
administrators and traffic analyzers. What really makes the reply traffic. This proof of concept was originally published
ICMP protocol a viable covert channel is the use of data fields in Phrack World News [3] to demonstrate the vulnerabilities
or payloads within certain messages. By generating packets within these protocols. This implementation is very easy to
based on specific message codes and embedding the actual deploy and thusly carries a security risk of ICMP tunneling.
covert channel message in the data field enables ICMP to serve Another implementation named Ping Tunnel [4], is focused
as an alternate use for covert channels. These simple factors on reliably tunneling TCP connections using ICMP echo
enable ICMP to be considered as stealth traffic. request and reply packets. This tool can be with an outside

503
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_87, © Springer Science+Business Media B.V. 2010
504 STOKES ET AL.

proxy to tunnel traffic using ICMP messages back to the remote host. This solution can be utilized as a covert channel
requesting client. The exterior proxy serves all TCP requests when the operating network is heavily restricted.
and forwards the data back to the client via ICMP echo reply. Each solution will be tested and further explored in the
Although this solution can be viewed as subverting typical following sections in terms of resiliency with modern security
network communication, it is also important to realize other devices such as firewalls and intrusion detection systems.
potential uses for ICMP covert communication.
ICMP-Chat [6] implements a basic console-based chat V. E XPERIMENT
mechanism that utilizes ICMP packets for communication. A minimalistic test environment was created in a lab en-
A unique aspect of this solution incorporates the use of an vironment using VMware Workstation. Two CentOS work-
encrypted data field using AES-256. By implementing an stations are used as the endpoints of the covert channel. A
encrypted payload this further secures the covert channel, but Checkpoint NGX R65 firewall separates the network between
adds increased suspicions on an abnormal ICMP payload. internal and external clients. The internal network simulates a
With the increasing awareness of IPv6, covert channel tools small business with a firewall and IDS in place. The external
further expanded into the realm of ICMPv6. A tool v00d00N3t network simulates an Internet node communicating to the
[7] was developed to operate specifically over IPv6. This business. ICMP communication is permitted to the interior
tool focuses on the infancy of IPv6 protection technology. node, all remaining traffic is dropped at the firewall.
Dual IPv4/IPv6 routers are utilized to route IPv6 traffic to The Checkpoint firewall serves as a network layer inspection
send messages and files using ICMPv6. Useful information is engine to detect fields and patterns in the network layer.
embedded into fields other than the data field. For instance, the This firewall provides sufficient inspection for a network layer
sequence number tells the covert channel receiver if it should covert channel.
read the packet and ID field is used to identify how many A Snort IDS [8] is deployed inline on the network listening
bytes out of the payload to read. in promiscuous mode. All traffic passing through the internal
Additional defensive research has been performed to further network is captured and analyzed by the IDS. The standard
limit the capabilities of ICMP covert channels. A Linux kernel Snort ruleset was updated to the latest version from Sourcefire.
module was developed to scan ICMP messages for specific The basis for this experiment focuses on the resiliency of
signatures such as passwd, root, etc, ls and dir [5]. If these ICMP as a viable covert channel. ICMP-Chat and Ping Tunnel
signatures were detected, the ICMP message was scrubbed will be used in this environment to test resiliency against a
by zeroing out the data field while being processed by the modern firewall and IDS. Resiliency will be examined for
network stack. This technique is similar to normalizing traffic, each tool on connection establishment, connection integrity
however the action of data scrubbing is performed on the and overall covert design.
end nodes. This solution can also be deemed as a very
VI. R ESULTS
computational intensive process involving bi-directional deep-
packet inspection. The added network processing overhead Given the provided experiment, both ICMP-Chat and Ping
may not be acceptable for performance reasons. Tunnel are examined for ICMP resiliency. Each tool was
With these current solutions outlined, little information is installed and tested for functionality on both internal and
available for actual survivability of the tools and general external workstations.
ICMP covert channel message resiliency with common se-
A. ICMP-Chat
curity appliances. The use of intrusion detection, intrusion
prevention and firewalls are commonplace within a production By default, ICMP-Chat uses echo reply packets for primary
environment. To fully understand the capabilities of ICMP communication. Figure 1 illustrates the network topology of
as a covert channel, it must contain a moderate amount of the experiment. When comparing standard operating system
resiliency in modern networks. echo reply packets to echo reply packets generated by the
ICMP-Chat application many parameters follow standard RFC
IV. E XISTING S OLUTIONS compliance.
The standard CentOS echo reply, shown in Figure 2 consists
ICMP-Chat [6] is based on a simple console-based chat in- of several key focus areas such as packet size and data
terface which uses ICMP packets for communication. Features content. A total of 98 bytes are captured on the wire with
of this solution include the use of AES-256 encryption for a 56 byte data field. The data field also standardizes on the
encrypting chat data across the channel. ICMP-Chat provides following sequence of characters: “!#$%&’()*+,-./01234567”.
several protection mechanisms such as password protecting the For a perfect covert channel, the ICMP communication should
channel using SHA-256 and supporting the ability to change match this similar format to reduce detection.
usage of ICMP codes within the application. This allows for An ICMP-Chat echo reply session was initiated, sent with
mobility between ICMP codes to further obfuscate the channel. the message “test” and captured in Figure 3.
Ping Tunnel [4] allows a user to tunnel TCP connections Major differences with the reply packet in Figure 3 centers
to a remote host using ICMP echo request and reply packets. on the overall packet size of 318 bytes with a data size of
This is achieved through the use of a proxy serving as the 276 bytes. The dramatic increase in data size contributes to
 ICMP COVERT CHANNEL RESILIENCY 505

an abnormal attribute for typical ICMP traffic. Large packet

size decreases the covertness of this solution. If the network re-
stricts large ICMP packets, ICMP-Chat will likely be blocked.
Firewall Resiliency. Session initiation of ICMP-Chat with
External
the Checkpoint firewall must follow request and reply struc-
CentOS 5.2 ture. By default, ICMP-Chat uses echo reply packets. Similar
192.168.202.153
to stateful firewall inspection, the Checkpoint firewall expects
192.168.202.136
an echo request then permits an echo reply. If both internal
and external nodes use the echo replies, the firewall does not
10.1.1.1
permit the communication. If the communication is changed
to an echo request and reply structure the Checkpoint firewall
Checkpoint Firewall Snort IDS Internal permits the communication. This similar structure must be
2.8.3.1 CentOS 5.2 followed for continued communication; an echo request must
In-line promiscuous 10.1.1.2
be received before an echo reply is permitted to traverse the
Fig. 1. ICMP-Chat covert channel environment firewall.
IDS Resiliency. The Snort IDS was unable to detect abnor-
mal ICMP traffic when conducting the covert channel. This
further confirms that abnormal ICMP packet sizes are not
Frame Frame 2397 (98 bytes on wire, 98 bytes captured) added to the standard IDS ruleset. Given that the data field
Ethernet II, Src: Vmware_3a:02:dd (00:50:56:3a:02:dd), Dst:
Vmware_3a:07:f8 (00:50:56:3a:07:f8) is encrypted, this adds to the level of complexity needed for
Internet Protocol, Src: 10.1.1.2 (10.1.1.2), Dst: 192.168.202.153
(192.168.202.153) IDS detection in covert channels.
Internet Control Message Protocol
Type: 0 (Echo (ping) reply)
Code: 0 () B. Ping Tunnel
Checksum: 0x2568 [correct]
Identifier: 0x2448
Sequence number: 1 (0x0001)
Ping Tunnel [4] serves as a covert tunneling tool to disguise
Data (56 bytes)
Data: CB9E0B4AEC62080008090A0B0C0D0E0F1011121314151617
TCP traffic in ICMP request and reply packets. The basis
0000 00 50 56 3a 07 f8 00 50 56 3a 02 dd 08 00 45 00 .PV:PV:.E.
of this technique is to disguise traffic as a wrapper protocol
0010
0020
00
ca
54
99
84
00
94
00
00
25
00
68
40
24
01
48
5f
00
d0
01
0a
cb
01
9e
01
0b
02
4a
c0
ec
a8
62
.T.@._.......
.%h$H..J.b
and bypass specific TCP filtering. This is achieved through
0030
0040
08
16
00
17
08
18
09
19
0a
1a
0b
1b
0c
1c
0d
1d
0e
1e
0f
1f
10
20
11
21
12
22
13
23
14
24
15
25
.............
....... !#$%
the use of an external proxy to translate transmitted ICMP
0050
0060
26
36
27
37
28 29 2a 2b 2c 2d 2e 2f 30 31 32 33 34 35 &()*+,-./012345
67
client packets back to standard TCP packets. Unlike ICMP-
Chat, Ping Tunnel strictly uses ICMP echo request and reply
Fig. 2. Standard CentOS echo reply packets for communication. Connections are very similar to
TCP in that lost packets are resent as necessary to allow for
reliability. Multiple connections are permitted through the use
of the ICMP identifier field. The identifier field is included
within the standard Ping Tunnel packet format and should not
Frame 491 (318 bytes on wire, 318 bytes captured)
Ethernet II, Src: Vmware_3a:02:db (00:50:56:3a:02:db), Dst: be confused with the ICMP sequence number field.
Vmware_3a:07:f7 (00:50:56:3a:07:f7)
Internet Protocol, Src: 192.168.202.153 (192.168.202.153), Dst: 10.1.1.2 The example test environment, illustrated in Figure 4, was
(10.1.1.2)
Internet Control Message Protocol designed to allow the client node to establish an SSH tunnel to
Type: 0 (Echo (ping) reply)
Code: 0 () an Internet based server. A known proxy address was provided
Checksum: 0x8857 [correct]
Identifier: 0x0000 to the client and listens on a local port which tunnels all traffic
Sequence number: 1280 (0x0500)
Data (276 bytes) via ICMP echo request/reply packets. Once the ICMP proxy
Data: 0000000000000000000000000000000000000000449569F8...
was established, an SSH connection was initiated from the
0000 00 50 56 3a 07 f7 00 50 56 3a 02 db 08 00 45 00 .PV:...PV:....E.
0010 01 30 eb f9 00 00 40 01 f7 8e c0 a8 ca 99 0a 01 .0....@......... client to the localhost port.
0020 01 02 00 00 88 57 00 00 05 00 00 00 00 00 00 00 .....W..........
0030 00 00 00 00 00 00 00 00 00 00 00 00 00 00 44 95 ..............D. Upon establishment of the connection, the standard SSH
0040 69 f8 98 69 2a cf 87 6c 41 eb bf 95 14 0c 7e b8 i..i*..lA.....˜.
0050 af 0c d4 17 ad d0 2e 76 b8 75 cf 29 a4 8f 4c f7 .......v.u.)..L. handshake can be viewed in captured ICMP echo request/reply
0060 75 fe 22 86 a4 96 a6 14 0d 03 3c 2a 0d c5 95 9d u.".......<*....
0070 ca cc 56 1d 39 f0 21 ab 1d a9 76 0c 32 ec 71 3a ..V.9.!...v.2.q: packets.
0080 11 66 17 46 0a dc 5d ce 74 41 f8 bf e7 ee a1 1e .f.F..].tA......
0090 c2 b9 88 07 6a 18 f2 88 3c 3d 5d 29 a4 b9 49 ac ....j...<=])..I. The packet capture in Figure 5 shows a standard OpenSSH
00a0 bd c3 91 ec 82 da 2b d9 eb 4d b7 8e 54 44 ef 04 ......+..M..TD..
00b0 d4 91 4d 30 d7 91 c8 36 0f 92 d1 47 4b 50 e5 1f ..M0...6...GKP.. version handshake embedded in the data field of a ICMP echo
00c0 8e 68 c6 90 1d 9c a0 46 bb b0 2f 64 17 ed 60 ca .h.....F../d..‘.
00d0 d9 81 89 b7 30 08 9d 80 c6 f6 cd 4e 7c 0f 82 67 ....0......N|..g reply.
00e0 15 04 fe e6 d9 8f 7b 20 a6 12 be 2c 90 0e b5 2a ......{ ...,...*
00f0 8d 16 56 60 d0 85 f8 60 1f 3d 54 25 d4 71 3b 95 ..V‘...‘.=T%.q;. Firewall Resiliency. The Checkpoint firewall permitted
0100 8e 24 82 68 b4 64 0d be 00 35 92 67 6e 4d 46 e0 .$.h.d...5.gnMF.
0110 40 85 f2 a7 89 76 23 75 d4 72 7a 7e 94 4e 09 3e @....v#u.rz˜.N.> Ping Tunnel traffic largely due to the adherence to the ICMP
0120 5f 29 43 c7 ce ec 45 a1 f6 1a 76 7c 79 af 87 dc _)C...E...v|y...
0130 d1 63 26 fd 28 48 b3 63 ca 43 5f 82 94 00 .c&.(H.c.C_... echo request and reply structure. If this request and reply
format is continued throughout the communication, the traffic
Fig. 3. ICMP-Chat echo reply will go largely unnoticed. Unsolicited connections will be
dropped by the firewall. Similar to ICMP-Chat, the Checkpoint
firewall did not specifically block large ICMP packets.
506 STOKES ET AL.

IDS Resiliency. The inline Snort IDS also failed to detect

this covert channel. Again, abnormally sized ICMP messages
are overlooked given this situation. It is possible to tailor a
custom ruleset to detect this activity, but the default ruleset
Client
fails to recognize frequent, abnormally large ICMP messages.
CentOS 5.2
192.168.234.132
VII. C ONCLUSION
Snort IDS Based on the experimental findings of this study it is very
2.8.3.1 clear that a simple ICMP based covert channel can easily
In-line
subvert many modern security appliances if general ICMP
promiscuous
traffic is permitted. Administrators and security researchers
should be aware of the capabilities of a seemingly helpful
protocol. Current covert channel tools are widely and freely
Checkpoint Firewall
available for use across a number of platforms.
ICMP Proxy Blocking all ICMP traffic may not be an acceptable business
CentOS 5.2 practice in many cases. Steps can be taken to further reduce the
192.168.234.131
risk of an ICMP covert channel. Limiting the overall packet
ICMP size of ICMP messages may disrupt communications. Ensuring
unsolicited ICMP messages are dropped at the perimeter can
Internet assist in preventing channel establishment, but as investigated,
this can be easily circumvented by following a request and
TCP reply structure.
An IDS can be further improved by monitoring ICMP traffic
flow. Ping Tunnel generated abnormally sized ICMP packets
and often produced constant or bursting traffic to the proxy
node. IDS technology can be implemented to flag abnormal
SSH Server
traffic for further investigation. The traffic analysis can be
compared to constant streams of ICMP traffic resulting in
Fig. 4. Ping Tunnel covert channel environment
varying packet sizes to a single destination.
No. Time
10 0.012010
Source
192.168.234.131
Destination
192.168.234.132
Protocol
ICMP
Even with many security mechanisms and precautions in
Frame 10 (92 bytes on wire, 92 bytes captured)
place, covert channels may still exist. This activity is essen-
Ethernet II, Src: Vmware_3a:02:dd (00:50:56:3a:02:dd), Dst: Vmware_3a:02:db
(00:50:56:3a:02:db)
tially what a covert channel fundamentally strives to be, an
Internet Protocol, Src: 192.168.234.131 (192.168.234.131), Dst: 192.168.234.132
(192.168.234.132)
undetectable channel of communication.
Internet Control Message Protocol
Type: 0 (Echo (ping) reply)
Code: 0 ()
R EFERENCES
Checksum: 0x195d [correct]
Identifier: 0xe510 [1] Postel, J. “INTERNET CONTROL MESSAGE PROTOCOL.” RFC 792.
Sequence number: 0 (0x0000)
Data (50 bytes)
Internet Draft Submission Tool. 30 Apr. 2009.
[2] Lampson, B. W. 1973. “A note on the confinement problem.” Commun.
0000
0010
d5
00
20 08 80 00 00 00 00 00 00 00 00 80 00 00 02
00 00 00 00 00 00 15 00 00 e5 10 53 53 48 2d
. ..............
............SSH-
ACM 16, 10 (Oct. 1973), 613-615.
0020 32 2e 30 2d 4f 70 65 6e 53 53 48 5f 35 2e 31 0d 2.0-OpenSSH_5.1. [3] Daemon9. 1997. “LOKI2.” Phrack Magazine, Vol. 7 (51)
0030 0a 20 .
Data: D52008800000000000000000800000020000000000000015...
http://www.phrack.com/issues.html?issue=51&id=1.
[4] Stodle, Daniel. “Ping Tunnel.” http://www.cs.uit.no/ daniels/PingTunnel.
[5] Singh, Abhishek, et al. 2003. “Malicious ICMP Tunneling: Defense
against the Vulnerability.” In: Lecture Notes in Computer Science, Vol.
Fig. 5. Ping Tunnel OpenSSH handshake 2727, Springer Berlin / Heidelberg, pp. 226-236.
[6] Muench, Martin J. 2003. “ICMP-Chat.”
http://icmpchat.sourceforge.net/index.html.
[7] Murphy, R. P. 2006. “IPv6 / ICMPv6 Covert Channels.” DEFCON 14.
https://forum.defcon.org/archive/index.php/t-7588.html.
[8] Sourcefire. “Snort IDS.” http://www.sourcefire.com.
 Chip-to-chip Free-space Optical Interconnection Using Liquid-crystal-over-
silicon Spatial light modulator
Sadia Sultana, Ferdous Mohammed Shahriar, Mohammad Khalad Hasan
Department of Computer Science & Engineering
Shahjalal University of Science and Technology
Sylhet-3114, Bangladesh.
[email protected], [email protected], [email protected]

Abstract—Free-space optical chip-to-chip interconnect interconnection in high-capacity ATM switch was

offers a promising solution to bandwidth requirement. To proposed [2]. Free-space switch demonstrator using
move a light beam from a chip to another chip we need reconfigurable hologram design was reported [3]. Free-
beam steerers which will steer the beam to desired space board-to-board optical interconnect structure
direction. Different beam steering technologies have employing a single ferroelectric liquid-crystal spatial
already been demonstrated to control the direction of light light modulator in conjunction with a free space optical
beam. These include different types of prisms, micro- polarizing beam splitter, halfwave plates and
electro-mechanical systems (MEMS), and Opto-VLSI collimating lenses has been reported [4]. Chip-to-chip
processors. But all of those technologies have the optical interconnection using MEMS mirrors was
limitations of having high optical losses & low signal-to- experimented by Tod Laurvick [5]. But still a complete
noise ratio. But the spatial light modulator (SLM) has & feasible technology for chip-to-chip free-space
several advantages over these. The large number of optical interconnection does not exist. In our model we
independently controllable elements in SLMs is expected have proposed the use of SLMs as beam steerer in
to offer low optical loss as they are dynamic, controllable, conjunction with Hologram design. As both of these
and repeatable without moving parts. In this paper, we
have several advantages over other technologies, we
have proposed the concept of chip-to-chip optical
interconnect that employs vertical cavity surface emitting
expect to have better performance using this model.
lasers (VCSEL) and photo detectors (PD) in conjunction The rest of the paper is organized as follow: at first we
with two SLMs and Hologram design. This model for introduce the functions of SLM & Hologram design in
chip-to-chip free space optical interconnects uses SLM as section II and III respectively. In Section IV, our
beam steerer which is easy to control. Using our proposed proposed model for chip-to-chip interconnection is
model, we can have low optical losses, low crosstalk. This presented. Section V concludes the paper.
proposed chip-to-chip optical interconnection model can
be applied in processor-memory and processor-processor II. THE LIQUID CRYSTAL OVER SILICON
communication. SPATIAL LIGHT MODULATOR
Keywords-Diffraction, Electrode, Hologram Design, Liquid
A spatial light modulator (SLM) is a device that
Crystal, Spatial light modulator.
imposes some form of spatially-varying modulation on
I. INTRODUCTION a beam of light. Usually, an SLM modulates the
intensity or phase of the light beam, or both the
In electrical interconnect, density is a great problem. intensity and the phase simultaneously. In our model
The smaller elements on a chip operate faster. As the the SLM only changes the phase of light without
number of elements in a chip increases, this requires changing the intensity and relation of polarization state.
more connections, and placing all of these connections A silicon backplane phase modulating SLM acts as a
on a 2-D surface is a difficult optimization problem. beam-steerer. It consists of a liquid-crystal-over-silicon
Over short distances, for high frequency clock (LCOS) backplane, a layer of nematic liquid crystal,
distribution or high-speed data transfer optical and a cover glass. The LCOS backplane is an array of
interconnect will be the most feasible solution. aluminum pixels which serve as both reflective mirrors
Reconfigurable free-space optical interconnect modules and electrodes. Each electrode is an independently
employing polarization-selective diffractive optical controllable pixel. The modulation is done according to
elements in combination with a liquid crystal based the alignment of liquid crystal which is controlled pixel
polarization controller have previously been reported by by pixel. This silicon backplane also contains
Goulet, et al. However, these modules result in high addressing circuitry for the pixels. The liquid crystal
optical losses especially when the number of output layer is considered to be divided into different block
ports increases [1]. In next, using Opto-VLSI each corresponds to an electrode pixel. With each
processors as beam steerers and multicasters for electrode an electric field can be applied from the
reconfigurable interchip optical interconnection was LCOS backplane across the liquid crystal block to
reported [1]. Optical beam direction compensating achieve optical modulation. The cover glass is coated
system for board-to-board free space optical with a transparent electrode. As light enters the LC

507
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_88, © Springer Science+Business Media B.V. 2010
508 SULTANA ET AL.

layer through cover glass, an electric field is applied to fundamental mechanisms of optical propagation is
the layer to change the refractive index that depends on through the process of diffraction. In phase only
molecular orientation of liquid crystal layer block. The hologram or grating diffraction is done without
index change induces a phase-only modulation of the blocking photons. By using a hologram or grating to
input light. It is then reflected from the aluminum steer the light by diffraction, we can correct for any
mirror and passes back through the liquid crystal layer motion or misalignments and create an adaptive optical
[7]. Multilevel phase modulation can be achieved using interconnect. The SLM must maintain DC balance
a nematic material with the correct birefringence and during the display of each corrective pattern. [8]. A
molecular orientation [8]. The SLM chip is mounted in controller algorithm is used to design a suitable
a ceramic Pin Grid Array (PGA) package. This PGA adaptive binary holograms to calculate the amount of
package connects to a signal controlling board for phase modulation required for alignment of the light
receiving and sending data, power and control signals beam. In order to achieve active alignment we need to
[9]. The signal is collimated by the lens onto the SLM. generate holograms in real time thus we can use a non-
By controlling the deflection angle by SLM, the beam iterative algorithm [8]. When the light beam incidents
may be made to return to a selected point by on the SLM then the phase modulation causes it to split
collimating again by a lens. into a number of reflected beams and a phase
Following figures shows a cross-sectional view of a modulation pattern is created. Each beam corresponds
liquid-crystal-on-silicon spatial light modulator to a different diffraction order of the phase modulation
illustrates how the liquid-crystal layer causes a voltage- of which only one is created intentionally to carry the
dependent phase shift upon entering light. Each pixel is signal into the selected output point. The other
fully programmable, offering a high-resolution pure diffraction orders are unwanted and they may lead to
phase-modulating array [6]. crosstalk and back reflection. The first-order diffraction
efficiency of a binary-phase hologram with LC layer
thickness, and molecular tilt angle θ is given by
equation (1), where Δn is the birefringence and λ is
the operating wavelength [3].

§ 2π Δ nd ·
η = sin 2 2θ sin 2 ¨ ¸ (1)
© λ ¹

The output angle of each diffraction order, measured

Fig. 1. A cross-sectional view of a liquid-crystal-on-silicon spatial from the optical axis is given by equation (2), where
light modulator. is the period of the phase modulation, usually known as
the hologram period, and m is an integer that identifies
a particular diffraction order and  is wavelength of the
laser light that is used [3].
mλ
sin φ out = sin φ IN + (2)
Λ

Fig. 2. Voltage-dependent phase shift upon entering light.

III. HOLOGRAM DESIGN

Fig. 3: Incident and Reflected Light Beam.
To route a beam to a single point binary phase
hologram is used. The hologram refers to the phase Figure 3 shows the mechanism of driving a light beam
modulation 2D pattern and period applied by the SLM. to a selected pixel by using phase hologram.
Usually the technique of holography is used to optically
store, retrieve and process information. By using a
reconfigurable hologram we can steer the beam to any
desired positions in the detector plane. One of the most
 CHIP-TO-CHIP FREE-SPACE OPTICAL INTERCONNECTION 509

Figure 5 shows the first interconnect scenario, where a

VCSEL in one side connects to a PD of another side.
Both SLMs are loaded with proper steering holograms
to steer the beams of VCSEL #2 to PD #2.

Fig. 4 Multiple Beam steering using Diffraction.

If a multicasting phase hologram is synthesized,

multiple beams can be generated (Figure 4), whose
intensities can be controlled by reconfiguring the phase
hologram [1]. And one of our main concerns is to create
multiple beams to communicate with multiple chips for
parallel communication.
Fig. 6. 1-to-N interconnect scenario.

IV. PROPOSED MODEL Figure 6 shows the second interconnect scenario, where
multicasting interconnects can be established by using
The descriptions and functions of SLM and hologram multicasting phase hologram. Both SLMs are loaded
design were stated above. The complete mechanism of with proper steering holograms to steer the beams of
our proposed model is give here: VCSEL #2 to PDs #1 and #2.
Each chip has a VCSEL & a Photo Detector (PD) with
it. To communicate with another chip, each chip send
light beam by using VCSEL & in the designated chip
this beam of light is detected by using a PD. For each
VCSEL an individual pixel of SLM is assigned. Each
PD also has an individual pixel assigned to it. Light
from a VCSEL is collimated and focused by using a
collimating lens and it incidents on the first SLM. The
addressing & other information are sent to the signal
controlling board. Then the hologram pattern is
generated using Fourier series analysis and the required
diffraction angle is calculated by using an appropriate
controller algorithm. Light beam is then phase shifted Fig. 7. N-to-N parallel interconnect scenario.
and directed toward the pixel of the second SLM which
Figure 7 shows the third interconnect scenario, where
is assigned to the destined PD. The second SLM again
parallel optical interconnects can be established
do the same things and direct the light beam to the
simultaneously by two SLMs using the VCSELs and
specific PD. A collimating lens is used to collimate and
the PDs. Both SLMs are loaded with proper steering
focus the light beam to the PD for a chip. The different
holograms to steer the beams of VCSEL #1, #2, and #3
scenarios of this model that can be implemented are as
to PD elements #3, #1, and #2.
follows:
V. CONCLUSION

In this paper, we have proposed a model based on the

performance analysis of others concept for chip-to-chip
free space optical interconnects. This uses Spatial Light
Modulators as beam steerer, Hologram design, VCSELs
as light sources & PD as light detectors. It has different
scenarios including 1-to-1, 1-to-N & N-to-N
interconnections. SLMs have the advantages of being
dynamic, controllable, and repeatable without moving
parts [6]. Thus it offers a good long-term reliability.
The deflection angle is a function of only the hologram
pattern displayed on the SLM giving repeatable and
predictable performance, and as holograms are
Fig. 5. 1-to-1 interconnect scenario.
510 SULTANA ET AL.

relatively insensitive to pixel errors there are excellent [2] Katsuhiko Hirabayashi, Tsuyoshi Yamamoto, Shigeki
prospects for high yield fabrication of the devices. It is Hino, Yoshitaka Kohama, and Kouta Tateno, “Optical
Beam Direction Compensating System for Board-to-
also possible to correct for optical aberrations in the Board Free Space Optical Interconnection in High-
system. Finally, the prospect of liquid crystal on silicon Capacity ATM Switch”. Journal of Lightwave
(LCOS) devices means that, in principle, the necessary Technology, Vol. 15, No. 5, May 1997.
control electronics could be integrated on to the device [3] W. A. Crossland, I. G. Manolis, M. M. Redmond, K. L.
to make a very compact system [4]. Tan, T. D. Wilkinson, M. J. Holmes, T. . Parker, H. H.
There are two key challenges of implementing this Chu, J. Croucher, V. A. Handerek, S. T. Warr, B.
Robertson, I. G. Bonas, R. Franklin, C. Stace, H. J.
model. First, as the SLM is pixellated, so the possible White, R. A. Woolley, and G. Henshall, “Holographic
hologram periods must be an integer multiple of the Optical Switching: The “ROSES” Demonstrator”;
pixel pitches. Second, the unwanted diffraction orders Journal of Lightwave Technology, Vol. 18, No. 12,
that can lead to crosstalk and back reflection must be pp 2,3, December 2000.
routed and controlled appropriately, otherwise these [4] C.J. Henderson, B. Robertson, D. Gil Leyvaa, T.D.
will greatly reduce the performance of the system. Wilkinson, D.C. O’Brien, G. Faulkner Cambridge
Univeristy Engineering Department, Trumpington St,
Finally, SLM consists of a layer of nematic liquid- Cambridge, CB2 1PZ, UK Department of Engineering
crystal, but the nematic devices have a significantly Science, University of Oxford, Parks Road, Oxford, OX1
slower switching speed than ferroelectric devices. 3PJ, UK. “Control of a free space adaptive optical
In this preliminary study we have considered only chip- interconnect using a VCSEL array transmitter and FLC
to-chip interconnection. This proposed model of optical SLM for beam steering”; [email protected], pp 1.
interconnects can also be applied in board-to-board [5] Tod Laurvick, ”Chip-To-Chip Optical Interconnection
Using MEMS Mirrors”, March 2009.
communication.
[6] http://www.laserfocusworld.com/articles/255505.
[7] http://jp.hamamatsu.com/en/rd/publication/scientific_am
REFERENCES erican/common/pdf/scientific_0806.pdf.
[8] T. D. Wilkinson, C. D. Henderson, D. Gil Leyva and W.
[1] Muhsen Aljada and Kamal E. Alameh , Yong-Tak Lee A. Crossland, “Phase modulation with the next
and Il-Sug Chung, “High-speed (2.5 Gbps) generation of liquid crystal over silicon technology”,
reconfigurable inter-chip optical interconnects using Journal of Materials Chemistry, pp 1-4, June 2006.
opto-VLSI processors”, Centre for
MicroPhotonicSystems, Electron Science Research [9] Boulder Nonlinear Systems, “32-channel Spatial Light
Institute Joondalup, WA, 6027, Australia; pp 3-8, 2006 Modulator Driver Subsystem”, Boulder Nonlinear
Optical Society of America. Systems, pp 2, June 1, 2001.
 GA Based Neural Network Retraining
Using Adaptive Genetic Operations
C.I.Bauer; H.Yu; B.Boffey
Faculty of Computing, Engineering and Technology, Staffordshire University, Stafford, England
[email protected]; [email protected]; [email protected]

Abstract—Within cellular systems prediction has proven to

be a potential solution to enhancing the handover procedure to II. RELATED WORK
guarantee constant Quality of Service to mobile users. By using GAs have been investigated and applied to a large variety
historical route information, the future movement of mobile
of problems over the years due to their ability to search the
devices is predicted in advance with the aim to reserve
resources prior to arrival of the device in a new cell. However, solution space efficiently delivering optimal solutions to
as the traffic patterns of devices in this environment change problems to be solved. This ability has been exploited by
over time this needs to be taken into consideration when numerous authors through combining GAs with other
designing a prediction system. Some mechanism has to be approaches including NNs and applications found in
provided to address this issue. This paper presents a Genetic publications are manifold. The concept of NN retraining
Algorithm (GA) based retraining scheme using adaptive layer- within this is not entirely new and was also recognised
based genetic operations for a Neural Network (NN) based previously by authors in other subject areas, including NN-
movement prediction systems in a cellular environment to based prediction systems resulting in a variety of schemes
enhance system performance in the presence of changing
suggested for different scenarios. Schemes that fall into this
traffic patterns. Experimentation has shown that using an
adaptive layer-based GA approach can provide a significant category include the approach by Nastac and Matei [1], as
improvement to the predictive properties of the NN based referred to and detailed in [2] and [3], that proposes a NN
prediction system in the presence of changing movement retraining scheme based on scaling the reference weights by
patterns. a defined scaling factor. They can then be used as initial
weights for a new training cycle to improve the network
Index Terms—Cellular systems, genetic algorithms, accuracy and the adaptable NN model for recursive non-
handover, prediction, neural network, retraining linear traffic predictions of communication networks
proposed in [4] that is based on estimating the NN weights to
I. INTRODUCTION change them according to the current network conditions.
Genetic algorithms (GAs) have over the years been However, the majority of authors concentrate on using GAs
applied to a variety of problems with great success. One area for implementing feature selection algorithms to reduce the
of application is the use of GAs for initial training and amount of training data fed to the NN to shorten training
tailoring the weights of neural networks (NN) to a given times and enhance the network response as proposed in [5];
problem. Research in this field has clearly shown that the use for the initial training of the NN weights as discussed in [6]
of GAs can be beneficial, providing equally good results in and [7]; to identify a suitable network design for a given
the fraction of the time required by conventional NN training problem and periodically redesign the network structure and
techniques through backpropagation algorithm. Hence, they determine the network weights such as the load forecasting
hold the potential to widen the field of applications NNs can model [8] and the short-term traffic flow forecasting scheme
be applied to. Traditional applications of NNs are based on in [9] for intelligent transportation systems (ITS). Although
systems that are likely to encounter information with the there are a large number of GA-based NN schemes for a
same characteristics that are fed to the NN for processing. variety of problems, the possibility of using GAs to retrain a
During training of a NN this information is used to adjust the network with a fixed network structure to maintain
weights accordingly to produce the required response. While performance levels in the presence of changing conditions
in most NN based systems the characteristics of the
appears so far to have been mostly overlooked.
information are not likely to change much, retraining is
rarely a consideration. However, depending on the problem
to which a NN is applied, this may not necessarily be the III. GENETIC ALGORITHM RETRAINING SCHEME
case. This paper presents a GA based retraining scheme The GA based NN retraining scheme discussed in this
using layer-based adaptive mutation and crossover chapter is a continuation of the work discussed in [10] and
operations for a NN based movement prediction systems in a assumes that the NN based movement prediction systems
cellular environment to enhance system performance under introduced is the same publication is used to predict the
changing traffic patterns. Starting with a review of related future movement trajectories of mobile devices. For this
research work combining GAs and NNs, the following
prediction system the network is trained with mobile device
sections will provide details of the experimental results of a
route information that has been accumulated over time and
GA with layer-based adaptive mutation and crossover
operations for NN retraining to the given problem. reflects the devices regular movement patterns. Using this
for training enables the NN-based movement prediction

511
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_89, © Springer Science+Business Media B.V. 2010
512 BAUER ET AL.

system to predict future movement, through the information To create the initial generation to start the GA the
in the distribution probability matrix – maintained for each weight information of the trained and out-of-date NN was
cell within the network grid – and the previous movement used as a basis, as it can be argued that the ideal set of
step, with a certain level of accuracy. However, as the weights for the new traffic patterns are very close to the new
movement patterns of mobile devices change over time this optimal solution. Hence this holds the potential to further
will affect the prediction success rate of the system. Without reduce training times. To generate the individuals of the
retraining with new route information that reflects the initial population, each gene of the original set of weights
changed movement pattern of mobile devices the prediction was modified by a small, randomly selected value.
success rate will drop over time. As full retraining using The remaining parameters used for the GA were a
traditional training mechanisms such as the backpropagation mutation rate of 1/N (with N being the length of the
algorithm can be very time-consuming and inconvenient at chromosome), single point crossover with a crossover rate
times, GAs can offer an alternative solution to the problem, of 50% using the roulette wheel selection algorithm with
or at least provide a temporary measure to allow the system elitism as well as the population size of 180 individuals. The
to continue with acceptable performance levels until full number of individuals selected for the next generations was
retraining is possible. This algorithm could be employed set to 50% of the population size. These parameters were
once the prediction success rate has dropped below a established through initial simulations to optimise the GA
defined threshold to improve the prediction success rate. retraining approach. The simulations were run with 10, 20,
50 and 100 generations and the results for the parameter set
When applying GAs to a problem there are two main optimised for the retraining approach compared against the
requirements as defined in [11]: firstly, a suitable way to standard settings as defined by DeJong [12]. The changing
represent the problem/solution needs to be found. Secondly, traffic patterns were simulated through data sets with
in order to be able to evaluate the solutions of each varying levels of known regular route information ranging
generation, a fitness function needs to be defined that allows between 100%–50% in steps of 10% the NN had
for selecting the best solutions from each generation with encountered during training. The remaining route
which to continue the evolutionary process. In addition to information within the data sets consisted of unknown
this parameters such as population size, number of regular route information that had not been experienced
generations, how to implement the genetic operations and previously. The results of the simulations can be seen in the
apply them to the defined chromosomes as well as any diagram below which is showing the achieved fitness for the
additional parameters that arise due to the scenario-specific different number of generations averaged across the
implementation have to be identified. In the case of the investigated data sets for ease of representation:
given scenario using NNs in combination with GAs, two
parameters are indirectly given through the design of the 82
NN and the way its performance is evaluated. As the aim of
the system is to successfully predict the next visited cell of a 81
mobile device, the prediction success rate is used as the
Fitness in %

fitness function to evaluate individuals of a generation. In 80

addition to the fitness function, elitism was used to carry
across the best solution to the next generation without 79
alteration. To represent the problem, the weight information
provided through the trained network was encoded in a 78
chromosome as depicted in the diagram below. Dejong Roulette 10 Roulette 20 Roulette 50 Roulette 100

Fig. 2. Performance comparison

As can be seen from the diagram the results received for the
customised parameter set using 20, 50 and 100 generations
for varying levels of new unknown regular route
information within the data set clearly outperformed the
standard DeJong settings for Neural Network retraining.
However, although the simulation results showed that the
GA retrained network can improve the fitness value
compared to the original network without retraining, the
Fig. 1. Conversion of NN weight information into chromosome results obtained were lower than expected. Therefore, the
characteristics of the provided route information were
This representation was chosen because the vital investigated further to identify a reason for this. The weight
information of a trained network is stored in the weights for information of networks trained with route information of
the connections between the different nodes and these are 100%, 95% and 90% of regular known routes was
the parameters that determine the prediction success rate for investigated. It was observed that with each progressing
a given traffic pattern. When the traffic patterns change this layer of hidden nodes, the weight difference of the network
is the information that needs updating to maintain prediction trained with 100% regular routes compared to 95% and 90%
performance levels. as well as the network trained with 95% regular routes
 GA BASED NEURAL NETWORK RETRAINING USING ADAPTIVE GENETIC OPERATIONS 513

compared to 90%, increased, as depicted in the diagram As in the case of the non-adaptive GA investigation, the
below: parameters for the adaptive implementation of GA for the
purpose of neural network retraining, were tested overall to
18 allow comparison and performance analysis. With the
Layer 3
16 exception of the parameters for the adaptive elements
Layer 2
14 identified through the preliminary investigation, all
Layer 1
12 remaining parameters remained the same as in the case of
Difference

10 the non-adaptive GA approach. For performance

8 comparison the adaptive GA implementation was run
6 against the standard DeJong settings as well as the non-
4 adaptive implementation investigated in the previous
2 section. In order to get a good response and representative
0 results for the comparison run with the standard DeJong
95 to 100 90 to 95 90 to 100 Average settings different intermediate steps were experimented with
Comparison
using 10, 20, 50 and 100 generations as previously. The
Fig. 3. Comparison of NN weight information simulations runs were repeated three times for both the non-
adaptive GA parameters and the DeJong parameters for the
This suggests that when performing the genetic different training data sets ranging from 50% – 100% in
operations on the chromosomes it may be beneficial to steps of 10%. The results were then compared against the
distinguish between the different layers of hidden nodes adaptive approach using an adaptive mutation and crossover
contained within them. For this reason the application of the operator as part of the scheme. The results of the simulation
mutation and crossover operations were reconsidered to take runs for the different data sets and training data sets are
into account the established pattern to investigate the effect depicted below showing the performance in the same format
of this on the GA results. as previously presented:
The mutation operation was hence adjusted to reflect
this by varying the amount the value of a gene is altered 84
when mutation is applied. In the experiment – if mutation 83
was applied to a gene of the chromosome – a random value
82
ranging between 0 and 1 was multiplied by a multiplier
Fitness in %

value and then randomly added or subtracted after being 81

scaled based on the hidden layer the gene belonged to. To 80

investigate different multiplier values and identify the most
79
suitable one this value was ranged between 1 and 200. The
investigation showed that a value of 30 for the multiplier 78
Adaptive Dejong Roulette 10 Roulette 20 Roulette 50 Roulette 100
produced the best results.
The crossover operation was also altered to take into Fig. 5. Performance comparison with adaptive operations
consideration that within a chromosome the weight
information (genes) is arranged per network layer and per As can be seen from the graph, the layer-based adaptive
node within each network layer. Hence the crossover mutation and crossover have provided significant
operator was applied differently. Only genes within a improvements to the achieved fitness. This indicates that
randomly selected node of a particular layer after a there is a clear benefit to taking into consideration the
randomly selected point within that node were crossed over structure of the weight information contained within a
as depicted below. To find the optimal crossover rate to chromosome when applying the mutation operation. This
perform the adaptive crossover operation were conducted, can also clearly be seen in the diagram below which shows
which identified a rate of 20 as the best value. the degradation of the prediction success rate of a trained
network in the presence of changing traffic patterns and the
prediction success rate of the network after retraining using
GA based NN retraining with adaptive operations.

100
Degradation over time
90 Adaptive Retraining
Fitness in %

80

70

60

50
100 90 80 70 60 50
Fig. 4. Chromosome with weights encoded per network node Percent of Known Regular Route Information

Fig. 6. Comparison adaptive GA retraining versus natural degradation

514 BAUER ET AL.

As can be seen in figure 6, the adaptive GA retraining

1000
algorithm was able to significantly improve the fitness GA
achieved, compared to the fitness values of the simulation 800 Neural Network

Time in Seconds
run without retraining. As expected the improvement is more 600
significant for data sets with a higher percentage of new
400
unknown regular route information as it is for data sets with
a lower percentage of new unknown regular route 200
information. A maximum of over 15% for the data set with 0
50% known route information was established. This is 50 60 70 80 90 100
confirmation that using an adaptive GA retraining scheme is Percent of Known Regular Route Information
a viable solution to provide temporary relief to NN-based Fig. 9. Comparison Training Times Neural Network versus GA
prediction systems in the presence of changing traffic
patterns until full retraining is possible. Looking at the time measurements for the two
retraining techniques in figure 9 the Backpropagation
approach shows steadily increasing time requirements with
As the improvement in the prediction success rate is
a rising percentage of unknown route information within the
only one of the deciding factors on the success of the GA
data sets. For the GA approach there also seemed to be
based NN retraining approach for the given problem further
rising time requirements for data sets with higher levels of
studies need to be conducted. To investigate whether the use
known route information which, after the first few data sets,
of GAs for retraining would provide a time benefit for the
however, seemed to level off. Overall seen, the retraining
prediction scheme, time measurements were taken and
time requirements of the GA approach were well below
compared to the prediction success rates. As a comparison
those of the Backpropagation approach clearly confirming
algorithm the original Backpropagation approach was used.
the time benefit of the GA scheme. This is down to the fact
Subject to the investigation were data sets with varying
that the GA algorithm in previous experiments seemed to
levels of known regular route information ranging from 50 –
converge after a relatively small number of generations as
100% in steps of 10%. To obtain representative time
can be see in figure 7 which shows the best and the average
measurements three simulation runs were conducted and the
population score for an example run of the GA.
results averaged across the three runs per data sets. Obtained
simulation results are shown below:
Considering the results of both graphs it can be concluded
that looking at the prediction success rate alone, the
90 Backpropagation algorithm appears to be the better choice
85 of retraining the network. However, considering the original
80 aim of the GA retraining, this approach has a clear
advantage on the training times. As the GA retraining is
Fitness in %

75
70 only meant to be used as a temporary measure to extend the
65 life of the Neural Network until full retraining is possible
60 again this proves that this is a viable way to achieve this. In
55 light of the obtained simulation results, taking into
50 consideration the time saving and the improved prediction
1 11 21 31 41 51 61 71 81 91 101 111 121 131 141 151 161 171 181 191
Number of Generations
success rate, it can further be concluded that when using the
GA approach for retraining to set a threshold value to a
Fig. 7. GA Training Cycle
suitable value to benefit from the better prediction success
rate achieved by the GA algorithm.
100

90
IV. CONCLUSION
Fitness in %

80
In this paper a GA-based retraining algorithm using
70 adaptive mutation and crossover to assist the NN based
60 NN success rate prediction systems in maintaining its performance levels in
50
GA Success Rate the presence of changing traffic patterns is presented. As
50 60 70 80 90 100 simulation work has shown, the adaptive layer-based
Percent of Known Regular Route Information implementation of the genetic operations provided
Fig. 8. Comparison NN versus GA prediction success rate improvements to the system’s capabilities to increase the
prediction success rate (fitness) compared to the standard
mutation genetic operations. This is a strong indication that
when applying genetic operations, the structure of the
weight information contained within a chromosome should
be taken into consideration.
 GA BASED NEURAL NETWORK RETRAINING USING ADAPTIVE GENETIC OPERATIONS 515

Further to the investigation of enhancing the prediction [4] Doulamis, A.D.; Doulamis, N.D.; Kollias, S.D., “An
Adaptable Neural-Network Model for Recursive Nonlinear
success rate through adaptive genetic operations when Prediction and Modeling of MPEG Video Sources”; IEEE
retraining a NN, time measurements were taken to compare Transaction on Neural Networks; Volume 14, Issue 1; January
the traditional Backpropagation algorithm and the GA based 2003 Page(s): 150 – 166.
retraining algorithm with respect to retraining time and [5] Ichibuchi, H.; Nakashima, T.; Nii, M., “Learning of Neural
achieved prediction success rate. Simulation showed that Networks with GA-Based Instance Selection”; Joint 9th IFSA
World Congress and 20th NAFIPS International Conference,
looking at the prediction success rate the Backpropagation 2001; Volume 4, 25-28 July 2001; Page(s): 2102 - 2107 vol.4.
algorithm clearly outperformed the GA retraining approach. [6] Ling, S.H.; Lam, H.K.; Leung, F.H.F.; Tam, P.K.S., “A
However, with respect to retraining times, the GA retraining Genetic Algorithm Based Neural-Tuned Neural Network”;
approach showed a clear time benefit over the traditional IECON ’03; The 29th Annual Conference of the IEEE –
retraining approach, although not achieving the same Industrial Electronics Society, 2003; Volume 3, 2-6 Nov.
2003 Page(s): 2423 – 2428 Vol.3.
accuracy levels as the Backpropagation approach. As the
[7] Zuo, Guoyu; Liu, Wenju; Ruan, Xiaogang, “Genetic
adaptive GA based NN retraining approach is intended as a Algorithm Based RBF Neural Network for Voice
temporary measure to bridge the time until full retraining is Conversion”; WCICA 2004; Fifth World Congress on
possible, again this proves that this is a viable way to Intelligent Control and Automation, 2004; Volume 5, 15-19
June 2004 Page(s): 4215 – 4218 Vol.5.
achieve this.
[8] Zhangang, Yang; Yanbo, Che; Cheng, K.W. Eric, “Genetic
Algorithm-Based RBF Neural Network Load Forecasting
Model”; IEEE Power Engineering Society General Meeting,
REFERENCES 2007; 24-28 June 2007 Page(s): 1 – 6.
[1] Nastac, I.; Matei, R. “Fast retraining of artificial neural [9] Ji, Tao; Pang, Qingle; Liu, Xinyun, “Study of Traffic Flow
networks”; Rough Sets, Fuzzy Sets, Data Mining and Forecasting Based on Genetic Neural Network”; Sixth
Granular Computing, Springer-Verlag in the series of Lecture International Conference on Intelligent Systems Design and
Notes in Artificial Intelligence; 2003; pp. 458.462; ISBN: Applications – ISDA ’06; Volume 1, Oct. 2006; Page(s): 960
978-3-540-14040-5. – 965.
[2] Nastac, I.; Costea, A. “A Retraining Neural Network [10] Bauer, C.I.; Yu, H.; Boffey, B. “Adaptive Genetic Algorithm
Technique for Glass Manufacturing Data Forecasting; for Neural Network Retraining”; International Joint
International Joint Conference on Neural Networks”, 2004; Conferences on Computer, Information, and System Sciences
Proceedings 2004; Volume 4; 25-29 July 2004 Page(s): 2753 – CISSE ’08; 5-13 December 2008
– 2758. [11] Núñez, Edwin; Banks, Edwin Roger; Agarwal, Paul, “High
[3] Nastac, I.; Cristea, P., “ANN Flexible Forecasting for the Performance Evolutionary Computing”; HPCMP Users Group
Adaptive Monitoring of Multi-Tube Reactor”; 6th EURASIP Conference 2006, June 2006; Page(s): 354 - 359.
Conference focused on Speech and Image Processing, [12] DeJong, K.A. and Spears, W.M. "An Analysis of the
Multimedia Communications and Services; 14th International Interacting Roles of Population Size and Crossover in Genetic
Workshop on Systems, Signals and Image Processing 2007; Algorithms," Proc. First Workshop Parallel Problem Solving
27-30 June 2007; Page(s): 193 – 196. from Nature, Springer-Verlag, Berlin, 1990. pp. 38 - 47.
Decision Support for IP Uplink and Routing Choice
A Whole View of the Internet

Peng Su Zhengping Wu
Department of Computer Science and Engineering Department of Computer Science and Engineering
University of Bridgeport University of Bridgeport
221 University Avenue, CT 06604, USA 221 University Avenue, CT 06604, USA
[email protected] [email protected]

Abstract—Although Border Gateway Protocol (BGP)

is broadly used among autonomous systems (ASes), the II PROBLEMS CAUSED BY POLICIES
topology of peer autonomous systems is often mysterious Traditionally, routing policies between autonomous
to some Internet Service Providers (ISPs) or systems are designed following local autonomous system’s
administrative domains. With unknown routing policies interests and not opened to other External Border Gateway
of BGP neighbors and their uplink ASes, it is uneasy to Protocol (EBGP) neighbors. Therefore, one BGP customer or
make a proper design when building a multihoming one ISP may be blind to the area out of their own AS and
environment. Sometimes it may even cause congestion neighboring ASes. After balancing interests and needs, one
problems. Some mistakenly announced prefixes of ASes institution may believe that some “desired” uplinks could
over the Internet will continuously increase routing table become its major outgoing and incoming interfaces because
and deteriorate policy-based routing. This paper analyzes that might work better. Other uplinks will only be used as
the adjacency relationships of ASes over the entire backup links. However, this is not always true. For example,
Internet and their announced prefixes. Using this from the view of an AS, one BGP user may know little about
information, our prototype system can automatically adjacency relationships out of the local domain, this BGP user
identify the more appropriate access point for individual may unwittingly suffer from packet loss and delay problems,
ASes, depict the relations among ASes, and provide when congestion occurred between its uplink AS and the
decision support to manage the improper address spaces destination AS. Figure1, 2, and 3 together explain this issue.
over the Internet. Here, Figure1 illustrates the normal situation of packet
Keywords-BGP; autonomous system; routing policy; forwarding between AS1 and C1 (C1 is a customer needs
adjacency relationship, topology uplinks). C1 chooses the link to AS2 as its default route to
AS1 and the Internet, the link to AS3 as its backup link.
However, when some congestion happens between AS 1 and
I INTRODUCTION AS2 (as illustrated in figure 2) the packets to C1 will be
As described in [1] and [8], the Border Gateway Protocol partially dropped. Especially, when the traffic load between
is an inter-autonomous-system routing protocol used between C1 and AS 2 is light, C1 will not switch its main link to AS3
Internet Service Providers, domains, regions, nations and etc. automatically and thus suffers from this remote congestion.
BGP exchanges reachability messages between neighbors or This situation will be more often and more complicated when
peers, maintains three information databases such as Adj-RIB- there is more than one AS between AS2 and AS1 (as
In, Adj-RIB-Out, Loc-RIB, as well as provides the routes that illustrated in figure 3).
BGP has selected using local routing policies. Furthermore,
this protocol supports Classless Inter-Domain Routing (CIDR
[9]) and Variable Length Subnet Mask (VLSM [10]), which
makes BGP peers free to aggregate IP addresses to an
appropriate scale and easy to advertise proper IP prefixes not
limited by classes.
A routing policy is a set of rules that regulates the traffic
as well as adjacency relationships between routers and other
hosts in term of route exchanges and protocol interactions.
Summarily, in a routing policy [6], a customer can define:
z A collection of prefixes the router will receive or not
from others.
z A collection of prefixes the router will advertise or
not to others.
z Redistribution of prefixes between various routing
protocols.
These three items can be defined in incoming policies, Figure1. Packet forwarding in normal condition with blindness to the
outgoing policies, route-maps and access lists. topology of uplink AS

517
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_90, © Springer Science+Business Media B.V. 2010
518 SU AND WU

ASes, because some ISPs not only aggregated the address

spaces, but also filtered the detecting packets on their
equipments. As illustrated in figure 4, C1 needs to send
packets from AS2 to AS1 where C1 chooses AS2 as its main
uplink, and AS3 as its backup link. But AS1 is imperceptible
to C1, so packets between AS1 and C1 will go through the
path C1-AS2-AS1, and AS1-AS2-AS3-C1. It is clear that
packets are affected by routing policies and suffer from
unnecessary hops. Besides, if there exist one or more
autonomous systems between AS2 and AS3, the forwarding
action will become more complicated with more uncertainty.
At the same time, packet loss, loop, and long delay of data
will appear at C1 with this design.
However, things will change when the customer, C1,
knows the topology outside its local area. The customer, C1,
can use AS3 as their major uplink, whereas the connection to
AS2 as a backup. Of course, many customers may not be
Figure 2. Packet forwarding in congestion condition with blindness to aware of the importance of topology outside their local AS.
the topology of uplink AS But this topology is helpful, sometimes necessary. Moreover,
nowadays many BGP routing tables have already reached
300,000 entries, which require extra hardware recourses to
support effective and efficient routing. This situation may be
caused by following reasons:
˄1˅ The increasing needs of IP addresses
˄2˅ Announced overlapping prefixes
˄3˅ Some announced prefixes whose network masks
are more than 24
Along with the requirements of network expansion,
routing entries in a router’s memory will grow. However, the
situation (1) and (2) can be avoided when some ISPs and
domains pay more attention on the whole view of the Internet.
For example:
Situation (1)
route-views.oregon-ix.net> show ip bgp 123.252.128.0/18
BGP routing table entry for 123.252.128.0/18,
Paths: 2905 702 6453 4755 17762
Situation (2)
route-views.oregon-ix.net> show ip bgp 123.252.129.0/24
Figure 3. Packet forwarding in congestion condition (multi-hop) with BGP routing table entry for 123.252.129.0/24
blindness to the topology of uplink AS Paths: 2905 702 6453 4755 17762
These two examples show one overlapping prefix
announced by AS 17762. Although this method may lead to
traffic redundancy, the increasing number of BGP routing
table entries makes convergence [7] time rise continuously.
Situation (3)
route-views.oregon-ix.net> show ip bgp 199.77.194.254
BGP routing table entry for 199.77.194.254/32
Paths: 3333 1103 20965 6509 10764 19401 10490
In this situation (3), one prefix whose network mask is 32
emerges in the BGP table. This situation may be caused by
some configuration errors or specific needs. Nevertheless, it
should be avoided whatever the considerations are. Therefore,
it is unwise to show every autonomous system’s route entries
on a router and check them one by one. Analyzing BGP
entries, making a statistics from all ASes, and then checking
route entries as well as making policies following the
statistics are more feasible and practical. The analysis result
Figure 4. Packet forwarding impacted by routing policy and blindness from the statistics is called the whole view of the Internet.
to the topology of uplink AS
III ACCESS CHOICE TO THE INTERNET
In another example, when the forwarding action is Access choice to the Internet can be reached through the
influenced by routing policies or routing control mechanisms database recording the whole view of the Internet, which
[3], there will be unnecessary hops between one AS and its contains two tables. The table “AS” collects the neighborhood
destination. Generally, it is difficult for a customer to detect of ASes and the table “ROUTES_BY_AS” stores the prefixes
and improve the quality of packet transferring in its uplink announced by each AS. Table “AS” consists of three fields,
 DECISION SUPPORT FOR IP UPLINK AND ROUTING CHOICE 519

AS_ID, Nei_num, and Nei_as, where AS_ID includes the AS Input

numbers over the entire Internet, Nei_num is the number of Whole BGP routing tables from 38 EBGP peers
EBGP peers, and Nei_as field describes the EBGP neighbors
of the autonomous system in AS_ID. Also, in the table of
Output
“ROUTES_BY_AS,” the fields of “Route_number” and
Database BGP, table “AS”
“Route_entries” describe the route entry number as well as the
content of each prefix. Like table “AS,” this table contains
one field named “AS_ID” too, which acts as the index for this Phase 1: regulate strings
table.
Read routing table into memory;
Each line of table “AS” contains binary relations of While (i<= some certain number)
autonomous systems. R=(X, Y, G(R)) represents a relation {if (!resource is at the end)
between X and Y, in which X consists of AS number {/* Because the amount of resource is huge,
identified in AS_ID field, Y consists of the AS number(s) of the certain number is one part of the resource.
the EBGP neighbors peering with the AS in AS_ID field, and */
G(R) means the graph of R. Equally, R ҧ X × Y = {<X, Y> |
truncate the route entries, and put them into
XęAS_ID, YęNei_as}. For a better way to access the
string array;
Internet, we can assume R’=(Y, Z, G (R’)), and R’ ҧ Y × Z
= {<Y, Z> | YęAS_ID, ZęNei_as}. Here, Y may contain }
one or many neighboring AS numbers. Simultaneously, in else break;
the lines indexed by AS_ID of Y, each element of Y can also i++;
have one EBGP peer or several EBGP peers. The collected }
EBGP peers or peers of every element of Y is Z. If putting Return string array in the certain pattern;
this action forward, the recursion will continue until the last
binary relation contains a proper Internet Exchange Point Phase 2: analyze strings and update database
(IXP) or local area IXP. Then the shortest path of ASes to
some IXP may be a better choice to access the Internet. For Connect database
example, X= {15430}, Y= {34, 981}, then R is the Cartesian while (! the string array is at the end)
product of X and Y. R= {<X, Y> | Xę (15430), Yę (34, {
981)} = {(15430, 34), (15430, 981)}. R’Y=34= {<Y34, ZY=34> | if (! pointer of one element is at the end)
Yę (34), ZY=34ę (2158, 7018)} = {(34, 2158), (34, 7018)}. {
R’Y=981= {<Y981, ZY=981> | Yę (981), ZY=981ę (36, 1209)} = Check and remove overlapped ones;
{(981, 36), (981, 1209)}. In this example, AS 7018, AT&T if (this AS number_n exists in table "AS")
WorldNet Services, could be a proper IXP, therefore, AS {
15430 may choose AS 34 as its main link and another one as locate the line, check to see whether it
the backup link. contains this one (AS number)
Each line of table “ROUTES_BY_AS” is indexed by before/behind AS number_n;
AS_ID, and includes the prefixes originated from this AS. if (exist)
Let Mi = {i.prefix1, i.prefix2 … | i = AS_ID}, M’ ‫ ؿ‬Mi, Nj =
continue;
{j.prefix1, j.prefix2 … | j = AS_ID}, and N’ ‫ ؿ‬Ni. Then M’
else
¤N’ = .
{
update this line and continue;
IV A WHOLE VIEW OF THE INTERNET }
}
The construction of a whole view of the Internet is
else
discussed in this section.
{
A. Algorithm description insert one line, and update this line;
The data source, containing 38 BGP peers, reaches 9 }
million lines of route entries. To discover the adjacency }
relationships and prefix information, these two following else
algorithms are designed to analyze all BGP routing entries, {
regulate them, form topology among ASes, and put each read next string from the array;
prefix’s information into database by each AS number. Here, }
the algorithm for analyzing AS adjacency relationships is }
named “Analysis Algorithm,” as illustrated in figure 5. The return the disposed string array;
major goals are: Close database connection;
˄1˅To filter peer relationships of ASes and remove
unhelpful messages, like the metric and weight. Figure 5. Analysis Algorithm for regulating and analyzing BGP routes

˄2˅To get rid of the repeated AS number generated by The second algorithm is designed to regulate and classify
AS prepending (append AS_NUM to AS path), the BGP prefixes. It is named “Class Algorithm,” as
search the database by AS number, and update the illustrated in figure 6. The tasks of this set of programs can
database. be described as:
520 SU AND WU

Autonomous systems belonged to this area mainly contain

Input
backbone IXP, such as 701, 7018, 3356, and etc.
Whole BGP routing tables from 38 EBGP peers

Output
Database BGP, table "ROUTES_BY_AS"

Phase 1: regulate strings

Read prefixes into memory;
while (i<= some certain number)
{
if (!resource is at the end)
{
truncate entries to the beginning of prefixes;
}
else
break;
i++;
}
Load lines into array and return prefix array; Figure 7. Distribution of Autonomous Systems
whose EBGP peers are more than 100
Phase 2: classify route entries by AS number and update
database When comparing to the total number of ASes, the
numbers of autonomous systems go up as their EBGP
Connect database;
neighbors fall. For example, stub ASes take more than half of
while (!prefix array meets the end)
the total AS entities. Its percentage reaches 57.27%. In the
{ view of autonomous systems, these stub areas do not respond
read one line from prefix array; to data relay, while those ASes that have over 10 EBGP peers
if (this AS number exists in table ) are mainly responsible for the connectivity of other areas,
{ though they take a comparatively small part, 2.67%, over the
update database by increasing the route entire Internet.
entries number and adding the prefix in this
line; C. The statistics of prefixes provided by database
} When analyzing the prefixes originated from different
else autonomous systems, the algorithm can also handle irregular
{ address spaces on the Internet. The common characteristic of
insert one new line in table " these address spaces is their masks are over 24. For example,
ROUTES_BY_AS"; the number of prefixes whose net masks are 32 reaches 20
update database by adding the prefix in this and the number of those entries whose net masks are 25 is
line; 110. Although these entries could be used among
} autonomous systems, it is unwise to receive and advertise
} them in routing policies. Therefore, these “bugs” are entirely
and explicitly examined by the algorithm, which can provide
Figure 6. Class Algorithm for regulating and classifying BGP routes policy decision for routing control.

B. The statistics of AS adjacency relationship V DECISION SUPPORT SYSTEM

Based on the analysis result, it is not difficult to make a A. Decision of access
statistics of AS adjacency relationships and their scale. The As mentioned above, traditionally when constructing a
following figures, tables, and data are used to illustrate the network by BGP, the top factors in consideration [5] are
current status of adjacency relationships among autonomous scalability, stability, robustness [2], and business interests.
systems, their topology, and the characteristics of the entire However, the whole adjacency relationships of uplink ASes
Internet. could easily be neglected, not only because some designers
Here, there are two data granularities on figure7 in that the have not recognized the importance of neighborhood outside
number of ASes with more than 500 EBGP peers is much their uplink ASes, but also these areas are blind to network
less than that with 100 to 500 EBGP peers. At the same time, designers. This paper tries to demonstrate the significance of
these ASes whose EBGP neighbors range from 100 to 500, knowing the whole adjacency relationships of ASes over the
are much less than the number of ASes with less than 100 entire Internet, including local uplink ASes. By the database
EBGP peers. Figure 7 illustrates the relationship between the of the whole view of the Internet, some designers can depict
number of ASes and their EBGP peers. X axis is the number the topology of their outgoing interface, discover which ones
of EBGP peers, and Y axis is the number of ASes. are appropriate IXP ASes, and avoid remote congestions as
As the figure 7 shows, the number of AS entities whose well as redundant hops impacted by routing policies. Suppose
EBGP neighbors are more than 500 is only 11, and the the situation of figure3 has occurred and AS1 is one
number of AS entities whose EBGP peers are more than backbone IXP. C1 does not know the topology of other ASes
1000 is 5. Comparing with the total number of autonomous either because of receiving aggregated routes or default route
systems, 30583, these 11 AS entities count for only 0.036%.
 DECISION SUPPORT FOR IP UPLINK AND ROUTING CHOICE 521

configurations [8]. When C1 sets its main link to AS2 and issue. As a result, the customer stub ASes suffer from
undergoes some remote congestion or oscillations [4], this them.
˄2˅Even though one stub AS uses more than two links to
AS will not automatically switch its link to AS3. However, if
the uplink AS, these links functions only load balance
putting this situation to the support decision system, the fruits
will be like this: [12] and failover, which will not be helpful to the
previous issues.
Key = AS1 ˄3˅In many cases, the traffic between two stub ASes is
Neighbor number = n huge, which hold the most bandwidth of these areas.
Neighbor information of AS1: <AS1, ASn>, <AS1, AS3>, However, these packets have to go through the hub
<AS1, ASj> … <AS1, ASk> area and affected by traffic policies and unexpected
routing questions.
Key=C1 The first and second questions may occur when some
Neighbor number = 2 uplink ASes (maybe some ISPs or domains) conceal the
Neighbor information of C1: <C1, AS3>, <C1, AS2> topology by summary action or stub areas only use default
route. Then these stub ASes are blind to the topology outside
Key=AS2 local area. Nevertheless, in our decision support system, the
Neighbor number=2 adjacency relationships of stub ASes and other IXP are
Neighbor information of AS2: <AS2, C1>, <AS2, ASn> obvious. Thus, as described in the previous section, stub
ASes can get the nearest IXP information, at least in the
Key=AS3 aspect of AS-PATH, and then make a proper decision for the
Neighbor number=2 network design.
Neighbor information of AS3: <AS3, C1>, <AS3, AS1> Nowadays, it is pretty common for the third question over
the Internet. Consequently, traffic has to endure unnecessary
Key=ASn hops between stub areas. Figure 8 and 10 illustrate this issue.
Neighbor number=2 Figure 9 and 11 provide a solution.
Neighbor information of AS3: <ASn, AS1>, <ASn, AS2>
; ;
Let ‫׊‬A and ‫׊‬B ‫{ א‬Neighbor information of C1, AS2,
AS3, ASn}, suppose that x A ¤ B if and only if x A and
x B, and let M = A ¤ B, x=AS1, then (xęM) Æ (‫ ׊‬ęǡ
šę) Æ AS1 ę Neighbor information of AS3 OR AS1 ę
Neighbor information of ASn. < = < =
From the calculation of ASes’ adjacency relationships, it is
obvious that the congestion may occurred either in <AS1, Figure 8. Unnecessary hops between Figure 9. Refined network
ASn> or in <AS2, ASn>. Furthermore, AS3 is closer to AS1, two stub ASes and one hub AS architecture between stub ASes
from the AS-PATH aspect. Thus, changing main link of C1
will solve the problem. For another question, as illustrated in Figure 8 presents a simple “hub and spoke” topology,
figure 4, although C1 may not be sensitive to the impacts of where X is the hub AS, and stub area Y as well as Z work as
AS2’s routing policies, after checking the topology between spoke. When X forwards the packets from Y to Z, the hops
source and destination, C1 still can avoid the redundant hops between Y and Z is 2. However, when the decision support
by switch its main link to its backup link. Therefore, by system shows that AS Y and AS Z connect to the same hub
introducing the decision support system using the whole view AS - AS X, a more appropriate solution emerges, that is to
of the Internet, these two previous issues can be resolved. add a link between Y and Z, as shown in figure 9.
Besides, in the view of IXP, there will be similar issues to This situation could be more complicate when introducing
AS1, as described in figure 3 and figure 4. While the more hub ASes into this scenario, as figure 10 shows.
difference between C1 and AS1, in this scenario, is that the
situations of IXP would be more complex and more difficult ; 1 ˞
to deal with. However, following the normal forms explained
in the previous example, the adjacency relationships, EBGP
peer numbers, route entry scales make the relationships
among neighbors and the entire Internet topology clearer.
Moreover, when take the ranks of ASes, the stub areas, and < = ˟ ˠ
stability of ASes into account, this decision system is able to
provide more useful ranking information to make decisions. Figure 10. Unnecessary hops between two stub ASes and various hub
Let H(i, j) defines the hops between two different ASes.
B. The future of stub AS Suppose the traffic between AS Y and AS  holds a great
According to the discovery of whole adjacency amount of each AS. When Y needs to access , then the least
relationships of ASes over the Internet, the percentage of stub hops between from Y to  are:
AS over the Internet reaches 57.27%, so that the star
topology is still dominant and prevalent in the Internet. But
as a reality, this is an unstable structure, for below reasons:
˄1˅Many stub ASes use only a default route to other Therefore, each hop along the road from Y to  could
ASes, whereas their unique uplink ASes could arise influence the quality of transmission. As mentioned above,
congestion, redundant hops, as well as packet loss after knowing the adjacency relationships between Y and ,
some solutions, as figure 11 shows, could be either
522 SU AND WU

connecting to the AS  to which  links or directly accessing VII CONCLUSION

 by adding a new neighbor to . At the beginning of this paper, some common issues from
; 1 ˞ Internet Service Providers and domains are examined.
Because of the common use for prefixes aggregation, security
mechanisms and default route configurations, some BGP
users are blind to the entire topology outside their
autonomous systems and the prefix’s appropriateness on the
< = ˟ ˠ entire Internet, which make problems above unsolvable.
Therefore, there is an idea to build the whole view of the
Figure 11. Refined network architecture between stub Internet, including adjacency relationships among ASes and
ASes and various hub ASes prefixes announced from them. Then two algorithms to
discover the adjacency relationships and the announced route
Apparently, in both situation showed in figure 9 and 11, entries are explained. After that, a decision support system
hops are greatly reduced to 1or 2, and traffic between source based on the discovery is illustrated in the part V, which
and destination is free from routing policies, damping issues, focuses on researching the adjacency relations among ASes,
broadcast flood, and convergence questions. designing a new and more efficient network, and improving
the current quality of service. Then the application of this
C. Building a system free from improper address space decision support system on actual network design is
Since a lot of overlapping and irregular prefixes existing illustrated, which includes designing secure policies, uplink
on the Internet, some convergence problems and security AS choices for Internet Service Providers and stub AS
issues may increase, although some of them exist for backup optimization.
functions. Especially, some domains have to receive these
route prefixes for the reason that it is impossible to check REFERENCES
every prefix redundant condition and their lengths on a [1] Y. Rekhter and T. Li. A Border Gateway Protocol 4 (BGP-4). RFC
device. Therefore, by checking with our decision support 4271, “The Internet Society,” 2006
system, the overlapping address spaces and improper address [2] K. Sriram, O. Borchert, O. Kim, and P. Gleichmann, and D.
will be on the desk. Thus a method to make routing policies Montgomery, “A Comparative Analysis of BGP Anomaly Detection
and improve devices’ performance is provided. and Robustness Algorithms,” Proceedings of the Cybersecurity
Applications and Technology Conference for Homeland Security
(CATCH), Washington D.C., pp. 30-36, March 3-4, 2009
VI DISCUSSION
[3] Osamu Akashia, Kensuke Fukudab, Toshio Hirotsuc and Toshiharu
Fortunately, there are several BGP analysis tools available Sugawarad, “Policy-based BGP-control architecture for inter-AS
for researchers, network designers and Inter Service routing adjustment,” Computer Communications Volume 31, Issue 13,
Providers. Most used systems among these open software are 15 August 2008, Pages 2996-3002, pp. 2997-3001, 2008
“BGPath” [13], “AS ranking” [14], and “BGP inspect” [15]. [4] Alex Fabrikant, Christos H. Papadimitriou, “The complexity of game
dynamics: BGP oscillations, sink equlibria, and beyond,” Symposium
These tools have their special focus individually. on Discrete Algorithms Proceedings of the nineteenth annual ACM-
As for “BGP inspect,” this service provides some main SIAM symposium on Discrete algorithms, pp.5-7, 2008
IXP’s newly updated routing entries, such as ATT, Sprint, [5] Yi Wang, Michael Schapira, Jennifer Rexford, “Neighbor-specific
UUNET and so on. For instance, customers can submit their BGP: more flexible routing policies while improving global stability,”
queries like “Global Summary Queries” according various Proceedings of the eleventh international joint conference on
IXP, and obtain the corresponding results. Of course, it is Measurement and modeling of computer systems, pp. 224-227, June
2009
very useful for the network designers and researchers to track
[6] Randy Zhang, Micah Bartell, “BGP Design and Implementation,”
those new prefixes and making an appropriate routing policy, Cisco Press, March 2003
but when designing a network, researching the topology of
[7] S. Murphy, “BGP Security Vulnerabilities Analysis,” RFC 4272,
ASes or ranking the ASes, it cannot help.
January 2006
“AS ranking” is a supportive tool that can sort autonomous
[8] D. McPherson, “Experience with the BGP-4 Protocol,” RFC 4277,
systems based on AS number, name, country, ISP Cisco Systems, January 2006
information. However, this software can only provide [9] V. Fuller, “Classless Inter-Domain Routing (CIDR),” RFC1519,
ranking utility without a topology view of the entire Internet. September 1993
Besides, it is slow, 10 minutes for producing top 50 AS [10] T. Pummill, “Variable Length Subnet Table For IPv4,” RFC1878,
rankings. December 1995
From the aspect of topology view, the “BGPath” is the [11] Bruce Eckel, “Thinking in Java, 4th edition,” Prentice Hall, February,
popular software to show some part of the adjacencies by the pp 157-566, 2006
customers’ instructions. Comparing with “BGPath,” the [12] X Liu, L Xiao, "Inbound Traffic Load Balancing in BGP Multi-homed
decision support system illustrated in this paper depicts a Stub Networks", The 28th International Conference on Distributed
Computing Systems, pp 369-372, 2008
whole topology over the Internet and routing entries of each
[13] A. Campisano, L. Cittadini, G. Di Battista, T. Refice, C. Sasso.
AS, not some part of the current network. Besides, some “Tracking Back the Root Cause of a Path Change in Interdomain
subject autonomous systems, such as stub areas cannot be Routing”. IEEE/IFIP Network Operations & Management Symposium,
included on “BGPath” interface. Of course, it is neat, but not 2008
enough. Hence, a complete and accurate decision system [14] Xenofontas Dimitropoulos, Dmitri Krioukov, Bradley Huffaker, kc
based on the relations of total autonomous systems, their claffy, George Riley, “Inferring AS Relationships: Dead End or
routing prefixes as well as the statistics is required. Lively Beginning,” Networking and Internet Architecture, 2005
[15] D. Blazakis, M. Karir, and J. S. Baras, “BGP-Inspect - Extracting
Information from Raw BGP Data,” 10th IEEE/IFIP Network
Operations and Management Symposium, Vancouver, Canada, April
3-7, 2006.
 Differential Cryptanalysis of Blow-CAST-Fish
Xiaoling Sun1, Meiqin Wang2
1. Institute of Disaster Prevention Science and Technology, Yanjiao, Hebei, 065200
2. Key Laboratory of Cryptologic Technology and Information Security,
Ministry of Education, Shandong University, Jinan, 250100

Abstract. Blow-CAST-Fish is a sixteen-round Feistel cipher in the key, we will analyze the properties of F function under
which the F function is a part of the private key. In this paper, we weak key assumptions. We will try to identify differential
show that the disclosure of F allows perform a differential characteristics with high probability for round function F in
cryptanalysis against a number of rounds. We firstly identify the which a non-zero inputxor resulted in a zero outputxor under
properties of F function with three active S-boxes to construct some weak keys, then the differential cryptanalysis for
the 6-round differential characteristic with the probability 2 -61
8-round and 16-round of Blow-CAST-Fish will be given. As
and the differential weak keys occur with the probability 2-12. the attack assumption in [4], we also assume the opponent
Then we identify the properties of F function with only one knows the part of the key which describes the F function, that
active S-box to construct the 14-round differential characteristic is the four S-boxes. And the main idea of the 8-round attack is
with the probability 2-49 and the differential weak keys occur from reference [9].
with the probability 2-52. With the above differential The paper is organized as follows. Section 2 introduces the
characteristics, we can attack 8-round and 16-round description of Blow-CAST-Fish. We present how to attack
Blow-CAST-Fish respectively. 8-round and 16-round Blow-Cast-Fish in Section 3 and
 Section 4 respectively. We conclude this paper in Section 5.
Introduction 
Blow-CAST-Fish is a block cipher designed by Description of Blow-Cast-Fish
Krishnamurthy G.N et.al in 2008[1]. It uses the features of As a Feistel block cipher, Blow-Cast-Fish uses a block size
Blowfish [2] and CAST-128 [3]. of 64 bits, and the key size can vary from 32 bits to 448 bits.
S.Vaudenay presented the differential cryptanalysis for The cipher consists of two parts: a key expansion part and a
Blowfish based on the differential characteristics for round data encryption part.
function with non-zero inputxor and a zero outputxor [4]. J. Key expansion converts a key of at most 448 bits into
Nakahara Jr and M. Rasmussen presented the first concrete several subkey arrays totaling 4168 bytes following a
linear cryptanalysis on reduced round CAST-128 and scheduling scheme which works as a pseudo random
CAST-256 in [6], for CAST-128, they can recover the subkey generator. As this scheme is very complicated, one has to
for the 4th round with 237 known plaintexts and 272.5 times of store definitely the expanded key. Our cryptanalysis is not
4-round CAST-128 encryption. M. Q. Wang and X. Y. Wang related to the key schedule, so we don’t present it in detail.
et.al presented the linear cryptanalytic results for 6-round The expanded key consists of:
CAST-128 with 253.96 known plaintexts and 288.51 times of z 18 32-bit round subkeys P1,ĂĂP18˗
6-round encryption in [8]. Haruki Seki and Toshinobu z Four arrays of 256 32-bit values which describe
Kaneko found the 2-round differential characteristics for F2 four S-boxes S1, S2, S3 and S4 with 8-bit inputs and
round function of CAST-128 in [5], in which a non-zero 32-bit outputs.
inputxor resulted in a zero outputxor, then they could attack The four S-boxes define 32-bit to 32-bit round dependent
5-round CAST-128. function F as follows:
As we know, there is no attack on Blow-CAST-Fish until (Divide the input of F into four 8-bit quarters: a, b, c and d)
now. Due to the S-boxes of Blow-CAST-Fish are related with Rounds˖1ˈ4ˈ7ˈ10ˈ13ˈ16

Corresponding author (email:[email protected])

Supported by National Basic Research Program of China (GrantNo.2007CB807902) and National Natural Science Foundation of China (GrantNo.60525201 and
60910118).

523
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_91, © Springer Science+Business Media B.V. 2010
524 SUN AND WANG

+ S2[b])-S3[c])+S4[d]
F1= ((S1[a]ƻ function in round 2 is as follows:
Rounds˖2ˈ5ˈ8ˈ11ˈ14 I=IaýIbýIcýId=(L1ƻ + P2)<<<kr2
+ S4[d]
F2= ((S1[a]-S2[b])+S3[c])ƻ + S4[Id]
F2= ((S1[Ia]-S2[Ib])+S3[Ic])ƻ
Rounds˖3ˈ6ˈ9ˈ12ˈ15 The mapping (Ia,Ib ė61>,a@62>,b@ is a 16 to 32 bits
+ S3[c])-S4[d]
F3= ((S1[a]+S2[b])ƻ function, so it may have a collision61>,a@62>,b@ 61>,a’@
Where “ ” is the bit-wise xor, “+” is the addition modulo 62>,b’@with high probability.
232 and “ˉ” is the subtraction modulo 232. Let ¥=Ia+Ia’, =Ib+Ib’, assuming there is only one collision
The encryption process is defined as follows: for S1-S2ˈso the probability of characteristic ¥00ė0000
z Split the plaintext into left and right 32-bit halves + S4[Id] is 2-15. So 2-round
from I to ((S1[Ia]-S2[Ib])+S3[Ic])ƻ
XL and XR . differential characteristic for Blow-Cast-Fish with the
z For 1¡i¡16 probability 2-15 can be produced,
XL=XLƻ + Pi R1: (0000ý(¥00>>>kr2))((¥00>>>kr2ý0000)
XL=XL<<<Kri R2: ((¥00>>>kr2) 0000)(0000ý(¥00>>>kr2))
(Kri=last 5-bit of Pi) We search for 210 keys randomly, each key produces four
XR=F(XL)ƻ + XR S-boxes according to the key schedule of the cipher, then
Swap XL and XR (Undo the last swap) test the number of keys which make S1-S2 have a collision.
z XR=XRƻ
+ P17 We tested for twenty times to know that about half of every
+ P18
XL=XLƻ 210 random keys will produce S-boxes which have a
Recombine XL and XR as the ciphertext. collision with S1-S2. The value of ¥ý is different for
The encryption process is shown in figure 1. differential keys, and for most time there is only one value
of ¥ý for the same weak key. So the weak keys that cause
 Differential attack on 8-round Blow-Cast-Fish differential characteristic ¥00 ė 0000 for the round
For convenience, we denote the input of ith round as function in the second round occurred with probability 2-1.
(Li-1,Ri-1). We consider the second round at first, the round We will try to extend the above 2-round differential
characteristic to 6 rounds. First we only consider kr2 as zero,
Plaintext the other cases will be considered later.
Round 1

3.1 Differential Characteristic for 6-round Blow-Cast-

P1 Kr1
<<< F1 Fish
First, we extend the above 2-round differential character-
istic to 4 rounds. The inputxor for the 3rd round is zero, so
Round 2
the outputxor of the 3rd round must be zero and the inputxor
P2 Kr2
for round function F1 in the 4th round must be ¥00( kr2=0),
<<< F2
and the outputxor has not been decided. The differential
characteristic of F1 in the 4th round can be denoted as
Round 16
follows,
I=IaýIbýIcýId=(L3ƻ + P4)<<<kr4
P16 Kr16
<<< F1 I’=Ia’ýIb’ýIc’ýId’=((L3ƻ + ¥00)ƻ + P4)<<<kr4
+ + (((S1[Ia’] ƻ
1=(((S1[Ia] ƻ S2[Ib])-S3[Ic])+S4[Id]) ƻ + S2[Ib’])-
P18 P17
S3[Ic’])+ S4[Id’])
Ciphertext In order to further extend the differential characteristic to
Fig 1˖Feistel network showing encryption more rounds, we only consider the case that 1 equals to zero.
 DIFFERENTIAL CRYPTANALYSIS OF BLOW-CAST-FISH 525

As Iƻ+ I’=¥00<<<kr4 and kr4 varies from 0 to 31, we find 2. For every four S-boxes which have a collision with
+ I’ be zero, i.e. one of
out that there is always a byte of Iƻ S1-S2 and the corresponding inputxor ¥ý , we exhausted
the S-boxes always be non-active. So the probability of the 232 values of L3 and the 25 values of kr4 in the 4th round.
differential characteristic ¥00<<<kr4 ė 0000 from I to If  , the key used is the weak key that results the
+ S2[Ib])-S3[Ic])+S4[Id] is 2-23 and the probability of
((S1[Ia]ƻ differential characteristic ¥00<<<kr4 ė0000 for the 4th
4-round differential characteristic is 2-38 . round.
We will test the proportion of weak keys together with 3. For the keys which satisfy both condition 1 and 2, we
the 6th round later. exhausted the 232 values of L5 and the 25 values of kr6 in the
6th round. If   , the key used is weak key that results the
Then we extend the above 4-round differential differential characteristic ¥00<<<kr6 ė0000 for the 6th
characteristic to 6 rounds. The inputxor of the 5th round is round.
zero, so the outputxor of the 5th round must be zero and the The key which satisfies the three conditions is the weak
inputxor of round function F3 in the 6 th
round must be key for the 6-round differential characteristic.
¥00(kr2=0), and the outputxor has not been decided. The 4. Test for twenty times to get the steady proportion of
differential characteristic of F3 in the 6 th
round can be weak keys.
denoted as follows,
I=IaýIbýIcýId=(L5ƻ + P6)<<<kr6 We tested that the weak keys occurred with the
’ ’ ’ ’ ’
+ ¥00)ƻ
I =Ia ýIb ýIc ýId =((L5ƻ + P6)<<<kr6 probability 2-2. There is at least one solution for kr4 which
+ (((S1[Ia’]+S2[Ib’]) ƻ
+ S3[Ic])-S4[Id])ƻ
2=(((S1[Ia]+S2[Ib])ƻ + satisfies the differential characteristic in the 4th round for
’ ’
S3[Ic ])-S4[Id ]) the same keys, so the satisfied kr4 occurred with probability
We consider the case that 2 equals to zero. 2-5. Similarly, the satisfied kr6 occurred with probability 2-5.
+ I’=¥00<<<kr6 and kr6 varies from 0 to 31, we find
As Iƻ The probability for the above 6-round differential
’
+ I be zero, i.e. one of
out that there is always a byte of Iƻ characteristic is 2-61 (see figure 2), and the 6-round
the S-boxes always be non-active. So the probability of differential weak keys occur with the probability 2-5™2-5™2-5
differential characteristic ¥00<<<kr6 ė 0000 from I to ™2-2 2-17 .
+ S3[Ic])-S4[Id] is 2-23 and the probability of
((S1[Ia]+S2[Ib])ƻ In order to further increase the proportion of the
the 6-round differential characteristic is 2-61 (see figure 2).
differential weak keys for 6-round differential characteristic,
we try to test the differential characteristics with the
Due to S-boxes are key dependent, in order to obtain the
inputxor ¥00>>>kr2 and zero outputxor for round
proportion of weak keys for the 6-round differential
functions in the 4th and 6th round as kr2 is not zero.
characteristic, we need to identify the weak keys which
As kr Į 0, the differential characteristics for round
cause the following three conditions :
functions in the 4th and 6th round are (¥00>>>kr2)<<<kr4ė
1. There is a collision for S1-S2 with differential
0000 and (¥00>>>kr2)<<<kr6 ė 0000 respectively.
characteristic ¥ýė0ý0. 5
Obviously, for each value of kr ę 2 , there must be
2. 1=0 for the 4th round differential characteristic.
solutions of kr4 and kr6 that satisfy the differential
3. 2=0 for the 6th round differential characteristic.
characteristics. It means that for each non-zero kr, the
We test the proportion of weak keys for the 6-round
proportion of weak keys in the 4th and 6th round is same.
differential characteristic as follows:
So the differential weak keys which produce the 6-round
1. Select 27 keys of Blow-Cast-Fish randomly, and differential characteristic occur with the probability 25 ™
identify the keys produce four S-boxes which have a 2-17 2-12.
collision ¥ýė0ý0 with S1-S2 .
526 SUN AND WANG

0000 δμ 00
3.3 Complexity Analysis

P1
In this attack, k = 37 is the total subkey bits of the 8th round,
Kr1

<<<
0000
F1 p=2-61,  =2-32 because the ciphertext pairs can be filtered
according to their left halves XOR value ¥00>>>kr2. For a
p=2-38
random ciphertext pair, we denote C=(CLýCR), since we
have
0000 δμ 00 F2((CLƻ + F2((CL’ƻ
+ P10)<<<kr8)ƻ + P10)<<<kr8) =xyzt
P5 Kr5
We can exhaustively try all the 237 possible values for (P10,
0000 p=1
<<< F2 kr8) until the above equation holds. Given the ciphertext pair,
on the average there are 25 solutions for the equation, so the
δμ 00 0000
average count of keys per analyzed pair  is equal to 25.
P6 Kr6
Consequently,
βγσθ
<<< F3 p=2-23 2−61 × 237
S N= = 23
25 × 2−32
0000 δμ 00 The data complexity is 263 plaintexts. The time complexity

is about 2 ™237 ™232 274 times of one-round encryption

Fig2: Differential characteristic for 6-round Blow-CAST-
Fish (kr= 0) which is equivalent to 271 times of 8-round encryption. The
attack is efficient for Blow-CAST-Fish with key sizes more
3.2 Differential Attack on 8-round Blow-Cast- Fish than 71 bits.
With the differential characteristic of 6-round, we can
attack 8-round Blow-CAST-Fish (see figure 3). The attack 4 Differential attack on 16-round Blow-Cast- Fish
procedure is implemented as follows, In order to increase the number of round that can be attack-
63
1. Choose 2 values of the plaintext. ed, we need to improve the probability of the differential cha-
2. Loop for candidate i of kr(i = 0, ĂĂ, 31): racteristic for each round. So we consider the F function with
(a) Collect all those plaintext pairs whose XOR is only one active S-box, see S1 for analysis.
¥00>>>i at the right halves R0. There are 262 such The mapping aėS1[a] may have a collision S1[a]=S1[a’]
+ a’ˈkr2= ˙ˈˈĂˈ˅
under some keys. Let ¥=aƻ ˈ
pairs, and the number of right pairs is 262™2- 2.
assuming there is only one collision for S1 with input
(b) Loop for 237 values of P10 and kr8 in round 8. differential ¥, the probability of characteristic ¥000ė0000 is
i. Loop for remaining pairs: 2-7. So 2-round differential characteristic for Blow-Cast-Fish
For all 231 remaining ciphertext pairs with the probability 2-7 can be produced,
CLƻ + CL’=¥00>>>i R1:(0000ý(¥000>>>))((¥000>>>ý0000)
calculate the outputxor of round function F2 in the 8th round. R2:((¥000>>>) 0000)(0000ý(¥000>>>))
If the values of P10 and kr make the next equation holds,
increase the corresponding key counter [P10ýkr] by 1. 4.1 Differential Characteristic for 14-round Blow-Cast-
F2((CLƻ+ P10)<<<kr8)ƻ + F2((CL’ƻ
+ P10)<<<kr8) =CRƻ + Fish
CR’=xyzt We iterated the above 2-round characteristic seven times
where C=(CLýCR) and C’=(CL’ýCR’) are a pair of cipher- as shown on figure 4 (assuming kr2=kr4=kr6=kr8=kr10=kr12
texts. If the key counter is equal to 2, the subkey values of P10, =kr14˙, ‰¥ǃxyzt represent undetermined values). The
kr8, and kr2 are the right subkey with high probability. resulting characteristic has probability 2-49.
 DIFFERENTIAL CRYPTANALYSIS OF BLOW-CAST-FISH 527

0000 δμ 00 >>> kr2 0000 ¥000>>>¢

P1 Kr1 P1 Kr1
0000 0000
<<< F1 <<< F1 p=1

p = 2−61

¥000>>>¢ 0000
δμ 00 >>> kr2 0000 P2 Kr2=¢
<<< ¥000 F2
-7
p=2
P6 Kr6
βγσθ
<<< F3

0000 δμ 00 >>> kr2 ¥000>>>¢ 0000

P7 Kr7
P14 Kr14=¢ -7

<<<
0000
F1 p =1 <<< ¥000 F2 p=2

δμ 00 >>> kr2 0000 0000 ¥000>>>¢

P8 Kr8
βγσθ
<<< F2 p =1 Fig 4: Differential characteristic for 14-round Blow-CAST-
P10 P9 Fish
δμ 00 >>> kr2 xyzt

2. Loop for candidate  of kr2 ( = 0, ĂĂ31):

Fig 3: 8-round Blow-CAST-Fish (kr¦0) (a) Collect all those plaintext pairs whose XOR is ¥000>>>
 at the right halves R0. One structure proposes 231 such pairs,
20
We search for 2 keys randomly, each key produces four thus 220 structures propose 251 such pairs. The number of right
S-boxes according to the key schedule of the cipher, then pairs is 251×2-49=22.
test the number of keys which make S1 have a collision. We (b) Loop for 237 values of P18 and kr16 in round 16.
3
tested for twenty times to know that at least 2 keys will i. Loop for remaining pairs:
produce S-boxes which have a collision with S1. The value For all 219 remaining ciphertext pairs
of ¥ is different for differential keys, and for most time there CLƻ + CL’=¥000>>>
is only one value of ¥ for the same key. Then the weak keys calculate the outputxor of round function F1 in the 16th round.
which can produce the differential characteristic If the values of P18 and kr16 make the next equation holds,
¥0000000 for F functions occurred with probability 2-17, increase the corresponding key counter [P18ýkr] by 1.
kr2=kr4=kr6=kr8=kr10=kr12=kr14= occurred with probability F1((CLƻ + F1((CL’ƻ
+ P18)<<<kr16)ƻ + CR’
+ P18)<<<kr16)= CRƻ
2-5×7=2-35, so the proportion of weak keys for 14-round =xyzt
-35 -17 -52
differential characteristic is 2 ×2 =2 . where C=(CLýCR) and C’=(CL’ýCR’) are a pair of cipher-
texts. If the key counter is equal to 4, the subkey values P18,
4.2 Differential Attack on 16-round Blow-Cast-Fish kr16, and kr2 are the right subkey with high probability.
With the fourteen rounds differential characteristic, we
can attack 16-round Blow-CAST-Fish (see figure 5). The 4.3 Complexity Analysis
attack procedure is implemented as follows, As the analysis of 8-round, the signal to noise can be
1. Choose all 232 values for the right halves of the computed as follows,
plaintext R0, which constitute a structure. Choose all 220
values for the left halves of the plaintext L0, so 220
structures are produced.
528 SUN AND WANG

0000 ¥000>>>¢ CAST-Fish, 14-round differential characteristic with the

P1 probability 2-49 can be identified which are resulted by 2-52
Kr1
0000 p=1 of the total key space. Based on the 14-round differential
<<< F1
characteristics, we can attack 16-round Blow-CAST-Fish.
We can also use S2, S3 or S4 for analysis, the probability of
¥000>>>¢ 0000
P2
weak keys which can produce collision with S2 is nearly to
Kr2=¢
<<< ¥000 F2 p=2
-7 the one with S1. But the probability of weak keys which can
produce collision with S3 or S4 is lower than that of S1.

0000 ¥000>>>¢
References
P15
[1] Krishnamurthy G.N, Dr. V. Ramaswamy, Leela G.H
Kr15 p=1
<<<
0000
F3 and Ashalatha M.E, “Blow- CAST-Fish: A New 64-bit
Block Cipher”, Bapuji Institute of Engineering and
Techology, Davangere-577004, Karnataka, India.
¥000>>>¢ 0000
[2] B.Schneier, “Description of a New Variable-Length Key,
P16 Kr16
βγσθ p=1 64-Bit Block Cipher (Blowfish)”,Fast Software
<<< F1
xyzt Encryption, Cambirdge Security Workshop proceedings
P18 P17 (December 1993), Springer-Verlag, 1994, pp.191-204.
¥000>>>¢ xyzt [3] Adams, C. “The CAST-128 Encryption Algorithm”.
RFC 2144, May 1997.
Figure 5: 16-round Blow-CAST-Fish [4] S. Vaudenay, “On the weak keys of Blowfish,” Fast
Software Encryption (FSE’96), LNCS 1039, D. Gollma-
2−49 × 237 nn, Ed., Springer-Verlag, 1996, pp. 27–32.
S N = 5 −32 = 215
2 ×2 [5] Seki. H and Kanekko. T, “Differential Cryptanalysis of
where k=37,p=2-49,=25,=2-32 ,.The data complexity is CAST-256 Reduced to Nine Quad-rounds”, IEICE
52
2 chosen plaintexts. The time complexity is about Transactions on Fundamentals of Electronics Communi-
25×237×220=262 times of one-round encryption which is cations and Computer Sciences, E84A(4), pp.913-918,
equivalent to 258 times of 16-round encryption. The attack 2001.
is efficient for Blow-CAST-Fish with key sizes lager [6] J. Nakahara Jr and M. Rasmussen, “Linear Analysis of
than 58 bits. And the proportion of weak keys is 2-52. Reduced-round CAST-128 and CAST-256”,
SBSEG2007, pp.45-55, 2007.
5 Summary [7] E. Biham and A. Shamir, “Differential Cryptanalysis of
Until now, there is no analysis result for the algorithm. In DES-like Cryptosystems”, Journal of Cryptology, vol.4,
this paper, we first studied the properties of round function no.1, pp.3-72, Springer, 1991.
F1 and F3 with non-zero inputxor and zero outputxor in [8] M. Q.Wang, X. Y.Wang, C. H. Hu, New Linear
the 4th and 6th round of Blow-CAST-Fish separately, then Cryptanalytic Results of Reduced-Round of CAST-128
6-round differential characteristic with the probability 2-61 and CAST-256, appeared on SAC’2008.
-12
can be identified which are resulted by 2 of the total key [9] M. Q.Wang, X. Y.Wang, K.P. Chow, Lucas C.K. Hui,
space. Based on the 6-round differential characteristics, we Differential Crypanalysis of Reduced-Round of
can attack 8-round Blow-CAST-Fish. CAST-128, Preprint.
Then we studied the properties of S1 with non-zero
inputxor and zero outputxor in round function of Blow-
 A Keyless Polyalphabetic Cipher
William J. Spezeski
Department of Computer Science
Massachusetts College of Liberal Arts
375 Church Street, North Adams MA 01247 USA
Phone: 413.662.5591
Email: [email protected]

Abstract-A polyalphabetic cipher is one in which the letters of the standard alphabet as shown in Figure1
multiple alphabets (monoalphabets) are used to encode below. To encode the word ‘CIPHER’, we embark on a
a plaintext message. Each letter of plaintext is encoded ‘search and count’ process.
by selecting one of the alphabets with which to
determine a substitution. A key word or phrase is often For illustration, we begin at position 0 (the letter ‘A’); any
used to select different alphabets to encrypt the pre-designated letter position will do.
plaintext. This paper considers three concepts that,
The first letter of plaintext is obtained by counting the
when combined, lead to a keyless polyalphabetic cipher.
number of letters that we pass over as we go (left to right)
INTRODUCTION – POLYPHABETIC CIPHERS from the starting position ‘A’ until we find the first
plaintext letter ‘C’. It is coded by the resulting count, in
There are many variations of polyalphabetic cipher types this case, 2.
with varying degrees of complexity. Most rely on a key or
keyword to form the substance of the ciphertext. Consider The second letter of plaintext is obtained by counting the
this very simple example: A polyalphabetic cipher with the number of letters that we pass over as we go from the first
key BLUEBIRD would encode its first letter of plaintext plaintext letter ‘C’ to the second plaintext letter ‘I’. The
using its B-alphabet, its second letter using its L-alphabet, second letter is coded by the second count 6. And so on. If
its third letter using its U-alphabet and so on. The ninth the search process takes us beyond letter25 (Z), we return to
letter thru the sixteenth letters of plaintext would reuse the letter0 (A) and continue the count from right to left. Thus
B,L,U,E,B,I,R,D-alphabets for encipherment, respectively. letter26 = letter0 in our circular count.
Every eight letters of plaintext would use the key for
encipherment until the plaintext message is exhausted. A B C D E F G H I J K L MN O P Q R S T U VWX Y Z
Both the sender and receiver would need seven distinct 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25
alphabets available designated by B, L, U, E, I, R, D, C 2
respectively. Some classic multiple-alphabet ciphers I 6
include the Vigenere Cipher [6], Gronsfeld Cipher [3], P 7
Porta Cipher [3] and the Enigma Cipher of WWII [4]. H 17
E 23
A KEYLESS POLYALPHABETIC CIPHER
R 13
The polyalphabetic cipher being considered here is the Fig. 1 The Search and Count Algorithm
byproduct of combining three unrelated concepts. 

Consequently, we will look at the methodology of this Using this scheme, the plaintext word ‘CIPHER’ is encoded
cipher in three stages. as the substitution ciphertext 2 6 7 17 23 13. This is
the general idea that we shall use for substituting characters
A. A Monoalphabet Substitution Cipher
of plaintext with numbers to form the ciphertext. Note that
To illustrate the concept behind a keyless alphabetic cipher,
it can be made more obtuse by setting up patterns of
we first consider a simple alphabetic cipher which relies on
alternating clockwise and counter-clockwise counting.

529
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_92, © Springer Science+Business Media B.V. 2010
530 SPEZESKI

Some examples of various monoalphabets are shown on. Better yet, an arbitrary (random) number can determine
below. what occurrence of each letter of plaintext we will search
A1 = defghijklmnopqrstuvwxyzabc for in the polyalphabet. A computer can randomly generate
A2 = jlpawiqbctrzydskegfxhuonvm these integers or they can be pulled off-the-top of the
A3 = juliscaertvwxyzbdfghkmnopq sender’s head. The selection of these numbers results in
A4 = xzavoidbygerspcfhjklmnqtuw different versions of the encoded plaintext, but does not
A5 = fzbvkixaymeplsdhjorgnqcutw affect the decoding of the ciphertext. The original plaintext
and its corresponding decoded ciphertext are always the
B. A Substitution Cipher with a Polyalphabet same. For example, each ciphertext below
To develop our polyalphabet, we string several alphabets,
say {A1, A2, A3, A4, A5}, together to form the Plaintext: happy birthday
polyalphabet. The same basic assignment method of
converting plaintext to ciphertext shown above is used with Ciphertexts:
the polyalphabet. A starting position must be identified;
1. 36 75 16 55 10 57 57 35 64 30 16 78 52 52
then the ‘search and count’ process begins. If the
2. 9 48 16 28 37 30 3 62 10 3 43 51 79 25
polyalphabet is a string of length N, then after reaching
3. 63 75 43 55 37 57 3 8 64 57 16 78 25 79
letterN-1 in a count, the count continues with letter0 , letter1 ,
letter2 and so on; thus letterN = letter0 . is obtained by using the “Search and Count” algorithm with
a standard 26- letter alphabet plus a ‘space’. If we use five
27-character unsorted alphabets to construct a polyalphabet,
A1 A2 A3 A4 A5
some resulting ciphertexts for ‘happy birthday’ appear as
Fig. 2 – Polyalphabet Structure
follows:
The alphabets making up the polyalphabet need not be
1. 48 14 60 29 10 33 17 45 60 96 51 76 74 8
ordered or of the same length. The number of alphabets
2. 48 37 12 54 56 14 9 26 60 27 11 63 24 67
that constitute the polyalphabet may vary as well. Below is
3. 5 27 90 29 56 41 4 58 61 60 32 41 72 17
an example of a polyalphabet formed by concatenating
4. 74 86 56 17 81 69 4 58 31 43 79 41 79 60
alphabets A1, A2 and A3 above.
All four ciphertexts above decrypt back to the original
defghijklmnopqrstuvwxyzabcjlpawiqbctrzydskegfxhuonvm
plaintext. Notice that the last series of plaintext ciphers
juliscaertvwxyzbdfghkmnopq
consists of (with leading zeros) all 2-digit numbers with the
largest number being 96, even though there are 135
In some sense, this polyalphabet can be thought of as a characters in the polyalphabet that is being used. This is
large monoalphabet. Because its construction (as I use it) being controlled by the fact that the random number choices
makes use of several existing monoalphabets , I use the are restricted to 1, 2, or 3. This helps to disguise the
term polyalphabet to describe it. number of ‘alphabets’ that form the polyalphabet as well as
the length of the polyalphabet.
C. A Keyless Polyalphabetic Cipher
Most polyalphabetic ciphers require a key - a word or DECODING THE CIPHERTEXT
phrase that distinguishes which particular monoalphabet a
Deciphering the coded message is very easy. One starts at
particular plaintext letter is being enciphered with. The
the pre-designated letter position of the same polyalphabet
classic Vignere Cipher [5] is a case in point. However,
and simply counts, associating the final number of each
using the ‘search and count’ process mentioned above,
count with the associated plaintext letter. Figure 1 above
there is no need for a key.
illustrates both how the word CIPHER can be encoded and
The ‘search and count’ process essentially looks for the how the ciphered word can be decoded. In the encipherment
first occurrence of the plaintext letter as it traverses the process, searching for a plaintext letter results in a count ,
polyalphabet. A match results in a ciphertext number. which in turn becomes the ciphered substitute; in the
However, there is no need to look for only first occurrences decipherment process, the count becomes the means of
of a plaintext letter. We could search for the second locating the corresponding plaintext letter. As mentioned
occurrence of a plaintext letter, a third occurrence, and so
 A KEYLESS POLYALPHABETIC CIPHER 531

before, an enciphered text can have many possible variants. deciphering algorithms in a computer language like Java is
However, each variant will decipher to the same plaintext. a straight-forward process using String arrays. Here are
some things to keep in mind:
SOME OBSERVATIONS
The number of alphabets used to form the polyalphabet
Let us suppose that a five letter word is to be encoded with
should be a prime number. This is to ensure that the
a 100-character polyalphabet using the random search
searches do not get confined to any cyclic subset of
‘modifiers’ 1,2 and 3. Then there will be 100 possible
alphabets comprising the polyalphabet.
starting positions and

100 x 3 x 32 x 33 x 34 x 35 or 100 x 315 The number of distinct random numbers that get generated
should be strictly less than the number of alphabets making
possible variants of the enciphered five letter word. More up the polyalphabets. For example, assume that a particular
generally, if a plaintext message consisting of N characters polyalphabet consisted of 5 alphabets each having 25
(including spaces) is encoded with a P-character characters. Assume further, that the distinct random
polyalphabet using K random search modifiers, then the numbers that get generated are 1, 2, 3. The largest possible
number of possible enciphered variants is numeric substitution for a plaintext letter would be 100,
even though there are actually 125 characters in the
P KN(N + 1)/ 2 . polyalphabet. This helps to conceal the actual string length
of the polyalphabet. In fact, the 5 alphabets mentioned can
If a 135-character polyalphabet to code ‘happy birthday’
be rearranged so that only1- or 2-digit numbers will be
using 3 random search modifiers, the number of variants
generated for substitution. If the ciphertext is converted to
would be 135 x 3105 . Four of the variants were shown
ASCII characters , 3-digit numbers (< 256) can be
above. For any fixed starting point, the number of variant
accommodated without the ciphertext getting unwieldy.
ciphertexts for ‘happy birthday’ is 3105 .

IMPLEMENTATION Use the Java string method length() (or equivalent) to make
any reference to the length of the active polyalphabet. It
The implementation of the Search and Count Cipher (SCC) helps to further conceal its actual string length.
requires that the both the sender and receiver use the same
polyalphabet as well as an agreed upon starting position The alphabets that comprise the polyalphabet can vary in
within the polyalphabet. There are 26! (over 1026 ) regular number, length, letter arrangement and order of appearance
26-letter monoalphabets available for building a within the polyalphabet. The details, of course, must be
polyalphabet. That number grows rapidly with an increase agreed upon in advance by both sender and receiver.
in alphabet size. Below is the ciphertext that is generated from a
polyalphabet formed by the 5 alphabets above (A1, A2, A3,
Polyalphabets can be easily constructed by having a A4, A5 each with an extra ‘space’) using modifiers 1, 2, 3
collection of monoalphabets available for concatenation. and starting at the first position of alphabet A1.
The computer language Java, for example, provides string
concatenation (+) that allows a wide variety of 74 52 17 78 38 44 73 25 34 41 29 5 14 57 48 77 67 26 18
polyalphabets to be built very simply from existing 61 79 64 13 25 39 1 20 75 23 30 41 11 3 29 14 72 78 30 4
monoalphabets. Again, this is a primary reason for 78 20 8 63 56 32 42 78 40 79 64 68 36 52 23 47 23 19 55
referring to the cipher as a polyalphabetic cipher. In turn, 76 34 15 36 63 79 23 16 27 26 41 3 59 39 72 15 9 64 39 8
multiple alphabets can be constructed from a single one by 21 50 13 2 68 28 56 38 30 57 4 80 80 21 80 68 23 9 25 23
using a shuffle algorithm to rearrange an alphabet’s letters. 12 4 2 63 51 63 37 12 8 21 40 51 46 41 15
If both sender and receiver have the same set of numerically
identifiable alphabets at hand, the sequence of alphabets It is the ciphertext for the verse
used to construct the operable polyalphabet can be encoded
speak roughly to your little boy
within the numerical ciphertext. The starting position can
and beat him when he sneezes
be part of the ciphertext as well. And like the polyalphabet
he only does it to annoy
construction, the plaintext message string can be built by
because he knows it teases
using concatenation. Coding the enciphering and
532 SPEZESKI

FIELD USE Frequency Count Ciphertext Random Sequence

This cipher system works very efficiently using a computer 0 21 22
to encode and decode messages. It can also work very 1 29 26
efficiently in the field. It would be helpful to have a cipher 2 15 18
disk [4] or some graph paper and limit the number of 3 9 10
alphabets making up the polyalphabets to 3, 5 or 7. The 4 4 4
random integers can be picked at will without hindering the
5 0 0
deciphering process. A drawback to SCC is that an error in
6 1 0
counting will corrupt the rest of the ciphertext. A slide
rule-like device could be very helpful to minimize this Fig. 3 - Distribution of 108 Integers in the Interval [1, 80]
problem.
For this ciphertext, there were 21 values in [1, 80] that did
A polyalphabet constructed with five 30-character not appear; There were 22 values in [1, 80] that did not
alphabets for field use would have 150 characters while a appear in the random sequence. There were 29 values of
Vigenere table [4] will have 676 (26x26) characters. ciphertext that had a frequency count of 1 while the random
sequence had 26. And so on. The outlier is one value with
CONCLUSION
frequency count 6.
In 1949, Claude Shannon [5], widely known as the Father
of the Information Age, proposed the following Also of significance is that there is only one digram and no
characteristics of a good cipher: trigrams in this ciphertext. This, coupled with the random-
ness of the cipher and its keyless feature, helps to increase
1. The amount of secrecy needed should determine the effort of any analysis to break the code. All this speaks
the amount of labor for encryption and to Shannon’s first requisite. That the SCC meets requisites
decryption. two and three are clear. The SCC, however, does not meet
2. The set of keys and the enciphering algorithm the requirement four. An error in counting will compromise
should be ‘free from complexity’. the rest of the ciphertext. On the other hand, computers are
3. The implementation of the process should be as pretty reliable. The final requisite is not met either. The
simple as possible. use of numbers for ciphertext will at least double the length
4. Errors in ciphering should not propagate and cause of the message. If, on the other hand, the ciphertext is
corruption of further information in the message. converted and sent as ASCII code characters, the ciphertext
5. The size of the encoded message should be no length will be roughly the same. Of course, it adds a step to
larger than the text of the original message. the message encryption and decryption.

The SCC produces random-like sequences of numbers for Will the SCC survive the test of cryptanalysis? That
ciphertext that help to diffuse the process of cryptanalysis. question remains.
The randomness of the ciphering algorithm reduces the REFERENCES
recognizable statistical patterns that a cryptanalyst might
use to a minimum. The lack of a key further frustrates [1] Baldoni, Ciliberto & Cattaneo, Elementary Number Theory,
Cryptography and Codes, Springer-Verlag, (2009).
cryptanalysis efforts.
[2] Friedman, William F., Military Cryptanalysis - Part II: Simpler
Varieties of Polyalphabetic Substitution Systems, Filiquarian Publishing,
In the previously encoded verse above, there are 108 LCC./Qontro (1938).
integers in the ciphertext whose values fall in the interval [3] Gaines, Helen F., Cryptanalysis, Dover Publications, (1956)
[4] Kahn, David, The Code-breakers, Macmillan Publishing, (1979)
[1, 80]. Figure 3 illustrates the frequency counts of numeric
[5] Shannon, Claude, “Communication Theory of Secrecy Systems”, Bell
ciphertext in the interval [1, 80] and compares it with 108 System Technical Journal, (1949).
randomly generated numbers in the same interval. [6] Singh, Simon, The Code Book, Fourth Estate Ltd., (1999).
[7] Sinkov, Abraham, Elementary Cryptanalysis,2nd edition, The
Mathematical Association of America, (2009).
 Vertical Handover between WiFi Network and
WiMAX Network According to IEEE 802.21 Standard
Mariem Thaalbi, Nabil Tabbane
Research Unit MEDIATRON
SUP’COM
Tunis, Tunisia
[email protected] [email protected]

Abstract—IEEE 802.21 is one of the emerged standards designed BSSs connected to the same wired LAN so called Distribution
to build handover between heterogeneous technologies while System (DS).
optimizing session continuity. In this paper we provide a
performance evaluation of the vertical handover between WiFi In ad hoc mode, devices can communicate directly with
and WiMAX using IEEE802.21 for different traffic flows: VoIP, each other without Access Point support.
video streaming and FTP. The performance metrics used for this In our paper, we will focus on infrastructure mode.
evaluation include throughput, end-to-end delay and packet loss
rate. The performance evaluation is done using the Network
Simulator NS2.

Keywords: WiFi; WiMAX; vertical handover; IEEE802.21;

MIH

I. INTRODUCTION
The rapid expansion of heterogeneous wireless and mobile
networks raises interesting issues such as vertical handover. To
insure vertical handoff between WiFi and WiMAX networks,
we will propose, in this paper, the IEEE 802.21 as an alternative
to maintain uninterrupted user sessions during handovers
between these two different networks. The IEEE 802.21
supports media Independent handover (MIH) in order to
provide seamless mobility between heterogeneous networks.
The purpose of this work is to evaluate the performances of
IEEE 802.21 in vertical handover between WiFi and WiMAX
technologies. The remainder of this paper is organized as
follows: The background of WiFi and WiMAX are briefly
viewed on Section II and III. An overview of IEEE 802.21 is
given in Section IV. Simulation Scenario and experimental
Fig.1. WiFi infrastructure mode.
results are presented in Section V. The paper is concluded in
Section VI. III. WIMAX NETWORKS
Worldwide Interoperability for Microwave Access
II. WIFI NETWORKS (WiMAX) [2] is a Wireless Metropolitan Access Network
Wireless Fidelity (WiFi) [1] is a wireless local area network (WMAN). It is used generically to describe wireless systems
over limited range. It is also known under IEEE 802.11, and based on IEEE 802.16.
aims to connect devices such as Personal Computer, PDA,
WiMAX offers high data rates and large area coverage. It
laptop, printers, etc.
supports fixed and mobile broadband access. In our work, we
IEEE 802.11 allows connectivity in infrastructure mode or will focus on Mobile WiMAX (IEEE 802.16e) [3].
ad hoc mode. In infrastructure mode, the WiFi stations
The Mobile WiMAX Network, described in Fig.2, is
communicate through the Access Point (AP). In this mode, the
divided into three parts: the Mobile Station (MS) which is the
802.11 network architecture, illustrated in Fig.1, is hierarchical.
user’s device, the Access Service Network (ASN) is the radio
Its basic element is the Basic Service Set (BSS), which is the
access and includes one or more Base Stations (BS) and one or
group of stations covered on the area covered by one AP. A
more ASN Gateways (ASN-GW). The Connectivity Service
BSS may also be part of larger network element called
Network (CSN) is the core of WiMAX Network. It provides IP
Extended Service Set (ESS). The ESS consists of one or more
connectivity to the MS.

533
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_93, © Springer Science+Business Media B.V. 2010
534 THAALBI AND TABBANE

In order to provide mobility, 802.16e supports handover Commands originate from the upper layers down to the MIHF.
mechanisms. The IEEE 802.16e defines three different types of Link Commands are specific to the lower layers. MIIS provides
handover: Hard Handover (HHO), Fast Base Station Switching a framework by which MIHF can discover homogenous and
(FBSS) and Macro Diversity Handover (MDHO). The HHO is heterogeneous network information existing within a
often referred to as a break-before-make handover: first the MS geographical area to facilitate seamless handover when roaming
disconnects from the serving BS and then connects to the target across these networks. The MIIS provides a bidirectional way
BS. In FBSS and MDHO, the MS maintains a diversity set for the two layers to share information such as current QoS,
which includes numerous active BSs in its range. The BS, to performance information and availability of service. The figure
which the MS has the connection, is called the anchor BS. In bellow illustrates the MIH architecture:
FBSS, the MS transmits to and receives data from a single
serving BS. The transition from the serving anchor BS to the
target anchor BS in FBSS is done without invocation of the
normal handover procedure, only the anchor BS update
procedure is needed. In MDHO, the MS maintains a
connection to one or more BSs in the diversity set
simultaneously.

Fig.3. MIH architecture.

V. SIMULATION SCENARIO AND EXPERIMENTAL RESULTS
Fig.2. WiMAX architecture. A. Simulation environement
IV. 802.21 OVERVIEW The simulation results presented in this paper are obtained
using Network Simulator NS-2 [5] with the MIH patch [6] and
IEEE 802.21 [4] referred as Media Independent Handover WiMAX patch [7].
(MIH) aims to enable seamless handover among heterogeneous
networks. MIH defines a logical entity, Media Independent The network simulator is used to define an appropriate
Handover Function (MIHF), located on layer 2.5 between link topology that allows vertical handover between 802.16e cell
layer and network layer of the OSI model. It provides a and 802.11b cell. The WLAN and WiMAX parameters used
framework that allows interaction between higher layers and during simulation are shown in TABLE I and TABLE II.
lower layers. The MIHF supports three types of services: Media TABLE I. WLAN cell's parameters
Independent Event Service (MIES), Media Independent
Command Service (MICS), and Media Independent IEEE 802.11b
Information Service (MIIS). The MIES aims to provide and to
Coverage Raduis 100 m
predict link changes such as LINK_UP, LINK_DOWN,
LINK_GOWING_DOWN, etc. These events are propagated Radio Propagation Model Two- RayGround
from lower layers to upper layers through the MIH layer. MIES
Frequency 2.4 GHz
is divided into two categories, link events and MIH events.
Link events are generated from the lower layer and transmitted Transmission Power (Pt_) 0,0027 W
to MIH layer. The MIH events are the events forwarded from 2.64504e-10 W
Receiving Threshold (RXThres)
MIH to upper layers. MICS refers to the commands, such as
initiate handover and complete handover, sent from higher
layers to lower layers. It allows enabling handover mechanism.
MICS includes MIH command and Link command. MIH
 VERTICAL HANDOVER BETWEEN WIFI NETWORKAND WIMAX NETWORK 535

The simulation duration was set to 250 s, the mobile node

TABLE II. WiMAX cell's parameters starts moving 10 s after the beginning of simulation time. The
flow traffic starts from the beginning of the simulation.
IEEE 802.16e
Coverage Raduis 1 Km C. Performance Criteria
Radio Propagation Model Two-RayGround In this paper, we focus on evaluating handover
performance when the MS moves from WiMAX cell to WiFi
Frequency 3.5 GHz
cell and also the inverse direction from WiFi cell to WiMAX
Transmission Power (Pt_) 15 W cell. The performance criteria adopted on our evaluation are:
Receiving Threshold (RXThres)
7.59375e-11W Throughput, end-to-end delay and packet loss ratio.
Sensitvity to link degradation
(lgd_factor_)
1.1 D. Results
Throughput
The throughput is the average rate of successful data
B. Proposed scenario delivery over a communication channel per unit time. It is
measured in Kbits per second (Kb/s).
The simulation topology, as shown in Fig.4, consists of one
In this section we will calculate the throughput during
802.11b AP, one 802.16e BS and one mobile station: MS. The
MS is moving between the two cells with speed of 1 m/s, and simulation time for the two mobile speeds 1 m/s and 10 m/s
with the two scenarios: handover from WiMAX to WiFi and
then we increase the speed to 10 m/s.
from WiFi to WiMAX.
The throughputs during handover cases for different traffic are
shown in Fig.5 (a) - (b) - (c).

(a) Throughput (Kb/s) - VoIP traffic

Fig.4. Simulation scenario

During simulations, the MS exchanges with its

correspondent three types of traffic: VoIP, Streaming and File
Transfer Protocol (FTP).The traffic flows parameters are
shown on TABLE III.

TABLE III. Simulated traffic parameters

Traffic type Packet size Delay interval Data rate

VoIP 160 bytes 20 ms 64 Kb/s

Streaming 2000 bytes 100 ms 160 Kb/s
FTP 1500 bytes 100 ms 120 Kb/s
(b) Throughput (Kb/s) - Video streaming traffic
536 THAALBI AND TABBANE

(c) Throughput (Kb/s) - FTP traffic

Fig.5. The throughput for different types of traffic

(a) End-to-end delay (ms) - VoIP traffic
From these figures, we notice that the MS speed affects the
effective throughput at handover moment. The thresholds fixed
here to evaluate the throughput are equal to 4 Kb/s [8] for the
VoIP traffic, 5 Kb/s [8] for the video streaming traffic and 10
kb/s [8] for the FTP traffic. With mobile speed equal to 1 m/s
we obtain good results respecting the fixed thresholds. With
MS speed fixed to 10 m/s the results obtained are not
acceptable at handover moment. WiMAX network provides
better throughput than WiFi network with mobility speed equal
to 10 m/s.

End-to-end delay
The end to end delay refers to the time taken by data packet
to be transmitted across a network from source to destination.
We calculate in this section the end-to-end delay with the same
cases already described. The average delay for different
simulated traffic is shown in Fig.6 (a) - (b) - (c).
When the MS speed is equal to 10 m/s, the delays degrade (b) End-to-end delay (ms) - Video streaming traffic
in WIFI network because it supports only the low mobility
speed. The curves obtained with VoIP and FTP traffic during
the handover process exceed slightly the threshold which is
fixed to 100 ms for the VoIP and FTP traffics [8]. The curves
obtained with streaming video traffic exceed very largely the
threshold which is fixed to 100 ms [8].
We note that, with the MS speed (10 m/s) the delays
offered by WIMAX network are better than those offered by
WIFI network because the mobile WIMAX network can
support the low and the high mobility speed.

(c) End-to-end delay (ms) - FTP traffic

Fig.6. End-to-end delay for different types of traffic

 VERTICAL HANDOVER BETWEEN WIFI NETWORKAND WIMAX NETWORK 537

Packet loss rate In this figure, we see that with MS speed equal to 1 m/s, the
two handover cases curves reflect good results, unlike the
The packet loss rate parameter allows to determine the lost medium mobility speed equal to 10 m/s the handover curves
packet rate compared to those sent. The percentage of packet exceed the threshold (fixed to 1%) for the VoIP and video
loss for different simulated traffics is illustrated in Fig.7 (a) - streaming traffics [8].
(b) - (c). With low mobility, the packet loss rate is less in WIFI
network and the better results are obtained with the case when
the MS makes a handover from WIMAX to WIFI. With MS
speed equal to 10 m/s, the results are better in WIMAX
network and the handover case from WIFI to WIMAX gives
better packet loss rate values than the opposite case.
VI. CONCLUSION
The interoperability and the vertical handover between
different networks present currently a real challenge to
overcome.
In this paper, we have evaluated the performance of
vertical handover between WiMAX and WiFi according to
802.21 for FTP and real-time traffic such as VoIP and video
streaming. Seeing the results obtained, we can notice that the
speed of the MS affects the handover performance. Also, the
(a) Packet loss rate (%) - VoIP traffic WiFi supports only the lox mobility speed. Finally, we can
conclude that with the low speed the handover from WiMAX
to WiFi generate best results than the opposite case of
handover. But with medium speed, it is the otherwise because
the 802.16e supports better the mobility management.

REFERENCES

[1] IEEE Std 802.11™-2007, IEEE Standard for Wireless LAN Medium
Access Control (MAC) and Physical Layer (PHY) Specifications, June
2007.
[2] IEEE Std 802.16™-2004, IEEE Standard for Local and Metropolitan
Area Networks—Part 16: Air Interface for Fixed Broadband Wireless
Access Systems, October 2004.
[3] IEEE Std 802.16e™-2005, IEEE Standard for Local and Metropolitan
(b) Packet loss rate (%) - Video streaming traffic Area Networks—Part 16: Air Interface for Fixed and Mobile Broadband
Wireless Access Systems, February 2006.
[4] IEEE Std 802.21™-2008, IEEE Standard for Local and Metropolitan
Area Networks- Part 21: Media Independent Handover, January 2009.
[5] Network Simulator- NS2- http://www.isi.edu/nsnam/ns, February 2009.
[6] National Institute of Standards and Technology (NIST), “The Network
Simulator NS-2 NIST mobility package add-on,”, March 2009.
[7] National Institute of Standards and Technology (NIST), “The Network
Simulator NS-2 NIST IEEE 802.16 Implementation,”, March 2009.
[8] Jeffery G. Andrews, Arunabha Gosh, Rias Muhamed, “Fundamentals of
WiMAX, 1.7.3 Quality of Service”, June 2007.

(c) Packet loss rate (%) - FTP traffic

Fig.7. Packet loss rate for different type of traffic
Study on the Impact of Node Density and Sink Location
in WSN
Plamen Z. Zahariev, Georgi V. Hristov, Teodor B. Iliev
Department of Communication Systems and Technologies, University of Ruse, Ruse 7017, Bulgaria
E-mail: {pzahariev, ghristov, tiliev}@ccnrgroup.com

Abstract – In this paper we analyze the total energy energy and present the setting of the energy model. We
consumption of a wireless sensor network (WSN) per single consider a sensor network that is hierarchically clustered.
communication round as a function of the node density and base The LEACH (Low Energy Adaptive Clustering Hierarchy)
station coordinates. We provide performance results of several protocol maintains such clustering hierarchy [2, 3]. In LEACH,
simulations with WSN using both single hop and multi hop
mechanisms for data transfer, and than discuss the relationships
the clusters are re-established in each round. New cluster heads
between the node density, the base station location and the total are elected in each round and as a result the load is well
energy dissipation per round. distributed and balanced among the nodes of the network.
Keywords – WSN, Energy dissipation, Cluster, Cluster Head Moreover each node transmits to the closest cluster head so as
to split the communication cost to the sink [4]. Only the cluster
I. INTRODUCTION head has to report to the base station and may expend a large
amount of energy, but this happens periodically for each node.
Sensor networks play a major role in many aspects of In LEACH there are an optimal predetermined percentage
society including home automation, consumer electronics, p opt of nodes that has to become cluster heads in each round
military application, agriculture, environmental monitoring, assuming uniform distribution of nodes in space. Since the
health monitoring and geophysical measurement. Usually nodes are homogeneous, which means that all the nodes in the
sensor devices are small and inexpensive, so they can be field have the same initial energy, the LEACH protocol
produced and deployed in large numbers. Their resources of guarantees that everyone of them will become a cluster head
energy, memory, computational speed and bandwidth are exactly once every 1 / popt rounds. Throughout this paper we
severely constrained. Therefore, it is important to design refer to this number of rounds as epoch of the clustered sensor
sensor networks aiming to maximize their life expectancy. network. Initially each node can become a cluster head with a
Different aspects of sensor networks such as data aggregation probability popt [5]. On average, n × p opt nodes must become
or fusion, packet size optimization, target localization, design
challenges, network protocols are discussed in the literature cluster heads per round per epoch. Nodes that are elected to be
with respect to crucial energy limitations and network lifetime cluster heads in the current round can no longer become cluster
maximization [1]. heads in the same epoch. The non-elected nodes belong to the
This paper presents a study focused on the total energy set G and in order to maintain a steady number of cluster heads
dissipation of a sensor network per a communication round. per round, the probability of nodes ∈ G to become a cluster
For this purpose we have simulated a WSN with a various head increases after each round in the same epoch. The
number of sensors and with a base station, which coordinates decision is made at the beginning of each round by each node
are first at the center of the sensor field, than at the edge of the s ∈ G independently choosing a random number in [0,1]. If the
field and finally outside of the field. random number is less than a threshold T (s ) then the node
The rest of the paper is organized as follows. In Section 2 becomes a cluster head in the current round. The threshold is
we have provided the communication model, which is being set as:
used, together with the energy model. Section 3 presents the ­ popt
performance measures used in this paper. Section 4 provides ° if s ∈G
° § ·
the simulation scenarios and the various settings we use for the T (s ) = ®1 − popt ¨ r mod 1 ¸ (1)
simulations. In Section 5 we provide the results of the ° ¨ popt ¸¹
©
conducted simulations as well as a discussion on the obtained ° 0 if s∉G
¯
results. Section 6 contains the conclusions on the conducted
studies and directions for future work. where r is the current round number. The election probability
of nodes ∈ G to become cluster heads increases in each round
II. MODEL OVERVIEW in the same epoch and becomes equal to one in the last round
of the epoch. Note that by round we define a time interval
In this section we describe our model of a wireless sensor where all clusters members have to transmit to the cluster head
network with nodes, homogenous in their initial amount of once. For the purpose of our study we use the energy model
presented in Figure 1.

539
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_94, © Springer Science+Business Media B.V. 2010
540 ZAHARIEV ET AL.

III. PERFORMANCE MEASURES

d

We define here the measures used in this paper to evaluate

B bit
ETX (B,d) ERX(B)
B bit
the impact of the node density and base station location.
packet Transmit Receive packet
Amplifier
Electronics Electronics

• Node density – since we assume a constant area of 100

x 100 square meters, this is the relationship between the
Fig.1 Radio Energy Dissipation Model number of nodes in this field and the surface of the
field. Since the surface of the field is constant, higher
According to the radio energy dissipation model illustrated node density means less distance between nodes [9].
in Figure 1, in order to achieve an acceptable Signal-to-Noise • Round – the time interval for which all sensor nodes in
Ratio (SNR) in transmitting a message with size of B bits, a cluster transmit their packets to the cluster head.
over a distance d , the energy expended by the radio is given
by: • Cluster head energy – the energy dissipated by a cluster
head for one round. This energy can be obtained using
­° B × Eelect + B × E fs × d 2 equation (3), and is equal to the sum of the energy for
if d < do
ETX (B, d ) = ® (2) receiving B bits from all sensor nodes in the cluster,
°̄ B × Eelect + B × E ms × d 4 if d ≥ do the energy for data aggregation of the received B bits
from all sensor nodes in the cluster and the energy for
where E elect is the energy dissipated per bit of the transmitter transmitting B bit of information to the base station or
a cluster head closer to the base station.
or receiver, E fs and E ms depend on the transmitter amplifier
model used and d is the distance between the sender and the • Sensor node energy – the energy dissipated of a regular
receiver [6]. By equating the two expressions at d = d 0 , we sensor in the cluster. The value of this measure can be
obtained using equation (4), and is equal to the energy
E fs of transmitting B bits to the cluster head.
have d 0 = . The energy dissipated to receive a message
 ms
• Total energy per round – this measure is used to
with size B bits can be given by E RX ( B) = B × Eelect [7]. describe the whole energy dissipation by all devices in a
For our simulation studies we assume an area of wireless sensor network per round. The value of this
100×100 square meters with n number of nodes that are measure can be obtained using equation (5).
randomly distributed over the area. The energy dissipated in
every cluster head node can be given using the following IV. SIMULATION SCENARIOS
equation:
The goal of our study is to evaluate the impact of node
§ §n · n· density and base station location in wireless sensor networks.
E CH
= B × ¨¨ E elect ¨ − 1¸ + E DA ¸¸ + ETX (B, d BS ) , (3)
© © k ¹ k ¹ In order to complete our goal we have conducted several
simulations with both single hop and multi hop scenarios. We
where k is the number of cluster, E DA is the energy have created two separate MatLab models, one for the single
hop (Figure 2.) and one for the multi hop scenario (Figure 3.).
consumption for processing of a bit per signal and d BS is the
distance between the cluster head and the base station [8]. The
energy dissipation of a node in the wireless sensor network can
be give using the following:

E Sensor = ETX (B, d CH ) ,

CHn
(4)
CH2

where d CH is the distance between the sensor node and the

cluster head. The total energy dissipation in the network per BS
round can be given by:
N SH
CH1 2

n−k k
ETotal = ¦ EiSensor + ¦ ECH
j (5)
1
BS

i =1 j =1

Fig.2 Single hop wireless sensor network

 STUDY ON THE IMPACT OF NODE DENSITY AND SINK LOCATION IN WSN 541

Base station x=50, y=50

1

0.9

Total energy dissipation per round, E Total, J

0.8

CHn 0.7
Multi hop
CH2 0.6

0.5

0.4
BS
0.3

MH
CH1 N 0.2
2
0.1 Single hop

BS 0
0 100 200 300 400 500 600 700 800 900 1000
1
Number of nodes, n

Fig.3 Multi hop wireless sensor network Fig.4 Total energy dissipation of the network, when the base station is in the
center of the network

The area we use is 100 ×100 square meters and we randomly

deploy n sensors in this area, where 20 ≤ n ≤ 1000 . For all the multi hop scenario increases greatly compared to the level
simulations we assume that the nodes are homogenous in their of increase of the energy dissipation in the single hop scenario,
initial amount of energy and that the size of the transmitted due to the many retransmissions between cluster heads.
information is B = 4000 bits. Initially we place the base station The simulation results obtained with the base station at the
at x = 50, y = 50 meters, which is in the center of the area. For edge ( x = 50 , y = 100 ) of the sensor network are given on
Figure 5. It is easy noticeable that the value of the total energy
the second scenario we move the base station at
consumption for the single hop scenario is almost not affected
x = 50, y = 100 meters, which is at the edge of the area. For
by the repositioning of the base station. The reason for this is
the last two scenarios the base station is outside the sensor that only few cluster heads have distance to the base station
network area at coordinates x = 50, y = 200 and larger than d 0 .
x = 50, y = 400 meters, respectively. For all four scenarios we The total energy dissipated in the multi hop scenario, on the
record the information about the performance measures of other hand, is greatly affected by the repositioning of the base
interest, shown above. The rest of the parameters used are station. Analyzing the basic idea behind multi hop, we can
given in Table 1. come to the conclusion that, by moving the base station from
the center to the edge of the network, the number of hops, for
TABLE I the data transmitted from the cluster heads close to the other
SIMULATION PARAMETERS
edges of the network, will increase by approximately two
Parameter Description Value
times. This can be observed on Figure 5 and is the main reason
E elect Energy dissipation: electronics 50 nJ/bit for the quite larger value of the energy dissipated by the
E fs Energy dissipation: amplifier 10 pJ/bit/m2 network devices in the multi hop scenario.
Energy dissipation: amplifier 13.10-4 pJ/bit/m2 Figure 6 presents the results for the total energy dissipation
E ms
as the base station is being relocated outside the wireless
B Transmit packet size 4000 Bits
n Nodes number 20 ≤ n ≤ 600 Base station x=50, y=100
2

1.8

V. SIMULATION RESULTS
Total energy dissipation per round, E Total, J

1.6

1.4

On Figure 4 we present the results of the simulation, whit 1.2 Multi hop
the base station in the center of the sensor network 1
( x = 50 , y = 50 ). It can be clearly seen that when the node 0.8
density is relatively small ( n < 100 ), the total energy 0.6 Single hop

consumption of the network for both multi hop and single hop 0.4
scenarios is almost the same. This is due to the fact that the 0.2
distance of every node in the field is smaller than d 0 and due
0
to the fact that the number of clusters is relatively small, which 0 100 200 300 400 500 600
Number of nodes, n
700 800 900 1000

means that there can be only few retransmissions between

cluster heads. With the increase of the number of nodes the Fig.5 Total energy dissipation of the network, when the base station is at the
number of cluster also increases and the energy dissipation for edge of the network
542 ZAHARIEV ET AL.

Base station x=50, y=200 transmit to the base station compared to the single hop
2.4 scenario, where all do. Compared to the results from the
2.2
previous simulation here even the cancellation effect of the
Total energy dissipation per round, E Total, J many hops that the data travels from the edge cluster heads to
2

1.8

1.6
ones closer to the base station shows little effect, due to the
1.4 Multi hop larger distance to the base station.
1.2

1
Single hop VI. CONCLUSIONS AND FURTHER WORK
0.8

The contribution of this paper is a study of the impact of

0.6

0.4

0.2
node density and base station location in wireless sensor
0 networks. From the obtained simulation results it is clear to
notice that with the relocation of the base station away from
0 100 200 300 400 500 600 700 800 900 1000
Number of nodes, n

the sensor network area, the value of the energy dissipated in

Fig.6 Total energy dissipation of the network, when the base station is at the singe hop scenario increases dramatically, compared to the
coordinates (x=50, y=200) multi hop scenario. The node density also has a critical impact
on the total dissipated energy per round. This is especially
sensor network at x=50, y=200, which means that the distance evident in the multi hop scenarios shown on Figures 4, 5, and
of all sensors in the network to the base station will be > d 0 . 6. A potential goal for further research can be the evaluation of
The results shown are quite interesting especially in the part node density and base station location in heterogeneous
where the density of the nodes is relatively small ( n < 100 ). networks and design of a new approach for determining the
range of the clusters and their optimal number.
Unlike the previous results the multi hop scenario shows better
results than the single hop. This can be explaned by the fact
that the cluster number is relatively small and basically only ACKNOWLEDGMENT
few cluster heads will transmit to the base station (the other
will transmit to a closer cluster head towards the base station), This work is a part of the project BG051PO001/07/3.3-02/8–
compared to the single hop scenario where all cluster heads “MEQSIS”, funded by scheme “Support of the development of
will transmit to the base station. Again the situation changes PhD students, postdoctoral, post-graduate and young
with the increase of the number of sensors. The main scientists” from the program “Development of human
difference is that both lines are closer than in the previous resources” of the “European social fund”.
simulations. This is due to the fact that the larger transmission
distance for the cluster heads in the single hop scenario is REFERENCES
partially canceling the effect caused by the many
retransmissions in the multi hop scenario. [1] G. Smaragdakis, I. Matta, A. Bestavros, SEP: A stable Election Protocol
Figure 7 presents the simulation results, obtained after the for heterogeneous wireless sensor networks, In Proceedings of
relocation of the base station even further away from the International Workshop on SANPA, 2004.
[2] W. Heinzelman, A. Chandrakasan, H. Balakrishnan, Energy-efficient
sensor network at x=50, y=400. It is obvious that the total communication protocol for wireless microsensor networks, In
energy dissipated in the single hop scenario is many times Proceedings of the 33rd Hawaii International Conference on System
grater than the one in the multi hop scenario. This is due to the Sciences (HICSS-33), 2000.
fact that in the multi hop scenario only few cluster heads [3] A. Chandrakasan, A. Smith, W. Heinzelman, An application-specific
protocol architecture for wireless microsensor networks, In IEEE
Transactions on Wireless Communications, vol. 1, pp. 660-670, 2002.
9
Base station x=50, y=400
[4] I. Akyildiz, W. Su, Y. Sankarasubramaniam, E. Cayirci, A survey on
sensor networks. IEEE Communications Magazine, vol. 8, pp.102–114,
8
August 2002.
Total energy dissipation per round, E Total, J

7 [5] S. Bandyopadhyay, E. J. Coyle, An energy efficient hierarchical

Single hop
clustering algorithm for wireless sensor networks. In Proceedings of the
6 22nd Annual Joint Conference of the IEEE Computer and
5
Communications Societies (Infocom 2003), April 2003.
[6] S. Bandyopadhyay and E. J. Coyle. Minimizing communication costs in
4 hierarchically-clustered networks of wireless sensors, Comput. Networks,
3 Multi hop
vol. 1, pp. 1–16, January 2004.
[7] C.-Y. Wan, A. T. Campbell, L. Krishnamurthy, PSFQ: A reliable
2 transport protocol for wireless sensor networks, In Proceedings of the
2nd IEEE Workshop on Applications and Services in Wireless Networks
1
(WSNA 2002), pp. 1–11, July 2002.
0
0 100 200 300 400 500 600 700 800 900 1000
[8] C. Lin , M. Gerla, Adaptive clustering for mobile wireless networks, In
Number of nodes, n IEEE Journal on Selected Areas in Communications, vol. 15 pp. 1265-
1275, 1997.
[9] E. H. Callaway, Wireless Sensor Networks, Architectures and Protocols,
Fig.7 Total energy dissipation of the network, when the base station is at Auerbach Publications, 2004.
coordinates (x=50, y=400)
 An Algorithm for Synthesis of Families of Generalized
Orthogonal Complementary Pairs
B. Y. Bedzhev, Member, IEEE
University of Shumen “Bishop Konstantin Preslavsky”
115 Universitetska Str.
Shumen, Bulgaria 9712, EU
M. P. Iliev
University of Ruse “Angel Kanchev”
8 Studentska Str.
Ruse, Bulgaria 7017, EU

Abstract-Families of generalized orthogonal complementary A promising candidate for employment in the future wireless
pairs (FGOCPs) are called families of pairs of radio signals, which communication systems are the so – named Families of
autocorrelation functions (ACFs) and cross-correlation functions generalized orthogonal complementary pairs (FGOCPs). They
(CCFs) among every two pairs of a family have so-called zero are families of pairs of radio signals, which ACFs and CCFs
correlation windows (ZCWs) in the area around the zero time shift.
They allow the users to communicate practically without multiple
among every two pairs of a family have so-called zero correlation
access interferences in a multipath environment if the wireless windows (ZCWs) in the area around the zero time shift [4], [5],
communication systems works in a quasisynchronous mode. As a [6]. They allow the users to communicate practically without
result the employment of the FGOCPs in the future wireless MAI in a multipath environment if the wireless communication
communication systems will significantly increase the data systems work in a quasisynchronous mode. It should be
transmission rate. Due to this reason in the paper an algorithm for mentioned that the so – named families of orthogonal
synthesis of FGOCPs is proposed. It distinguishes by the fact that it complementary codes (FOCCs) provides possibility the users to
provides small level of the lobes of the CCFs without of the ZCW, communicate practically without any SI and MAI in a truly
which is a valuable feature. asynchronous mode [1], [4]. Anyway, in a comparison with
Index Terms – Synthesis of signals, Families of quasiorthogonal FOCCs valuable advantages of the FGOCPs are the simplicity of
complementary pairs. their technical realization and the higher resistance to the
frequency selective fading.
I. INTRODUCTION
With regard to above described situation, our paper aims to
Today users and industry expect a significant enhancement of develop an algorithm for synthesis of FGOCPs, which
the quality of the services and the data transmission rate, offered distinguishes by the fact that it provides small level of the lobes
by the wireless communication systems. This is a very hard of the CCFs without of the ZCW. The importance of the studied
technical challenge for the theorists and developers. From the problem results from the following facts. First, today it is
information in the open literature [4], [5], [9] it can be concluded possible to provide a quasisynchronous mode of operation of the
that a critical role for the whole performance of the wireless wireless communication systems by a real time measurement of
communication systems has the system of employed radio the location of the users relatively to the base stations of the cells.
signals. Due to this reason great efforts are directed to finding of Second, despite of taken efforts many problems in the field of the
families of radio signals possessing both autocorrelation synthesis of the FGOCPs are open still [4], [5], [6].
functions (ACFs) with a small level of the side-lobes and small Paper is organized as follows. First, the basics of the FGOCPs
cross-correlation functions (CCFs) among all pairs of members are recalled. Second, our algorithm for synthesis of FGOCPs is
of a family. The focusing of the research activities on this suggested. At the end some important conclusions are given.
problem can be explained by the following facts [1], [2], [4], [5],
[6], [9], [10], [11], [13], [14], [15], [16], [17]. First, the pointed II. BASICS OF THE SYNTHESIS OF THE FAMILIES OF
out families of radio signals allow the so-named self-interference GENERALIZED ORTHOGONAL COMPLEMENTARY PAIRS
(SI), caused by multipath spreading of electromagnetic waves, to Let K ≥ 2, N ≥ 1, be integers and let us consider a family
be reduced by a separate processing of the direct and reflected W(K, 2, N, Z0), comprising K pairs as every pair consists of two
signals. Second, it is possible the negative effect of simultaneous sequences with (code) length N:
transmission of numerous users, named multi user interference
(MAI), to be minimized.

543
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_95, © Springer Science+Business Media B.V. 2010
544 BEDZHEV AND ILIEV

{ak (m, n)}nN=−01 = (1)

Ak , k = 0,1, ..., K − 1 is the complex envelope of the n-th
= {ak (m, 0), ak (m, 1), ..., ak (m, N − 1)}, m = 0, 1 elementary pulse of the m-th sequence. In fact the m-th sequence
describes mathematically the law of the amplitude and phase
modulation of the m-th subcarrier frequency during the every
Then W(K, 2, N, Z0) is called family of generalized orthogonal
clock interval in the process of the spreading of the spectrum of
complementary pairs (FGOCPs) if the aggregated ACFs of the
the radio signals. Most often only phase manipulation (Q-phase
sequences of every pair and the aggregated CCFs of the
shift keying) is used, i.e.
sequences of every two pairs have so-called zero correlation
windows (ZCWs) in the area around the zero time shift [4], [5],
[6]. ∀ak (m, n) ∈ {exp[(2πil ) / Q]; l = 0, 1, ..., Q − 1}, i = − 1 , (5)
These conditions can be mathematically described as follow:
but this restriction will not be used in the sequel.
­2.N , k = l ∩ r = 0; It should be pointed out that for a FGOCP the number of the
° subcarrier frequencies is M = 2 . Due to this reason, the both
R Ak , Al (r ) = ® 0, k ≠ l ∩ r = 0; (2) signals of every complementary pair can be simultaneously
° 0, 0 < r ≤ Z , conveyed to the receivers by means of so – named quadrature
¯ 0
modulation [4]. This peculiarity is a very valuable feature of the
where: FGOCPs, because it makes possible to use only one carrier
- r is the time shift, r = −( N − 1), ( N − 2), ..., − 1, 0,1, ..., N − 1 ; frequency with two orthogonal phase shifts (0 and π / 2
- Z0 is the width of the ZCW; respectively) instead of two subcarrier frequencies. As a result,
- Ak , k = 0,1, ..., K − 1 is the matrix, denoting the k-th pair of the technical complexity of the wireless communication system
and the negative influence of the so – named Doppler shift of the
the family W(K, 2, N), i.e.: subcarrier frequencies, caused by the moving of the users, can be
significantly reduced. Consequently, from a philosophical point
ªa (0, 0) ak (0,1) ... ak (0, N − 1)º of view the FGOCPs combine the advantages of the orthogonal
Ak = « k » (3)
¬ ak (1, 0) ak (1,1) ... ak (1, N − 1) ¼ complementary codes [1], [4] and the simplicity of the existing
direct spreading (DS) CDMA wireless communication systems
- R Ak , Al (r ) is the aggregated CCF (ACF if k = l) of the rows [4], [5], [6], [9]. All these arguments explain the interest to the
methods for synthesis of FGOCPs.
of the k-th and l-th pairs (matrices), k , l = 0,1, ..., K − 1 . For brevity but without loss of generality we shall suppose that
The aggregated R Ak , Al (r ) CCF (ACF if k = l) is evaluated by 0 ≤ r ≤ N − 1 in (3). Then, after changing the order of
summations, (4) can be transformed as follows:
the well known formula [1], [2], [3], [4], [7], [8], [9], [15], [16]:
N −1− r 1
R Ak , Al (r ) = R Ak , Al (r ) = ¦ ¦ ak (m, n).al* (m, n + r ) =
n =0 m =0
­ 1 N −1− r (6)
N −1− r
° ¦ ¦ ak* (m, n).al (m, n + r ), − ( N − 1) ≤ r ≤ 0, (4)
°
= ®m = 0 1n =N0 −1− r
= ¦ CAk (n) ⊗ CAl* (n + r ).
n=0
°
°¯ m¦ ¦ ak (m, n).al* (m, n + r ), 0 ≤ r ≤ N − 1.
=0 n =0 Here CAk (n) is the n-th vector – column of the matrix Ak and
the symbol matrix “ ⊗ ” denotes the inner product of the n-th
Here the symbol “*” means “complex conjugation”. vector – column of the matrix Ak and the complex conjugated
It should be mentioned that FGOCPs are mathematical models
of the radio signals for the code-division multiple access (n+r)-th vector – column of the matrix Al . Now it should be
(CDMA) wireless communication systems working in recalled that the evaluation of the CCFs can be simplified by
quasisynchronous mode. In such a system K user can means of multiplication of polynomials [3]. More specifically, let
simultaneously communicate practically without SI and MAI if us introduce the following polynomials
the following conditions are available. The k-th set (i.e. the
matrix Ak , k = 0,1, ..., K − 1 ) of the exploited FGOCP is assigned N −1

to the k-th user as its signature code. The entry ak (m, n) ,

Fk ( x) = ¦ CAk (n).x n , k = 0,1, ..., K − 1 , (7)
n =0
m = 0,1, ..., M − 1 , n = 0,1, ..., N − 1 of the matrix
 AN ALGORITHM FOR SYNTHESIS OF FAMILIES 545

N −1 Second, a big part of known algorithms are based on specific

Fl* ( x −1 ) = ¦ CAl* ( j ).x − j , l = 0,1, ..., K − 1 , (8) techniques [4] so that it is hard or impossible to be unified.
j =0
With regard to above shortcomings in this part of the paper an
algorithm for synthesis of FGOCPs will be suggested. It
then the coefficients of their polynomial product Fk ( x).Fl* ( x −1 ) generates FGOCP, which aperiodic correlation properties not
will be the aggregated cross-correlations R Ak , Al (r ) for all time only meet the conditions (2), but also provides small level of the
correlation lobes for all non-zero time shifts. Besides our
shifts r = −( N − 1), − ( N − 2), ..., − 1, 0,1, ..., N − 1 according to (6), algorithm allows a significant part of known today techniques for
i.e.: synthesis of FGOCPs to be viewed from a common theoretical
base.
N −1 The algorithm for synthesis of FGOCPs, presented in the
Fk ( x).Fl* ( x −1 ) = ¦ R Ak , Al (r ).x r (9) sequel, is based on the following proposition.
r = − ( n −1)
Proposition 3.1: Let A be a set of M = 2 complementary
sequences with length N and let D be a matrix
It should be stressed that multiplication of the coefficients in
the left side of (9) is performed as multiplication of vectors, i.e.
ª d (0, 0) d (0,1) ... d (0, K − 1) º
1
« d (1, 0) d (1,1) ... d (1, K − 1) »»
CAk (n) ⊗ CAl* (n + r ) = ¦ ak (m, n).al* (m, n + r ) (10) D=« (11)
« ... ... ... ... »
m=0
« »
¬d ( K − 1, 0) d ( K − 1,1) ... d ( K − 1, K − 1)¼
and due to this reason it is possible CAk (n) ⊗ CAl* (n + r ) = 0 .
Besides, the coefficients R Ak , Al (r ) of the right side of (9) satisfy which columns are mutually orthogonal, i.e.
the conditions (2). K −1 ­K , n = l;
III.
AN ALGORITHM FOR SYNTHESIS OF FAMILIES OF ¦ d (k , n).d * (k , l ) = ® 0, n ≠ l , (12)
k =0 ¯
GENERALIZED ORTHOGONAL COMPLEMENTARY PAIRS
The complementary codes were invented by M. Golay in 1949. for 0 ≤ n, l ≤ K − 1 . Then the row - submatrices of the tensor
In his origin works [7], [8] K = 1, M = 2 and product
∀a (m, n) ∈ {− 1, + 1}, m = 0,1, n = 0,1, ..., N − 1 , which gave the
reason these codes to be named complementary pairs. Later, in ª d (0, 0). A d (0,1). A ... d (0, K − 1). A º
1972, Tseng and Liu [16] extended the Golay’s conception and « d (1, 0). A d (1,1). A ... d (1, K − 1). A »»
introduced the so - named mutually orthogonal complementary E=« (13)
sets, consisting of M ≥ 2 complementary sequences. « ... ... ... ... »
« »
Due to their unique correlation properties the complementary ¬d ( K − 1, 0). A d ( K − 1,1). A ... d ( K − 1, K − 1). A¼
codes have been intensively studied. For instance the study in [4]
is based on over 300 journal papers and conference reports,
are a FGOCP with parameters: family size – K, length of the
concerning this theme. The complementary codes have also
sequences – N’ = K.N, width of the ZCW – Z0 = N.
exploited in some types radars and specialized communication
systems and obtained experience has proved the practical Proof: According to (7) and (8) let F (x) and F * ( x −1 ) be the
effectiveness of these signals. As a result at the beginning of the polynomials of (N-1)-th degree which coefficients are the vector
new century it was suggested the FGOCPs, described in the – columns of the matrix A. As A is a set of M = 2 complementary
previous section, to be the one of the base tools for multiple sequences with length N ([4], [7], [8]):
access to the communication channel in the next generations
wireless communication systems [4], [5], [6], [9]. F ( x).F * ( x −1 ) = M .N = 2.N . (14)
At the moment the used algorithms for synthesis of families of
signals with ZCWs have the following shortcomings.
Analogously, let H m (x) and H l* ( x −1 ) be the polynomials of
First, very often these algorithms generate only periodic
signals with desirable correlation properties [5], [6], [9]. It should (K-1)-th degree which coefficients are the entries of the m-th and
be stressed that in the real wireless communication systems the l-th rows of the matrix D respectively:
aperiodic correlation properties of the exploited signals are much
more important [4].
546 BEDZHEV AND ILIEV

N −1 On the base of the Proposition 3.1, FGOCPs can be

H m ( x) = ¦ d (m, n).x n , m = 0,1, ..., K − 1, synthesized by the following Algorithm.
n=0 Algorithm 3.2:
N −1
, (15)
1) Choose an initial complementary pair A:
H l* ( x −1 ) = ¦ d * (l , j ).x − j , l = 0,1, ..., K − 1
j =0
ªa(0, 0) a(0,1) ... a(0, N − 1)º
A=« » (22)
Using (15), the CCF of the m-th and l-th rows of the matrix D ¬ a(1, 0) a(1,1) ... a(1, N − 1) ¼
can be presented in the following polynomial form:
so that
K −1
PH m , H l ( x) = H m ( x).H l* ( x −1 ) = ¦ RH m , H l (r ).x . (16)
r
N = Z0 ,
r = − ( k −1)

a(m, n) ∈ {exp[(2πil ) / q]; l = 0,1, ..., q − 1}, i = − 1, (23)

It is not hard to prove that the rows of the square matrix D are
m = 0,1, n = 0,1, ..., N − 1
mutually orthogonal as a consequence of (12). Due to this reason

R H m , H l ( 0) = 0 . (17) In (23) Z0 is the desirable width of the ZCW, q should be a

divisor of Q, where Q denotes the type of phase manipulation,
which will be used in the developed wireless communication
Let REm (k ) be the m-th row – submatrix of the tensor product system.
E: 2) Choose a square orthogonal matrix D, which size is
equivalent to the desirable number K of the users of the
REm (k ) = [d (m, 0). A d (m,1). A ... d (m, K − 1). A] . (18) developed wireless communication system. Besides, the entries
of the matrix D must belong to the set
After taking into account (14) - (18), the CCF of the m-th and
l-th row - submatrices REm (k ) and REl (k ) of the tensor product d (m, n) ∈ {exp[(2πil ) / s ]; l = 0,1, ..., s − 1}, i = − 1,
(24)
(13) can be evaluated as follows m = 0,1, ..., K − 1, n = 0,1, ..., K − 1,

ª K −1 ª K −1 º where s is a divisor of Q.
k .N º
« ¦ d (m, k ).F ( x).x » « ¦ d * (l , j ).F * ( x ) x
−1 − j . N
»= It should be mentioned that in the open literature detailed
¬k =0 ¼ ¬« j = 0 ¼» tables with complementary pairs and orthogonal matrices are
given [4], [8].
ª K −1 º ª K −1 º
= F ( x).F * ( x −1 ) « ¦ d (m, k ).x k . N » « ¦ d * (l , j ).x − j. N » (19) 3) Form the tensor product E, according to (13).
¬k =0 ¼ ¬« j = 0 ¼» The rows of E are a FGOCP with parameters: family size – K,
length of the sequences – N’ = K.N, width of the ZCW – Z0 = N.
= 2 N .H m ( x N ).H l* ( x − N ) = 2.N .PH m , H l ( x N ).
IV. RESULTS
Equation (19) proves the Proposition 3.1, because the It should be stressed that the above Algorithm 3.2 gives the
coefficients of the polynomial in the right side of (19) satisfy the valuable opportunity to minimize correlation lobes without of the
following restrictions ZCW of the synthesized FGOCPs by an appropriate selection of
the matrix D. With regard to the known upper bounds of the level
of the lobes of the CCFs of the binary sequences [12], an
­° 0, s ≠ r.N ;
2.N .RH m , H l ( s) = ® exhaustive examination of the correlation properties of the square
°̄2.N .RH m , H l (r ), s = r.N , (20) orthogonal matrices have been conducted. As a result we have
s = −( K − 1).N , − ( K − 1).N + 1, ..., − 1, 0,1, ..., ( K − 1).N found that the most appropriate matrices D are these, which are
generated by the so – named Frank’s method [2]. This conclusion
and issues from the following propositions.
Proposition 4.1: Let g be a K-th primitive root of the unity:
­ K , r = 0 ∩ l = m;
RH m , H l ( r ) = ® (21) g = exp[(2πil ) / K ], i = − 1 (25)
¯ 0, r = 0 ∩ l ≠ m,
where l and K are relatively prime numbers.
as the rows of the matrix D are mutually orthogonal.
 AN ALGORITHM FOR SYNTHESIS OF FAMILIES 547

Then the square matrix, obtained by the Frank’s method: K −1− ( K − r )

RCG n ,CGl ( K − r ) = ¦ g k .n .g − ( k + r ).l ≤
k =0
ª g ( 0.0 )
g ( 0.1)
... g ( 0.( K −1)) º . (32)
« (1.0) » r −1
K
G=«
g g (1.1) ... g (1.( K −1)) »
(26)
≤ ¦ g k .n .g − ( k + r ).l = r ≤
2
« ... ... ... ... » k =0
« (( K −1).0) »
«¬ g g (( K −1).1) ... g (( K −1).( K −1)) »¼ This proves the Proposition 4.2.
In fact, we have found that the aperiodic ACFs and CCFs of
the rows (columns) of the matrix (26) have side – lobes which are
has cyclically orthogonal rows (and columns) for all shifts.
smaller than the upper bound, found in the Preposition 4.2.
Proof: Let us consider the periodic CCF of the n-th and l-th
Namely, for large number of values, applicable in the practice,
columns of the matrix (26) for an arbitrary shift r = 0, 1, ..., K − 1
the maximum side – lobes do not exceed K .
This conclusion will be explained by the following example.
K −1
Example 4.3: Let K = 4. Then the matrix (24) has the form:
PCGn ,CGl (r ) = ¦ g k .n .g − ( k + r ).l =
k =0
(27)
K −1 ­° K .g − n.r , n = l ; ª1 1 1 1 º
= g − l .r ¦ g (n −l ).k =®°̄ 0, n ≠ l.
«1 − i − 1 i »
k =0 G=« », i = − 1 (33)
«1 − 1 1 − 1»
Equation (27) proves the Proposition 4.1. « »
¬1 i − 1 − i ¼
As the rows (and columns) of the matrix (26) are cyclically
orthogonal for all shifts, the matrix G satisfies the condition (12).
Consequently, we can use the matrix D = G in the Algorithm 3.2, The lobes of the aperiodic CCFs of the rows of the matrix (33)
in order to synthesize a FGOCP. are given in the following table:
Proposition 4.2: The lobes of the aperiodic CCFs of the
columns (rows) of the matrix G do not exceed K / 2 . TABLE I
Proof: From (4) and (27) follows that the periodic CCF of the LOBES OF CCFS OF THE ROWS OF THE MATRIX (33)
n-th and l-th columns of the matrix (26) is r -3 -2 -1 0 1 2 3
R1,2 1 1+i i 0 -1 -1-i i
PCG n ,CGl (r ) = RCG n , CGl (r ) + RCG n , CGl ( K − r ) . (28) R1,3 1 0 1 0 1 0 1
R1,4 1 1-i i 0 -1 -1+i i
R2,3 i -1-i 1 0 -i 1+i -1
Here RCGn ,CGl (r ) and RCGn ,CGl ( K − r ) are the aperiodic
R2,4 i 0 i 0 -i 0 i
CCFs of the n-th and l-th columns of the matrix G for shifts r and R3,4 -1 1+i -i 0 1 -1-i i
K-r respectively. Without loss of generality let us suppose that
In Table I r is the time shift and Rk,l is the CCF of the k-th and
r ≤ K −r . (29) l-th row of the matrix (33).

Consequently V. CONCLUSION
In the paper an algorithm for synthesis of FGOCPs is
K suggested on the base of Proposition 3.1. It allows the users to
r≤ . (30) communicate practically without MAI in a multipath
2
environment if the wireless communication systems work in a
From (27) and (28) follows: quasisynchronous mode. It should be stressed that a
quasisynchronous mode can be provided by a measurement of
the locations of the users respectively to the base station and
− RCG n ,CGl (r ) = RCGn ,CGl ( K − r ) . (31)
introducing a time schedule for sending messages to the base
station. This is not a hard technical task and as a result it is
The magnitude of the right side of (31) can be estimated from possible to exploit quasisynchronous mode in the next
the chain of inequalities: generations wireless communication systems.
The FOCCs provides possibility the users to communicate
practically without any SI and MAI in a truly asynchronous mode
548 BEDZHEV AND ILIEV

[1], [4], [5], [6]. Anyway, in a comparison with FOCCs valuable REFERENCES
advantages of the FGOCPs are the simplicity of their technical [1] B. Y. Bedzhev and M. P. Iliev, “A general method for synthesis of families of
realization and the higher resistance to the frequency selective orthogonal complementary codes,” unpublished
fading. [2] J. J. Benedetto and J. J. Donatelli, “Ambiguity function and frame-theoretic
properties of periodic zero-autocorrelation waveforms,” IEEE J.of Selected
The algorithm for synthesis of FGOCPs, suggested in the Topics in Signal Processing, vol.1, No.1, pp. 6-20, June 2007.
paper, has the following positive features. [3] R .E. Blahut, Theory and Practice of Error Control Coding, Addison-Wesley,
First, it gives a general approach to the synthesis of the Reading, Mass., USA, 1987
FGOCPs. Namely it is a direct construction for generating [4] H.-H. Chen, Next Generation CDMA Technologies, John Wiley & Sons,
Chichester, West Sussex, England, 2007
FGOCPs, providing both the desirable number of users K and the [5] P. Fan, “Spreading sequence design and theoretical limits for
width Z0 of the ZCW. quasisynchronous CDMA systems,” EURASIP J. on Wireless
Second, in contrast with large number of constructions [5], [6] Communications and Networking, pp. 19 – 31, no. 1, 2004
it is developed on the base of the aperiodic correlations which is [6] L. Feng, P. Fan, X. Tang, and K.-K. Loo, “Generalized pairwise
Z-complementary codes,” IEEE Signal Processing Letters, vol. 15, pp. 377-
the typical case for the present wireless communication systems. 380, 2008
Third, it allows minimizing of the correlation lobes without of [7] M. J. E. Golay, “Multislit spectroscopy,” J. Opt. Soc. Amer., vol. 39, pp. 437 –444,
the ZCW of the synthesized FGOCPs by an appropriate selection 1949
of the matrix D. For example, if we use matrix D = G in [8] M. J. E. Golay, “Complementary series,” IRE Trans. Inform. Theory, vol. 7,
pp. 82–87, £2, 1961.
Algorithm 3.2, as G is generated by the so – named Frank’s [9] L. Hanzo, M. Munster, B. J. Choi, and T. Keller, OFDM and MC-CDMA for
method [2] (i.e. by (26)), then the maximum side – lobes do not Broadband Multi-User Communications, WLANs and Broadcasting, John
Wiley & Sons and IEEE Press, New York, NY, USA, 2003
exceed K .N . Here K is the size of the matrix D, determining
[10] S. P. Maity, M. Mukherjee, “A high capacity CI/MC-CDMA system for
the number of the users of the developed wireless communication reduction in PAPR,” in Proc. Seventh Annual Communication Networks and
system, N is the length of the used pair of complementary Services Research Conference, pp.297-304, Moncton, New Brunswick,
sequences and K.N is the amplitude of the central peak of the Canada, 11-13 May, 2009
[11] A. Milev and V. Mutkov, “Research over correlation properties of families
aperiodic ACFs of the FGOCP, synthesized by the optical orthogonal codes,” in Proc. International Conference on
Algorithm 3.2. CompSysTech’09, pp. III.20.1- III.20.6, Ruse, Bulgaria, EU, June 2009
With regard to all above stated we hope the algorithm for [12] D. Y. Peng and P. Z. Fan, “Generalized Sarvate bounds on periodic
synthesis of FGOCPs, suggested in the paper, to be useful in the autocorrelations and crosscorelations of binary sequences,” Electronics
Letters, vol. 38, no. 24, pp. 1521 – 1523, 2002
process of development of the new generation wireless [13] T. R. Qureshi, M. D. Zoltowski and R. Calderbank, “A MIMO-OFDM
communication system, which aim to provide a radical channel estimation scheme utilizing complementary sequences,” in Proc.
enhancement of the data transmission rate. The 2009 IEEE International Conference on Acoustics, Speech and Signal
Processing, pp.2677-2680, Taipei, Taiwan, 19-24 April, 2009
ACKNOWLEDGMENT [14] R. G. Tsakov, V. A. Mutkov, N. R. Nikolov and L. . Staneva, “A survey of
The authors wish to thank Prof. Dr Kunchev, Prof. E. the correlation properties of the generalized Barker codes”, unpublished
[15] R. G. Tsakov, V. A. Mutkov, N. R. Nikolov and L. . Staneva, “An
Ferdinandov, Prof. Dr H. Hristov and Prof. Dr B. Tsankov. improved method for synthesis of families of Costas arrays”, unpublished
Finally, the authors appreciate the thoughtful critique by the [16] C.-C. Tseng and C. L. Liu, “Complementary set of sequences,” IEEE Trans.
referees. Inf. Theory, vol. 18, pp. 206–209, Sept. 1972.
[17] M. D. Zoltowski, T. R. Qureshi and R. Calderbank, “Complementary codes
based channel estimation for MIMO-OFDM systems,” in Proc. Forty-Sixth
Annual Allerton Conference, pp. 133-138, Allerton House, UIUC, Illinois,
USA, 23-26 Sept., 2008
 Low Message Overhead Clustering Algorithm
for Wireless Sensor Networks
Batista, Othon Marcelo Nunes; Giozza, William Ferreira

Grupo de Redes Ópticas e Wireless (GROW) – Universidade Salvador (UNIFACS)

Rua Ponciano de Oliveira, 126 – Rio Vermelho
CEP: 41.950-275 – Salvador – BA – Brazil
[email protected], [email protected]

Abstract—A Wireless Sensor Network (WSN) is defined by a deaths occur from the farthest to the nearest ones. Direct
set of sensor nodes and a Base Station (BS). Typically, sensor diffusion forms a plain topology.
nodes are distributed over a geographical area, sense some Another way to send information to the BS is via multi hop
information, process it and send it to the BS via a wireless link.
routing. Each sensor must configure its transmission power to
Since sensor nodes are small, may be distributed in large scale or
in a dangerous area, their battery are small and may not be reach one or more neighbor sensor node as close as possible.
recharged or replaced. So the network lifetime is prolonged when The sensed information is sent to the BS via intermediate
the battery energy is used wisely. The function of the BS is to sensor nodes, in a multi hop way. The energy spent to send an
receive information sensed by the nodes, therefore it's a sink. information to the BS is the summation of the energies spent
One way to send information is to cluster the WSN, dividing the to send the information to the sensor nodes in a route toward
sensor nodes in two exclusive sets: Cluster Heads (CH) or Cluster
the BS. If a sensor node is very near to the BS, it spends more
Members (CM). Therefore, the clustered WSN presents two
types of communication: one from CM to CH, inter-cluster energy than the others, since it aggregates informations from
communication, and other from CH to BS, intra-cluster all other nodes. The sensor nodes deaths occur from the
communication. Once a WSN is clustered, energy consumption nearest to the farthest ones.
and coverage are improved. The first step in clustering a WSN is Coverage is the degree of the geographical area in which
a CH election algorithm. We propose a distributed clustering the sensor nodes were distributed that are in fact being sensed.
algorithm for WSN named Low Message Overhead Clustering
As the sensor nodes die, coverage is gradually lost. When the
Algorithm (LMOCA) that has the goal to spend as minimum
energy as possible on the clustering process. Beyond that it is information is sent in one hop or multi hop, the coverage is
very intuitive and easy to implement. lost really fast, only varying the direction, from the farthest
sensor node to the base station or from the nearest sensor node
Index Terms— Wireless Sensor Networks, Clustering to the base station, respectively.
Algorithm, Simulation Results. Another way to send information to the BS is by clustering
the WSN. Clustering a WSN means to divide it in two sets of
nodes: Cluster Heads (CH) and Cluster Members (CM). This
I. INTRODUCTION division happens also in the communication level: one

A Wireless Sensor Network (WSN) is defined by a set

of sensor nodes and a Base Station (BS). Typically,
communication occurs from CM to CH, intra-cluster
communication, and the other from CH to BS, inter-cluster
sensor nodes are distributed over a geographical area, sense communication. There are several techniques used to cluster a
some information, process it and send it to the BS via a WSN. They may be divided in two groups: localized and
wireless link. Since sensor nodes are small, may be distributed approaches. In localized approaches, nodes decide
distributed in large scale and in a dangerous area, their battery locally, without exchanging messages, if they are CH or CM.
are small and may not be recharged by any means. So the In distributed approaches, nodes decide if they are CH or CM
network lifetime is prolonged when the battery energy is used by exchanging messages with a distributed algorithm. Once a
wisely. The BS may be a computer with a Universal Serial WSN is clustered, energy consumption and coverage are
Bus (USB) dongle to communicate with the sensor nodes. The better than direct diffusion or multi hop routing.
function of the BS is to receive information sensed by the In this sense, we propose a distributed clustering algorithm for
nodes, therefore it's a sink. WSN named Low Message Overhead Clustering Algorithm
The simplest way to send information to the BS is (LMOCA). Its goal is to spend as minimum as possible energy
denominated direct diffusion. As soon as a sensor node senses in the clustering process by exchanging few messages between
any information, it is sent directly to the BS in a one hop sensor nodes. Beyond that, it is very easy to implement, what is
communication. Each sensor node must configure its good for sensor nodes, given their small computing power and
transmission power to reach the BS to implement direct resources.
diffusion. If a sensor node is too far, much energy is spent and
the probability of a collision is bigger. The sensor nodes

549
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_96, © Springer Science+Business Media B.V. 2010
550 BATISTA ET AL.

II. RELATED WORKS A. Assumptions

Over twenty clustering schemes were proposed from 2001 to The clustering algorithm was simulated with this assumptions:
2005. They may be classified in three types: energy-efficient, • sensor nodes are homogeneous;
k-tree and management schemes. They are compared to Time • sensor nodes have different transmission power
Division Multiple Access (TDMA) dynamic cluster scheme. levels, at most 10, but use only one at time;
This scheme is suitable only sensor nodes distributed in an • frequency used is 2.4 GHz;
area forming a grid. Its goals are: to locate CH at the center of • antenna height is 0.03048m;
a cluster, to elect CH periodically to average power • receiver sensitivity is -98dBm;
consumption among sensor nodes and not occur duplicated • Friis and Two ray ground are the propagation models
transmission of data through multiple routes [11]. used;
Low Energy Clustering Hierarchy (LEACH) elects CH with
• all sensor nodes are initially CM;
a probabilistic localized approach. This is very energy
• there are neither collisions or loss of packages, since
effective, since the radio is not used. It uses TDMA in intra
the objective is to test the algorithm;
cluster communications and Direct Sequence Spread
• BS is located at coordinate (0, 0).
Spectrum (DSSS) in inter cluster ones. With LEACH, the
Some of these parameters were gathered from MICA sensor
lifetime of the WSN is divided in turns, each turn is formed by
nodes specifications sheet available at [18].
a cluster head election and some intra and inter cluster
A WSN may be designed as a graph, defined by G = (V, E).
communication. LEACH spends about eight times less energy
V is a finite set of nodes and E is a finite set of edges that
than direct diffusion or multi-hop routing communications
connects nodes from V. To apply this definition to WSN, it
[12].
must be stated that sensor nodes are elements of V and
There is an improvement of LEACH denominated Energy
connections based in the sensor node transmission range are
Efficient Clustering Scheme (EECS) that produces a near
the elements of E. A common assumption is that a sensor node
uniform distribution of CH and balances the loading among
transmission range is a Unit Disk Graph (UDG), a circle with
them. EECS showed to prolong network lifetime as much as
the sensor node located at the center and the limits of its
35% of LEACH [19].
transmission range in the circumference. This defines two
Clustering is not always the best scheme to make good use
distinct areas: with connectivity (inside the circle until the
of energy in WSN. Clusters must be within isoclusters of
circumference) and without connectivity (outside the circle).
monitored phenomenon. An analytical analysis shows that
UDG in wireless networks may be simulated using Friis or
five hops clusters, instead of usual single hop clusters, can
two-ray ground propagation models [8].
provide near to optimal network performance [13].
A comparison of representative clustering techniques and B. Clustering Problem Statement
recent developments in connectivity, Medium Access Control Considering a WSN as a graph G may lead us to define
(MAC) design, cluster head election periodicity, node duty clusters by dividing the network in two complementary sets:
cycle, optimal cluster size and node synchronization are CH set and CM set. CH set is formed by sensor nodes that are
presented in [9]. the logical center of clusters. The quantity of elements in the
Hybrid Energy-Efficient Distributed clustering (HEED) CH set is the quantity of clusters. CH are regular sensor nodes
doesn't make any assumption about node distribution, node with additional responsibility to aggregate information of a
density or node location. It terminates the clustering process in cluster and send it towards BS. In the other hand, elements of
O(1) iterations and is independent of WSN topology or size. It CM set, in therms of communication, only send information to
has a low message exchange and processing cycles and a good its CH.
cluster head distribution across the WSN [14].
An algorithm for electing cluster heads using maximum IV. CLUSTER HEAD ELECTION ALGORITHM
residual energy with the goal is to even distribute energy
The first step in a clustering protocol for WSN is a CH
consumption at the overall WSN to obtain the longest lifetime
election algorithm. Since there may be a huge amount of
is showed in [15]. The lifetime is expressed as to both the
sensor nodes and energy consumption must be minimum, a
maximum last node dying time and the minimum time
distributed approach with few message exchanges is desired.
difference between the last node dying and the first node
LMOCA has the goal to elect cluster heads with a low
dying. Results showed that lifetime can be prolonged when
message exchange overhead. For each node only three
CH are elected with an optimal value of a predefined electing
messages are sent to decide if they are CH or CM. At the
coefficient.
beginning of the process, node degree information is used to
estimate if a sensor node is a CH or a CM. After the cluster
heads are elected, a rearrangement of the network is done, by
III. PROBLEM STATEMENT
grouping CM to their nearest CH. Hence, CM transmission
Initially, we present the the assumptions made in LMOCA and range is optimized to reach only its CH.
state the problem of clustering a WSN.
 LOW MESSAGE OVERHEAD CLUSTERING ALGORITHM 551

The algorithm is divided in five phases: on to get the node degree messages.
phase 1: node degree calculation; Received node degrees are stored in an array in a position
phase 2: node degree exchange; correspondent to the node identification number in the
phase 3: initial clusters; neighbors array. At the end of this phase, all sensor nodes
phase 4: network rearrangement; know their neighbors and neighbour degrees.
phase 5: cluster member transmission range adaptation.
C. Initial Clusters
For testing purposes, the phases were configured to run for
5 seconds each. In the future, we intend to find the best time The initial clusters are formed by nodes with higher degree.
that suits the protocol. Since all sensor nodes are initialized as CM, they must choose
This algorithm was simulated in Castalia, a discrete event a node as a CH, if one exists. Each sensor node, checks its
WSN [16] simulator based in OMNET++ [17]. For a better node degree array and sends a message to the highest degree
explanation, one example is presented in a step-by-step basis sensor node. This message is called DOMINATE ME.
with the aid of the WSN graph shown in Fig. 1. This graph Once a sensor node receives a DOMINATE ME message, it
represents 100 sensor nodes in an area of 900m2 (30m x 30m). changes its status to CH and stores sensor node identification
Sensor nodes are white circles with a number (sensor node number in an array of CM sensor nodes.
identification number). The red circle in coordinate (0, 0) is At the end of this phase if a node neither generate or
the BS. receive a DOMINATE ME message, it self elects CH. This
Sensor nodes have a initial transmission power of 0 dBm. guarantees that, all nodes are either CM or CH at the end of
This transmission power applied to Friis and Two Ray Ground the clustering phase.
propagation models lead to a range of approximately 8.59m. The algorithm could stop after this phase, however, as Fig.
This range indicates the distance until the transmission power 2 shows, the clusters are not well distributed in the
is reduced to the receiver sensitivity of -98 dBm. Also, it geographical area, what will raise the interferences and
indicates the radius of each cluster. collisions between clusters. So, there's a needing to reorganize
The five phases describing the operation of LMOCA are clusters to minimize interference and maximize energy usage.
presented as follows. Notice that cluster heads are the red triangles and cluster
members are blue circles.

Fig. 1. Example WSN Graph.

Fig. 2. Initial CDS of the graph presented in Fig. 1.
A. Node Degree Calculation
The first phase consists of one message sent by every sensor D. Network Rearrangement
node. This message contains only the sensor node The clustered WSN obtained until now has some clusters
identification number. After sending the message, all sensor formed by members very dispersed in the area. So, cluster
nodes keep their receivers on to get messages from the heads must send a message that will be received by all sensor
neighbors. nodes in their range. This message will be used by the cluster
Each message received by a sensor node is counted and the members to verify, by the received signal strength intensities,
node identification numbers are stored in an array of neighbors. which is the nearest cluster head.
The quantity of messages received is the node degree.
E. Cluster Member Transmission Range Adaptation
B. Node Degree Propagation Once the messages of CH are received, the CM adjust their
After calculating degree, each sensor node must propagate it transmission power to cover the distance to the nearest CH
to its neighbors. To do so, a sensor node sends a message with and send a DOMINATE ME message to it. The WSN
the previously calculated degree and the node identification obtained at the end of this phase contains very well distributed
number. Once more, all sensor nodes must keep their receivers
552 BATISTA ET AL.

clusters, which must aid in the future inter and intra-cluster Hence, message complexity for the whole network in the
communications, regarding that the interferences decrease worst case is O(n). n is the total number of nodes in the
with the transmission power adjustment (Fig. 3). network.

Lemma 3. At the end of LMOCA, a sensor node is either a

cluster head or a cluster member.
Proof. At the end of the clustering phase, if a sensor node
neither generate or receive a DOMINATE ME message, it self
elects a CH. This guarantees that, all nodes are either CM or
CH at the end of the clustering phase.
So, at the end of the execution of LMOCA, all nodes are
either CH or CM.

VI. SIMULATION RESULTS

LMOCA was executed 100 times in five different scenarios.
Simulations were executed with sensor nodes located in
different positions. For each scenario, the quantity of CH was
observed to be related to the transmission power. Parameters
Fig. 3. WSN at the final phase of the proposed algorithm.
used in simulations were the same presented in section III of
this paper. The only differences are one additional power level
V. ALGORITHM ANALYSIS
and the quantity of sensor nodes per scenarios: 100, 50, 10
In this section, we present and prove some lemmas regarding and 2, respectively. Graphs 1, 2, 3 and 4 have in axis X the
LMOCA. cluster radius, which is proportional to transmission power
applied to Friis and Two Ray Ground propagation models,
Lemma 1. LMOCA message complexity is O(1) for each and in axis Y the quantity of CH obtained by LMOCA.
node. The results of the first simulated scenario are presented in
Graph 1. In this scenario, 100 sensor nodes were used, each
Proof. The quantity of messages sent by each node is, at most, simulation in a aleatory different position. It shows that the
four. CM nodes send three messages, which are: quantity of CH is inversely proportional to transmission
• an initial message used to calculate node degree; power. For instance, if we increase the transmission power to
• a second message to inform calculated node degree; the maximum, which leads to a cluster radius of 48,31m, we
• a third message to be a member of a cluster decrease the quantity of CH to approximately 2.
(DOMINATE ME);
In the other hand, CH nodes send four messages, which are
these three for CM nodes and and additional message to
construct well distributed clusters.
74,22

80
Therefore, message complexity for each sensor node is
O(1). 70

Lemma 2. LMOCA message complexity over the network is 60 LMOCA Average

Cluster Quantity

O(n). 44,63

50

Proof. According to Lemma 1, in the worst case, a sensor 32,58

node sends four messages until the network is clustered. CM 40

25,84

nodes send three messages and CH nodes send four messages. 22,01
20,62

The total quantity of messages sent by all nodes is given by

30 19,16
17,94

formula 1. In formula 1, QuantityMessages is the total of message 10

11,94

for the network, QuantityCH is the number of CH and

8,12

QuantityCM is the quantity of CM.

2,45

0
2,04 3,62 4,83 5,74 6,44 7,23 8,11 8,59 11,46 15,28 48,31

Graph 1. Quantity of CH after 100 simulations for 100 sensor nodes.

Formula 1.
Quantity of messages sent to
cluster a wireless sensor In Graph 2, results obtained in second simulation are
presented. Scenario for this graph is composed with 50 sensor
network.
nodes in different positions for each simulation. The relation
Quantity Messages= 4*QuantityCH + 3*QuantityCM
of the quantity of CH to be inversely proportional to
transmission power still holds.
 LOW MESSAGE OVERHEAD CLUSTERING ALGORITHM 553

50 1,2

45
43,59

1 1 1 1 1 1 1 1 1 1 1
1
40

35
0,8

30,5 LMOCA Average LMOCA Average

Cluster Quantity Cluster Quantity
30

25 0,6
23,19

19,98
20

16,69
0,4
15 14,49
13,87
13,51

10 9,38

0,2

4,54
5
2,55

0 0
2,04 3,62 4,83 5,74 6,44 7,23 8,11 8,59 11,46 15,28 48,31 2,04 3,62 4,83 5,74 6,44 7,23 8,11 8,59 11,46 15,28 48,31

Graph 2. Quantity of CH after 100 simulations for 50 sensor nodes. Graph 4. Quantity of CH after 100 simulations for 2 sensor nodes.

The results of the third simulated scenario are presented in Although not graphically shown in this paper, the scenario
Graph 3. In this scenario, 10 sensor nodes were used, each with only one sensor node were simulated. All the times the
simulation in a aleatory different position. The quantity of CH quantity of CH were 1. This also shows that LMOCA works
is inversely proportional to transmission power. for this special situation.

VII. CONCLUSION
10

This paper presented a proposal for an algorithm to elect

cluster heads in a WSN. For each sensor node, at most, four
8,94
9

messages are sent to form well distributed clusters, which

8,23
8,09
8 7,83

minimizes the interferences, collisions and, consequently,

7,54
7,33
7,16

saves energy. Some lemmas were proved for the LMOCA and
7
6,64

6
LMOCA Average
Cluster Quantity
simulation results were presented. In these results, we
observed that the quantity of CH depends on the transmission
5,46

5
power. Bigger transmission powers lead to fewer CH.
4 LMOCA is part of a clustering protocol and by this way,
future works consist of finding a CH rotation policy based in
3,47

residual energy and the integration of an intra-cluster and

inter-cluster communication scheme to this algorithm.
2,03
2

REFERENCES
[1] Brooks, R. R., Ramanathan, P. and Sayeed, A. M. (2003) “Distributed
0
2,04 3,62 4,83 5,74 6,44 7,23 8,11 8,59 11,46 15,28 48,31

Target Classification and Tracking in Sensor Networks” Proceedings of

Graph 3. Quantity of CH after 100 simulations for 10 sensor nodes. the IEEE, Vol. 91, NO. 8.
[2] Werner-Allen, G., Johnson, J., Ruiz, M., Lees, J. and Welsh, M. (2006)
The fourth and last scenario analyzed has only 2 sensor “Monitoring Volcanic Eruptions with a Wireless Sensor Network”,
nodes. This is to show that LMOCA runs correctly even with Proceedings of the 7th USENIX Symposium on Operating Systems
Design and Implementation (OSDI 2006). Seattle, USA.
the minimum quantity of sensor nodes. Only one sensor node [3] Liu, T., Sadler, C. M., Zhang, P. and Martonosi, M. (2004)
is CH. The graph 4 shows the results for this situation. “Implementing Software on Resource-Constrained Mobile Sensors:
Experiences with Impala and ZebraNet”, Proceedings of the The Second
International Conference on Mobile Systems, Applications, and Services
(MobiSYS'04). Boston, Massachusetts, USA.
554 BATISTA ET AL.

[4] Krishnamurthy, L., Adler, R., Buonadonna, P., Chhabra, J., Flanigan, [11] Dali, W. and Chan, A. C. (2005) “A Survey on Cluster Schemes in Ad
M., Kushalnagar, N., Nachman, L. and Yarvis, M. (2005) “Design and Hoc Wireless Networks”, Proceedings of the IEEE Mobility Conference.
Deployment of Industrial Sensor Networks: Experiences from a Guang Zhou, China.
Semiconductor Plant and the North Sea”, Proceedings of the 3rd ACM [12] Heinzelman, W. R., Chandrakasan, A. and Balakrishnan, H. (2000)
Conference on Embedded Networked Sensor Systems (SenSys). San “Energy-Efficient Communication Protocol for Wireless Sensor
Diego, California, USA. Networks”. Proceedings of the 33rd Hawaii International Conference on
[5] Szewczyk, R., Mainwaring, A., Polastre, J., Anderson, J. and Culler, D. Systems Sciences .
(2004) “An Analysis of a Large Scale Habitat Monitoring Application”, [13] Vlajic, N. and Xia, D. (2006) “Wireless Sensor Networks: To Cluster or
Proceedings of the 2rd ACM Conference on Embedded Networked Not to Cluster?”, Proceedings of the 2006 International Symposium on a
Sensor Systems (SenSys). Maryland, USA. World of Wireless, Mobile and Multimedia Networks (WoWMoM'06).
[6] Su, H. and Zhang, X. (2006) “Optimal Transmission Range for Cluster- [14] Younis, O. and Fahmy, S. (2004) “Distributed Clustering in Ad-hoc
Based Wireless Sensor Networks with Mixed Communication Modes”, Sensor Networks: A Hybrid, Energy-Efficient Approach”, Proceedings
Proceedings of the 2006 International Symposium on a World of of the. Twenty-third Annual Joint Conference of the IEEE Computer and
Wireless, Mobile and Multimedia Networks (WoWMoM'06). Communications Societies, vol. 1, p. 640.
[7] Heinzelman, W. B. and Chandrakasan, A. P. (2002) “An Application- [15] Wang, W. and Jantsch. A (2006) “A new protocol for electing cluster
Specific Protocol Architecture for Wireless Microsensor Networks”, head based on maximum residual energy”. In Proceedings of the IEEE
IEEE Transactions on Wireless Communications, volume I, Issue 4, Oct International Cross-Layer Designs and Protocols Symposium.
2002, p.p. 660 - 670. [16] Pham, H. N, Pediaditakis, D. and Boulis, A. (2007) “From Simulation to
[8] Kuhn, F., Moscibroda, T. and Wattenhofer, R. (2004) “Unit Disk Graph Real Deployments and Back”. in Proceedings of t2pWSN.
Approximation”, Proceedings of the 2004 Joint Workshop on [17] Omnet++ (2009) “OMNET++ Discrete Event Simulation System”.
Foundations of Mobile Computing, Philadelphia, PA, USA, p.p. 17 - 23. http://www.omnetpp.org.
[9] Younis, O., Krunz, M. and Ramasubramanian, S. (2006) “Node [18] XBow (2009) “Crossbow Technology”. http://www.xbow.com. San
Clustering in Wireless Sensor Networks: Recent Developments and Jose, CA. USA.
Deployment Challenges”, IEEE Network. May/June. [19] Ye, M., Li, C., Chen, G. and Wu, J. (2005) “EECS: An Energy Efficient
[10] Funke, S., Kesselman, A., Meyer, U. and Segal, M. (2006) “A Simple Clustering Scheme in Wireless Sensor Networks”. the 24th IEEE
Improved Distributed Algorithm for Minimum CDS in Unit Disk International Performance, Computing, and Communications
Graphs”, ACM Transactions on Sensor Networks, Vol. 2, No. 3, August, Conference. p.p. 535-540.
p.p. 444-453.
 On the Design of SOHO Networks
Lawan Ahmed Mohammed
King Fahd University of Petroleum and Minerals
HBCC, 31991 Hafr Batin, Saudi Arabia
[email protected]

Abstract — Small networks such as small office/ home office and portable computing nodes into existing fixed
(SOHO) face the same threats as large enterprise networks. infrastructure based networks increases the heterogeneity of
However, they also contend with the challenge of limited the resulting network, making the problem of intrusion
resources and budgets for IT expenditures. In such networks detection even harder especially as these devices can be
environment, the role of administering the system often falls on connected to each other and also to Internet by using wired or
the business owner or on the default setup when the system was wireless technology. For example, wireless technology like
installed. In most cases, the owners do not usually have the time,
Bluetooth, 802.11, IrDA can provide LAN and Internet access
resources or expertise to work on complex security problems.
Similarly, most of the default setup use to have some loopholes
to a handheld computer wirelessly through access points. The
which were not initially identified. This paper describes the mobility aspect of these devices makes it harder to protect
primary security challenges facing SOHO networks today, and them. Further, different components and networking
suggests simple easy to use security solutions to resolve these technologies support different security mechanisms, which
challenges. may not be compatible and which have different security
characteristics. This implies the need for multi vendor’s
Keywords— Network security, Design of SOHO, Flooding attack, device and security management. For instance, usually,
Security measures. Access Point (APs) from manufacturers comes with a set of
default configuration parameters. These default parameters
I. INTRODUCTION leave some loopholes for attacks. For example (depending on
the manufacturer), most APs have a default administrator
The development of cost-effective networking password, SSID, channels, authentication/encryption settings,
technologies has provided large amount of new possibilities SNMP read/write community strings, etc. Since these default
for the users in small office and home (SOHO) environments. values are available in user manuals, vendor’s website and
Some of the benefits of these networks include: Improved installation guide, they are well known to the general public
efficiency and access to mobile employees, Offers mobility and may be used by hackers to compromise WLAN security.
and flexibility, Increase productivity and reduced costs, Furthermore, these environments involves different category
Optimizing business processes, Increase competitive of users with different access right and privacy requirements.
advantage, Improve customer satisfaction and retention etc. For example, home users may include parents, children,
As with all types of networking environments, SOHO guests, and external administrators while offices may have
environments and technologies are also vulnerable. Security employees with different roles. Finally, as the system may
issues such as authentication, authorization, and access control involve different type of users, individual skills and expertise
requirements should be considered when designing services challenges should also be considered.
for such networks. As SOHO networks can be either wired or As SOHO is generally small and require fewer
wireless, peculiar vulnerabilities associated with the chosen infrastructures, attack on this network can easily make the
technology should also be taken into consideration. In view of entire system inoperable. In view of these, the focus of the
these, implementing and supporting scalable SOHO faces paper is to examine the security threats and requirements
many challenges. For instance, these environments consist of especially in the design of wireless SOHO networks. First, the
variable hardware and software components, coming from paper provides an overview of SOHO design. Later, it
different vendors or manufacturers some being more discusses security threats and attacks for these environments.
trustworthy than others. In this case issues to be managed may Finally, it presents some methods of preventing the attacks.
include integration management, rapid technological
expansion and changes etc. II. OVERVIEW OF SOHO DESIGN
Lightweight SOHO computing devices like Personal
Digital Assistants (PDAs), mobile phones, and laptops are Wired SOHO uses either dialup or Ethernet. This consists
also prone to large set of software attacks like the denial of of network adapters, hub, and simple configuration software.
service (DOS) or buffer overflow attacks. These devices when Unlike in wired SOHO, wireless SOHO uses WLAN
connected to infrastructure based networks like office LAN, technology based on the IEEE 802.11 standard. Users can
make them vulnerable to the attacks similar to that on freely move inside or outside the building and still remain
infrastructure based networks. Addition of mobile, handheld, connected to the network and other hardware devices on the
network. This network components can be set up anywhere in

555
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_97, © Springer Science+Business Media B.V. 2010
556 MOHAMMED

the building. Further, wireless SOHO is very flexible and also On the negative side, some of the key challenges can be
allows users to move computers and other devices without the summarized as follows:
need to reconfigure the network. Flexibility comes in as users • Integration management
can get connected through simple steps of authentication • Multi vendors device and security management
without the hassle of running cables. Also compared to the • Rapid technological expansion and changes
wired network, wireless network installation costs are minimal • Skill and expertise challenges
as the number of interface hardware is minimal. As illustrated • Future proofing challenges etc
in table 1 below, WLANs are less expensive to install even for
relatively modest workforces. Those savings grow along with III. SECURITY THREATS
the size of the workforce as the cost associated with WLAN
controllers is further amortized. Further, wireless SOHO does SOHO computing environment share all of the security
not require any computer running Windows server 2003, it issues of traditional networked applications. These include
fully supported in Windows XP (even home edition), it is authentication of devices and users, privacy of data or
relatively easy for even a novice user to deploy, it does not information, defence against malicious code such as viruses,
require the purchase of a server certificate, and also provides worms, Trojan horses etc, and access control mechanisms.
simple methods to control or manage workgroup members. However, the SOHO computing environment combines both
These reasons make wireless SOHO more popular than the wired and wireless network security challenges as the system
wired SOHO. can be based on both types of environment. When wireless
TABLE 1 network is employed, information that is usually confined
COST OF WIRED VS WLAN (Source: www.Sonicwall.com) behind a corporate firewall will be winging its way through
Equipments Ethernet WLAN the air, possibly spending some time on hosted servers or
Ethernet/PoE Port $35 $35 wireless gateways. Further, the techniques of hacking
Installation $200 $250 wireless and mobile devices such as laptops, cell phones,
AP - $1,300 (50 users) PDAs etc is already spreading. In view of these, adding
Controller - $15,000 (50 APs) security to such environment presents challenges at different
Installed costs - 100 users $23,500 $16,585 levels. For instance, having a central authority for a single
Installed costs - 1,000 users $235,000 $46,700 building or even a group of rooms is infeasible because every
possible access right will have to be specified for every user.
In a typical WLAN configuration, a transmitter/receiver Authenticating the identity certificate of a previously
(transceiver) or AP, connects to the wired network from a unknown user doesn’t provide any access control information.
fixed location using standard Ethernet cable. At a minimum, Simple authentication and access control are only effective if
the Access Point receives, buffers, and transmits data between the system knows in advance which users are going to access
the WLAN and the wired network infrastructure. A single a particular subject or stored information and what their
Access Point can support a small group of users and can access rights are. Portable handheld and embedded devices
function within a range of less than one hundred to several have severely limited processing power, memory capacities,
hundred feet. Fig. 1 below shows the basic components of a software support, and bandwidth characteristics. Also,
SOHO wireless network infrastructure. hardware and software environments are becoming
increasingly heterogeneous [1]. In view of these, the security
requirements can be roughly divided into three main
categories; network security requirements, device security
requirements, and finally the user security requirements.
However, in this paper we focus more on the network security
requirements.
SOHO computing aims to provide users with computer-
supported capabilities like traditional network environment. In
order to achieve this, correct identification and verification of
users is paramount. Authentication is equally important when
allowing a user access to a secure file space. Its purpose is to
ensure that each entity is correctly matched to its
Fig 1 Basic Wireless SOHO Infrastructure corresponding privileges. If someone is not accurately
identified then their identification is left unverified. In
Some of the well-known SOHO benefits, including:
wireless SOHO computing environment, the risk of someone
• Improved efficiency and access to mobile employees
fraudulently performing input using authorized user’s login
• Offers mobility and flexibilty details is greater and, for reasons such as this, access to these
• Increase productivity and reduced costs workspaces and the computers within them may require more
• Optimizing business processes physical and information security measures than the normal
• Increase competitive advantage wired network environment. Details of WLAN vulnerabilities
• Improve customer satisfaction and retention. etc can be found in [2], [3] and [4]. Moreover, one of the main
 ON THE DESIGN OF SOHO NETWORKS 557

difficulties in designing a secure SOHO environment is in amount of network activity serves as an indicator for the
ensuring that the functionality of the environment is not over occurrence of a large event. Secondly, the attacker can find
constrained by security. Brief descriptions of some specific information about the location of APs in the surrounding area.
attacks targeting the entire network and its infrastructure when This is because unless turned off, APs broadcast their service
the devices are connected to the SOHO network are given set identifiers (SSIDs) for identification. Thirdly, the attacker
below: may learn the type of protocols being used in the transmission.
Web Attacks – The web browser is a complex application This knowledge can be obtained based on the size, type and
that can be extended through entire programming languages number of packets in transmission over a period of time.
such as Javascript, Java applets, Flash applications and Analysis of TCP three-way handshake is described in [5].
ActiveX controls. While these technologies make for a richer Unauthorized Access – this attack involves gaining free
web experience for the end user, each of them is a new attack access to the network and also using the AP to bypass the
vector in a program running inside a firewall. Filtering firewall and access the internal network. Once an attacker has
network traffic cannot protect against all attacks. Further, even access to the network, he/she can then launch additional
Web sites that are legitimate for use in a business context can attacks or just enjoy free network use. Although free network
serve as a source of threats. use may not be a significant threat to many networks, however,
Impersonation attacks - An intruder assumes the access is a key step in ARP based man-in-the-middle attacks.
identity and privileges of another node in order to consume AP Overloaded - This attack is based on the observation
resources or to disturb normal network operation. An attacker that in 802.11 a client must be successfully authenticated and
node achieves impersonation by misrepresenting its identity. associated to an AP before using wireless communication
This can be done for example by changing its own MAC services. AP maintains the client state information in a client-
address to that of some other legitimate node. Strong association table. When the table reaches the permitted level
authentication procedures can be used to stop attacks by of associated clients, the AP start rejecting new association
impersonation. requests. This attack is facilitated by the open system
Eavesdropping – This implies the interception of authentication method in 802.11where an AP authenticates
information/data being transmitted over the network. In case anyone who requests authentication. An attack can be
of wireless SOHO, when the wireless link is not encrypted, an launched if the adversary does a large number of associations
attacker can eavesdrop the communication even from some with an AP, using random MAC addresses. Since an AP can
few miles away. The attacker can gain two types of maintain a limited number of associations, this will prevent
information from this attack. The attacker can read the data other stations from joining the AP.
transmitted in the session and can also gather information Layer-Level Attack - Every year, network attacks
indirectly by examining the packets in the session, specifically become more widespread, more intelligent and more difficult
their source, destination, size, number, and time of to detect. Because many of today's attacks are blended attacks
transmission. Eavesdropping can also be active; in this case which use multiple techniques at different layers to try to
the attacker actively injects messages into the communication infiltrate the network, they can bypass outdated firewalls that
medium in order to assist him/her in determining the contents only provide packet inspection for network traffic. With
of messages. packet inspection, firewalls mainly ensure that connections are
Poor network administration - A surprising number of valid. However, they perform absolutely no inspection beyond
machines are configured with an empty or easy to guess the first few networking layers and thus do not concern
root/administrator password. One of the first things an themselves with the content carried within these data streams,
attacker will do on a network is to scan all machines for empty allowing both desired and malicious data through. Further, as
or commonly used passwords. the amount of traffic being scanned by firewall as well as the
System Intrusion – Also known as “Privilege increasing amount of threats and malware lurking in that
Escalation”, this type of hacking assumes the attacker already traffic is quickly becoming more than many firewalls can
has a low-privilege user account on the system. If the system handle. Inspecting every byte of every packet can overwhelm
doesn't have the latest security patches, there is a good chance some firewalls and bottleneck network performance. This not
the attacker will be able to use a known exploit in order to only keeps users from getting the most out of the available
gain additional administrative privileges. bandwidth, but also degrades and disrupts streaming
TCP/IP protocol flaws - The TCP/IP protocol was applications like Voice over IP (VoIP) phone and video.
designed before we had much experience with the wide scale Modern Malware - The goal of malware is money, not
hacking we see today. As a result, there are a number of recognition, and stealth is highly prized. The methods and
design flaws that lead to possible security problems. Some techniques for making malware pay expand and evolve at a
examples include smurf attacks, ICMP Unreachable rapid pace, with the more general risks of infected systems
disconnects, IP spoofing, and SYN floods. being absorbed into botnets for use in spam campaigns and
Traffic Analysis – Traffic analysis allows the attacker to DDoS attacks, draining bandwidth and resources, and halt
obtain three forms of information. The attack primarily device performance, supplemented by the more personal
identifies that there is activity on the network. Similar to dangers of data theft. A more dangerous malicious program
standard radio communications, a significant increase in the can transmit itself across the network, bypassing some of the
558 MOHAMMED

network security systems, and potentially damaging other

components of the network.

IV. ATTACK SCENARIO

In this section, we demonstrate a simple scenario to show
how easily a network with limited resources such as SOHO
can be disrupted by ping flooding attack. First, we briefly
explain the purpose of ICMP and then show how the
vulnerability associated with this protocol and leads to Fig. 4 The response time at the end of the flood attack
flooding attack.
Internet Control Protocol or ICMP is used to report the
delivery of Internet Protocol (IP) echo packets within an IP
network. It can be used for network trouble shooting purposes
to show when a particular end station is not responding, when
an IP network is not reachable, when a node is overloaded or
when an error occurs in the IP header information etc [6].

PING flood attack: One of the features of TCP/IP is

fragmentation; it allows a single IP packet to be broken down
into smaller segments. Attackers take advantage of that
feature when they found that a packet broken down into Fig. 5. The status of a victim before and after the flood attack.
fragments could add up to more than the allowed 65,536 bytes.
Many operating systems cannot handle oversized received Some types of attacks that are associated with ICMP are
packet, so they froze, crashed, or rebooted.. We demonstrated documented in [7], [8], and [9]:
this as shown in Fig. 2, 3, and 4 below. ICMP DOS Attack: This is a denial-of-service
attack in the sense that the attacker could use either the ICMP
“Time exceeded” or “Destination unreachable” messages to
force a host to immediately drop a connection. An attacker
can make use of this by simply forging one of these ICMP
messages, and sending it to one or both of the communicating
hosts. Their connection will then be broken. The ICMP
"Redirect" message is commonly used by gateways when a
host has mistakenly assumed the destination is not on the local
network. If an attacker forges an ICMP "Redirect" message, it
can cause another host to send packets for certain connections
Fig. 2 The response time at the beginning of the flood attack
through the attacker's host.
ICMP packet magnification (or ICMP Smurf): In
Smurf attack, which is one of the most effective in the
category of network-level attacks against hosts, the attacker
sends a large amount of ICMP echo (ping) traffic at IP
broadcast addresses, all of it having a spoofed source address
of a victim. If the routing device delivering traffic to those
broadcast addresses performs the IP broadcast, most hosts on
that IP network will take the ICMP echo request and reply to
it with an echo reply each, multiplying the traffic by the
number of hosts responding. On a multi-access broadcast
Fig. 3 The response time at the middle of the flood attack network, there could potentially be hundreds of machines to
reply to each packet. The “smurf” attack's close relative is
The excess and unwanted ICMP packets that flood the target called “fraggle”, which uses UDP echo packets in the same
computer buffers have caused this lack of response as shown way as the ICMP echo packets; it was a simple re-write of
in fig 5below. “smurf”.
Inverse Mapping - Inverse Mapping is a technique used
to map internal networks or hosts that are protected by a
filtering device. In this attack an attacker sends an ICMP reply
message to a range of IP addresses presumably behind a
filtering device. Upon receiving the series of ICMP reply
messages, since the filtering device does not keep state of the
 ON THE DESIGN OF SOHO NETWORKS 559

list of ICMP requests, it will allow these packets to their Against Spoofing - Attackers launching spoofing usually
destination. If there is an internal router, the router will hide the identity of machines they used to carry out an attack
respond with a ICMP “Host Unreachable” for every host that by falsifying the source address of the network
it cannot reach, thus giving the attacker knowledge of all hosts communication. This makes it more difficult to identify the
which are present behind the filtering device. sources of attack traffic. It is therefore important to use
Crafted ICMP Messages: An attacker can send a network switches that have MAC binding features that store
specially crafted ICMP message to an affected system. An the first MAC address that appears on a port and do not allow
attacker who successfully exploited this vulnerability could this mapping to be altered without authentication. To prevent
cause the affected system to reset existing TCP connections, IP spoofing, disable source routing on all internal routers and
reduce the throughput in existing TCP connections, or use ingress filtering. DNS spoofing can be prevented by
consume large amounts of CPU and memory resources. These securing the DNS servers and by implementing anti-IP
attacks, which only affect sessions terminating or originating address spoofing measures. Some vendors have added access
on a device itself, can be of three types: 1. Attacks that use control lists (ACL), implemented through MAC address
ICMP "hard" error messages, 2. Attacks that use ICMP filtering, to increase security. MAC address filtering amounts
"fragmentation needed and Don't Fragment (DF) bit set" to allowing predetermined clients with specific MAC
messages, also known as Path Maximum Transmission Unit addresses to authenticate and associate. While the addition of
Discovery (PMTUD) attacks, 3. Attacks that use ICMP MAC addresses filtering increases security, it is not a perfect
"source quench" messages. Successful attacks may cause solution given that MAC addresses can be spoofed. Also, the
connection resets or reduction of throughput in existing process of manually maintaining a list of all MAC addresses
connections, depending on the attack type. can be time consuming and error prone. Therefore MAC
address filtering is probably best left for only small and fairly
V. DEFENSE MECHANISMS static networks [10].
Against Flooding Attack - A combination of Host-
The following countermeasures can be taken to minimize or based Intrusion Detection System (HIDS) and Network-based
thwart the attacks mentioned above. Intrusion Detection System (NIDS) can greatly help against
this attack. HIDS can be placed on critical servers and NIDS
Against Unwanted Web Contents - An effective way that can be placed on one or more network segments. Signature
can be implemented to block Web-based threats is content detection scheme would be good at detecting any known
filtering designed to block unwanted file types. Blocking file attacks. Alerts arising from any suspicious activity can be
types based on their content can be useful in preventing some intimated to the administrator immediately. NIDS reactions
types of Web threats from entering a network, particularly can also be TCP resets, IP session logging and Blocking.
files that are traditionally known to be associated with HIDS approach looks into log files, file checksums and
malware, such as .scr or .pif. These systems can also block file intercepting requests to the operating system for system
types that are generally not used in a legitimate business resources before they can proceed. Signatures and generic
context, such as .mp3, .jpg or .mov files. rules help in anomaly detection. Open server ports can also be
Against Access to Harmful Websites - Another option monitored for excess or abnormal traffic. Firewalls are an
that should be considered is the deployment of URL filtering excellent form of protection; however, they must leave some
tools that will block access to non-approved Web sites. Many ports open to allow the operation of the web, mail, ftp, and
organizations have deployed these filters, albeit with varying other Internet based services, and which are the paths
levels of success. While URL filters can be useful, they can exploited by most of the vulnerabilities. A more effective
rarely keep up with the new threats that enter the Web on an measure may involve a hybrid IDS system. This is a
hourly basis and for which no signature has been created in distributed system where each and every node in the network
the tool. Further, URL filters can generate significant levels of collectively cooperated to form a team of IDS agents for the
false positives – blocking Web sites that appear to be entire network. This can easily be done by placing an IDS
suspicious but might have a legitimate business purpose. intelligent agent on every node in the network. Each IDS
Against Malware - A new approach is required to thwart agent runs independently and monitors local activities (user,
the escalating threats posed by targeted malware and botnets system, or communication) in the network. It will then
that render traditional security techniques obsolete. Deploying program to detect any intrusion from local traces (especially
an anti-botnet protection system that uses global intelligence communication activities). If any anomaly or suspected
combined with local analysis and control ensures that botnets activity is detected, a report is send to the host server and
do not promulgate within the enterprise network. An adequate other nodes. These individual IDS agents form the system’s
protection system scans all network traffic and executes IDS to defend the entire network.
relevant traffic streams in virtual victim machines – safe from Against Layer-Level Attack - One of the simple was to
real clients – and conclusively determines if botnet malware is secure SOHO network today is the use of Unified Threat
present. This is the only real defense against network-borne Management (UTM) tools that layer on antimalware
attempts at spamming, spear-phishing, coordinated attacks, protection, content filtering, antispam and intrusion prevention,
identity theft, or theft of intellectual property. It eliminates the because deploying a single, multi-function device reduces
underlying cause for all these events—botnets
560 MOHAMMED

costs and simplifies configuration. Thus, UTM is a network capabilities. SSL VPNs are best suited for secure remote
security tool providing a new level of scrutiny to network access during an emergency because they allow workers and
traffic passing into an organization. Simply put, UTM partners to connect safely to corporate network resources
firewalls combine the effectiveness of various point defences using a Web portal, without having to pre-install clients.
to add protection at every networking layer. The power of Modern continuous data protection (CDP) solutions can
UTM comes from its simplicity: a single appliance that automatically backup data and applications to discs, avoiding
provides a powerful defence against a wide range of security the complexity and human error involved with tape backup.
threats, while extending beyond, the capabilities of regular Backup to a secure secondary business location or third-party
packet inspection firewalls. This makes network protection site means business systems can be restored and operational
more complete, affordable and easy to manage. For optimal even in the primary site is compromised. Bare metal recovery
performance while maintaining maximum security, solutions (BMR) technology enables entire operating systems, such as
such as reassembly-free deep packet inspection UTM help to database or file servers, be recovered to new or different
deliver throughput that won't bog down performance. Clever hardware platforms if the original device can’t be restored.
algorithms help to reduce latencies to non-noticeable levels.
Also, advances in microprocessor technology, especially those VI. CONCLUSION
designed for network processing, allow UTM appliances
designed for SOHO environment to gain incredible network This paper described some vulnerabilities associated with
efficiency. SOHO networks. It shows that implementation of wireless
Alternatively, a firewall can be considered when SOHO posses more loopholes than wired SOHO if proper
connecting that network to a second network (such as virtual security measures are not taken into consideration. The paper
private network or VPN) which has a less stringent security discuses some essential areas required for a complete and
policy. Implementing firewall along with VPN over the robust protection policy. These include anti-malware
SOHO LAN/WLAN can strengthen the system’s security. protection on all platforms, along with firewalling, intrusion
IPSec can be used as the security protocol and the secure prevention, network access control, data leak prevention and
tunneling with authentication and encryption. In [11], we data encryption among others.
demonstrated this idea using a simple network topology. Two
nodes A and B had to access database and web pages on a ACKNOWLEDGMENT
Server through a network with firewall and VPN server The author wishes to acknowledge Hafr Al-Batin
implemented. The firewall would not permit database access, Community College and King Fahd University of Petroleum
but would allow HTTP or web service access. A compulsory and Minerals (KFUPM) Saudi Arabia for their support in
VPN tunnel is implemented between the routers (source router providing the various facilities utilized in the presentation of
and destination routers) that connect node A to the Server. So this paper.
even if the firewall blocks a service, a node with VPN tunnel
enabled can go through the firewall as data communication REFERENCES
happens as encapsulated packets (packet within a packet). [1] L. Kagal, F. Tim, and J. Anupam, “Trust-Based Security in Pervasive
Against Wireless Breaches -When wireless network is Computing Environments”, Computer v34, n12, Dec 2001, pp:154-157T.
deployed, it is recommended to check all default settings, [2] T. Takahashi, “WPA passive dictionary attack overview” [Online]:
including administration password and access path, and http://www.3c.com.vn/Uploaded/huyenctt/New%20Folder/WPA_Passiv
change them or disable functions as needed. Turn off all e_Dictionary_Attack_Overview.pdf
access point services that are not being used (ftp, http, etc…). [3] P. Liza, “WPA PSK Crackers: Loose Lips Sink Ships”, [Online].
Also WEP encryption must be turned on and must use 128 bit www.wi-fiplanet.com/tutorials/article.php/3667586
keys. Moreover, SSIDs must be changed from defaults, must [4] E. Tews ‘Gone in 900 Seconds, Some Crypto Issues with WPA’,
PacSec 2008 Conference, Nov. 2008, Tokyo, Japan.
not reflect any information about the organization or location
[5] W. Donald, “Wireless Security Threat Taxonomy”, IEEE Workshop on
of the Access Point, and SSID cloaking must be turned on. Information Assurance. US Military Academic West Point N.Y, June
MAC address filtering must be used to restrict wireless access 2003, pp: 76-83.
to only approved hardware addresses. The AP locations [6] Internet Control Message Protocol (ICMP)., [Online]:
should be in interior positions and away from exterior walls http://www.erg.abdn.ac.uk/users/gorry/course/inet-pages/ icmp.html
and windows, ensure they are placed in secured locations to [7] ICMP Attack, [Online]: http://www.javvin.com/networksecurity/
prevent unauthorized physical access. ICMPAttacks.html
[8] Cisco White Paper: “Crafted ICMP Messages Can Cause Denial of
Disaster Recover Plan - No matter how secure the Service “ [Online]: http://www.cisco.com/warp/public/707/cisco-sa-
network is, the need for a disaster recovery solution is 20050412-icmp.shtml
essential. Major disasters have demonstrated how exposed [9] SANS White Paper, “ICMP Attacks Illustrated” [Online]:
small businesses can be to unexpected events. Building fires, http://www.sans.org/info/36914
[10] L. A. Mohammed and B. Issac, “Detailed DoS Attacks in Wireless
broken water pipes, power outages, equipment failures, Networks and Countermeasures” International Journal of Ad Hoc and
terrorist, or even lost or stolen handheld devices, can mean Ubiquitous Computing, (IJAHUC), Vol 2. No 3, pp. 157-166, 2007
disaster for small businesses networks such as SOHO. Modern [11] B. Issac and L. A. Mohammed , “War Driving and WLAN Security
UTM appliances can feature integrated IPSec or SSL VPN Issues Attacks, Security Design and Remedies”, Information Systems
Management, Vol 24, pp. 289-298, 2007
 Study of UMTS Handover Scenarios
B. Issac
Swinburne University of Technology (Sarawak Campus), Malaysia
[email protected]

well as service providers and that will create added interest and
Abstract— Universal Mobile Telephone system (UMTS) is a it can be expected that different forms of service providers will
third generation mobile and wireless communications standard play a more definite role in the future. This could in turn boost
which is also being developed into a fourth generation technology.
We wanted to study the performance of such a wireless cellular new business models and new forms of revenue sharing
network during soft and hard handover scenarios with single and through the advent of new services.
multiple towers. An attempt is thus made to use existing The content services that will be provided by UMTS will
simulation models to study the related performance graphs as the evolve from the services provided by GSM. These services
mobile node moves around. offered by UMTS can be classified as “mobile Internet”, where
display text or other contents are optimized for the limited
graphical capabilities of the mobile devices, along with the
I. INTRODUCTION location awareness functionality. A range of services have
been suggested, for example: Images and video: streaming
T he European countries have chosen Universal Mobile
Telephone system (UMTS) as one of the standards for the
third generation mobile communication system. The third
pictures and video, entertainment, games, lottery, gambling,
and video-conferencing, Short range and Location based
generation in the Nordic countries has been preceded by the services: Information of nearby restaurants, shops and special
first generation mobile telephone system NMT (Nordic Mobile offers, Context based services, or push: Notification of
Telephone), and the second generation GSM (Global system interesting sports event if you are in front of the TV,
for Mobile Communication). UMTS is planned to coexist or advertising if you are close to the advertiser, Simultaneous
work with 2G for several years, as NMT coexisted with GSM transfer of speech, data, text, pictures, audio and video, High-
from 1992 to 2002, when NMT finally was phased out. As speed, mobile access to the Internet, Customized infotainment,
with NMT and GSM, GSM handsets will not be compatible Travel information: congested roads, flight departures, and if
with the UMTS. On the other hand, the different networks will you get lost, find your current location etc.
not be totally separated as UMTS handsets will be compatible When mobile users are roaming (moving from one area or
with GSM providing a dual mode of communication. This is cell to another) the handset must attach itself to the closest
because UMTS, as NMT, will provide less coverage than base-station. When one base-station hands over the user to
GSM, probably only rural areas with high density of users. another base-station, multiple transactions must be made. If the
UMTS is defined by European Commission [1] as: “... a user is roaming among base-stations belonging to the
third-generation mobile and wireless communications system subscribed operator, this is only a technical issue. But when
capable of supporting in particular innovative multimedia roaming implies other operators, national or international, the
services, beyond the capability of second generation systems picture becomes more complicated. The vision of UMTS
such as GSM ...” . This definition is not that specific on what presupposes that it will be possible for users to roam freely
new services can be expected by the users. International with their handsets and have affordable access to equivalent
Telecommunication Union is slightly more specific as they products and services globally by end-users utilizing the most
define UMTS by higher capacity and enhanced network proper network despite which operator it belongs to. Today,
functionalities that would allow more demanding services and technically and efficient roaming exists for the GSM standard
applications, like multimedia. For an end-user, the major throughout Europe, parts of Africa, Asia and South America.
difference between 3G and 2G is underlined by these However, it is far from global: for instance, roaming to and
definitions to be the increased user value created by a broad from North America and Japan requires special measures and
service offering. UMTS provides key features for end-users users may even incur custom duties and taxes relating to the
provided as – Options or facilities for Internet and other use of their handsets abroad. Moreover, the complexity and
multimedia-rich applications, a good range of content services non-transparency of pricing for roaming services have been
and global roaming capability. problematic for 2G or 2.5G users [2]-[8].
UMTS will provide basic services as with GSM and they are – Some related works on UMTS can be seen in [2]-[7]. The
voice, Short Messaging Service (SMS), roaming and billing. paper is organized as follows. Section 2 discusses the features
So multimedia services would be developed by operators as of UMTS, section 3 is the architecture details, section 4 is the

561
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_98, © Springer Science+Business Media B.V. 2010
562 ISSAC

simulation of two UMTS scenarios where we use different propagation constraints or other economic reasons. In order to
network topologies and section 5 is the conclusion. enable subscribers to always use their terminal, services will be
adaptive to different data rate availability and other Quality of
II. UMTS FEATURES Service (QoS) parameters. In the early stages of deployment,
A. Fast access UMTS coverage can be limited. UMTS can enable roaming
UMTS is better than 2G or 2.5 G cellular technology for its with other networks like GSM system operated by the same
potential to support 2 Mbit/s data rates. This capability, operator or with other GSM or 3G systems of other operators,
together with the built-in IP support, merges powerfully to including UMTS compatible satellites.
allow interactive multimedia services and new wideband
multimedia-rich applications, such as video telephony, video F. UMTS Services global availability through Satellite
conferencing, network video gaming etc. Satellite technology can give enhanced global coverage and
services and it is expected to play an important role for UMTS
B. Packet transmission and data rate on demand world-wide coverage. UMTS is being standardized to ensure
Current cellular systems use circuit switched technology for an efficient and effective roaming and handover between
data. UMTS combines packet and circuit data transmission satellite and terrestrial networks.
with the advantage of virtual connectivity to the network at all
times and alternative ways of billing. For example, pay-per-bit G. Spectrum for UMTS
option, to pay per session, to pay flat rate etc as demanded by WRC-92 identified the frequency bands 1885-2025 MHz
many upcoming data services. The operation of the system can and 2110-2200 MHz for future IMT-2000 systems, with the
be made much cheaper as UMTS is also being designed to bands 1980-2010 MHz and 2170-2200 MHz intended for the
offer data rate on demand which can work in combination with satellite part of these systems. These bands can be made
packet data transmission. available for UMTS licenses to enable operators for network
deployments. Wideband Code-Division Multiple-Access (W-
C. Friendly and consistent service environment CDMA) is one of the main technologies for the
UMTS services are based on having the user face common implementation of third-generation (3G) cellular systems. In
capabilities throughout all UMTS user and radio environments. W-CDMA interface different users can simultaneously
When roaming from the user’s network to other UMTS transmit at different data rates and data rates can even vary in
operators, a personal user will experience a consistent set of time. UMTS networks need to support all current 2G/3G
services and this would give a feeling or sense of his or her services [5]-[7].
home network (termed as “Virtual Home Environment” or
VHE). VHE will ensure the delivery of the service provider’s III. UMTS ARCHITECTURE
full environment. VHE will enable terminals to handshake and A UMTS network consists of three domains that interact
agree on some functionalities with the visited network, and with each other, namely – Core Network (CN), UMTS
"home network like" services will be presented with full Terrestrial Radio Access Network (UTRAN) and User
security and transparency across a mix of access and core Equipment (UE) [8]. The main function of the core network is
networks. to provide switching, routing and transit for user traffic. Core
network also contains the databases and network management
D. Mobility and Coverage functions. The basic Core Network architecture for UMTS is
UMTS is implemented as a global system, combining both based on GSM network with GPRS. All equipment has to be
terrestrial and satellite communication network components. modified for UMTS operation and services. The UTRAN
Multi-mode terminals operating also via 2.5 or 3G systems will provides the air interface access method for User Equipment.
further extend the reach of many UMTS services. With these Base Station is referred as Node-B and control equipment for
terminals a user will be able to move around and roam from a Node-B's is called Radio Network Controller (RNC).
private network into a micro-cellular public network, then into It is important for a network to know the approximate
a wide area macro-cellular network (like a 2.5 or 3G network), location in order to be able to contact the user equipment.
and then to a satellite mobile network, with minimal break in Some of the list of system areas from largest to smallest –
communication and packet loss. UMTS systems (including satellite), Public Land Mobile
Network (PLMN), MSC/VLR (Mobile Switching
E. Radio Technology for all environments Center)/(Visitor Location Register) or SGSN (Serving GPRS
The UMTS radio interface UTRA (UMTS Terrestrial Radio Support Node), Location Area, Routing Area (PS domain),
Air interface) can help operation with high spectral efficiency UTRAN Registration Area (PS domain), Cell and Sub cell etc.
and service quality. In practical implementations, UMTS
terminals might be unable to function at the peak data rates at A. Core Network
all times, and in remote or heavily congested areas system The Core Network consists of circuit switched and packet
services might only support lower rates due to radio switched domains. Some of the circuit switched elements are
 STUDY OF UMTS HANDOVER SCENARIOS 563

Mobile services Switching Centre (MSC), Visitor location services to be offered over the PS domain (like VoIP). CS
register (VLR) and Gateway MSC. Packet switched elements mode of operation: The MS is attached to the CS domain only
are Serving GPRS Support Node (SGSN) and Gateway GPRS and may only operate services of the CS domain.
Support Node (GGSN). Network elements like EIR UMTS IC card has similar physical characteristics as GSM
(Equipment Identity Register), HLR (Home Location SIM card. It has various functions as follows – support of one
Register), VLR (Visitor Location Register) and AUC User Service Identity Module (USIM) application (optionally
(Authentication Center) are shared by both domains. The more that one), support of one or more user profile on the
Asynchronous Transfer Mode (ATM) is defined for UMTS USIM, Update USIM specific information over the air,
core transmission. ATM Adaptation Layer type 2 (AAL2) security functions, user authentication, and optional inclusion
handles circuit switched connection and packet connection of payment methods and optional secure downloading of new
protocol AAL5 is designed for data delivery. The architecture applications [8]-[9].
of the Core Network may change when new services and
features are introduced. Number Portability Database (NPDB) IV. UMTS HANDOVER
will be used to enable user to change the network while There are following categories of UMTS handover (also
keeping their old phone number. Gateway Location Register referred to as handoff) that can happen:
(GLR) may be used to optimize the subscriber handling
between network boundaries. A. Hard Handover
Hard handover means that all the old radio links in the UE
B. Radio Access are removed before the new radio links are established. Hard
Wide band CDMA technology was opted for UTRAN air handover can be seamless or non-seamless. Seamless hard
interface. UMTS WCDMA uses Direct Sequence CDMA handover means that the handover is not perceptible to the
where user data is multiplied with quasi-random bits derived user. In practice a handover that requires a change of the
from WCDMA spreading codes. In UMTS, in addition to carrier frequency, known as inter-frequency handover is
channelization, codes are used for synchronization and always performed as hard handover.
scrambling. WCDMA has two basic modes of operation:
Frequency Division Duplex (FDD) and Time Division Duplex B. Soft Handover
(TDD). The functions of Node-B are stated as follows: Air Soft handover means that the radio links are added and
interface Transmission/Reception, Modulation/Demodulation, removed in a way that the UE always keeps at least one radio
CDMA Physical Channel coding, Micro Diversity, Error link to the UTRAN. Soft handover is performed by means of
Handing and Closed loop power control. The functions of macro diversity, which refers to the condition that several
RNC are – Radio Resource Control, Admission Control, radio links are active at the same time. Normally soft handover
Channel Allocation, Power Control Settings, Handover Control, can be used when cells operated on the same frequency are
Macro Diversity, Ciphering, Segmentation/Reassembly, changed.
Broadcast Signaling and Open Loop Power Control.
C. Softer handover
C. User Equipment Softer handover is a special case of soft handover where the
The UMTS standard does not hinder or cripple the radio links that are added and removed belong to the same
functionality of the User Equipment. Terminals work as an air Node B (i.e. the site of co-located base stations from which
interface counter part for Node-B and have many different several sector/cells are served). In softer handover, macro
types of identities which are taken directly from GSM diversity with maximum ratio combining can be performed in
specifications. They are International Mobile Subscriber the Node B, whereas generally in soft handover on the
Identity (IMSI), Temporary Mobile Subscriber Identity downlink, macro diversity with selection combining is applied.
(TMSI), Packet Temporary Mobile Subscriber Identity (P- Active Set is defined as the set of Node-Bs the UE is
TMSI), Temporary Logical Link Identity (TLLI), Mobile simultaneously connected to (i.e., the UMTS Terrestrial Radio
station ISDN (MSISDN), International Mobile Station Access cells currently assigning a downlink DPCH (Downlink
Equipment Identity (IMEI), International Mobile Station Dedicated Physical Channel) to the UE).
Equipment Identity and Software Number (IMEISV).
UMTS mobile station (MS) can operate in one of three V. SIMULATION DETAILS
modes: PS/CS (Packet Switched/Circuit Switched) mode of A. Scenario 1 (Soft and hard handovers between UMTS
operation: The MS is attached to both the PS domain and CS networks)
domain, and the MS is capable of simultaneously operating PS We used existing simulation models and only verified the
services and CS services. PS mode of operation: The MS is existing results and no modifications were done to the model.
attached to the PS domain only and may only operate services Our aim is to conduct a performance comparison between soft
of the PS domain. However, this does not prevent CS-like and hard handovers between UMTS networks (with different
564 ISSAC

towers). Consider fig. 1. The configurations done are given handovers compared to the soft handovers. The difference of
below. In these scenarios, the two UEs (User Equipment) transmission power is around 3 dBm roughly. Thus soft
move around and make zigzag (back and forth) transition handovers enable the UEs to perform the handovers at power
between two Node-Bs (UMTS towers) creating multiple levels that are low, which is an added benefit for other UEs
handovers from one cell to the other. Continuous upload of around when we consider the issue of interference.
large FTP files to the server was done by the two UEs during To achieve equal performance benefits with lower power,
the simulation, which makes the traffic move in the uplink soft handovers benefit from an idea that can be termed as "soft
direction in the UMTS part of the network. UMTS QoS 3 is handover gain". As a radio bearer is aided by more than one
used for the traffic associated with FTP. radio links during the soft handovers, the network can lower
The configurations done and the network topology of both the quality (BLER, Block Error Rate) of these links below the
scenarios are same except that in hard handover case, the soft requested level as long as their combined or integrated quality
handovers are disabled at the RNC (Radio Network is enough to provide the needed or requested quality for the
Controller), so that all the handovers done are hard handovers. radio bearer. Thus the transmission power can be maintained at
To study the soft handover case, they are all configured as soft lower levels during soft handovers, as a result of lowering the
handovers. radio link quality, while still providing the same quality to
higher layer as the results show us. Have a look at fig. 2 to 5.

Fig. 2. Uplink transmission power (UE 0)

Fig. 1. Network Scenario 1

In this study, we observed the results for two statistics that

are used for QoS 3 traffic: The FTP Upload response time and
the Uplink Transmission Power of the physical channels in the
UE (User Equipment). The ASC (Access Service Class) 6 is
selected here for UMTS.
ASC classes are defined at the RNC node. There are a
maximum of 8 access service classes available. As we
investigated and compared application response times from the
two scenarios, it showed that there is not much difference
between hard and soft handovers in terms of their effect and Fig. 3. FTP Upload response time
impacts on application response times.
On the other hand, for both UMTS user equipments, soft
handover produced better results in terms of uplink
transmission power. While observing the uplink transmission
graphs, the peaks or highs indicate handovers. The peaks are
smoother during soft handovers. The UEs go away from their
Node-Bs, which increases the transmission power and after
handovers they are connected to the nearest Node-B resulting
in lower transmission power. The advantage of soft handovers
over hard handovers can be seen as follows. When the line
graph of these two handovers is compared, the UE
transmission power reaches to higher values for the hard Fig. 4. Uplink transmission power (UE 1)
 STUDY OF UMTS HANDOVER SCENARIOS 565

Some simulation parameters or settings are as follows: The

RNC (Radio Network Controller) UMTS handover parameter
– Soft Handover is set to “Supported” in the softer handover
scenario and to “Not Supported” in the hard handover
scenario. The configurations done and the network topology of
both scenarios are same except that in hard handover case, the
soft handovers are disabled at the RNC (Radio Network
Controller), so that all the handovers done are hard handovers.
To study the soft handover case, they are all configured as soft
handovers as explained above. The Active Set Size parameter is
set to 2.
Fig. 5. FTP traffic received by Server (bytes/sec)

B. Scenario 2 (Softer and hard handover between three

sectors)
Our aim is to demonstrate the softer and hard handover
process that happens to mobile user equipment while it is
around a UMTS Node B with three sectors or cells. We also
want to study the user equipment’s (UE) transmission power
between the softer and hard handover scenarios. Softer
Handover takes place when the UE is positioned in an area
where sectors are overlapped. Hard handovers take place
when the UE is moving from one sector to another.
Fig. 7. Uplink transmission power

Fig. 8. UMTS Cell/Sector Uplink throughput

Fig. 6. Network Scenario 2

Consider fig.6. The configuration used is as follows. The

Node B has three directional antennas that cover an angle of
160 degrees. The sectors/cells are positioned every 120
degrees starting from 0 degrees and so it would be 0, 120, and
240 degrees. Thus an inter sector/cell overlap of 40 degrees is
achieved between neighboring cells.
Fig. 9. UE active set details
In these simulation scenarios, the UEs follow its pre-defined
path around the Node B causing multiple handovers as it Here we discuss some statistics and to illustrate the softer
moves from one sector to another sector. Using UMTS QoS 3 handover process. These statistics show the number of cells in
service level, the UE also uploads files through FTP to a user equipment’s Active Set, showing which cells are included
remote server at a distant network. and discarded from that set during the simulation performed.
566 ISSAC

As the UE travels around the Node B, it notes the pilot channel But during the hard handover case, only one sector receives
signal strength that emanates from all the three sectors. These traffic. See fig.7, 8 and 9 to help understand the scenarios
signal strength measurements allows it to change its active set. fully.
Let’s have a look at the softer handover events that occurred
VI. CONCLUSION
during simulation. The UE path is from sector 1 to sector 0,
followed by sector 0 to sector 2 and finally from sector 2 to The Universal Mobile Telecommunications System (UMTS)
sector 1. At the start of simulation, based on the strongest is one of the third-generation mobile telecommunications
signal from sector 1, this sector is included in the initial active technologies, which is currently worked upon to be developed
set. The UE starts moving at 115 sec. After some time when it into a 4G technology. The technology was proposed by 3GPP
reaches the edge of sector 0 at 175 sec, adding sector 0 into the (3rd Generation Partnership Project) and is part of the global
active set. The UE is now in softer-handover between sectors 0 International Telecommunication Union IMT-2000
and 1 while it remains in this position for around 60 seconds. (International Mobile Telecommunications-2000) standard.
Now the UE traverses toward the middle of sector 0 at 235 sec 3GPP specifications are based on evolved Global System for
(remaining in sector 0 for 60 seconds) and thus UE leaves Mobile Communications (GSM) specifications. A study is
sector 1 deleting it from its active set at 255 sec. At 375 sec the done on the different scenarios of UMTS network and its
UE moves to sector 2. While UE is in the overlapped area performance evaluation is done when the mobile node is on the
between sectors 1 and 2, it gets into softer-handover state once move. We were able to look into two network scenarios where
again at 369 sec. The UE continue moving along its path to the different types of handovers were investigated and the
middle portion of sector 2, deleting or removing sector 0 and performance graphs studied. This would help to understand the
ending its softer-handover state at 412 sec. In the last traversal, issues in soft and hard handover scenarios if one were to look
the UE comes back to sector 1 and enters into a softer into the details of UMTS technology.
handover state when it lies on the area shared by both sectors
(1 and 2) at 605 sec. Then the UE goes to the center of sector REFERENCES
1, dropping sector 2 from its active set at 650 sec. [1] EU, Decision No 128/1999/EC of the European parliament and of the
Thus Softer Handover takes place when the UE is council, of the 14 December 1998, on the coordinated introduction of a
third-generation mobile and wireless communications system (UMTS) in
positioned in an area where sectors are overlapped. Hard the Community, 1998, available [online]: http://europa.eu.int/eur-
handovers take place when the UE is moving from one sector lex/pri/en/oj/dat/1999/l_017/ l_01719990122en00010007.pdf
to another. In both scenarios, the results for several statistic [2] Saravanan, N.; Sreenivasulu, N.; Jayaram, D.; Chockalingam, A.,
groups are collected like: UE active set details, Node Bs "Design and Performance Evaluation of an Inter-System Handover
Algorithm in UMTS/GSM Networks", Proceedings of TENCON 2005,
throughput, UE RLC/MAC uplink transmission power etc. 2005, pp.1-6.
Let’s have a look at the hard handover events that occurred [3] Oudelaar, J., "Evolution towards UMTS", Proceedings of 5th IEEE
during simulation. At the start of simulation, based on the International Symposium on Personal, Indoor and Mobile Radio
strongest signal from sector 1, this sector is included in the Communications, 1994. Wireless Networks - Catching the Mobile
Future, 1994, pp.852-856.
initial active set. UE starts moving at 115 sec. After some time [4] Bosch, P.; Mullender, S.; Samuel, S., "Mobility in UMTS packet IP
when it reaches the edge of sector 0 at 175 sec, a hard networks", Proceedings of Wireless Communications and Networking
handover happens. Sector 0 gets added into the active set while Conference, 2006, pp. 545 - 550.
[5] Pereira, A.; Gameiro, A., "Simulation-based Evaluation of Spectrum
sector 1 is taken off. At 355 sec, the UE traverses to sector 2 Opportunities in UMTS Cellular Networks", Proceedings of Mobile and
and a hard handover happens from sector 0 to sector 2 at 369 Wireless Communications Summit, 2007, pp.1-5.
sec. Finally the UE returns to sector 1 and performing another [6] Website: http://www.cellular.co.za/umts_technical.htm, Retrieved March
hard handover state, but now from sector 2 to 1 at 605 sec. 15, 2009.
[7] Website: http://www.umtsworld.com/technology/wcdma.htm, Retrieved
Then the UE goes to the middle of sector 1 and remains there March 20, 2009.
at 650 sec. [8] Petter Nielsen, Introduction of the third generation of mobile
As we compared the UE transmission power between both communication system in Norway, Department of informatics, University
of Oslo, Norway. Website: http://heim.ifi.uio.no/~pnielsen/PHD/Ni.doc,
softer and hard handover scenarios, we found that the UE in
Retrieved March 25, 2009.
the softer handover scenario uses less transmission power, thus [9] UMTS World, Overview of The Universal Mobile Telecommunication
reducing the cell interference. But on the other hand during System, Online: http://www.umtsworld.com/ technology/overview.htm,
softer-handovers the UE is taking more network resources. Retrieved February 2009.
Note that during the softer handover scenario, the UMTS
Node B is getting traffic through two sectors at the same time.
 Reduction of Energy Consumption in WSN using an
Approach Geocast
Harilton da S. Araújo Wagner L. T. de Castro Raimir Holanda Filho
Technology Sciences Center Technology Sciences Center Technology Sciences Center
Master in Applied Computer Computer Science Dept. Master in Applied Computer
University of Fortaleza University of Fortaleza University of Fortaleza
Fortaleza – Ceará – Brasil Fortaleza – Ceará – Brasil Fortaleza – Ceará – Brasil
[email protected] [email protected] [email protected]

Abstract – Wireless sensor networks have deserved special The energy consumption is the main requisite of the
attention in the last years due to their intrinsic characteristics network and it affects all of the phases of its life cycle, not
that become them attractive in various applications, such as in being dependant upon the application. The communications in
the industrial activities, in the environment monitoring and
security, among others. In general, this kind of network can be WSN consume more energy than the processing and the
utilized as powerful tools in order to monitor and, eventually, sensor coverage performed by the network nodes. This
control an environment. The routing protocols, for the WSN, characteristic requires routing protocols that make possible
must have self-configuration characteristics aimed at discovering that the sensor nodes communicate themselves in an efficient
which is the best way for transferring the information, with and effective way with minimum energy consumption.
delivery assurance and with minimum energy consumption,
among the nodes that compose the network. This article proposes The routing protocols for sensor networks must have self-
a modification of the Directed Diffusion routing protocol so as to configuration characteristics in order to discover which the
reduce the energy consumption in the network when there is an best way is for transferring the information from a node to the
occurrence of failures. The proposal utilizes a Geocast approach
other one, with delivery assurance and minimum energy
to repair broken paths by constructing a new routing tree. The
performed simulations demonstrated that our proposal is more consumption, having the purpose of extend the lifetime of the
efficient from the energy viewpoint. network. If a node fails, through which a given information
flows, a new routing will have to be made in order that the
Keywords - wireless sensor; networks; routing; geocast information to be collected is able to reach the destination
node. The communications, among the network nodes, must
I. INTRODUCTION
be made so as the energy consumption is optimized. In this
A network of sensors is composed of a large number of sense, some protocols have the essential role of increasing the
nodes that are utilized as per request at an area of interest. time of the network useful life.
Each node has one or more sensors, in addition to the capacity
for processing, storage and communication. This article proposes a modification of the Directed
Diffusion routing protocol, having the purpose of reducing the
The Wireless Sensor Networks have aroused the interest of energy consumption in the network by means of focusing on
the scientific community due to their applicability that reaches the presence of failures. A Geocast approach was used in order
various areas such as the military, environmental, medical and to reduce the number of sent and received control messages.
industrial ones. Those networks differ in several aspects from That reduction was reached through the construction of a new
the traditional networks. Some characteristics are: they are routing tree being limited and directed towards a specific
composed of a large number of sensor nodes; the sensors have region of the network, focusing the occurrence of failures.
high energy limitations, and low processing and memory
capacity. In addition to that, some WSN applications require The article is organized in the following way: the Section II
characteristics of self-organization, being able to adjust presents the linked works. The Section III describes the
themselves, in an autonomous way, to the possible structural modifications that we made in the Directed Diffusion protocol.
changes due to external interventions such as modifications of The evaluation of the algorithm is discussed on the Section IV,
the topology caused by failures, inclusion of nodes or as per a followed by conclusions on the Section V.
request submitted by an external entity (user).
II. RELATED WORKS
The purpose of the WSN is to collect data of a region being
A proposal for optimization of the energy consumption is
covered by sensors and to allow the extraction of that data, by
analyzed on [1]. The conception of this proposal is to put
an external entity, through the sink node.
some specific nodes in the sleep (disconnected) mode in order

567
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_99, © Springer Science+Business Media B.V. 2010
568 ARAÚJO ET AL.

to save energy, while keeping the connectivity of the network utilized in order to reduce the energy consumption. The
thus assuring the communications. algorithm is described in detail on the following Section.
The EAR (Energy Aware Routing) protocol is described III. PROPOSAL FOR AN ENERGY AWARE PROTOCOL
on [2]. The EAR protocol is a reactive one that looks for paths
We propose, on this article, an extension towards an
of minimum energy consumption through the network. The
existing data centric routing protocol, Directed Diffusion [6].
basic operation consists of the occasional use of a set of paths
chosen through a probability function (which is dependant The main purpose of our proposal is to develop a protocol
upon the energy consumption of each path). In this way, it being tolerant to failures and having a minimum of energy
avoids that the best path has its energy depleted, thus consumption. In order to simultaneously reach those
increasing the network utilization time. It is assumed that each requisites, an approach, based on the Geocast routing method,
node has an address and information about its localization. was utilized in our algorithm. The approach consists in
repairing broken paths by constructing a new, limited and
An algorithm is proposed on [3], which utilizes GPS in
directed routing tree. The approach used in the proposed
order to restrict the flooding area of the packages. This
routing algorithm is described as follows:
algorithm uses two approaches in order to determine which
nodes must propagate the control messages. The first one is A. Limited and Directed Geocast Approach
determined based on the position and the size of the area. The
Based on [5], [7] and [8], the limited and directed geocast
second one is established according to the distance and the
approach was created in order to perform the construction of a
localization of the destination.
new routing tree being limited and directed towards a specific
A routing protocol, that limits the research aimed at looking region of the network denominated the geocast region. In
for new routes in ad hoc networks, is described on [4]. It is order to do that, it becomes necessary the use of geographical
utilized, in that proposal, an approach based on flooding and localization information regarding the nodes, which allows to
on the localization of the nodes in order to limit the flooding direct the routing of the messages. The approach, as adopted
area. When sending a message, the emitter node defines an on this article, restricts, in the rectangular format, the region
“expected zone” of the destination node. If any node, that that contains the incidence area of events. At the time of
receives the message, is outside that zone, it is discarded. The receiving a message, each node verifies whether it belongs to
purpose is to reduce the cost of the retransmissions of each the geocast region. If it belongs, the node sends the message
message through the whole network. again to its neighbors; in the contrary case, the node disregards
the control message. It is assumed that all of the nodes know
Alternative definitions of requisition zones, being aware of
their geographical positions.
the localization, are analyzed on [5]. Those definitions have
the purpose of optimizing the performance by delimiting the The scheme that restricts the geocast region is similar to the
areas of interest. In order to do that, the areas are approached optimization scheme as quoted on [7]. This approach restricts
based on the rectangular, circular or conic format. the flooding area by the straight lines r2,r3,r4,e r5. It assumes
An algorithm for data dissemination (Directed Diffusion) is that the node A1 (Sink) knows the localization (XA2,YA2) of the
described on [6], where it is proposed a routing scheme being node A2 (Source). Being r1 the straight line that passes through
centered on data, there not being semantics of directing. It also the points A1 and A2. The distance, between the points A1 and
attempts to take care of dynamic networks through a is represented by DA1,A2.P is the mid-point between A1 and A2.
negotiation scheme with dissemination of interests and The radios R is the distance utilized to draw the parallel
reinforcements of paths, allowing the network to converge in straight lines(r2 er3)and the perpendicular straight lines (r4 er5)
the presence of any topologic modification. In addition to that, towards the straight line. The distance between and is
the Directed Diffusion utilizes, as metrics for the selection of represented by:
the route, that one that has the lower delay, which generally
leads to the route with a lower quantity of hops. A
disadvantage of this approach is the high cost of
communications for repairs of paths when there is occurrence
of failures, due to the need of periodically performing a If the distance between and is represented by ,
flooding in the network in order to reinforce other paths. then the distance between and P is indicated by:
We present, on this article, a proposal of routing protocol
for wireless sensor networks. The proposal is implemented as
a modification of the Directed Diffusion protocol, and it
differs from the solutions presented in the literature for the
fact that it uses an approach to minimize power consumption The figure 1 shows an example of the restricted geocast
during recovery from failures. Localization information is area.
 REDUCTION OF ENERGY CONSUMPTION IN WSN USING AN APPROACH GEOCAST 569

The propagation of messages performs the following steps:

• The information, needed to create the geocast region is
sent on the control message;
• At the time of receiving a control message, the node
verifies whether it belongs to the restricted area;
• If it belongs, it propagates the message to the
neighboring nodes, in the contrary case, it discards the
message as it is the case of the J and K nodes.
B. Proposed Routing Algorithm
Figure 1. Restriction scheme in the geocast region
Our routing algorithm is implemented in four phases:
A field, denominated (number of hops), was
1) Configuration Phase
created in order to calculate the number of hops as from the
“sink” node. The value of the radios is calculated as being In each run of the protocol, the sink node performs
floodings with messages of interest for its neighbors. The
the distance between the sink node and the most distant node
that could be reached with “X” hops. Where “X” is defined by messages of interest describe the task by means of an
the user. The radios construction message is sent by utilizing a attribute-value pair [6]. When a node receives an interest, it
determined integer number that will be stored in the verifies whether the interest exists in the cache. If there is no
linkage with any of the various stored interests, the node
field, which will be decreased one unit for each
creates a new entry of interest. Each entry in the cache of the
retransmission and when reaching zero, the radios
node has a gradient that points the interest for the neighbor
construction message will be discarded and a message,
that sent it. By utilizing interests and gradients, paths are
informing the geographical position of those nodes, is sent to
established among the sink node and the source nodes.
A1 (Sink) in order to determine which the most distant node is.
In that way, the radios construction message will have met its Upon receiving an interest, a node resends this interest by
target, that is to say, reaching the nodes by meeting the utilizing a gradient. For their neighbors, the interest appears as
number of hops as predetermined in the configuration phase. having been originated by the node that is the emitter of the
The size of the geocast region is calculated on the basis of message, although it may come from another distant node. In
and DA1,A2. The reduction of the area, for the construction of a this way, this protocol performs just local interactions for the
dissemination of the interests.
new routing tree, gets advantages as the number of control
messages to be interchanged will be reduced. The lower the 2) Phase of Route Establishment
is, the lower the size of the broadcast zone will be. When the interest arrives at an appropriate region, through
Following that restriction scheme of the area, the nodes the exploratory messages, one or more sensors are enabled,
must propagate the control messages only for the nodes that thus becoming sources. The source nodes send data to the
belong to the geocast region as it is shown on the figure 2. neighbors that have a gradient. The first data are called
exploratory data. Within the directed diffusion, the nodes
resend the first exploratory data, that were received, up to
arriving at the sink node, favoring the paths of low latency.
According to what was stated on [9], the route, that has low
latency, is not an effective metrics in WSN. In our protocol,
each node inserts, in the exploratory message, the number of
hops and the sum of the residual energies of each node, which
composes the path between the source and the destination. The
data message is immediately sent to its neighbors. The
exploratory messages are stored at a local cache in order that
the selection of routes is performed. A timer is created in each
node of the network and it is utilized after the reception of the
exploratory messages. The use of the timer is appropriate in
order to allow the exploratory messages, from the most distant
nodes, to arrive and that they are considered for the selection
of the path.
The selection of the routes is performed by utilizing the
Figure 2. Propagation of messages
same mechanism as adopted on [10], which allows the nodes
570 ARAÚJO ET AL.

to locally determine the route along which the data will be established. In order to do it, basically, the Directed Diffusion
conveyed without incurring in high cost related to the protocol restarts the reinforcement mechanism to search for
knowledge of the whole network topology. The selection other paths. However, the repair scheme of paths in the
process of routes can be represented as follows. Being Directed Diffusion has a high cost as regards to the energy
a directed graph, where is a set of vertexes consumption as it requires a flooding of the network to
(sensor nodes) and is the set of pairs , not ordered of reinforce other paths.
elements in (connections). In case that a message is In the case of occurrence of failures in the nodes that
conveyed through a path ) in a graph complete the stretches, our algorithm utilizes a Geocast
, where , ... , are vertexes and , ... , ) approach instead of performing a flooding of the network for
are edges, then each node in loses some of its energy linked reinforcement of other paths. That approach has the purpose of
to the cost for sending messages. Being the cost for repairing broken paths by constructing a routing tree being
sending the message, , the initial energy, and the limited and directed towards a specific region of the network
residual energy of after sending a message. Hence, in which the set of source nodes is located, thus minimizing
for . Consequently, the the number of control messages being sent and received. The
approach restricts, in the rectangular format, the area where
energy of the path is indicated by , where:
the new routing tree will be constructed. At the time of
receiving the package, each node verifies whether it belongs to
the geocast region. If the package belongs, the node sends the
message again to its neighbors; in the contrary case, the node
The rule, as implemented in the protocol, does not choose disregards the control message.
just the path that has the highest value of available residual
energy for all of the nodes, but also the number of nodes that
compose the path. As the solution utilizes just localized
interactions, the reason is used for getting the best path. The
reason is got by means of the division between the sum of
the residual energy ( ) and the number of hops that compose
the path ( ), according to what the equation shows .

Consequently, it is considered that the best path to be

chosen is that one which, amongst all of the available paths of
a given pair of vertexes of origin and destination, gets the
highest value corresponding to the reason .
On the basis of the exploratory messages and the
mechanism for selection of routes, the best path is selected, Figure 3. Reconstruction of paths
which means that it is the path that has the highest residual At the time of the definition regarding the geocast region,
energy with the lowest number of hops. the protocol disregards the node that failed, in order that this
3) Phase of Data Communication node does not compose the new routing tree that will be
created. At the end of the transmission, referred to the event
The phase of data communication occurs by utilizing a
that was interrupted due to failure of the route, the protocol
gradient between the receiver node and the node that sent the
restarts the configuration phase. The failures in the routes are
message. The gradient is established by the exploratory
identified when the sink node stop receiving the events
message. As soon as the source node starts sending its data,
associated to the interest served.
the messages are conveyed by the intermediary nodes
following the direction as pointed out by the gradient as The approach, as utilized in that algorithm, assures the
predetermined in the prior phase. Each one of the intermediary quick reconstruction of paths in the presence of failures and
nodes conveys the package of data up to arriving at the sink the reduction of the energy consumption as it avoids the
node. flooding in the whole network for the reconstruction of new
routes. The main difference, between the approach as utilized
4) Reconstruction Phase of Paths
in the Directed Diffusion and what is being proposed on this
In case of failure occurrence regarding a path between the article, is the construction of the routing tree by means of the
source node and the sink node, an alternative route must be
 REDUCTION OF ENERGY CONSUMPTION IN WSN USING AN APPROACH GEOCAST 571

occurrence of failures. In the first approach, the broadcast for broadcasts for the reconstruction of routes at the Directed
the construction of the tree has maximum reaching, that is to Diffusion, while in our proposal the broadcasts are limited to
say, it reaches all of the network nodes. According to the the geocast region and they are directed to the region of
second approach, the limited and directed construction of the interest.
routing tree is performed, having the purpose of reaching just
the nodes that are close to the occurrence of events.
IV. EVALUATION OF PERFORMANCE
In order to evaluate the performance of our proposal, we
utilized the Sinalgo simulator [11]. Sinalgo is a framework,
written in Java, that allows the simulation of wireless
networks and abstracting itself from the lower layers of the
network stack.
A. Simulation Scenario and Metrics Utilized
The simulation scenario was constructed in order to allow a
didactic understanding of the proposed algorithm. The
topology of our scenario allows that each node communicates Figure 5. Energy map (Directed Diffusion)
with up to 8 neighboring nodes. There are 121 nodes, in an 11
x 11 network topology.

Figure 4. Topology of the Scenario

There is a data flow that consists of a source and a
destination (Sink). The source node generates an event and it Figure 6. Energy map (Directed Diffusion + Geocast
is located as it is shown on the figure 4. The total simulation Approach)
time was 500 seconds and we repeated the simulation three
times. The messages were simulated with packages of 500 The depressions, as shown on the figures 5 and 6, refer to
bytes. The duration of the interest was 150 seconds. The the region between the sink node and the source node. It is
energy dissipation pattern, that was adopted, was the same as perceived that as regards to the initial energy status of the
the one utilized in [12]. The initial energy of the nodes was nodes (1,5J) and not regarding the positioning of the mesh, our
adjusted to 1,5J. Three metrics were chosen in order to proposal shows a less accentuated depression as a result from
evaluate the performance of our proposal as compared with the mechanism for selection of routes that is utilized, which
the Directed Diffusion protocol: network energy map, energy propitiates a balancing of the energy consumption.
consumption and cost of receiving messages. The network
energy map shows the residual energy of each sensor node.
The consumption of energy measures the total energy rate that
is available at all of the network nodes. The cost of receiving
messages measures how many messages are delivered for each
unit of energy (J) used by all network nodes.
B. Results Got
The energetic distributions of the network along the time
t=450s are shown on the figures 5 and 6. Analyzing the
surface, we observe the following results: on the Directed
Diffusion (Figure 5), the energy is consumed at all of the
network nodes and because of that, it can be observed that the
mesh is positioned at a lower level than the one of our
proposal (Figure 6), that occurs due to the periodical Figure 7. Energy consumption
572 ARAÚJO ET AL.

The Figure 7 shows that the energy consumption, in the It is intended to evaluate, in future works, the cost for
whole network, is reduced by utilizing our proposal. This is receiving messages, which measures how many messages are
got due to the use of the geocast approach regarding the delivered for each energy unit (J) used for all of the network
mechanism for selection of routes, which chooses the path that nodes. In addition to that, we will also evaluate the reliability
has the highest residual energy with the lowest number of of the network, which will be measured based on the delivery
hops. rate.
ACKNOWLEDGMENT
This study was partially sponsored by CNPq (National
Counsel of Technological and Scientific Development)
process Nr. 135654/2008-4 and 112816/2009-6.

REFERENCES
[1] Chen, B., Jamieson, K., Balakrishnan, H. and Morris, R. Span: An
energy-efficient coordination algorithm for topology maintenance in ad
hoc wireless networks. In 7th Annual International Conference on
Mobile Computing and Networking (July 2001).
[2] Shah, R. C. and Rabaey, J. Energy aware routing for low energy ad hoc
sensor networks. IEEE WCNC (2002).
[3] Lee, Sung-Ju; Hsu, J.; Hayashida, R. Gerla, M. et al. Selecting a Routing
Strategy for your Ad Hoc Network. University of California, Los
Angeles. Computer Communications, Vol. 26, Issue 7, pp 723-733, May
2003, Elsevier Science.
[4] Y.-B. Ko and N. H. Vaidya. Location-aided routing (LAR) in mobile ad
hoc networks. Wireless Networks, Hingham, MA, USA, v.6, n.4, p. 307–
Figure 8. Cost of receiving messages 321, 2000.
[5] Y.-B. Ko and N. H. Vaidya. Location-aided routing (LAR) in mobile ad
The Figure 8 shows that the cost of receiving messages is hoc networks. In ACM/IEEE International Conference on Mobile
reduced by approximately 50 percent, using our proposal. This Computing and Networking (MobiCom’98), October 1998.
is due to the reduction in the number of control messages and [6] C. Intanagonwiwat, R. Govindan, and D. Estrin, “Directed Diffusion: A
scalable and robust communication paradigm for sensor networks”,
the selection of routes that have a smaller number of hops with Proceedings of ACM MobiCom'00, Boston, MA, pp. 56-67, Aug. 2000.
higher residual energy. [7] Y.-B. Ko and N. H. Vaidya. Optimizations for location-aided routing
(LAR) in mobile ad hoc networks. Technical Report 98-023, CS Dept.,
Texas A&M University, November 1998.
V. CONCLUSION [8] Y.-B. Ko and N. H. Vaidya. Location-based multicast in mobile ad hoc
We presented, on this article, a routing algorithm that aims networks. Technical Report 98-018, CS Dept., Texas A&M University,
September 1998.
at reducing the energy consumption of the network. The [9] I. F. Akyildiz, W. Su, Y. Sankarasubramaniam, and E. Cayirici, “A
proposal consists in the modification of the Directed Diffusion survey on sensor networks”, Communications of the ACM, pp. 102-114,
protocol. Two modifications were implemented in the original Aug. 2002. A.P.R. Canázio, Automatic Cork Processing System,
Reference Manual. Addinson Wesley Publishing Company, 2002.
protocol: the first one is a geocast approach and the second
[10] Teixeira, I., de Rezende, J. F., and Pedroza, A.C.P. - "Wireless Sensor
one is a mechanism for selection of routes. The modifications Network: Improving the Network Energy Consumption", in XXI
have the purpose of reducing the flooding caused by the Simpósio Brasileiro de Telecomunicações - SBT'2004, Belém, Brazil,
Directed Diffusion in order to recover broken paths and to September 2004.
[11] Sinalgo (2007). http://dcg.ethz.ch/projects/sinalgo/.
determine the best route for delivery of the collected event. [12] W. Heinzelman, A. Chandrakasan, and H. Balakrishnan,“Energy-
Efficient Communication Protocol for Wireless Microsensor Networks”.
The evaluation of the proposal, by means of simulation,
Hawaii International Conference on System Sciences, January 4-7, 2000,
demonstrated that the utilization of the geocast approach and Maui, Hawaii.
of the mechanism for selection of routes shows positive results
for reducing and balancing the energy consumption of the
network.
 Conflicts in Online Social Network Privacy Policy
Analysis and Resloution

Zhengping Wu Yuanyao Liu

Department of Computer Science and Department of Computer Science and
Engineering, University of Bridgeport Engineering, University of Bridgeport
221 University Avenue, Bridgeport, CT 221 University Avenue, Bridgeport, CT
06604, USA 06604, USA
[email protected] [email protected]

Abstract—Social network sites have the potential to OpenSocial is a set of common application programming
fundamentally change the character of our social lives, interfaces (APIs) for web-based social network applications,
both on an interpersonal and a community level. In the developed by Google along with MySpace and a number of
past few years, social network sites have become integrated other social networks. Applications that build with OpenSocial
into the daily practices of millions of users. Social APIs can be implemented into different social network sites,
applications provide a convenient approach where users which support OpenSocial APIs. Social applications are
can communicate and share their personal information. information exchange utilities, which will acquire user’s
However, privacy and information security have been information from social network sites, where users put their
information, and then provide some service base on this
concerned as a critical issue in development social
information. Social applications usually provide some privacy
applications. Policy-based management, which is a
control settings, which will be enforce into social network site
management approach, has been used to control the where the application is installed. It means if a same user
information access privilege. Users’ policies are more installs one application on two different social network sites,
flexible and ease to change. However, different policies each site will have one privacy control setting. If this
may cause conflicts that affect execution of social application is used to publish user signed information, two
applications. In this paper, we propose a semantic different sites may publish different information, which belong
temporal logic mechanism to detect policy conflict in social to a same user. This is a problem.
network environment.
Privacy concern has been a critical issue since social
Keywords-component; network sites appear. Privacy control that social network site
provide is limited to usage of social network site self. The
I. INTRODUCTION privacy control in social applications has to be set by users or
follow the social network site’s setting. Since online social
Social network sites have become increasingly popular; privacy is harder to guarantee [1], privacy concern has gotten
many people are using Facebook, Myspace, and Twitter to more attention from society. Policy-based management has
communicate with others. Social network sites are not a been used in social network sites. The privacy settings in social
platform for communication; it has become a collaboration network sites are a kind of policy rules. These policy rules are
platform, since social applications provide a great extension. easy generated into machine-readable format. Policy-based
People share their information with others on social network management is an administrative approach to manage system
sites, but the privacy control is simple and limited through a behaviors and rules within an information domain. A policy
number of selections that social network sites provide. describes the set of behaviors that a party is required or allowed
The main motivation for the users to create accounts and to take. In social network environment, these behaviors are
participate in these services is sharing their information with usually access actions. There are two main types of policies in
others for some purposes. People usually sharing two types of social network sites, one is global policy, which is used to
information: professional information and private information. control entire information set within user account, and another
In the first case, social network sites are as a utility to establish type is application policies. Application policies are used to
and entertain business task. In the second case, social network control information that will be used in application processing.
sites are used to share private information such as personal Policy conflicts can be classified into static conflicts and
contract details, personal photos and videos. However, by using dynamic conflicts. Analysis of static policy conflicts is time-
social applications, social network sites are not just used to independent. These conflicts can be resolved by simple logic
share information, but to communicate with specifically people. reasoning or compiling techniques. Dynamic policy analysis is
Sharing information has become information exchange. time-dependent, which means the applicability of policies and
information will change over time. The major issue about

573
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_100, © Springer Science+Business Media B.V. 2010
574 WU AND LIU

dynamic policy analysis is the changing information. The On online social network sites, users define policies; they
information covered by policies can be identified as different may not have ability to make sure policies on different online
types of elements. Relations between different elements are social network sites are consistent. If there are positive and
critical to conflict detection and resolution. negative authorization or obligation policies applied on the
same subjects, targets and actions, modality conflicts may arise
There are two platform-independent privacy languages: [8]. Modality conflicts can be detected through static analysis
IBM’s Enterprise Privacy Authorization Language (EPAL) [2] of the policy specifications. Once all policy-driven actions of a
and the OASIS eXtensible Access Control Markup Language system have been analyzed and these different types of
(XACML) [3] along with Privacy policy profile of XACML conflicts that can arise have been identified, it is possible to
[4]. XACML is a large, complex language, and this can make define rules that can be used to recognize conflict situations in a
formal analysis more difficult. While EPAL and XACML are policy set. Then these rules can be invoked during a conflict
very similar, and share a common authorization model, they detection process prior to policy deployment and enforcement
differ in important ways. In almost every area, the to identify potential inconsistencies. This is known as static
functionality of XACML 2.0 is a superset of EPAL 1.2 [5]. conflict detection and takes place at specification time. Most
Where the two languages differ, the EPAL differences usually existing work in detecting conflicts are related to modality
result in less functionality than XACML has. In many cases, conflicts. But sometimes application-specific conflicts cannot
the EPAL 1.2 differences from XACML make construction of be detected directly from policy specifications without
flexible or scalable privacy policies impossible or difficult. additional information. Conflicts may arise from the semantics
However, there is not a specification for social applications. In of the policies as well. Thus, a metapolicy [9] concept is
this paper, we will use a simple XML format to present introduced to represent this additional information, which is a
privacy policies. constraint or context associated with related policies. Yet there
is still a possibility of conflicts even when there is not any
II. RELATED WORK shared subject, targets or actions between two policies. There
may be implicit relationships between subjects, targets and
Privacy within online social network sites is often not
actions when policies are applied to them. Thus a metapolicy-
expected or is undefined [6]. Dwyer [1] gives us an image that
based approach cannot resolve all conflicts, because an
different online social network sites make users have different
administrator would have to specify metapolicies for all
privacy concern level. The social application, which is an add- possible runtime conflict situations. It is almost impossible to
on for social network site, also affects the reputation of social predict all implicit relationships that may cause conflicts in the
network sites. In another hand, the social application is installed system by a user.
into user’s interface (social applications will retrieve user’s
identify information and set up a connection between user and In classical logic, propositional logic is based on a set of
application server if it is needed). The social application always elementary facts by using a set of logic operators. It indicates a
has more privilege to access user’s information, like Boolean value set. First order logic (FOL) [10] is an extension
Facebook’s APIs give developers more rights to access user’s of propositional logic. Higher order logic [11] is an extension
friends list. In this case, social applications can get more of first order logic. Temporal logic [12] is an extension of
information than user allowed. Therefore, privacy concern on classical propositional logic, which is built for representing the
social applications is more than online social network sites set of domain elementary facts by using a set of logic operators
themselves. [12]. And ontology [13] is an explicit specification of a
Police-based systems are suitable solutions for automated conceptualization, which represents a set of entities and the
online collaborations across domain boundaries, but policies relationships between these entities in a specific domain.
become more and more complex for cross-domain activities, Entities are abstracted from real world objects in that domain;
which require more adaptive management. Since a request or relationships are contacts or effects between objects. The
an action from a foreign domain may not have an exact match ontology describing this type of information uses a set of
with an existing policy, several compatible policies can be formal terms and operators.
applied to it. Choosing one of these policies may lead to a
conflict with another related policy or violate the principle of III. POLICY MODEL
least privilege [7]. Even a similar request or action with
different parameters may lead to difficulties when applying a 3.1 General Policy Model
previously compatible policy. In general, whenever multiple As a policy can contain one or more rules, which describe
policies apply to an action or request, there is a potential for relations between policy executants and requester (subject and
some form of conflict, but it is essential that multiple policies object), we focus on a simple policy that contains one rule. For
should be applied in order to cover the diversity of management more complex policies, they can easily be divided into simple
needs and of different groups of users. There may be different policy segments. A policy segment describes a behavior and
policies related to security, privacy, or other management attributes about this behavior. For each behavior, there is an
requirements, which are applied to different aspects reflecting executor or a type of executors, an object or a type of objects,
different management needs. So policy conflict analysis and some constraints, which limit and describe the action. So
(detection and/or resolution) becomes an essential part in in each policy, there are four major components: subject,
management of policy-based systems. object (resource), action and context constraints. A policy
segment describes a complete behavior, which contains subject,
 CONFLICTS IN ONLINE SOCIAL NETWORK PRIVACY POLICY 575

object, action, and context constraints. These elements are <?xml version="1.0" encoding="utf-8"?>
consisted of attributes, where detail information is stored. The <Policy> Policy 1
policy’s subject is a set of attributes (A) identifying the <Workspace>
action’s executor, it can be denoted as s={A}; the object is a <Update>
set of attributes (A) that describes an action’s target, it can be <Permit>
denoted as o={A}; context constraints include other attributes <Group name="Colleague"/>
within the policy, such as network contexts, execution <User name="Allen"/>
</Permit>
environments, and etc. We distinguish this type of context
<Period start="08:00:00" end="13:00:00"/>
information to other attributes that are about subjects and </Update>
objects. A policy’s action actually represents a temporary </Workspace>
binding between subjects and objects, which can be denoted as </Policy>
A(c)=s×o. Following notations show the components of a
policy segment in this general policy model:
<?xml version="1.0" encoding="utf-8"?>
A: attribute the attributes that describe the subjects and <Policy> Policy 2
objects. <Workspace>
S: {As} the subject of a segment, the set of attributes <Update>
<Permit>
that limit policy subject.
<Group name="University"/>
O: {As} the object of a object, the set of attributes </Permit>
that limit policy objects. <Period start="12:00:00" end="17:00:00"/>
A(c)=S×O the action of segment, it indicates the </Update>
relation between subject and object. </Workspace>
Context the information that hide in the network or </Policy>
environment.
In this policy model, each element is denoted in a set of In policy 1, the subject has attributes, which is role
attributes. In another words, a policy is a set of attributes with information. It indicates the request from the “Colleague”
different constrain conditions. group members and specific user “Allen” will be permitted
within a time period. In policy 2, the subject has the same type
attributes, which is also role information. It indicates the
3.2 Privacy Control Policy Model request from the “University” group members will be permitted
A privacy control policy should contain a resource (object), within a different time period compared to the first policy.
a requester (subject), an action and a set of information that Actions in these two policies are the same. The action reflects
limited these elements. The subject is the information requester, the relation between subjects and objects. In this case, action
who wants to access the information; the object is the resource “Update” is showing that resource will be changed by the
or the information; the action is usually access or other action subject. Obviously, there is a conflict between these two
that requester want to perform. Therefore, the privacy policy policies.
model can be presents as follow: Between two policies, the dynamic conflict will involve any
s = {As} s: information requester. element within those policies. We generated some situations
o = {As} o: the user’s information. that can be used in conflict analysis:
A(c) = s×o A(c): the action that requester a) Identity may be replaced by role information to act as
want to perform. the subject ;
Context: Context: the additional b) The identity information is provisional;
Additional information to describe s, o and c) The resource is provisional;
information action. d) The relationships (action) between a resource and an
Rule = Rule: a functional segment of a identity are provisional;
{s×o×A(c)×Context} policy e) The relationships (action) are constrained by network
Table 1 Privacy Control Policy Model or security domain contexts
A rule is a functional segment of a policy; it indicates a
condition, which is used to filter requests. Here are two privacy IV. SEMENTIC TEMPORAL LOGIC
control policies that are used to check requester’s role
information and time information. Here is an example scenario: 4.1 Temporal Logic and Ontology
an organization collaborates with a research group in a One major issue about dynamic policy analysis is that
university. They use online social network site as the information fluent will change over time. The dynamic policy
collaboration platform. The social application provides the analysis has to keep track of the change of information fluent.
information exchange functionality. Here are two policies about
the update action of different participants. It is easy to represent static states and values using classical
logic. However, time-dependent situations cannot be
576 WU AND LIU

represented by classical logic. Temporal logic also called tense Update.permit={true};

logic can handle these time-dependent situations. Temporal University.time={12:00,17:00};
logic is an extension of classical propositional logic, which Update.permit={true};
represents a set of domain elementary facts using a set of logic The relationship between “Colleague” and “University” is that
operators [14]. It has been broadly used to cover all they play mutually exclusive actions, and there are some
approaches to the representation of temporal information restrictions.
within a logical framework [15]. Relation: not_Update (Colleague, University, t);
In Policy 1& 2, time period is a very important attribute. To build the representation foundation, we use ontology to
This attribute affects the effect of these policies. describe the model of policy domain and relations within that
08:00 12:00 13:00 17:00 policy domain. This ontology keeps track of the relations
between elements in the policy domain. Ontology describes the
Policy 1 elements in a domain and also the relationships between these
elements. Different ontology languages provide different
Policy 2 facilities. Using OWL-Full, the above policy example,
“Colleague” is a class, which has some properties such as
“Update” and “from 08:00 to 13:00”. The “University” is
Figure 1 Policy 1&2 Time Period 
another class, which has properties such as “Update” and “from
The Figure 1 is time period of two policies. In traditional 12:00 to 17:00”. The relation between “Colleague” and
logic, it is hard to express this attribute. However, in temporal “University” is “they play same action”, and the property on
logic, we can present this situation as: this relation is mutually exclusive (12:00 to 13:00). And we can
see there is another element in Policy 1, user “Allen” is also
HoldsAt(colleague,Workspace,update(),t1); allowed to update the object during “Colleague” time, however,
Terminate(colleague, t3) we do not know other attributes about this special element.
HoldsAt(University,Workspace,update(),t2); Therefore, when we analyze these policies, we have to check
Terminate(colleague, t4) properties of “Colleague”, “University” and user “Allen”:
t1=08:00:00, t2=12:00:00, t3=13:00:00, t4=17:00:00.
The temporal logic can denote the state of elements and the Policy= {“Colleague”, “Allen”}×”Workspace” ×”Update”
changing of states. However, temporal logic does not contain ×Period;
information about relation with these policies. We need Policy= “University” ×”Workspace” ×”Update” ×Period;
something that can express and store relation information about
elements with a policy domain. The policy domain is a
collection of elements and services, administered in a V. DYNAMIC POLICY ANALYSIS
coordinated fashion [14]. 5.1 Dynamic Policy Analysis
Ontology defines a set of representations of classes, Dynamic policy conflict analysis is then based on logic
instances, attributes, and relationships among them. Therefore, reasoning and Boolean function comparison. The complexity
the ontology-based knowledge representation is a good tool for of policy rules and the environment fluent always affect the
the purpose of information transferring, sharing, and analysis. result of analysis. And also, the policy conflict analysis needs
It is a good support for our proposed Semantic Temporal lots of manual work to adjust semantic relationships for
Logic (STL). accuracy, especially distinguishing subjects and objects in
policy segments. Temporal logic can be used to perform
4.2 Semantic Temporal Logic (STL) automatic policy conflict analysis for dynamic policies, but the
accuracy cannot be guaranteed due to the lack of capability to
Using temporal logic to process cross-domain or multi- support semantic relationships among entities. The proposed
domain policies, semantic relationships among different Semantic Temporal Logic (STL) uses ontology as the
components is unavoidable, which reduce the accuracy and knowledge supplier, which improves accuracy and reduces
efficiency of policy analysis. STL provides a bridge over this ambiguity in analysis results. The ontology, which records the
barrier by applying an ontology-based knowledge
policy model and relationships within the policy model, helps
representation onto temporal logic itself. This representation
provides not only information of individual entities from the logical reasoning mechanism identify elements and
specific domains but also the relationships among these relationships between elements. For example, Policy 1 and
entities, which are key issues for policy analysis. Policy 2 are two privacy control policies express the relations
between two elements: Colleague and University.
The ontology-based knowledge representation stores the Logical expressions for this example are:
information of specific policy domain, which contains each HoldsAt(Colleague,Workspace,update(),t1);
entity’s information and relationships among several entities. HoldsAt(University,Workspace,udpdate(),t2);
Back to Policy 1 and Policy 2 described in the previous section, 08:00 < t1 <13:00;
there are some attributes within these policies: 12:00 < t2<17:00;
Relation: not_update(colleague, university, t3);
Attributes : Colleague.time={08:00,13:00}; 12:00<t3<13:00 ;
 CONFLICTS IN ONLINE SOCIAL NETWORK PRIVACY POLICY 577

So the STL rules for this example become: Trajectory(permit(Sub,Obj2,A2(c),t2),t,deny(Sub,Obj1,A1(c),t1)

HoldsAt(colleague, Workspace,Update, t1) ,d)
Terminate(colleague, t2) Relation: HoldsAt(sub,obj1,A1(c),t1);
HoldsAt(University, Workspace, Update,t3) change(obj1,obj2,t);
Conflict(Update(colleague, Workspace)&&Update(University, t1<t<t2;
Workspace),t) [The situation c: a subject (sub) may perform different actions (A1(c)
and A2(c)) toward a changing object (authentication secret) in the
So in a rule of STL, semantic relationships are combined time period between t1 and t2. A conflict of interest may occur. Then
only one action will be permitted.]
with logic predicates to represent policies from multiple
domains with dynamic elements. This rule in the above
example uses the relationships between different entities to VI. IMPLEMENTATION AND EXPERIMENTS
detect dynamic conflicts among different policies.
6.1 Privacy Policy Conflict Analysis
Privacy-control policies often place requirements on the
actions of more than one subject. The dynamic environment We design a prototype system that can dynamically analyze
could affect the enforceability of policies. This dynamic privacy policy to detect potential conflicts within an online
environment could be expressed by ontology. The ontology social network collaboration environment. The system is
contains relations, attributes and other restrictions. The implemented in Microsoft .NET platform. We use its subset
information within this ontology supports the temporal logic OWL-Full to store and retrieve our ontology-based temporal
evaluations. We use the privacy policy model discussed before logic STL.
as the foundation of dynamic policy analysis.
This prototype is designed to analyze cross-domain privacy
Here are STL rules that are used to detect conflicts; these control policies and indicate whether there is any conflict
rules are also used to return conflict type. when a cross-domain collaboration activity happens. The
¾ HoldsAt(permit(Role1(sub),obj,A1(c)),t) ġ prototype contains two parts, one is temporal logic evaluation,
HoldsAt(permit(Role2(sub),obj,A2(c)),t) ġ A1<>A2 ė which is the logic mechanism, and the other one is ontology
HoldsAt(overlapConflict(conflictOfDifsubject,overlaps(permit( model, which contains relations within policy model.
Role1(sub),obj,A1(c)),permit(Role2(sub),obj,A2(c)),t) ė
Trajectory(permit(Role1(sub),obj,A1(c)),t,deny(Role2(sub),obj, 6.2. Experiments
A2(c)),d) Ģ
Trajectory(permit(Role2(sub),obj,A2(c)),t,deny(Role1(sub),obj, The policy set used in the experiment contains 5 pairs of
A1(c)),d) privacy control policies. Each pair of policies indicates one
Relation: sub.attribute.role1Įsub.attribute.role2; type of conflicts we mentioned before. And there are 45
begin<t<finial; random combinations, which are not all conflict. The number
[The situation a: the subject (sub) can have different two roles of conflicts of these 45 pairs is 21. The prototype analyzes
(Role1(sub) and Role2(sub)). When different roles perform different these policy pairs based on different types of conflict, which
actions (A1(c) and A2(c)) toward the same object (obj) at time t, then
we discussed before. If use temporal logic to analyze these
an overlap conflict may occur. Only one action will be permitted.]
¾ HoldsAt(doAction(sub1,obj,A1(c)),t) ġ
policy pairs without ontology part, the accuracy will decrease.
The Figure 2 illustrates a comparison of the accuracy when
HoldsAt(doAction(sub2,obj,A2(c)),t) ġ A1<>A2 ė
HoldsAt(dynamicConflict(conflictOfDifsubject,Overlaps(doActi using STL and using general temporal logic analysis result.
on(sub1,obj,A1(c)),doAction(sub2,obj,A2(c)),t),t) ė From this figure, we can see that STL could easily analyze
Trajectory(permit(sub1,obj,A1(c)),t,deny(sub2,obj,A2(c)),d) Ģ both designed policy pairs and random policy pairs with higher
Trajectory(permit(sub2,obj,A2(c)),t,deny(sub1,obj,A1(c)),d) accuracy than general temporal logic.
Relation: at time t,sub1.attribute.IdĮsub2.attribute.Id;

HoldsAt(sub1,obj,A1(c),t); 
HoldsAt(sub2,obj,A2(c),t); 
begin<t<finalt 
[The situation b: when two different subjects (sub1 and sub2) use 
same authentication secret (obj) perform different action toward the 1XPEHURI 5HSRUWHGE\67/ 5HSRUWHGE\7/
same target at the same time t, another type of overlap conflict may FRQIOLFWV
occur. Only one action will be performed on the object.] 'HVLJQHG3ULYDF\3ROLF\3DLUV
¾ HoldsAt(permit(Sub,Obj1,A1(c),t1)) ġ 5DQGRP3ULYDF\3ROLF\3DLU
HoldsAt(permit(Sub,Obj2,A2(c),t2)) ġ t1<=t<=t2 ġ
Clipped(t1,authorize(sub,secret,t),t2) ė Figure 2 Comparison of the accuracy
HoldsAt(dynamicConflict(conflictofInterest,overlaps(permit(Su
b,Obj1,A1(c),t1),permit(Sub,Obj2,A2(c),t2),Clipped(t1,authoriz VII. DISCUSSION
e(sub,secret,t),t2)),t),t) ė Temporal logic has been used to analyze required
Trajectory(permit(Sub,Obj1,A1(c),t1),t,deny(Sub,Obj2,A2(c),t2)
properties in trust policies, such as the FTPL mechanism [16],
,d) Ģ
which use first order temporal logic to check satisfaction of
578 WU AND LIU

states. In 2003, Event Calculus was proposed to analyze a [4] Moses, T., ed., Privacy policy profile of XACML v2.0; OASIS
Standard, February 1, 2005; http://docs.oasis-
combination of authorization and management policies [17]. open.org/xacml/2.0/access_control-xacml-2.0-privacy_profile-
But the authors do not identify the special capability of Event spec-os.pdf.
Calculus for transient properties in various activities. We apply [5] Anderson, A., Comparing Two Privacy Policy Languages:
Event Calculus in STL for analyzing transient properties and EPAL and XACML, Sun Microsystems Laboratories Technical
actions (relationships) in collaboration environment and find Report 2005-147, 2005;
its capability to detect and resolve dynamic conflicts. http://research.sun.com/techrep/2005/smli_tr-2005-
147/TRCompareEPALandXACML.html.
Meanwhile, static and dynamic conflict detection and [6] Dwyer, C. (2007) “Digital Relationships in the 'MySpace'
resolution in policy-based management for large open Generation: Results From a Qualitative Study.” Proceedings of
the 40th Hawaii International Conference on System Sciences
distributed systems are discussed in [18]. A general model of (HICSS), Hawaii, 2007.
security policies has been discussed in [19]. Detection and [7] Liang Chen, Jason Crampton, “Inter-domain role mapping and
reconciliation of security policy conflicts following that model least privilege,” Proceedings of the 12th ACM symposium on
are restrained by the complexity of the policy set to be Access control models and technologies, pp. 157-162, 2007.
reconciled. And only two-party conflict reconciliation can be [8] M.Sloman, “Policy Specification for Programmable Networks,”
tractable. Table 2 illustrates these comparisons. Proceedings of the First International Working Conference on
Active Networks, pp: 73 – 84, 1999.
Dynamic Policy Conflict detection [9] M. Sloman, E. Lupu, “Security and Management Policy
ity domain and resolution Specification,” IEEE Network, vol. 16, no. 2, pp. 10-19,
March/April 2002.
STL Dynamic Multiple Detection and
and resolution [10] Nir Friedman, Joseph Y. Halpern, Daphne Koller, “First-order
Static suggestion conditional logic for default reasoning revisited,” ACM
Transactions on Computational Logic (TOCL), Vol. 1, Issue 2,
FTPL Static Single Detection PP. 175 – 207, 2000
Conflict Static Single Detection and [11] Andrews P.B., “An Introduction to Mathematical Logic and
detection and resolution Type Theory: To Truth Through Proof, 2nd ed.” Kluwer
in [18] Partial suggestion Academic Publishers, PP: 201-203, 2002
dynamici [12] P. Bellini, R. Mattolini, and P. Nesi, ”Temporal logics for real-
ty time system specification,” ACM Comput. Surv., vol. 32, no. 1,
Ismene in N/A Multiple Detection and pp. 12-42, March 2000.
[19] resolution [13] B. Orgun, M. Dras, A. Nayak, G. James, “Approaches for
suggestion semantic interoperability between domain ontologies,”
Table 2 Comparison of Different Policy Analysis Proceedings of the second Australasian workshop on Advances
Mechanism in ontologies, Vol. 72, PP: 41-50, 2006
[14] A. Westerinen, J. Schnizlein, “RFC3198 - Terminology for
Policy-Based Management,”
VIII. CONCLUSIONS http://www.faqs.org/rfcs/rfc3198.html 2001
In this paper, we use Semantic Temporal Logic for [15] Edward N. Zalta, "Temporal Logic", the Stanford Encyclopedia
of Philosophy, fall 2008 Edition,
analysis of privacy policy conflict in an online social network
http://plato.stanford.edu/archives/fall2008/entries/logic-
collaboration environment. Through the experiments on our temporal/
prototype system, the improvement on capability and accuracy [16] Arun K. Eamani, A. Prasad Sistla, “Language based policy
of Semantic Temporal Logic for dynamic privacy policy analysis in a SPKI Trust Management System,” Journal of
analysis is confirmed. A comparison of different policy Computer Security, Vol. 14, No. 4, pp. 327-357, 2006.
analysis mechanisms shows that the ontology-based policy [17] A.K. Bandara, E.C. Lupu, and A. Russo, “Using event calculus
analysis system is more flexible, accurate, and applicable. to formalise policy specification and analysis,” Proceedings of
IEEE 4th International Workshop on Policies for Distributed
Systems and Networks, pp. 26- 39, 2003.
REFERENCES [18] Nicole Dunlop, Jadwiga Indulska, Kerry Raymond, “Methods
[1] Catherine Dwyer, Starr Roxanne Hiltz, Katia Passerini , “Trust for Conflict Resolution in Policy-Based Management Systems,”
and privacy concern within social networking sites: A Proceedings of 7th International Conference on Enterprise
comparison of Facebook and MySpace”, Proceedings of the Distributed Object Computing, pp. 98-109, 2003.
Thirteenth Americas Conference on Information Systems, [19] Patrick McDaniel and Atul Prakash, “Methods and limitations
Keystone, Colorado August 09 - 12 2007. of security policy reconciliation,” ACM Transactions on
[2] IBM, Enterprise Privacy Authorization Language (EPAL), Information and System Security, Vol. 9, No. 3, pp. 259-291,
Version 1.2, 2003; 2006.
http://www.w3.org/Submission/2003/SUBM-EPAL- 20031110/.
[3] Moses, T., ed., eXtensible Access Control Markup Language
(XACML), Version 2.0; OASIS Standard, February 1, 2005;
http://www.oasis- open.org/committees/
tc_home.php?wg_abbrev=xacml.
 Developing a Feedback Model for Managing
Congestion in Packet Switched Wide Area Networks
By

*E.F. Ogbimi, A.O. Oluwatope, G.A. Aderounmu

Department of Computer Science and Engineering
Obafemi Awolowo University,
Ile-Ife, Osun State, Nigeria
[email protected]

ABSTRACT continue to conduct research on how to improve performance

The objectives of the paper was to design a feedback model for of packet switched WAN.
congestion control in packet switched WAN and carry out the
performance evaluation of the model. A feedback approach was 2 LITERATURE REVIEW
generated by designing a feedback model. Control theory was 2.1 Related Research Works
used to assess the stability of the bandwidth. The stability of
flow of bandwidth was determined by finding the transfer
Janarthanan, et al. (2003) worked on Enhancement of
function of the routers. A fluid flow model was developed using Feedback congestion Control by deploying active congestion
a queuing model. Secondly, the flow of bandwidth was control. They used a scheme called active congestion control
optimized and the stability was checked. The conditions for to make the feedback congestion control more responsive to
stability in a packet switched WAN were established. network congestion, using the active network technology.
The work was done by designing the finite state machine
1 INTRODCTION models of every single component in the modules. The model
1.1 Background of Packet Switched WAN was decomposed into set of states and working of the model
The network evolution has been sparkled by the increased embedded in the transition between states.
application of exponential improvements in fundamental General AIMD (Addition Increase Multiplicative Decrease)
software and hardware technologies. These technologies are congestion control was written by [25]. They proposed a
enabling more distributed network architecture characterised window adjustment strategy where the window increases and
by the adding intelligence to the core of WAN, integrated decreases value used as parameters depending on the ACKS
multiprotocol/multiservice transport capability and variety of (Acknowledgements) received or the packet lost. The
new services and service capabilities. The rapid growth of approach used was GAIMD (General Addition Increase
WAN in recent years is part of the struggle for network Multiplicative Decrease). The analytic expression for sending
administrators to monitor the network for demand resources rate of GAIMD was derived as
such as voice, video and data. Most users continue to add = (1)
more networking devices of higher capacities to satisfy the
needs of network users. Yet, the network remains congested
and application software continue to run slowly and Mahajan and Floyd (2001), wrote on RED-PD (Random
experience poor performance of networks. New breed of Early Detection with Preferential Dropping). They used
software application and information technology drivers like packet drop at the router to the preferentially drop packets of
academic, industrialists and business tycoons contribute to high-bandwidth flow to address congestion. The approach
this problem by installing new software which may not be used was a partial state flow. In the approach, it identified
favourable to the networking environment [9]. Academics, high bandwidth during times of congestion and designed an
Industrialists and business tycoons demand for bandwidth at architecture for RED-PD. Preferential dropping was done by
the same time which leads to the flow of bandwidth computing probabilistically dropped packets from monitored
instability. The bandwidth demand for resources is higher flows before entering the output queue.
than the capacity of the network in this case. In view of the P (dp) = 1 – (2)
various efforts of the network providers to sustain network
performance, congestion and poor performance are still
common features of networks. Hence, there is need to § = drop rate, R = Target Response function

579
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_101, © Springer Science+Business Media B.V. 2010
580 OGBIMI ET AL.

2.2 A new feedback scheme acknowledgement will indicate that the packet has been sent
The new scheme being proposed was called dynamic explicit to the receiver. Some of the packets could be lost, delayed.
feedback model which will keep the bandwidth stable. It will The feedback controller passes a feedback message to any of
reduce packet drops to its barest minimum. It will also reduce the routers that there is congestion somewhere along the link,
bandwidth delay in packet switched WAN. It will balance the then the router adjusts the speed of transmitting packets. The
flow of transmission through paths. A feedback mechanism is stability of flow of packets will then be regulated.
added to monitor the acknowledgement of packets from
receiver to sender. It will also make bandwidth utilisation
high, converge to fairness and minimize amplitude of
oscillations and maintain high responsiveness.

3 METHODOLOGY
A feedback approach was generated by designing a feedback
model for congestion control in packet switch WAN. The
principle adopted was control theory. Control theory was
used in this case to assess the stability of flow of bandwidth
in a packet switched wide area network, when congested and
when not congested. The stability of the bandwidth was
determined by finding the transfer function of the flow of
bandwidth of the network. The transfer function is done by
finding the feedback gain of the routers in the network. In
this paper, three major things were done. Firstly, a fluid flow
model for packet switched WAN was developed using
queuing model. Secondly the flow of bandwidth will be
optimized and the stability was checked. Thirdly, the
conditions for stability in a packet switched WAN were
established. The result was validated using MATLAB. The
model was implemented in MATLAB.
3.1 Control Theory 3.3 Queuing Theory
Control theory is defined as the mathematical study of how to Queuing theory is the mathematical study of waiting lines.
manipulate the parameters affecting the behaviour of a Queuing theory is applicable in packet switched WAN
system to produce the desired or optimal outcome [21]. because queues can be chained to form queuing networks.
The main conceptions of control theory are dynamic systems, The use of queuing in packet switched WAN allows the users
desired outcome, its state and feedback. Control theory was to wait for their request from any of the web servers to be
concentrated mostly on the problems of stabilization. completed and idle routers will be available for packets to be
The stability of the bandwidth was evaluated using feedback distributed.
control theory. This was done by finding the transfer function 3.4 Notation
of the arrival rate of packets. Transfer function is defined as Kendall’s notation was adopted in this paper. It is notated as
the ratio of the output of the system to the input of the system A‫פ‬B‫פ‬C; it denoted a queuing system having A as interarrival
[17]. time distribution, B as service time distribution and C as
Mathematically, Transfer function is defined as number of servers or routers [20].
H(s) = (3) A convenient notation for summarizing the characteristics of
queuing situations in a system is given by the following
Where Y(s) = Output function of the system in Laplace
format: (K‫פפ‬C): (GD‫[ ) פ פ‬19], where:
domain
X(s) = Output function into the system in Laplace domain K = arrival distribution
3.2. Model Construction M = Service distribution
The figure below shows the block diagram which illustrates C = Number of servers in the network
what this study was set out to achieve. The diagram shows GD = General discipline (all types of discipline)
transmission of packets from source (s) to destination using a First = Indefinite number of network users accessing the
feedback model. A message is transmitted to the first router network
which acts as a buffer. The message remains in the buffer for Second = Capacity of resources accessed or downloaded
a while until there is an available router to send packets until from the web
the packets get to their destinations. The message is broken The steady-state measures of performance in a queuing
into packets by means of packetization. The packets are then situation are:.
distributed to the second and then third routers. There could ૃ = rate of arrival of packets
also be many routers depending on the topology of the μ = service rate (rate of transmitting packets)
network and the environment the network is used. An Ls = Average number of users on the network
 DEVELOPING A FEEDBACK MODEL FOR MANAGING CONGESTION 581

Lq = Average number of users requesting for resources from Ls = Lq + § (7)

a router in the network (in the queue).
Wq = Expected time users spend in requesting for resources Ls = P0 +§
from a router on the network.
Ws = Expected time users spend on accessing the network = P0 +§( (8)
Pn = Probability of routers that do not accept packets
= service time (time taken to transmit packets) These are the measurements of performances in a multiple
= Expected number of routers that do not accept packets server and multiple queue in a packet switched wide area
networks.
c = number of routers § = network utilization
The number of network users requesting for resources from
the web and network users on the network will be
The model was applied to the situation of packet switched
differentiated with respect to the arrival rate of packets. Ls
WAN. The parameters listed above were applied to the
and Lq will be differentiated with respect to ૃ.
situation of packet switched WAN accordingly. The queuing
model for packet flow in packet switched WAN was Lq = 0 (9)
modelled by the following parameters above.
Assuming c =1 for a single server to many number of users
The difference between average number of users on the
network and average number of people requesting for Lq = P0 (10)
resources from the web server is denoted by this equation
Network utilization = § = ; insert § = in equation (10)
Ls - Lq = ૃeff/μ (4)
Network utilization = /c [19] Lq = P0
For a single server model (c =1), there is no limit on
maximum number in the network. The model assumes a
= Po
capacity source. Arrivals occur at rate ૃ user per unit time. §
< 1 and ૃ < μ. If arrival rate is higher than the service rate,
then the geometric series will not converge and the steady-
= Po
state probability will not exist. The queue length will
continually increase and no steady state will be reached in the
network.  Lq = Po (11)
For multiple server models which is the area of
concentration. There are c parallel servers (routers). The Computation of transfer function
arrival rate is ૃ and the service rate per server is μ. There is The number of users requesting for network resources from
no limit on the number of routers in the network. With routers and number of users on the network (Lq and Ls) will
Kendall’s notation (K‫פ‬M‫פ‬C) : (GD‫)¨פ¨פ‬ be differentiated with respect to arrival rate of packets (ૃ)Ǥ
K is the arrival distribution, M is the service distribution and The differentiated equation• will be linearized and
C is the number of servers in the network. The queue differentiatedǤThe Laplace’s transform for the differentiated
discipline is generalized distribution that satisfies all linear functions will be found and the transfer functions will
conditions. The arrival rate and number of servers will be be found based on the conditions and principles for finding
optimized in this case. transfer functions.
If § = and assuming < 1, the value determined Properties of transfer function
A transfer function has the following properties
from
• The transfer function is defined for a linear time
invariant system.
• The transfer function between a pair of input and
output variables is the ratio of Laplace transforms of
Lq is determined in MIMO (Multiple Input Multiple Output)
the differentiated functions of the output to the
systems
input.
If § = (n = 0, 1, 2....., n) • All initial conditions of the system are set to zero.
• The transfer function is independent of the input of
Then the system [17]
The input and the output functions in both equations are not
Pn = * P0 n<c linear with respect to arrival rate (ૃ) and could not be used to
compute the transfer function.
= * Po n However, one of the prominent processes of stability analysis
in non linear systems is the linearization of the equation. A
non linear function could be made a linear one through the
Lq = * Po (6) process linearization. Linearization is the method of
 582 OGBIMI ET AL.

accessing the local stability of an equilibrium point of a If xo = 0

system of non linear differential equations or discrete
dynamical systems [10]. The typical equation for a linear
function is
y = f (ૃ) + f '(ૃ)(x – ૃ)
where y could be interpreted as the output of the system and 
x the input of the system.
Quotient rule of differentiation will be used to differentiate For the number of users in the network:
both the number of users in both the network and requesting
for web resources from the server with respect to arrival rate •ൌP0 +§( 
of packets.
From: Lq = Po For a single server model to many users
Assuming c = 1
Using Quotient rule of differentiation with respect to arrival
rate of packets •ൌ 
 ൌ ሺͳʹሻ

y = f (ૃ) + f '(ૃ) (x – ૃ)
at equilibrium point, f(ૃ) = 0 

If y = 0, then ሺͳ͵ሻ 

f'(ૃሻሺšȂૃሻൌͲ ൌૃʹɊ൅ͲૃʹɊǦʹૃʹɊ൅ ǢൌɊʹ൅ʹૃɊ൅ૃʹ

•‹‰ “—‘–‹‡– ”—Ž‡ ‘ˆ †‹ˆˆ‡”‡–‹ƒ–‹‘ ™‹–Š ”‡•’‡ – –‘
ƒ””‹˜ƒŽ”ƒ–‡‘ˆ’ƒ ‡–•Ǥ

ሺͳͶሻ
›ൌˆሺૃሻ൅ˆ̵ሺૃሻሺšȂૃሻ
x = ૃ 
›ൌˆሺšሻ൅ˆ̵ሺૃሻሺšȂૃሻ ƒ––Š‡‡“—‹Ž‹„”‹—’‘‹–ˆሺૃሻൌͲ

ˆሺšሻൌͲ

y = 2P0xμ(x – x0) 
(μ - x)3 ࣦ = SY(s) – Y(0

Y(s) =
If y = 0, then (μ - •)3 (17)
= ૃ(
x=ૃ
y = f(x) + f '(x)(x – x0)

(15)
(μ - ૃ)3

if x0 = 0, then
y= (16)
(μ - ૃ)3
4 RESULTS AND DISCUSSION
Y(s) = 4.1 Bode Plot
A bode plot is a graph of the logarithmic of the transfer
(μ - •)3
function of a linear, time invariant system versus frequency,
sY(s) =
plotted with a log-frequency axis, to show the system’s
s(μ - •)3 frequency response. It is a combination of bode magnitude
 DEVELOPING A FEEDBACK MODEL FOR MANAGING CONGESTION 583

plot (expressed in dB of gain) and a bode phase plot (the

phase is the imaginary part of the complex logarithm of the Impulse Response
0
complex transfer function. The magnitude axis of the Bode
plot is usually expressed as decibels of power that is by 20 System: h1
-2
log rule: 20 times the common logarithm of the amplitude Time (sec): 0.0109
Amplitude: -1
gain.
A bode phase plot is a graph versus frequency, also plotted
-4

on a log-frequency axis, usually used in conjunction with

the magnitude plot, to evaluate how much a frequency will -6

Am p litu d e
be phase-shifted , for example a signal described by: A sin
(wt) may be attenuated but also phase-shifted. -8

4.2 Impulse Response

In signal processing impulse response or impulse response -10

function of a dynamic system is its output when presented

with a brief input signal called an impulse. An impulse -12

response refers to the reaction of any dynamic system in

response to external changes. In both cases impulse -14
0 0.005 0.01 0.015 0.02 0.025
response describes the reaction of the system as a function Time (sec)

of time (or possibly as a function of some other independent

variables that parameterizes the dynamic behaviour of the
Figure 4.2.1: Impulse response of transfer function (h1) with a service rate
system. In this case, packet switched wide area networks is of 100 and probability of the number of users is 0.1.
a dynamic system, the external system may be messages and
the output may be information produced from the network. From figure 4.1.1, the axes moved up with a magnitude of -
In this case, the impulse response refers to the change in 98dB and a frequency of 0.1 Hz until it got to a magnitude
information over time in response to the initial message. of -62.3dB with a frequency of 10.1Hz. The frequency of
In control theory, impulse response is the response of a the bandwidth was stable at 10.1 Hz until it got to a
system to a Dirac delta input. This is also useful in the frequency of 12.4 Hz. For the phase angle, the axis moved
analysis of dynamic systems. The Laplace transform of the down with a frequency of 977 Hz and a phase angle of 0°
delta function is 1, so the impulse response is equivalent to and moved down till it got to a phase angle of -267° at a
the Inverse Laplace transform of the system’s transfer frequency of 0.322 Hz all through till it got to a frequency
function of 0.105 Hz. From figure 4.2.1, the Impulse response is a
dirac delta which represents the limiting case of a pulse
made in a very short time while giving a high peak. In this
Bode Diagram
-60 case, when the Amplitude is -1, the time taken to react to an
external change is 0.0109 seconds.
-80
M a g n itu d e ( d B )

Bode Diagram
-60
-100
-80
System: h2
M agnitude (dB)

-120 -100 Frequency (Hz): 1.51

Magnitude (dB): -80.5
-120
-140
0 -140

-160
-90 0
Phas e (deg)

System: h2
Frequency (Hz)
-90 Phase (deg): -1.
Phas e (deg)

-180

-180
-270
-1 0 1 2 3
10 10 10 10 10
-270
Frequency (Hz) 0 1 2 3 4
10 10 10 10 10
Frequency (Hz)

Figure 4.1.1: Bode plot of transfer function (h1) with a service rate of 100
and probability of number of users at 0.1. Figure 4.1.2: Bode plot of the transfer function with a
h1 = -20 s service rate of 200 and probability of 0.1.
-------------------------------------- h2 = -80 s
S3 - 300 s2 + 29980 s – 1000000 -----------------------------------------
s3 - 600 s2 + 119920 s – 8000000
 584 OGBIMI ET AL.

Figure 4.2.2: The impulse response of the transfer function number of users were varied. The bode plots and impulse
with a service rate of 200 and probability of number of users responses were drawn with MATLAB 7.0.
in the network was 0.2
REFERENCES
[1] Borisenok, S.V., (2005), Modern Control Theory and Its Application;
Impulse Response Department of Physics, Herzen State Pedagogical University, Moika
0
River Embankment, 48, 191186, St Petersburg, Russia.
System: h2 [2] Bosco, H.L. and D.C. Dowden, (2000), Evolution of Wide Area
Time (sec): 0.00401
-5
Amplitude: -1 Networks, Bellab Technical Journals.
[3] Bose, S.J., (2002), Chapter 1 – An Introduction to Queuing systems:
Kluwer/Plenum Publishers.
-10 [4] Daintith, J. (1992), Dictionary of Mathematics Series; International
Book Production, London.
Amplitude

[5] Dorf, R. C. and R.H. Bishop., (2004), Modern Control Systems;

-15
Pearson Education, Singapore.
[6] Flood, J.E., (1998) Telecommunications Switching, Traffic and
-20
Networks, Chapter 4: Telecommunications Traffic, New York:
Prentice Hall.
[7] Ghosh, S.,(2004), Control Systems: Theory and Applications; Pearson
-25
0 0.002 0.004 0.006 0.008 0.01 0.012
Education, Singapore.
Time (sec) [8] Hellstein, J. and S. Parekh., (2008), An Introduction to Control Theory
with Applications to Computer Science, IBM; T.J. Watson Research
From figure 4.1.2, the axes sloped up at a frequency of 1 Hz Centre.
at -84 dB and then moved up until it became stable at a [9] Hicks, M., (2004), Managing Gigabit Networks, Applications and
magnitude of -62.3dB with a frequency of 20.1 Hz and was Services – Compuware; Questnet 2204 Conference Networking Far
and Wide Held at Cairns International Hotel, Cairns, Austrailia.
stable till the axes got to the frequency of 24.8 Hz. The axes [10] http://en.wikipedia.org/linearisation.
moved down and touched the frequency axis of 8000 Hz. [11] Jain, R., (1998), Congestion Management in High Speed Networks,
Bandwidth became stable at the phase angle of -264°. The Issues and Trends, IEEE Magazine, pp 24-30.
frequency which the bandwidth was stable was between [12] Jain, R., (2006), Congestion Management in High Speed Networks,
Viewed from (http://www.cse.ohio.state.edu/~jain/cis788-
1.02 Hz and 1.19Hz. From the figure 4.2.2, the impulse
response is a Dirac delta which represents the limiting case 95/fp/atm_cong/index.htm) accessed 13th February, 2006.
[13] Janarthanan, Y., G. Minden, and J. Evans (2003), Enhancement of
of a pulse made in a very short time given high peak. When Feedback Congestion Control Mechanisms by Deploying Active
amplitude is -1 then the time taken for the system to react to Congestion Control; The University of Kansas Centre for Research,
external changes is 0.004 seconds. Inc., 2235 Irving Hill Road, Lawrence, KS 66044-7612. U.S.A.
The magnitude of the bandwidth flow was -62.3 dB with [14] Mahajan R., and S. Floyd, (2001), RED-with Preferential Dropping
frequency responses varying depending on the outcome of (RED-PD); University of Washington , AT&T Centre for Internet
the arrival rate, service rate and probability of number of Research at ICSI (ACIRI).
[15] Patterson, D.A., (2001), Network, Spring Lecture Slides; Department
users in the network. The phase angle of the bandwidth flow of Computer Science, University of California, Berkeley, U.S.A.
stability was -267°. The flow of bandwidth could be stable [16] Penittinen A, (2003), Queuing systems, Lecture Notes: S-38. 145-
with this condition and congestion could be reduced by Introduction to Teletraffic Theory.
using this model. The impulse function also shows that the [17] Philips, C. L. And R. D. Harbor, (2000), Feedback Control, Control
packet switched wide area networks can exhibit an impulse Systems, Fourth Edition, Prentice Hall, Upper Saddle River, New
within a small period of time in seconds depending on the Jersey, U.S.A
[18] Shor, M.H., Li Kung, J. Walpole, D.C., Steere, Pul Calton., (2003),
optimization of arrival rates, service rates and probability of Application of Control Theory to Modelling and Analysis of
number of users accessing the network. Computer Systems; Department of Electrical and Computer
Engineering, Oregon State University, Corvallis, U.S.A.
5 CONCLUSION [19] Taha, H.A., (2008), Operation Research: An introduction, Eight
The Conclusion of this work is that congestion can be Edition, Upper Saddle River, New Jersey, U.S.A. page 538
[20] Tijms, H.C.,(2003), Algorithmic Analysis of Queues, Chapter 9 in A
reduced by using a feedback model which is dynamic First Course in Stochastic Models, Wiley, Chichester.
explicit feedback model. This was done by writing a [21] Weisstein, E.W., (2003), CRC Concise Encyclopaedia of
queuing mathematical model for packet switched WAN. Mathematics; Boca Raton, London, New York: Chapman & Hall.
The bandwidth flow stability was measured by finding the [22] Williamson C., (2001), Congestion Control and Traffic Management
in High Speed Networks; Department of Computer Science,
transfer functions of the average rate of network users in the University of Calgary, 2500 University drive, New Calgary, AB,
network and accessing the network. This was done by Canada T2N, 1N-1.
differentiating the number of network users and the number [23] www.nationmaster.com/queuing_theory
of network users requesting for resources from the web with [24] www.sosmath.com/Linearization_Technique
[25] Yang, R.X., S.L. Lam., (2000), General AIMD Congestion Control;
respect to average rates of both network users accessing the Department of Computer Sciences, The University of Texas, Austin,
network and registered users in the network. Their Austin, U.S.A.
differentiated functions were linearised to make it a linear [26] Zabczyk, J., (1992), Mathematical Control Theory: An Introduction;
function. The laplace transform were computed and transfer Springer Series, Institute of Mathematics, Polish Academy of
Science, Sniadeckich, office 512, 00-956 Warsaw, Poland.
function were also computed. The transfer function was then
found and service rate, arrival rate, and probability of
 Voice Gateway to IP-PSTN Integration and NGN Architecture
Geneflides Laureno da Silva, Raimir Holanda Filho

Master's Course in Applied Computer Sciences – University of Fortaleza (Unifor)

Av. Washington Soares, 1321 – 60811809 – Fortaleza – CE – Brazil
[email protected], [email protected]

Abstract- Over the last decade, the technological environment The industry has developed solutions for connection
has traveled paths of updates and consolidated the emergence of between IP and PSTN, but the vast majority are proprietary
proposals for new architectures, aimed at integrating the various solutions and almost entirely be based on the binomial SIP-
networks in the world of telecommunications and building
ISDN. We can mention [3] [4] [5] as proposals for developing
synergy between data, voice and image. The NGN networks are
services, standards and integration. In the industrial field, we
based on the exchange of data packets, which permit a flexible
and scalable environment where the key factor is the provisioning can mention the work [6] as reference produced by a company
of new applications, enabling widespread mobility and integration that has contributed to the development of gateways aimed at
of new services with legacy networks. integrating the IP world with the traditional telephone networks
as well as the insertion of such devices to the new paradigm of
To face with this reality, we proposed a solution which is networks, the IMS (IP Multimedia Subsystem). Such
adherent to the functional architecture of NGNs. According to the implementations have proven the existence of recurrent
layered model, at the application layer, we developed five
challenges that need to be addressed, especially if the focus
applications that permit the implementation of new services; at
moves towards the core of the majors telecommunications
the control layer we use the open ip telephony platform asterisk
and finally at the transport/access layer, our solution support the operators and consequently the provision of new services to
following protocols: SIP, IAX, ISDN, R2Digital and SS7 Over IP. their customers. One of the major challenges is being able to
keep parity between the existing legacy networks and integrate
In order to validate our proposed solution, we implemented a new services from the new networks with ease and speed of
Voice Gateway that provides an integration between IP-PSTN development for these structures.
networks.
To face with this reality, we proposed a solution which is
I. INTRODUCTION adherent to the functional architecture of NGNs. According to
the layered model, at the application layer, we developed five
Over the last decade, the technological environment has
applications that permit the implementation of new services; at
traveled paths of updates and consolidated the emergence of
the control layer we use the open ip telephony platform asterisk
proposals for new architectures [1] [2], aimed at integrating the
and finally at the transport/access layer, our solution support
various networks in the world of telecommunications and
the following protocols: SIP, IAX, ISDN, R2Digital and SS7
building synergy between data, voice and image.
Over IP.
Telecommunications operators have made progress in
modernizing it's network elements, trying to keep existing II. NGN Networks
equipments and legacy networks and integrating this context
The NGN networks are based on the exchange of data
with the new paradigm of next generation networks.
packets, capable of providing telecommunications services
Small and medium-sized enterprises end up getting on making use of multiple bandwidths [7]. Through technologies
the margins of these new structures due to the high cost of that incorporate parameters of quality of service, the functions
equipment required for this deployment, as well as the value of of service delivery and transport are treated independently. The
service provided by operators is high because the need for key factors sought by next generation networks is that the
provision of dedicated access so much to voice as for data. provision of new services, enabling widespread mobility and
When physical access does not become the point of difficulty integration of services with legacy networks.
involved, the price for adding new services is an important
The work[8] is an example which we can see the
factor to be increased by the entry of small enterprises in this
different aspects and workforce for NGN. We focused our
new technological order.
solution in the ITU-T standardization.

585
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_102, © Springer Science+Business Media B.V. 2010
586 SILVA AND FILHO

The ITU-T (International Telecommunication Union - developed. This layer do the sharing of control functions of the
Telecommunications) is the international body that initiated transport layer, and control functions of the application layer.
studies on standardization of next generation networks. These Availability of network resources, call setup, user terminals
activities began in 2003 and strengthened in 2004 with the registration, accounting and authorization are examples of
formation of FGNGN (Focus Group New Generation Network) functions performed by the control layer.
[9]. Is important to stress here, studies and referrals decided,
the break occurring at seven specific areas of study, namely: The operation of applications like Charging and
capacity, architecture and functional requirements, quality of Billing are managed and controlled in this layer and more
service, control, security, migration of existing networks to details about this kind of control can be found on the
NGN and requirements for future packet-based networks. The recommendation M3060 of ITU-T.
definition of the basic architecture, interface specifications and
The main focus of application layer is to adding value
implementation guidelines make up the document that became
to architecture, through the possibility of the provision of new
the ITU-T recommendation for NGN[10].
services to existing IP networks and legacy networks.
According to the work [11], we elaborate the Fig. 1
The work [12] is an excellent example of
below, that represents a summary of the functional architecture
consolidation and standardization of the services, permitting a
of NGN and a basic description of the functionality performed
better use of the integration of the telecommunication
by each layer.
environments and patterns of development of the sphere of
information technology.

The development of new applications quickly and

following standards of development established, allow the
creation of new services by operators in a way competitive and
modular. Following this standardization, the application layer
concentrates modules that can be connected or disconnected in
a transparent manner.

In our view, the architecture consolidates its strong

advantage and motivation in two main points: convergence
between IP networks and legacy networks, creation and
deployment of new services in way transparent and modular.

In the next session we describe our solution, that

Fig. 1 NGN Functional Architecture
intend to support these two main points of NGN's
contributions.

On the layered model used by the architecture of next III. OUR SOLUTION
generation networks, the transport layer has the responsibility
to perform all connections between existing different networks, In our solution, the Gateway is fully implemented
protocols and interfaces. based on open libraries and open platforms. The Fig. 2 below,
shows the adherence referred to the functional architecture of
The provisioning of these various access networks, NGNs, summarizing the resources used and the applications
management and control of information are functions that are developed.
shared directly with the control layer. The IP protocol becomes
the core technology to the world of next generation networks, For the application layer, five modules were
since most of the equipment used, regardless of its developed with focus on generating services to users of the
functionality in the layered structure, uses this pattern to Gateway.
exchange signaling.
In the control layer, we use the open source platform
The control layer is the link between the transport Asterisk, whose main objective was to perform the interfacing
layer and the features related to services and applications between the applications developed and their access networks.
 VOICE GATEWAY TO IP-PSTN INTEGRATION AND NGN ARCHITECTURE 587

In the transport layer, we use standards already set and substantial amount of options that are outlined when adopting
open libraries for providing interconnection with the following this kind of platform that integrates the world of telephony with
protocols: ISDN, R2Digital, SIP, IAX2 and SS7 over IP. the world of IT. With a single application can occur a variety of
services to be provided by the corporation.
Below is the Fig. 3 that shows the flow of operation.

Fig. 2 A NGN Voice Gateway IP-PSTN Integration

We mention the work [13] as an example to test the Fig. 3 Application – Automatic Dialer
use of multiple manufacturers and we showed that in our
solution the validation of several protocols used have proven In Step 01 of this flow, the Supervisor performs a load
the flexibility and integration point between the IP-PSTN on the database, with list of names, telephone numbers and ID
networks. code of each User Destination. It is noteworthy that in the
overall business where we implemented this solution, the
We developed five applications but here we will average size of these lists dial is around 3000 contacts.
describe only two applications, the libraries used on the
transport layer and the topology of the tests. In Step 02 the supervisor start the application.

The five applications developed are as follows: Agent Step 03 is the time that the Gateway will ask the
Authorization, Automatic Dialer, Charging, Callback and database a list of Target Users to be contacted.
Employee Control. We will describe Automatic Dialer and
In Step 04 the ratio is returned.
Callback applications.
In Step 05 the Gateway will hold an investigation to
The first application that we will describe is
see if there are attendants free in Call Center.
Automatic Dialer.
The sixth stage is the time when the gateway starts
Automatic Dialer is the applications developed which
dialing according to the data contained in the database and
a list with data from several users destination is passed to the
already stored in memory.
Gateway in order to be contacted. The Gateway will be
responsible for making connections and in accordance with the In Step 07 the status of the User "B" is returned to the
response, connect the attendants to the Users "B". Gateway. The possibilities for this state are: answered, busy
and unavailable. In this flow we are describing the status
This kind of application is widely used in corporations
answered.
that perform credit recovery of its delinquent customers. This
application can be used in cases of aggressive voice marketing In Step 08, the Gateway forwards the call to the
merge for existing products, polls to vote, research on customer Callcenter and in accordance with the strategy used, an
satisfaction, among others. operator will answer the call.

As regards the possibility of adding services to the When the call is established, occur the Step 09.
specific company, this application is an example of the
The next application that we will describe is Callback.
588 SILVA AND FILHO

The main function of this application is to provide the In Step 10, the User "B" responds with a state of 'not
return of the calls to the mobile phones that need to in use'.
communicate and pay small financial rates.
In Step 11, the Gateway informs to the Callback User
The provision of such service becomes profitable for the state of User "B" and waiting for attendance by the User
internal users within companies, specially here in Brazil that "B".
the mobile telephones rates still high.
Step 12 is the point in time when the Callback User
and User "B" remain connected.

Now we describe the tasks performed at the transport

layer and topology of tests.

For the protocols(ISDN, R2Digital, SIP and IAX) we

only needed to configure and integrate the respective libraries
with Asterisk Core. On this layer, we used the open library
chan_ss7 to validate the signaling SS7 Over IP to our Gateway.
In chan_ss7 we needed to change the source code to perfect
working of the signaling.

The complete library consists of the following

modules: chan_ss7.c, l4isup.c, mtp.c and tranport.c.

In this library, the main change made was to correct

Fig. 4 Application – Callback
the declaration of certain kind of variables that were defined as
'long int' and were changed to the 'int'. The other changes was
In Step 01 of this flow, the Callback User performs a the definition of data structure that not existed and we needed
call for a specific number of telephone connected to the to create and set up.
Gateway.
We also observed errors regarding the path of certain
During Step 02, the Gateway records the number of files in the directory tree of the server.
the Callback User and sends a signal to disconnect him.
This library is the only in which the installation was
Step 03 and Step 04 is when the number of the performed only in the laboratory, unlike the others that were
Callback User is checked and validates their existence inside performed in corporate environments.
the database.
The integration tests performed with the Gateway
In Step 05, the Gateway will performs the dial to the were divided into two distinct groups, namely: laboratory tests
Callback User, previously validated by the confirmation of and enterprise environments test.
their existence in the database.
Other topics of changes made in this module are
Step 06 is the confirmation of attendance by the correlated with the integration between it and the native library
Callback User. of Asterisk core, namely DAHDI.

In Step 07, the Gateway sends a dial tone to the The main work of correction was performed by
Callback User. removing a portion of the code that occurred redundant
definitions of functions and variables.
The Callback User receives a dial tone, sends the
number of the User "B" to the Gateway. This is Step 08. The feasibility of performing tests with SS7 Over IP
in an environment of integration with the carriers was not
In Step 09 the Gateway dial to the User "B" and possible due the unwillingness of the circuits for such
awaits response from their state. purposes.
 VOICE GATEWAY TO IP-PSTN INTEGRATION AND NGN ARCHITECTURE 589

Below in Fig. 5 we show the topology used in The interconnection with telecommunications
laboratory to validate the signaling SS7 Over IP. operators through the use of various protocols and the
flexibility to develop new applications in a modular fashion
allows us to confirm the actual adherence of the Voice Gateway
to the stage of NGN.

To test the signaling SS7 Over IP with carriers and

develop new applications for corporate environments is a good
place to continues with this work.

REFERENCES

[01] K. Hechmi, G. Jean-Charles, “IMS for Enterprises”, IEEE

Fig. 5 SS7 Over IP Test Topology Communications Magazine, July 2007.

[02] Hyeong Ho Lee, "Signaling architecture and protocols for the Next
In corporate environments tests, we validate real-
Generation Network", Advanced Communication Technology, February 2009.
world scenarios, the operation of the applications developed
and the following protocols: SIP, IAX2, ISDN and R2Digital. [03] S. Mittal, C. Dipanjan, S. Goyal and M. Sougata, “SewNet – A Framework
The tests with traditional carriers were ISDN and R2Digital for Creating Services Utilizing Telecom Functionality, April 2008.

and with the current entrants in the emerging market for VoIP [04] NetIQ, “How to Succeed on Your Journey to Unified Communications”,
and IP Telephony, were the protocols SIP and IAX2. September 2008.

Following Fig. 6 aims to summarize the scenarios for [05] de Souza Pereira, J.H., Guilherme, J., Rosa, P.F, "Development of MGs in
integration in corporate environments tests. a next generation network with MEGACO/H.248 support", 12th IEEE
International Conference on, November 2004.

[06] Tekelec, “SS7 Over IP – Standalone Signalling Gateways versus

Integrated and Distributed Signalling Functions”, 2007.

[07] Knightson, K.; Morita, N.; Towle, T. "NGN architecture: generic

principles, functional architecture, and implementation", IEEE
Communications Magazine, October 2005.

[08] Vaaramaki, T.; Puttonen, J.; Fekete, G. "Next Generation Network Related
Standardization - Enablers for the Convergence", Fourth Advanced
International Conference on, June 2008.

[09] ITU-T FGNGN-MR-0065.

Fig. 6 Corporate Environment Test Topology [10] ITU-T Recommendation Y.2001, December 2004.

[11] Chae-Sub Lee; Morita, N."Next Generation Network Standards in ITU-T",

BCN 2006 , The 1st International Workshop on, April 2006.
IV. CONCLUSIONS
[12] Schreiner, F.; Wahle, S.; Blum, N.; Magedanz, T."Modular exposure of
We concluded that is possible to develop a gateway next generation network services to enterprises and testbed federations", ICCE
that provides integration between IP-PSTN through the open 2008, Second International Conference on, June 2008.
libraries and open platforms. This gateway permit access to a
[13] Ayatollahi, Z.; Sarukhani, S.; Fayazi, F.; Roknabady, Z.A.; Madani, A.
voice telecommunications architecture that has strong "Interoperability problems in Next Generation Network protocols", K-INGN,
adherence and similarity to the functional NGN architecture. First ITU-T Kaleidoscope Academic Conference, May 2008.
 An Algorithm for Route Selection on Multi-sink
Wireless Sensor Network Using Fuzzy Logic
Líliam Barroso Leal #1, Marcus Vinicius de Sousa Lemos #2, Raimir Holanda Filho #3
# Mestrado em Informática Aplicada, University of Fortaleza
Av. Washington Soares, 1321, Edson Queiroz, CEP 60.811-905, Fortaleza CE, Brazil
1 [email protected]
2 [email protected]
3 [email protected]

Abstract—In this paper, we propose a new algorithm for Thus, a multiple sink approach was proposed in order to
route selection on Wireless Sensor Network (WSN) which allows overcome that problem. In this new paradigm, multiple
communication between sensor nodes and multiple sinks. For sensors can communicate with multiple sinks (many-to-
that, the sensor nodes use a fuzzy inference machine in order to many).
select the most appropriate sink. This inference will take into
account some network characteristics such as: energy, number of
The main goal of the algorithm proposed in this paper is to
hops, collisions, and delay. Simulations have shown that our promote communication between sensor nodes and multiple
algorithm is feasible not only in terms of routes selection, but also sinks, where each node will select the best route to an
in terms of energy consumption. appropriate sink node, localized somewhere in the network.
This selection will be performed by a fuzzy inference
I. INTRODUCTION machine. We have chosen this scenario because of the
In the last years, there was a considerable increase in increasing popularity of applications which require networks
research involving wireless sensor networks (WSN), due to its with multiple sinks, for example, applications that use
applicability in many areas such as security, health, wireless multimedia sensor network [3], actuator nodes [4]
agriculture, among others. The WSN comprise a large number and other applications in which there may be more than one
of small and intelligent sensor nodes capable of detecting and base station. For example, a medical application [5], in which
transmitting characteristics from the physical environment [1]. inside a hospital we could have sensors attached to patients
Intelligent sensors can be defined as a small circuit board and the doctors would be equipped with portable computing
containing an integrated sensor capable of processing and data devices such as smartphones, which would act as sinks. Such
communication [1]. scenarios require more appropriate solutions, which make the
Wireless sensor networks have unique characteristics very existing proposals, for networks that have a single sink,
different from traditional networks, especially regarding the inefficient.
number of nodes present in the networks and constraints in In our solution, the sensors will be able to select, wisely,
some resources like energy, processing power, memory and the best route in a given time, based on some network
bandwidth. These variables make impractical the re-use of characteristics, such as distance to the sinks, the energy from
most of the algorithms developed for traditional computer the neighbor nodes, among others.
systems, since the design of any solution for sensor networks The remainder of this paper is organized as follows: in
must deeply consider the consumption of energy and Section II, we describe some related works; we present our
computational constraints. proposed algorithm in Section III; in Section IV, we perform
The sensor nodes are deployed in a geographical area in the simulation of our algorithm; in Section V, the results are
order to observe events, such as atmospheric pressure, depicted and finally, in Section VI we show the conclusions
temperature, level of pollution, etc. These nodes can be and future works.
randomly deployed or following some topology (which, in
most of the cases, depend of the network designer and the area II. RELATED WORKS
to be monitored). The sensors nodes, after sensing some In this section, we will describe some related works which
event, must be able to send a report to a particular node, called use a scenario of multiple sinks, as well as problems related to
sink (base station), which will analyze the received data and, the approach of using a single sink.
according to the network goal, make some kind of decision. In [2], we have a wireless sensor network with multiple
Some of the most common communication paradigms in sinks arranged in clusters, where each sink monitors a region
WSN involve communication of multiple sensors placed in an with multiple events. The authors propose the use of a
area of observation, reporting information to a single sink management system database, distributed in each sink of the
(many-to-one). In [2] it was shown that a single sink is a network and require two changes on the routing protocol: the
bottleneck in a network, especially for real-time applications. first is that the protocol should allow the creation and storage

591
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_103, © Springer Science+Business Media B.V. 2010
592 LEAL ET AL.

of multiple routes for the spreading of a message, and the A.General Aspects
second requires that the network provides QoS, because it In our proposal, we consider that the sensors are placed
works with the prioritization in the delivery of messages for uniformly in the region to be observed, and we also consider
notification of event. Once an event of interest is detected by a multiple sinks arranged in order to maintain coverage
node, notification message is sent toward the sink. Paths are throughout the area. The sinks are devices with features far
created and stored in each clusterhead (CH): each clusterhead more advanced than the sensor nodes, having no limitations of
stores information about the event and the ID of the cluster power.
from which the message was received. Each sink located in the network, will be responsible for
The research in [6] proposes a scheme based on fuzzy logic receiving the notification messages of events at any point of it,
to select a better way and increase the survival time in for this, each sensor node will, through a fuzzy inference
MANETs. In that work, it has been mentioned that the machine, select the most appropriate sink for it in that
survival of the network can be achieved when security moment, that is, which the best route is taking into account the
measures, preventive, reactive and tolerant lines of defense, network characteristics, such as energy, number of hops,
work together. This proposal presents the following collisions and delay. We should emphasize that this study
contribution: a scheme of selection of adaptive path considers only some of these characteristics, but we can add
independent of protocol. Furthermore, the correlation of new features according to the needs of the network.
characteristics of network security and discretion are made It is important to emphasize that we need to select
through the use of fuzzy logic to improve the path selection previously a routing protocol, because our solution works in
minimizing security limitations found in routing protocols. conjunction with it. The choice of the routing protocol will be
The process of selecting path involves three phases: data made according to the needs of the network designer, and it is
collecting, the process of fuzzy inference and finally the only required that this protocol be capable of providing multi-
selection of adaptive path. path.
The work in [7] investigates use of fuzzy inference in order
to make the nodes select the best path in a flat wsn with only B.The proposed solution
are sink. The fuzzy inference system selects the best candidate We will now discuss the steps used in our algorithm to
node basead on some metrics, like, for example energy. To choose the best path for data transmission over the network.
carry out this goal some fuzzy rules are implemented, which Table I shows the notations used, while Algorithm 1 presents
use as metrics: the distance from the candidate node from the the proposed algorithm.
origin, the distance from the candidate node to the shortest
TABLE I
path, energy and link usage. One of the disadvantages of this
NOTATIONS OF THE PROPOSED ALGORITHM
protocol is the fact that it does not address the use of multiple
sinks, which makes it vulnerable to the problems mentioned ͆n The beacon message generated by the base station
above, such as bottleneck, and a much higher dissipation of n
energy of the nodes near the sink. A: The operation after the ":" The node is held in
Our proposal differs from the related works, because it was A*: A passes the message ͆n through broadcast
developed for wireless sensor networks with multiple sinks ͆n
AN: A passes the message ͆n to node N
and makes it possible to choose the best routes for sending
͆n
data to a sink. We propose an algorithm that works with the x« y The value y is stored in the variable x
routing protocol in order to assist a node in the process of
selecting the best path among the possible routes in a given
time. This process of route selecting is made by the use of a After the deployment, the nodes are initialized and the
fuzzy inference system. Every route will be analized and sinks generate a list containing the neighboring nodes (line 1
classified regarding some criteria such as energy and number of Algorithm 1). This list is used by the sinks to send a beacon
of hops. Whenever a node has to send some data, it will take message n. This process is depicted in lines 1 to 4 of
the best classified route. Hence, the network life will increase Algorithm 1. The beacon message sent by the sinks is
as bad routes (route with little energy and/or much propagated to the network through the neighboring nodes.
interference) will be avoided. Upon receiving a beacon message, a sensor node A will check
the following fields:
• Energy: this field is used to store the lowest energy
III. THE PROPOSED ALGORITHM level found in the path.
In this section, we will analyze in detail the proposed • Station_base: indicates the sink from which the
algorithm. Then, we will discuss general aspects of the beacon message originated.
algorithm in sub-section A, while in sub-section B, the • Number_hops: distance in hops from the origin sink
selection phase of the routes with the aid of fuzzy inferences of the message to this sensor node.
will be presented. • Node_predecessor: identifies the node from which
the node A received the message.
• Last_sequence: identifies the beacon message. Each
time the sinks require a route rebuilding, a new beacon
message is generated and the last_sequence field will be
 AN ALGORITHM FOR ROUTE SELECTION ON MULTI-SINK WIRELESS SENSOR NETWORK 593

incremented, so that the nodes can distinguish the we will explain this calculation in more detail. Finally, the
current message from the messages sent previously. node A tests if its energy is lower than the energy stored in the
received message, if so, it updates the field energy of the
Algorithm 1: beacon message with the value of its remaining energy,
1 : Base Station n: generates V list with neighboring nodes otherwise it goes to the next step and relays the message
2 : Base Station: n belonging to V through the network.
3: Base Station n: generates (n)
4: Base Station  n : n
5 : A: receives ( n ) ( n ) transmitted by neighboring
nodes
6: A: energy « read field energy ( n )
7: A: base_station: « read field base_station ( n )
8: A: number_hops « read field num_hops ( n )
9: A: node_predecessor « read field node_predecessor
( n )
10 : A: last_sequence « read field sequence_broadcast(n )
11 : A: if A does not use the route predecessor
12 : A: discard n Fig. 1. Disposal of Beacon Message
13 : A: exit
14 : A: end-if
15 : A: sequence_current = recover In our work, we have considered that each sink node sends
sequence_pkg_received (node_predecessor) beacon messages in order to make the sensor nodes update
16 : A: if sequence_current >= last_sequence their fuzzy levels in network, within a window of time ,
where  is a value to optimize the lifetime of the network. The
17 : A: discard ͆n
sensor nodes in the network check their own energy for each
18 : A: exit
time window . We have also set a safety limit of the amount
19 : A: end_if
of remaining energy in the node, represented by . This limit
20 : A: FL « calculate FL (energy, number_hops)
makes the node able to report a poor energy condition at any
21 : A: update_FL_route (node_predecessor, FL)
time, without the need to wait for the next beacon message
22 : A: if energy(A) < energy( ͆n ) then
from the sinks. The value of is informed by an alert
23 : A: store energy (A) in the field energy of ͆n
message propagated through the nodes with hop = 5, avoiding
24 : A: end_if
the wasting of energy by unnecessary broadcasts. All values
25 : A  * : ͆n
assigned to , and hop may also be defined by the network
26 : A: end_if
designer.
If a node D receives a beacon message whose value of field C.The Fuzzy Inference Machine
node_predecessor is different from some of the neighboring Fuzzy inference can be defined as a process of mapping a
nodes it uses as a current route, it will drop this message, as given input into an output through the use of fuzzy logic. This
illustrated in Figure 1. This measure is adopted so that node D mapping is carried out by rules that define the output
does not have to calculate a fuzzy level for a node it does not associated with a particular input [6]. For example, if the
use as a route. energy condition of a node A is bad, then do not forward
Aiming to prevent old messages from keeping on traveling packets to this sensor node. These rules have as input values
through the network, some routines are set according to lines defined by the linguistic variables, which in this work are
15 to 19 of the algorithm presented. Whenever node A represented by energy, figure 2 and number of hops, figure 3
receives a beacon message, it checks the field last_sequence and they can be altered according to the need for the network
to verify whether it refers to a new beacon message, or if it is designer. Each linguistic variables has a set of linguistic
just the spread of a message received earlier. The next step is values (fuzzy values) associated to it. As an example of
the calculation of fuzzy level (FL) of the route; this parameter linguistic values of energy we have low, medium and strong.
allows the node to select the best route in a given moment, Each linguistic variable has its membership function which
according to the last beacon message received. Each route that will convert crisp values (real values measured from the
a node can use, it is associated with a FL. Routes with a environment) into fuzzy values. And both (rules and
higher value of FL are considered preferential. The calculation membership functions) are responsible for the behavior of our
is carried out using the FL and fuzzy logic, and in our algorithm regarding the routing decision. It is important to
algorithm we have considered as parameters the linguistic point out that in fuzzy sets, the values can range from 0 to 1,
values energy and number of hops. In the sub-section below, with 1 indicating the value of absolute truth and 0 absolute
false.
594 LEAL ET AL.

• Scenario 1: without multiple sinks and without fuzzy

logic. In this scenario we have one sensor node
transmitting data to one sink.
• Scenario 2: without multiple sinks, without fuzzy
logic, and with route balancing. In this scenario,
whenever a sensor has to send packets toward the sink, it
will balance the packets between their routes. For
example, a node X has three routes represented by the
numbers 1, 2, and 3 and has 4 packages to send to the
sink, represented by the letters A, B, C, and D. The node
Fig. 2. Function Energy
will forward them as follows: The package A will be
forwarded using route 1, the packet B using route 2, the
package C using route 3 and the package D will be sent
through route 1.
• Scenario 3: without multiple sinks and with fuzzy
logic. In the scenario, one sensor node transmits data to
one sink, using together with the routing algorithm, the
proposed solution with the use of fuzzy levels to choose
alternative routes.
• Scenario 4: with multiple sinks and fuzzy logic. In
this last scenario we have one sensor node transmitting
data for two sinks using fuzzy levels for choosing the
best routes.
3) Routing Algorithm: For the communication between
nodes and sinks, we have created a variation of the Protocol
Destination-Sequenced Distance Vector (DSDV), [10] in
Fig. 3. Function Path Length
which all the sensor nodes will transmit the data collected
IV.SIMULATION to a sink. The system starts without any knowledge about
the identity or the topology of sensors which are present.
This section presents comments on the simulations
Each node knows only its own identity. The sinks broadcast
performed to validate the proposed routing algorithm.
routing update messages periodically. Devices that are in
A.Simulator direct range of the sink, upon receiving this message,
update their route tables. These nodes perform then a
In our experiments, we used the simulator Sinalgo [8].
broadcast of a new routing update message to any device
Sinalgo is a framework, written in Java, for testing and
within its range (Figure 4a), indicating that there is a path
validation of networks algorithms. Unlike others tools such as
to a sink through them (nodes which have sent the
Network Simulator 2 [9] which allows the simulation of
broadcast). To implement the multi-path routing scheme,
various network stack layers, Sinalgo focuses on the
the devices will store the first three routing update
verification of algorithms and abstracts from lower layers (like
messages which they receive (Figure 4b).
physical, link, network, among others).
The algorithm used, although not the most efficient, has
B.Network Features been shown satisfactory since, in this work, we have only
We will present the features of the network as follows: analyzed the behavior of our application under a multi-path
1) The network topology: the simulated network is flat, routing protocol, leaving for future work the analysis of our
fixed and contains only two types of nodes: sensor nodes solution using other routing protocols.
and sink nodes. Each sensor node has a single identification
and a fixed radio range.
2) Used scenarios: in our simulations, we have used four
scenarios with a similar environment. All scenarios are
comprised by 100 sensor nodes uniformly distributed along
the network, covering an area of 700 x 700 m2. We
simulated the following scenarios:
Fig. 4. Routing Algorithms
Used
 AN ALGORITHM FOR ROUTE SELECTION ON MULTI-SINK WIRELESS SENSOR NETWORK 595

V. RESULTS determine whether the fuzzy rules created have reached

A.Scenarios according to time expectancy. Finally, the white column shows the use of fuzzy
logic and multiple sinks, in which we have obtained a time of
The results of the proposed algorithm are presented
activity of the network higher than the previous scenarios.
regarding three metrics according to the experiments. Below,
We also evaluated the behavior of our solution according to
we have defined our metrics:
the amount of sink nodes that we have on the network, while
• Time of death of the first node: In Figure 5, it is
still maintaining a satisfactory result. Therefore, gradually add
expressed the time of death of the first node in the
more sink nodes in the environment and simulate their
network. Here, we want to analyze how long our
behavior as can be seen in Figure 7. We realize that to some
proposed work makes the nodes alive.
extent the number of sink nodes influence positively the time
• Time of death of the network: Here, we investigate
of death of the first node, however, as more sink nodes are
how long the network stay alive, that is, when the
added, the greater the number of beacon messages traveling
network could no longer maintain the necessary
on the network, affecting the life nodes.
communications, Figure 6.
• The numbers of the sink nodes: Here, in this metric Fig. 7. Influence of the number of sink nodes
we wants to see how the numbers of sink nodes can
influence our work, Figure 7.
B.Window of time 

Fig. 5. Death Time of the First Node As mentioned previously, in our algorithm we consider
that each sink expect a time window  for sends beacon
messages in order to make the sensor nodes updates its fuzzy
Fig. 6. Death Time the Network levels. It was performed simulations with different values of 
and we evaluated the performance based on death time of the
We can realize from Figure 5 that the scenario 1 (without first node of the network (Figure 8). We can see that the
fuzzy logic and without multiple sinks) has been shown more values near to the optimum for this set are between 500 and
fragile as regards the life of a node which participates actively 700 seconds.
in a route in use. We can also infer that in the scenario 2
(without multiple sinks, without fuzzy logic, and with route
balancing), represented by the dark gray column, showed a Fig. 8. Values for Time Window 
gain of approximately 42% compared to the scenario 1, black
column, because the route balancing. Scenario 3 (without VI. CONCLUSION
multiple sinks and with fuzzy logic), illustrated by the light This work proposed an algorithm for path selection using
gray column, compared with scenario 1 showed an increase of fuzzy logic on WSN with multiple sinks. The main goal of the
106%, because of the possibility of using alternative paths algorithm is to extend the network time life and this can be
inferred by FL. Finally, on scenario 4 (depicted by a white achieved by making the nodes avoid bad routes while sending
column), we have the representation of the proposed solution packets through the network. We used fuzzy logic to build up
with the use of fuzzy logic and multiple sinks, in which the a knowledge base that makes possible the process of finding
gain, compared to the scenario 1, is approximately 234%. This the best routes
gain was expected because of multiple sinks, increasing the Results obtained from simulations have shown that our
number of alternative routes by which the node can use for algorithm is feasible, increasing up to 479% the lifetime of the
data transmission and the fuzzy inference which make the network, since several routes can be used for message
nodes avoid the bad routes. forwarding.
In future works, we intend to analyze the behavior of
In Figure 6, we have the representation of the time of death algorithms in more complex scenarios, and select new
of the network, time which it prevents the transmission of data linguistic variables for fuzzy inference.
due to the death of some nodes in the path of routes. The black
column represents the scenario 1 without fuzzy logic and VII. REFERENCES
multiple sinks, in which the time of death of the network
occurred after the death of the first node of the network, [1]Karl, Holger and Willig, Andreas, Protocols and Architectures for
because this scenario does not have multiple routes and Wireless Sensor Networks, John Wiley & Sons, Ltd, 2005.
[2]Boukerche, Azzedine and Martirosyan, Anahit, An Energy Efficient and
multiple sinks. In the scenario 2, shown by the dark gray
Low Latency Multiple Events' Propagation Protocol For Wireless Sensor
column, the network was active for a longer time,
Networks With Multiple Sinks, In: PE-WASUN'07, 2007, Chania, Crete
approximately 222% more than the scenario 1; this gain was Island, Greece.
achieved by the use multiple routes. The light gray column [3]Akyildiz I. F., Melodia T., Chowdhury K. R., “A Survey on Wireless
depicts scenario 3, where the network was active for a longer Multimedia Sensor Networks” Computer Networks (Elsevier), Vol. 51, no. 4,
time, approximately 330% more than the scenario 1; this gain pp.921-960, March 2007.
was achieved by the use of fuzzy logic, although the feature of [4]Pietro Ciciriello and Luca Mottola and Gian Pietro Picco, Efficient Routing
multiple sinks has not been added. We have used this scenario from Multiple Sources to Multiple Sinks in Wireless Sensor Networks, In
to assess the choice of routes from the node, and thus
596 LEAL ET AL.

Proceedings of the 4th European Conference on Wireless Sensor Networks [7]Mohammad Abdul Azim and Abbas Jamalipour, Optimized Forwarding
(EWSN), 2007. for Wireless Sensor Networks by Fuzzy Inference System, Auswireless
[5]Kim, Yung Bok and Kim, Marie and Lee, Yong Joon, COSMOS: a Conference, 2006.
middleware platform for sensor networks and a u-healthcare service, SAC [8]Sinalgo (2007). http://dcg.ethz.ch/projects/sinalgo/
'08: Proceedings of the 2008 ACM symposium on Applied computing, 2008. [9]NS2 (2008). http://nsnam.isi.edu/nsnam/index.php/main_page.
[6]Michele Nogueira Lima and Helber W. da Silva and Aldri L. dos Santos [10]Perkins, C. E. and Bhagwat, P. (1994). Highly dynamic destination-
and Guy Pujolle, Survival multipath routing for MANETs, NOMS, 2008. sequenced distancevector routing (dsdv) for mobile computers. SIGCOMM
Comput. Commun. Rev., 24(4):234-244.
 Mining Mobile Sequential Patterns in Wireless Cellular Networks

Joshua G. Bradley and Sherif S. Rashad

Department of Mathematics and Computer Science
Morehead State University
Morehead, KY 40351
[email protected]
[email protected]

Abstract— In this paper, we have taken concepts of data allocate resources required by the mobile user evenly
partitioning methods and modified the well known throughout the surrounding cells. This allows for the
algorithm SPADE (Sequential PAttern Discovery using network to be readily prepared to continue the node’s data
Equivalence classes), which we have implemented over a connection if the mobile user continues using the mobile
mining model, called mining mobile sequential patterns, device and transfers to an adjacent cell. Disadvantages to
and called MobileSPADE. From this, we have devised this strategy include producing unnecessary overhead by
two innovative forms, called Fixed-MobileSPADE distributing resources to cell towers that will not be used by
algorithm and Dynamic-MobileSPADE algorithm, which the mobile user, thus not allowing the mobile user full
will mine for mobile sequential patterns from either employment of all available resources. One alternative
fixed-length or dynamic-length transactions. In mining solution that has been extensively researched has been the
for mobile sequential patterns in a mobile commerce mining for frequent sequential patterns. In this paper,
environment, we used tower ID data extracted from a frequent sequential patterns refer to the frequent movement
dataset that was constructed and managed by the patterns of a mobile user. Finding these frequent sequential
Massachusetts Institute of Technology (MIT) Media patterns could allow for a network carrier to calculate with
Lab: Reality Mining Project, which logged objective data some probability where a mobile user will go next, based
from 100 mobile device users for a period of nine upon his/her past movement history. Using this innovative
months, to represent mobile user’s movements. mobile sequential pattern model and looking at Figure 1
Prediction analyses results were used to measure the again, when a mobile user resides in cell A, the network
performance of Fixed-MobileSPADE and Dynamic- carrier could allocate more resources of a certain service
MobileSPADE algorithms on the mining mobile requested by the user to the cell most likely predicted to
sequential patterns model. Results show that as the contain the end user. For example, based upon a mobile
quantity of items that are taken into consideration for user’s past history, if the mobile user is calculated to have a
mining mobile sequential patterns increases, the 75% chance of going from cell A to cell D, then the network
prediction accuracy correspondingly increases. can allocate/distribute more of its resources to cell D as
compared to the surrounding cells, in order to provide better
service to the end-user via a more efficient use of its
I. INTRODUCTION resource allocation methods and routing protocols.

I n a world where mobile devices are becoming more

popular and one of the leading factors to an end-user in
choosing a mobile network carrier is cell coverage for a
particular area, advancements in wireless communication
technologies and routing protocols is often advantageous to
a network carrier in order to gain more customers. When
studying a cellular network, it is often best to view the
network as a grid of cells where each cell represents the
coverage area for a cell tower, similar to the grid in Figure 1.
Due to the extent of applications now available for mobile Figure 1 A sample grid of cells in cellular networks
devices, there have been many studies on Location-Based
Service (LBS) in the past couple of years, but one important The remainder of this paper is organized as follows. The
research issue that is becoming prevalent in upcoming next section reviews the preliminary concepts needed to
network technologies is the tracking and prediction of a comprehend the rest of this paper and section 3 will discuss
mobile user’s movement pattern in a cellular network. the concept of data partitioning methods and the original
Currently, a mobile network carrier uses the following SPADE algorithm. The dataset provided by the
strategy to maintain cell coverage and introduce a node to its Massachusetts Institute of Technology Reality Mining
network. Using Figure 1 as an example, when a mobile user Project is discussed in section 4. Section 5 discusses
is in cell A, the mobile network carrier will distribute and algorithm MobileSPADE for discovering frequent sequential

597
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_104, © Springer Science+Business Media B.V. 2010
598 BRADLEY AND RASHAD

patterns in a cellular network and its two derivatives, patterns, because this is one of the first research projects that
algorithm Fixed-MobileSPADE and algorithm Dynamic- successfully collected an extensive amount of objective data,
MobileSPADE. Section 7 presents the experiment results as opposed to subjective data. In this paper, objective data
obtained from algorithm Fixed-MobileSPADE and refers to the data collected by the mobile device logging
algorithm Dynamic-MobileSPADE and section 8 discusses software. Subjective data refers to previous methods used to
the conclusion on the experiment results. test algorithms, which include creating a virtual mobile
commerce environment and issuing survey’s to record
relative location.
There is great potential for this dataset and other similar
II. RELATED WORK AND PRELIMINARY CONCEPTS datasets to revolutionize the whole concept of social
In this section, we analyze an experiment conducted in [1] networking and advertising. Use of this dataset provides
on the mining of frequent itemsets via a data partitioning researchers with a data model that represents objective
approach which employs a modification of the algorithm in human behavior. Prospective exploits of this dataset include
[2]. We also discuss the SPADE algorithm and which the testing and modification of current cellular network
aspects were modified/ enhanced during the development of algorithms to improve their performance, re-defining current
Fixed-MobileSPADE and Dynamic-MobileSPADE. advertising techniques to make advertising targeted and
In [1], Nguyen and Orlowska enhance the data pre- more personalized to an individual mobile user, and
processing approach of the Partition algorithm [2] to developing innovative methods to find frequent sequential
increase the performance of frequent itemset computations. patterns.
A series of data pre-processing methods such as the
introduction of new clustering and similarity algorithms
were applied across the data to show that the performance of Notation used throughout the paper is as follows:
the algorithm discovering frequent itemsets increases by • Let T = {t1, t2…tn} be a set of n transactions over data set
means of data partitioning. The Partition algorithm in [2] D.
follows the concept that a fragment P D of the database is • Transactions can be of fixed or variable length, depending
a subset of transactions within the database. With reference on the algorithm used.
to an itemset, local support is defined as the fraction of • Each transaction consists of a sequentially ordered set of
transactions containing that itemset and local candidate items {i1, i2, i3…im} Tn, referenced as tower ID
itemset describes an itemset meeting the minimum support numbers in this paper.
for the whole database. Global support and global frequent
itemset are calculated similarly except with respect to the The goal of Fixed-MobileSPADE and Dynamic-
whole database. The principle behind [1] is that each MobileSPADE algorithms is to generate all the discovered
fragment is calculated in the main memory one at a time. frequent itemsets that meet a user-defined support threshold
In [3], Zaki presents SPADE, an algorithm designed for the and confidence threshold.
accelerated discovery of sequential patterns. Previous
algorithms have required multiple scans over a database for
their calculations, but SPADE utilizes combinatorics to
decompose the original problem into smaller sub-problems, IV. ALGORITHMS FOR MINING FREQUENT MOBILE
which are individually solved in the main memory and later SEQUENTIAL PATTERNS
combined. Similar to SPADE algorithm [3], algorithm MobileSPADE
is an Apriori-based sequential pattern algorithm that utilizes
characteristics of combinatorics and adapts a vertical data
III. MOBILITY DATASET AND PRELIMINARY CONCEPTS format of the original dataset. Furthermore, two defining
It is important to discuss the mobility dataset used to obtain characteristics of algorithm MobileSPADE lie within its
performance tests results on Fixed-MobileSPADE and ability to incorporate data attributes into a consecutive
Dynamic-MobileSPADE algoritms. The dataset used in this sequentially ordered list, based upon timestamp values, and
research is a result of the reality mining project carried out its capacity to only consider itemsets of length two or greater
by the MIT Media Lab [4]. This dataset was collected over a len(itemset)  2, in order to find frequent sequential itemsets.
nine month period and involved the use of 100 cellular When using the mobile dataset from the MIT Reality Mining
devices [4, 5]. As explained in [4] mobile device logging Project, the timestamp is defined by the values in the
software was designed and installed on 100 mobile devices, “starttime” column of the dataset. A mobile user’s
which recorded critical information about the mobile user, movement path is defined as the tower ID utilized for each
such as proximity, location, services used, length of calls logged event. Table 1shows a sample of the collated dataset
made and received, type of data used, etc onto the device’s with the corresponding starttime and celltower_oid data
memory card. The data for all 100 mobile users was columns for mobile user number 94 after sorting with
collected and organized into one dataset. This dataset has reference to timestamp value. We can see the mobile user
been of great interest recently to researchers involved in utilized cell_tower(38) at timestamp(2004-07-23 12:22:34)
mobile data mining and the discovery of frequent sequential
 MINING MOBILE SEQUENTIAL PATTERNS IN WIRELESS CELLULAR NETWORKS 599

and then cell_tower(39) 22 seconds later at timestamp(2004- Begin

07-23 12:23:56). 1) Sort database D in sequential order with reference to
timestamp value (starttime). The result will be as
shown in Table 1.
Table 1 A sample dataset with starttime and celltower_oid
data columns for user number 94 after sorting with reference
2) Divide database D into equal-length transactions;
to starttime value
each transaction containing the user-defined
User 94 transaction_length number of items. For example,
Starttime celltower_oid if transaction_length = 5, then T = {t1, t2, t3, t4}
2004-07-23 12:22:34 38 where
2004-07-23 12:23:56 39
t1 = {38, 39, 38, 39, 40}
2004-07-23 12:25:10 38 t2 = {38, 40, 39, 41, 40}
2004-07-23 12:26:15 39 t3 = {39, 38, 41, 39, 40}
2004-07-23 12:27:24 40 t4 = {41, 40, 38, 41, 40}
2004-07-23 12:28:36 38
3) For each transaction tn, find all possible sequential
2004-07-23 12:28:51 40 subsets with length • 2. Using the current example,
2004-07-23 12:29:14 39 a list of candidate sequential itemsets would
2004-07-23 12:33:40 41 include:
2004-07-23 12:34:37 40 t1 = (38, 39), (38, 39, 38), (38, 39, 38, 39), (38, 39,
2004-07-23 12:37:16 39 38, 39, 40)
t2 = (38, 40), (38, 40, 39), (38, 40, 39, 41), (38, 40,
2004-07-23 12:38:22 38 39, 41, 40)
2004-07-23 12:38:49 41 t3 = (39, 38), (39, 38, 41), (39, 38, 41, 39), (39, 38,
2004-07-23 12:40:21 39 41, 39, 40)
t4 = (41, 40), (41, 40, 38), (41, 40, 38, 41), (41, 40,
2004-07-23 12:41:01 40
38, 41, 40)
2004-07-23 12:48:37 41
2004-07-23 12:49:29 40 4) Calculate the support support(im) of each sequential
2004-07-23 12:51:50 38 candidate itemset when support(im) =
2004-07-23 12:53:20 41 frequency_count(im)/ number_of_transactions
2004-07-23 12:55:18 40 where frequency_count(im) = frequency of itemset
im in database D and number_of_transactions = |D|
/ transaction_length. If support(im) <
support_threshold, then remove the itemset from
The original SPADE algorithm was proposed by [3] and did
not account for consecutive sequential order as a parameter candidate sequential itemsets.
when finding frequent itemsets. SPADE algorithm was
designed to utilize all candidate subsets of a transaction 5) Calculate the confidence confidence(im) of each
when calculating the support and confidence of each itemset. sequential candidate itemset when confidence(im) =
In our problem, it is important to keep consecutive frequency_count( / frequency_count
sequential order of the items (the networks cells in this case). where A = itemset[0] the first element of the
It is important to note also that when calculating the
itemset and ( = itemset[0:] the set of all
confidence of each itemset , the item at the end of each
transaction is not considered by algorithm Fixed- items in the itemset. If confidence(im) <
MobileSPADE, because there is no succeeding movement confidence_threshold, then remove from candidate
beyond that item (or that cell). sequential itemsets.
Using the sample dataset shown in Table 1 as an example,
the Fixed-MobileSPADE algorithm and the Dynamic- End
MobileSPADE algorithm are described as follows:
B. Algorithm Dynamic-MobileSPADE
A. Algorithm Fixed-MobileSPADE Input: Transaction database: D; support_threshold;
Input: Transaction database: D; transaction_length; confidence_threshold
support_threshold; confidence_threshold Output: frequent sequential itemsets
Output: frequent sequential itemsets Begin
600 BRADLEY AND RASHAD

1) Sort Sort(D) database by sequential order, with MobileSPADE is the greater number of sequential itemsets
reference to timestamp value. *Refer to Figure 2 considered in algorithm Fixed-MobileSPADE. This occurs
due to differences found in the candidate sequential itemset
2) Sort T into tn transactions by creating non-repeating generation phase of each algorithm. Algorithm Fixed-
MobileSPADE takes all sequential subsets of each
sequences of itemsets. For example,
transaction into consideration while algorithm Dynamic-
MobileSPADE only considers each transaction as an
3) when using the database from Figure 3, T = {t1, t2, t3, itemset. Once both algorithms have generated all itemsets of
t4, t5, t6, t7} where: length two or greater that meet the minimum support and
confidence threshold, sequence rules can then be defined,
t1 = {38, 39} based upon the sequential itemset results.
t2 = {38, 39, 40}
t3 = {38, 40, 39, 41}
t4 = {40, 39, 38, 41} V. EXPERIMENT RESULTS AND PREDICTION ANALYSIS
t5 = {39, 40, 41}
We conducted multiple experiments with algorithm Fixed-
t6 = {40, 38, 41}
MobileSPADE and algorithm Dynamic-MobileSPADE,
t7 = {40}
using different parameters, on the MIT dataset. These
experiments are performed on a 2.5 GHz Intel CPU and
4) Consider each transaction as a sequential itemset,
4096 MB of memory. There were many results collected, but
where the length of each itemset must be greater because of the limited space and for the purpose of
than two len(itemset)  2. Using the current explanation, we will only use length-2 itemset results in this
example, a list of candidate sequential itemsets will paper.
be as follows: We tested Fixed-MobileSPADE algorithm over 1000
records from the MIT dataset and let transaction_length =
t1 = (38, 39) 10, confidence_threshold = 20%, and support_threshold =
t2 = (38, 39, 40) 20%. We found that 11 length-2 itemsets met these
t3 = (38, 40, 39, 41) parameters and are now considered to be frequent length-2
t4 = (40, 39, 38, 41) sequential patterns over 1000 examples by Fixed-
t5 = (39, 40, 41) MobileSPADE algorithm. We used also the same previous
t6 = (40, 38, 41) parameters over 2000 records from the dataset. The results
show that six itemsets met these specifications and are now
5) Calculate the support support(im) of each sequential considered to be frequent length-2 sequential patterns over
candidate itemset when support(im) = 2000 records by Fixed-MobileSPADE algorithm. We tested
also the Dynamic-MobileSPADE algorithm using different
frequency_count(im)/ number_of_transactions
values for confidence_threshold and support_threshold .
where frequency_count(im) = frequency of itemset After finding the frequent sequential itemsets that meet the
tn in database D and number_of_transactions = |D| / user-defined minimum support threshold and confidence
transaction_length. If support(im) < threshold, experiments were conducted on the dataset to test
support_threshold, then remove the itemset from for prediction accuracy of Fixed-MobileSPADE and
candidate sequential itemsets. Dynamic-MobileSPADE algorithms. Using a minimum
support threshold and confidence threshold of 20% for user
94 in the dataset, results for Fixed-MobileSPADE algorithm
6) Calculate the confidence confidence(im) of each
are graphed in Figure 2, where using 200 examples yielded
sequential candidate itemset when confidence(im) = the best results. Using a minimum support threshold and
frequency_count( / frequency_count confidence threshold of 5% for user 94 in the dataset, results
where A = itemset[0] the first element of the for Dynamic-MobileSPADE algorithm are shown in Figure
itemset and ( = itemset[0:] the set of all 3, where using 200 examples yielded the best results.
items in the itemset. If confidence(im) <
We can see that, overall, Fixed-MobileSPADE algorithm
confidence_threshold, then remove the itemset
performed better for user 94 due to the fact that user 94
from candidate sequential itemsets. exhibits more frequent behavior and the Fixed-
MobileSPADE algorithm considers more candidate
End sequential itemsets than Dynamic-MobileSPADE algorithm.
We can say that Fixed-MobileSPADE algorithm was
Algorithm Dynamic-MobileSPADE is similar to algorithm designed to perform best for a user who exhibits more
Fixed-MobileSPADE in that when calculating the frequent behavior, whereas Dynamic-MobileSPADE
confidence of each itemset, the last item in each transaction algorithm was designed to perform better under conditions
is not considered. One significant difference between where the user exhibits more sporadic behavior. This can
algorithm Fixed-MobileSPADE and algorithm Dynamic- especially be seen from the experiment results for the
 MINING MOBILE SEQUENTIAL PATTERNS IN WIRELESS CELLULAR NETWORKS 601

Dynamic-MobileSPADE algorithm in Figures 3 and Figure

4, where the minimum support threshold and minimum
confidence threshold have been lowered and the number of
records considered by the algorithm increased. Even with
these two parameters changed, prediction accuracy still did
not outperform Fixed-MobileSPADE algorithm. Again, this
occurs because algorithm Fixed-MobileSPADE takes more
itemsets into consideration than Dynamic-MobileSPADE
algorithm when determining candidate sequential itemsets.

Figure 4 Prediction accuracy using Dynamic-MobileSPADE

for Support Threshold = 10% and Confidence Threshold =
10%

VI. CONCLUSION AND FUTURE WORK

We have proposed in this paper two mobile data mining
algorithms that can be used to support the wireless mobile
Figure 2 Prediction accuracy using Fixed-MobileSPADE for networks. These algorithms are Fixed-MobileSPADE
algorithm and Dynamic-MobileSPADE algorithm. These
Support Threshold = 20% and Confidence Threshold = 20% algorithms are modified versions of the SPADE (Sequential
PAttern Discovery using Equivalence classes) algorithm.
These algorithms were used to predict future movements of
mobile users using mobility sequential patterns that are
generated by Fixed-MobileSPADE and Dynamic-
MobileSPADE algorithms. These algorithms were evaluated
using the mobility data of 100 mobile users for a period of 9
months. These data were collected by the MIT Reality
Mining project. Prediction analysis results show Fixed-
MobileSPADE algorithm can be used to extract mobility
sequential patterns of mobile users with frequent behavior
while Dynamic-MobileSPADE algorithm can be used to
extract mobility sequential patterns of mobile users with
sporadic behavior. Future development of Fixed-
MobileSPADE algorithm and Dynamic-MobileSPADE
algorithm includes extensions to consider other factors when
predicting a mobile user’s future movements, such as time of
day, services utilized in each cell, and an enhancement in the
time series analysis currently undertaken by algorithm
MobileSPADE.

REFERENCES
[1] S. N. Nguyen and M. E. Orlowska, “A Further Study in the Data
Partitioning Approach for Frequent Itemsets Mining,” in 17th
Figure 3 Prediction accuracy using Dynamic-MobileSPADE Australasian Database Conf., pp. 31-36, Hobart, Australia, 2006.
[2] A. Savasere, E. Omiecinski, and S. Navathe, “An Efficient Algorithm
usforing Support Threshold = 5% and Confidence Threshold for Mining Association Rules in Large Databases,” in Proc. Of the
= 5% 21st Very Large Databases Conf., pp. 432 – 444, Zurich, Switzerland,
1995.
602 BRADLEY AND RASHAD

[3] Zaki, M., “SPADE: An Efficient Algorithm for Mining Frequent [5] B. Congleton and S. Nainwal, “Mining the Mine - Exploratory Social
Sequences,” Machine Learning Journal, 42(1/2), pp.31-60, 2001. Network Analysis of the Reality Mining Dataset,” School of
[4] N. Eagle and A. Pentland, “Reality Mining: Sensing Complex Social Information, University of Michigan, Ann Arbor, Michigan.
Systems,” Personal and Ubiquitous Computing, Vol 10, #4, pp. 255- [6] J. Han and M. Kamber, Data Mining: Concepts and Techniques, 2nd
268, 2006. ed. San Francisco, MK, 2006, ch. 8, “SPADE: An Apriori-Based
Vertical Data format Sequential Pattern Mining Algorithm” , pp. 502-508.
 Agile Software Development Methods and its Advantages
Dr.Soly Mathew Biju
University of Wollongong in Dubai

The Agile software development method is based on

Abstract –There is a high trend observed in the the following principles:
industry today towards adopting Agile software
development method. This paper explains the • Individuals and interactions over processes
principles of Agile software development and tools.
approach and its benefits. It also exhibits how • Working software over comprehensive
Agile software development method reduces the documentation.
impact of frequently changing requirements • Customer collaboration over contract
throughout the development process. The paper negotiation.
will also describes the two agile development • Responding to change over following a plan.
methods used in the IT industry-Extreme
programming(XP) and SCRUM and the major
difference between their work styles and The Agile Software Development manifesto was
processes. It highlights the advantages of Agile published by a group of software practitioners and
software development over traditional software consultants in 2001 [3].
development lifecycle.
II. EXTREME PROGRAMMING (XP)
I. INTRODUCTION A. Introduction and benefits.

Traditional software development methods (water- “Extreme programming” is one of the commonly
fall, spiral, Iterative) were very cumbersome and used Agile software methods [7]. It is used in the
involved a lot of documentation. These methods were problem domains where the initial requirements are
very rigid, well planned and hence provided very likely to change to a large extent.
little or no flexibility. McCauley [4] states that “the
underlying philosophy of process-oriented methods is It is based on the following assumptions:
that the requirements of the software project one are
documented, they are locked and frozen before the • Be willing to do rework frequently and fast,
design and software development process takes as well as welcome customer feedback. It
place”. This process is not flexible enough to becomes a customer focused and customer
accommodate changes in the specifications at a later driven project rather than a plan driven
stage of the development. According to Highsmith project.
and Cockburn [3], “what is new about agile methods
is not the practices they use but their recognition of • Initial software is developed based on the
people as the primary drivers of project success, stories and requirements given by the user.
coupled with an intense focus on effectiveness and This software will contain the most
manoeuvrability. This yields a new combination of important, vital feature required by the user.
values and principles that define an agile world This method is found to be very effective as
view”. it involves the customer from the very
beginning. The customer writes the user
Project development would prove to be a disaster if stories comprising simple lines of text
the customer is not sure of the final product describing the features they want included in
requirements and can give the requirement only on an the first release of the working program. The
incremental basis. It follows the CCM (Capability acceptance test for the software is designed
Maturity Model), which is process based. The “Agile and based on these user stories. This will
Method” is people based rather than plan based. In keep a check on whether all the features
Agile software development the focus is on customer requested are implemented in the program.
satisfaction. It is an incremental/iterative software
development process which focuses on quick • The developer then estimates the time for
development of a working model. the development activity for which they will
have to meet the customer, reconfirm and

603
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_105, © Springer Science+Business Media B.V. 2010
604 BIJU

make detail note of the requirements. The B. Working style

first release should not take longer than The working style is slightly different. It may take
three weeks. sometime for the team members to implement this
style.
• Hence this method consists of small release
cycles for which the feedback of the • Pair programming is followed wherein two
customer/user stories is written which are programmers work together to develop a
then used for the next release of the particular functionality. This provides a better
software. quality product. There is no concept of peer
review.
• The length of the increments is decided and
is usually equal for all releases. • Stand up meetings are held to discuss the status
of the project. These short meetings are held on
• The developers are advised to keep their everyday basis.
code as simple as possible and the
technically latest version which in turn • Collective code ownership: everyone is
minimizes the need for documentation. responsible for the complete software. Anyone
can make changes to the program if they find
• Requirements can be added anytime during that it will improve the code. Unlike the
the project development. traditional project management where the project
manager would be responsible for the risks,
• New features can be added to the software complexity, deadlines etc., in the case of agile
as the next increments are provided. software development the team will organize
reorganise itself to handle all the pressures of the
• Before the next release the customer will project.
have to prioritise the requirements for that
release. The next release will focus on these • The major difference between Extreme
requirements. All the contents of the release programming style planning and incremental
are subject to change except for the new delivery, as in case of one of the most accepted
feature implemented [9]. “Unified Process” Framework, is the assumption
that Architecture and design can be done
• If analysis of the requirements shows that incrementally. Extreme programming spends
the increment will take longer than the very little time on defining the architecture and
decided increment length, it is broken down overall design. It starts directly with the first
into smaller increments. increment to deliver functionality.

• Each increment will have the developer XP undertakes the design in the following manner:
working on the analysis, design code
integration and testing phases iteratively [9]. • A system metaphor is used which names the
classes and the objects in a consistent
It follows an Iterative life cycle model with 5 manner which is agreed upon by all the team
activities: namely, Planning, design, coding, testing members. It also describes the overall
and integration. architecture.

• Planning occurs at the beginning of each • It follows the test first method where the
iteration programmer writes the code for testing the
• Design, coding, and testing are done program before writing the program. These
incrementally tests are written incrementally based on the
• Source code is continuously integrated into user requirements. These are the tests the
the main branch, one contribution at the time program unit must comply with.
• Unit tests for all integrated units. Unlike in
case of traditional software development Consider the following class Employee and its
methods regression testing is carried out at corresponding Junit test case :
all the iterations
 AGILE SOFTWARE DEVELOPMENT METHODS AND ITS ADVANTAGES 605

public class Employee { Another controversial assumption is that Analysis

private double Salary;
private String name;
can be done incrementally XP performs analysis in
the following manner:
public Employee() {
name=”xyz”;
Salary = 5000; • User stories, which contain brief description
} of the requirements, are discussed
interactively by a team of programmer and
public void increment () {
Salary + = 7000; the customer. The requirements are then
} formulated into acceptance test cases which
public int getSalary() {
the software should comply with.
return Salary;
} • Planning game is used to do release
}
planning. These are meetings held per
The test case in Junit is as follows: iteration. The customer and the
programmers participate in the planning
import junit.framework.TestCase; game.
public class MoneyTest extends TestCase {
The assumption that the product can be delivered to
public static void main(String[] the customer in an iterative manner makes it most
args) { beneficial to the customer. The customer can receive
junit.textui.TestRunner.run(EmployeeT and make use of the most important and valuable
est.class); functionally within a very short period of time. Thus
} the customers get a sooner return on their investment.
The order in which the feature should be added to the
public EmployeeTest(String arg0) {
super(arg0); software can be decided by the customer. The
} customer can thus verify whether what is given is
what they wanted at a very early stage and can
protected
Exception {
void setUp() throws
provide feedback to the programmers thus driving the
super.setUp(); project in the right direction.
}
C. Is Extreme programming always the best method
protected void tearDown() throws
Exception {
to follow?
super.tearDown();
} This depends on the project, the team and the
customer. If the assumptions of XP hold, there are
public void testEmployee() {
final Employee emp1 = newEmployee(); various benefits to be derived from incremental
assertEquals(5000, emp.getSalary()); software delivery. One of them is the early delivery
of first incremental model of useful software.
Feedback from the customer gives required
}
confidence to the programmer and can help drive the
} next incremental release. Incremental releases can be
scheduled hence there can be predictable delivery.
Simple code and minimum documentation saves a lot
Listing 1, JUnit test for the Employee object of time.

The test case in listing1 tests whether the constructor III. SCRUM
is invoked and values are initialized correctly.
Another agile software development method is
The code must be written so that it will satisfy these SCRUM.
tests.

• Refactoring reworks the design to iteratively Scrum and Extreme Programming (XP) are clearly
adapt the changes in the environment and aligned. In fact, if you walked in on a team doing one
of these processes you might have hard time quickly
requirements. It follows group design
techniques like Whiteboard design sessions, deciding whether you had walked in on a Scrum team
Class-Responsibility-Collaborator sessions
(CRC) etc.
606 BIJU

or an XP team. The differences are often quite subtle,

but they are important.

There are four main differences between Scrum and

XP [8]:

First, Scrum teams typically work in iterations (called

sprints) that are from two weeks to one month long.
XP teams typically work in iterations that are one or
two weeks long.

Second, in Scrum the teams do not allow changes

into their sprints. Once the sprint planning meeting is Figure 1: SCRUM [10]
completed and a commitment has been made to
delivering a set of product backlog items, that set of
items remains unchanged through the end of the
sprint. XP teams are much more flexible within their
iterations. As long as the team hasn’t started work on The following graph depicts key differences between
a particular feature, a new feature of equivalent size the traditional waterfall model and Agile method
can be swapped into the XP team’s iteration in
exchange for the unstarted feature.

Third, Extreme Programming teams work in a strict

priority order. Features to be developed are
prioritized by the customer (Scrum’s Product Owner)
and the team is required to work on them in that
order. By contrast, the Scrum product owner
prioritizes the product backlog but the team
determines the sequence in which they will develop
the backlog items.

Finally, whereas Scrum doesn’t prescribe any

engineering practices, XP does. XP engineering
practices include things like test-driven development,
the focus on automated testing, pair programming, Figure 2: Difference between traditional waterfall
simple design, refactoring, and so on. model and Agile method.

These are small and often subtle differences between

Scrum and XP. They can, however, have a profound
impact on the team. A possible approach to
maximising the advantages of both approaches is to
start with Scrum and then begin using XP
engineering practices, such as test-driven
development and pair programming.
 AGILE SOFTWARE DEVELOPMENT METHODS AND ITS ADVANTAGES 607

The graph clearly shows that Agile programming REFERENCES

provides added value required by the customer in an
incremental fashion while the traditional approach [1] Agile Alliance 2007, Available: URL: http://
fluctuates between the required value and minimum www.agilejournal.com/ [Accessed 30th Sept 2007]
required value.
[2] Agile Manifesto (2007), Available: URL: http://
IV. DRAWBACKS OF THE APPROACH www.agilemanifesto.org [Accessed 7th Oct 2007].

If we are adopting agile method for the first time we [3] Beck, K. & Andres, C., Extreme Programming
will face a challenge of changing the mindset of our Explained. Addison-Wesley, 2004.
team member and helping them understand the rules
of the new method [5]. A lot of rework may be [4] Cohn, M.,“User Stories Applied for Agile Software
involved which could have been avoided if a Development” Addison-Wesley, 2004.
complete requirement analysis is done initially. If a Construx, http://forums.construx.com/ [Accessed 30th
proper order of building the software is not followed, Dec 2007]
we might end up reworking from the beginning to
include another set of requirements. This is especially [5] Evan, I. “,Agile Delivery at British Telecom”
likely if we need to add certain security or CSI communications , July 2007.
performance features to the software. The complete
process may slow down if requirement analysis and [6] IBM http://www.ibm.com/developerworks/java/
design have to be performed for all iterations. library/j-xp042203/ [Accessed 9th June 2007]

V. CONCLUSION [7] Larman, C. (2004). Agile & Iterative Development.

Addison-Wesley.
The Agile software development process will
produce incremental release of the software which is [8] Mike (Mike Cohn’s blog) http://blog.mountaingo-
very useful to reduce product risks and to delivery a atsoftware.com/?p=8
product with customer satisfaction. Delivery of a
working model of the product can be delivered at a [9] Pascal Van C.” Going round and round and getting
very early stage. The software will truly be “agile” nowhere extremely fast?”, CSI Communications, July
that is subjected to change until the final delivery. It 2007.
is always better to have a little planning ahead so as
to avoid possible pitfalls. Certain import [10] Tamhane, S., IT seminar ‘Application of IT in
requirements related to security and performance heath care’. Dubai: University of Wollongong ,
features can be collected in advance. If one is December 2007.
familiar with the domain, has a stable design and has
worked on a similar environment then it is better to [11] Wells, J. Donovan , Extreme programming: a
follow the waterfall method. If a project is not gentle introduction. Available at www.extreme-
familiar, has unstable design and involves a lot of programming.org [Accessed 9th Oct 2007]
risk, it is always advisable to adapt to the Agile
software development method. Agile has largely
become a synonym for "modern software practices
that work," [4].
 Toward a High Performance Grid Site RO-14-ITIM
and the Contribution to WLCG
F. Farcas, G. Popeneciu,
National Institute for Research and Development of Isotopic and Molecular Technologies,
Cluj-Napoca, 400293, Romania, email: [email protected], [email protected]

Abstract

Grid Technology is the next phase in computer infrastructure (ii) During the last years computational simulations based on
that will fundamentally change the way we think about and use the atomic description of biological molecules have been re-
computing power. RO-14-ITIM is the Grid site of the National sulted in significant advances on the comprehension of biologi-
Institute for Research and Development of Isotopic and Mole- cal processes. A molecular system has a great number of con-
cular Technologies Cluj-Napoca (INCDTIM). In this paper we formations due to the number of degrees of freedom in the ro-
present the evolution and prospects of the site, his ability to tations around chemical bonds, leading to several local minima
integrate in huge Grid networks, the hierarchical structure as in the molecular energy hyper-surface. It has been proposed
well as his key in research work of the Nord-West part of Ro- that proteins, among the great number of possible conforma-
mania research field. RO-14-ITIM site support physic domains, tions, express their biological function when their structure is
HPC and Grid computing for distributed analysis. close to the conformation of global minimum energy [3].
This type of research involves a large amount of computing
I. INTRODUCTION power and fills to be a very suitable application for grid tech-
nology.
The Grid computing vision promises to provide the needed
(iii) Over the past few years, a sizable body of experimental
platform for a new and more demanding range of applications.
data on charge transport in nanoscopic structures has been ac-
To become true, a lot of work, including design and deploy-
cumulated.
ment of adequate resource management and information ser-
We face the birth of a whole new technological area: the mo-
vices, need to be overcome. In this context we are presenting
lecule-based and molecule-controlled electronic device re-
the structure of Grid computing that applies at the INCDTIM
search, often termed “molecular electronics” (ME). The sim-
institute in Cluj-Napoca, Romania.
plest molecular electronic device that can be imagined consists
In 1996 we start working at the installation of the ATLAS
of a molecule connected with two metallic nanoelectrodes.
LHC CERN Geneva detector with the construction of parts for
There is now a variety of approaches to form such nanojunc-
the Tile Calorimeter. As a next step, studying the forecast in
tions (NJ) [4] that differs in the types of electrode, materials
Grid we made the preparation to build up our own Grid site.
employed, the way in which the molecule-electrode contacts
The work seemed easy but great challenges were to be ex-
are established and the number of molecules contacted. Recent-
pected in the next years.
ly the realization of the first molecular memory was reported
Today we apply and develop the Grid system in many re-
[5].
search areas as follows:
The computation of the current-voltage characteristic of a
(i) The fundamental research in particle physics uses ad-
realistic nanodevice needs about 6-10 GB of memory and one
vanced methods in computing and data analysis. Particle phys-
week computing time, which bring to the idea of using a Grid
ics is the theory of fundamental constituents of matter and the
environment with high processing power and MPI support.
forces by which they interact and ask some of the most basic
questions about Nature. Some of these questions have far- II. DATA ABOUT ROMANIA GRID SITES
reaching implications for our understanding of the origin of the
Universe [1]. We can speak about Grid sites in Romania after 2001, but
In 2009 Large Hadron Collider accelerator (LHC) [2] with officially on 2006 when Romania through National Authority
ALICE, ATLAS, CMS and LHCb experiments, will start tak- for Scientific Research (ANCS) and The European Organiza-
ing data. LHC will collide protons at the highest energy tion for Nuclear Research (CERN) signed a “Memorandum of
¬s=14TeV and luminosity L=1034 cm-2s-1 among all the other Understanding”, for collaboration in development and exploita-
accelerators and due to these performances, high precision tion on the Worldwide LHC Computing Grid. [6] From that
measurements will be possible and the results means new phys- time, officially we begun to work and develop the Romanian
ics. Grid sites.
To fulfil these requirements a high performance distributed In Romania we have 2 Grid Consortium, RO-Grid found in
computing system is needed. 2002 and Tier 2 Federation (ROT2F) found in 2006. The

609
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_106, © Springer Science+Business Media B.V. 2010
610 FARCAS AND POPENECIU

founders of both consortiums are the most important Institutes For the improvement of our Grid site, this first step where
in Romania which were working in Grid projects years ago, continued through other projects, consisted in purchasing mi-
and we can mention the most important: National Institute for nimal technical equipment.
Physics and Nuclear Engineering “Horia Hulubei” (IFIN-HH), In parallel we had built a location that meets the highest
National Institute for R&D in Informatics ICI Bucharest, Uni- needs for processing and storing data, which became a type
versity ”Politehnica” of Bucharest, Technical University of Tier 2 Datacenter [9] with a good security and monitoring sys-
Cluj-Napoca, Western University of Timisoara and National tem as well as a backup system assured by a 220 kVA genera-
Institute for R&D of Isotopic and Molecular Technologies tor.
(INCDTIM). RO-Grid most important objective is to encour- For the final stage we continued to equip our system with a
age and facilitate the involvement of other interested and com- high capacity storage unit and many more computing servers
petent national institutions and ROT2F purpose is to provide which where included in the INCDTIM grid site, RO-14-ITIM,
the computing resources needed to process and analyze the because the grid system is needed to store and analyze big
data gathered by the Large Hadron Collider experiments. amount of information.
Among the first site’s in Romania were RO-01-ICI and RO-
02-NIPNE and the last one that have a contribution to WLCG IV. NOWADAYS GRID SITE RO-14-ITIM
is RO-14-ITIM. [7]
Through the “Memorandum of Understanding”, we assumed
III. FIRST STEPS TOWARD INCDTIM GRID SITE responsibilities to provide computing power and storage capac-
ity in collaboration with all other WLCG sites from Romania.
As part of the WLCG Romanian group, and signer of the The resources that all this system needs are: high speed con-
“Memorandum of Understanding” we started the building of nectivity, both local and wide area; huge computer throughput,
RO-14-ITIM site with 2 servers and a 10 Mbps link in 2005, great off-line data storage; human resources to manage local
through the first CEEX project between INCDTIM and IFIN- configuration and to implement GRID specific activities.
Nowadays at INCDTIM Cluj there are 3 main directions for
HH. INCDTIM site was integrated and tested as an extension
processing data:
of the IFIN-HH ATLAS-GRID site. The following tests have
• ATLAS data processing in WLCG project;
been done [8]:
• Numerical simulation of computational physics
(i) Transmission speed measurements between IFIN-HH and
with application in biophysics and nanostructures;
INCDTIM during sustained high speed data traffic, shown in
• Simulation in environment domains.
Fig. 1, was obtained using NFS copy. We can see a limited 10
(i) LCG project - LHC Computing Grid - was launched in
Mbps link, which is not satisfactory for GRID production pur- 2002 and it has the goal to make a distributed GRID to be used
poses. by ALICE, ATLAS, CMS and LHCb experiments. Till now we
(ii) Entire system functionality tests as a single grid configu- have in Romania 2 sites that is working for the ALICE project
ration, task distribution and execution real speed depreciation and 3 sites that are working for the ATLAS project. The
measurements between the remote nodes. amount of data that CERN has to process is big enough so that
only one site in the world could not work for. The computing
resources estimated to be necessary are presented in Tabel.1
[10].

Table 1. ATLAS computing resources estimates

CPU(MS12k) 2007 2008 2009 2010
CERN Tier0 0.91 4.06 4.06 4.61
CERN Tier1 0.49 2.65 4.18 7.00
All tier-1s 4.07 23.97 42.93 72.03
All tier-2s 3.65 19.94 31.77 53.01
Total 9.12 50.61 82.94 138.45
where 1 Athlon 3Ghz ~= 1 kSI2k.

Beside LCG project we intend to use our grid site for other
inner projects. Both of them are connected with the numerical
simulation of computational physics with direct application in
biophysics and nanostructures. As a general rule, the computa-
tional techniques used by numerical simulations of computa-
tional physics are based on the use of Unix-like operating sys-
tems. The codes used are mostly developed in Fortran. A vast
majority of these codes are developed under GNU license
Fig. 1. Traffic measurement between IFIN-HH Bucharest and INCDTIM (such as ABINIT) or academic license (SIESTA, GAMESS
Cluj on April 2006
 TOWARD A HIGH PERFORMANCE GRID SITE RO-14-ITIM AND THE CONTRIBUTION TO WLCG 611

etc). The commercial ones are represented by GAUSSIAN or

MOLPRO.
The systems under investigation by current applications, re-
quire a computing time that can exceeds several weeks [11].
For example, the computation of the current-voltage characte-
ristic of a realistic nanodevice needs about 6-10 GB of memory
and one week computing time on the last generation of Opteron
processors. Therefore, the development of parallel computa-
tional techniques is nowadays in the forefront of development
of the methods used by computational physics.
The MPI (Message Passing Interface) has proved to be a
very good candidate for this task. The modern developments of
the above mentioned codes are using the MPICH2 implementa- Fig. 2. RO-14-ITIM Site structure
tion of MPI in order to reach a good scaling of the computing
time with the number of processors. RO-14-TIM hardware configuration, placed in INCDTIM
Datacenter, is presented in Fig.3.
A. Structure and testbeds

To understand the technical support of INCDTIM pragmatic

achievements we point out some theoretical knowledge about
network Grid structure.
A Grid model is used, in which a hierarchical structure has
been proposed for processing and data storage. This model,
generally speaking is based on computing elements, monitoring
elements and worknodes as well as a storage element which
can be a simple computer or a farm of thousand of hard disks.
A functional testbed needs a scalable infrastructure, adequate
middleware and application, sustained by good teams that
manage the whole system.
A minimal functional WLCG Grid site has to fulfill some
elements as: one computing element, one monitoring element,
one testing element, one user interface, one storage element, Fig. 3. RO-14-ITIM Grid site configuration
one site-BDII element and as many as possible worknodes.
It is notable the site administrator role accredited through his The link capacity between IFIN-HH and INCDTIM was
valid digital certificate, which is to test, monitor and update the monitories in the last 2 years. The results show an increasing
Grid-site having access to all resources within the existing VO link capacity from 200 to 400 MBps after installation in 2009
in the limits assigned by the ROC staff through the creation of a Switch Layer 3 Cisco 350.
process of his site. Transmission speed measurements during sustained high
speed data traffic made in 2008, shown in Fig. 4., carry out a
B. RO-14-ITIM grid site network speed about 200 Mbps.

RO-14-ITIM Grid Site is an EGEE type, running gLite 3.1

as middleware for the public Network computing elements
with Scientific Linux 6.4 and gLite 3.2 for the Worknodes on
Scientific Linux 5.3 x86-64.
RO-14-ITIM Grid site is certificated for production and is
registered in Grid Operations Center – GOC [12].
The public network consist of one central element (CE), one
user interface (UI), one site-BDII server, one storage element
(SE) with 80 TB storage capacity a testing element with Na-
gios and one monitoring element (MON). Fig. 4. Traffic measurement on April 2008
The private network contain 54 dual processor quad core
servers with 16 GbRam, from which two HP Blade system and
one IBM blade system. After the installation of Switch Layer 3 in 2009, transmis-
The wide area connection is plan to work at 10 Gbps in the sion speed measurements, Fig.5, show an increasing to 400-
near future. INCDTIM local and wide area network operates 450 Mbps.
now at 1 Gbps.
RO-14-ITIM Grid Site configuration is given in Fig. 2.
612 FARCAS AND POPENECIU

Traffic measurement

500

400
S peed [Mbps]

300

200

100

0
0 4 8 12 16 20 24 28 32 36 40 44 48
Time [h]

Fig. 8. RO-LCG statistic last 6 month

Fig. 5. Traffic measurement on April 2009
Fig. 9 shows INCDTIM Datacenter which hosts RO-14-
After 2009 until now the site was monitorised and compared ITIM Grid site (left side of the picture).
with other sites. Functionality tests show that the site provides
a very good availability and reliability, over 90% for the last 3
months, as shown in Fig. 6.

Fig. 6. RO-14-ITIM availability last 3 month

The statistics for the last 4 month are promising, there where
over 500 job done, as shown in the Fig. 7. Fig. 9. INCDTIM Datacenter

V. FUTURE WORK INTO INCDTIM GRID SITE

Since 2007, when the Datacenter of INCDTIM was finished,

National Institute for Research and Development of Isotopic
and Molecular Technologies had started for new more envi-
ronmental friendly resources and power saving computers and
storage elements. As a result of this futuristic concept big steps
were done in writing and gaining projects. So, through the In-
ternational project “POS-CCE, Improve of the capacity and
reliability of INCDTIM GRID center for its integration into
international Grid networks, 2009 – 2010” the institute would
be able to improve the storage capacity and processing power
Fig. 7. 500 ATLAS jobs done last 4 month of RO-14-ITIM site.
In parallel, a second target is to improve our internal and ex-
In comparison with other RO-LCG sites that have a great ternal bandwidth through the acquisition of a Cisco 6500-E
contribution to WLCG, RO-14-ITIM data are presented in router which would improve the network link to 5 - 10 Gbps
Fig.8. with other Grid sites.
As a result of this work in August 2009 the site RO-14-ITIM To have a reliable Datacenter, with an increased efficiency,
was accepted in the RO-LCG community, most of all because we are looking for a water cooling solution, which will also
of site great availability. minimize the inside noise.
Reaching these objectives, the side RO-14-ITIM will be able
to satisfy both the WLCG specifications and other computing
power demands.
 TOWARD A HIGH PERFORMANCE GRID SITE RO-14-ITIM AND THE CONTRIBUTION TO WLCG 613

VI. CONCLUSIONS [5] J.E. Green, et all, Heath: A 160-kilobit molecular electron-
ic memory patterned at 1011 bits per square centimeter,
RO-14-ITIM site was designed and developed along almost Nature 445(7126), 414 (2007).
5 years, from a basic structure, and with a continuous effort, to [6] Memorandum of Understanding for Collaboration in the
a competitive structure, well integrated in TIER-2 Federation. Deployment and Exploitation of the Worldwide LHC
The RO-14-ITIM contribution to WLCG with his 18.567 Computing Grid, CERN-C-RRB-2005-01, November
CPU time [13] which represents 0.45% of the whole time dedi- 2005.
cated by all 5 sites to the CERN project and 3.55% for the [7] F. Farcas, I. Szabo, A. Bulinschi, L. Miclea, A. Bot ,
ATLAS project. World Wide Romania Grid Database, 3rd International
We interact with new partners in international and national Workshop on High Performance Grid Middleware, 17th In-
programs and projects, and this give us the opportunity to de- ternational conference on control system and computer
velop our specialists and infrastructure of grid system, to reach science, 26 – 29 May 2009, Proceedings, CSCS – 17 Vol
our mission: INCDTIM Datacenter, to be the most important 2, pp 461-464, Copyright © 2009, Editura Politehnica
Grid centre in Transylvania capable to sustain the Nord – West Press
research and economical area of Romania. [8] C. Alexa, V. Boldea, M. Ciubancan, . Constantinescu, S.
Di­¯, D. Pantea, T. Preda, M. Jalobeanu, G. Popeneciu,
Acknowledgments IFIN-ITIM Distributed Grid System, ISPDC 2006, Pro-
ceedings 5th International Symposium on Parallel and Dis-
This work is supported by the Romanian Research and De- tributed Computing, Timisoara Romania, July 6-9, 2006,
velopment Agency through EU12/2009, EU15/2009, POS-CEE pp 215-219, IEEE Copyright 2006.
Ingrid/2008 projects. [9] TIA – 942, Data center standards overview.
[10] J. Knobloch and L. Robertson, LHC Computing Grid,
References Technical Design Report, CERN 2006.
[11] F. Farcas, G. Popeneciu, A. Bende, C. Morari, S. Belov, L.
[1] ATLAS Collaboration, Exploring the Mystery of Matter, Miclea, ITIM Distributed Grid System applied in high
Papadakis Publisher, UK, 2008. energy, biomolecular and nanotehnology physics, IEEE
[2] LHC Homepage, http://cern.ch/lhc-new-homepage/. International Conference on Automation, quality and test-
[3] S. P. Brown, S. W. Muchmore, J. Chem. Inf. Model., 46, ing, robotics (AQTR) May 22-25 2008, theta 16th edition,
999 (2006). vol III, Proceedings pp 343-346 2008.
[4] A. Salomon et all, Comparison of Electronic Transport [12] https://goc.gridops.org/site/list?id=1655004
Measurements on Organic Molecules, Adv. Mater. 15(22), [13] https://www3.egee.cesga.es/gridsite/accounting/CESGA/
1881-1890 (2003). tier2_view.html
A MFI4OR-based Approach of Semantic Annotation
for Web Service
Cheng Zeng Keqing He, Bing Li, Chong Wang, Yangfan He
1
State Key Laboratory of Software Engineering State Key Laboratory of Software Engineering
Wuhan University Wuhan University
Wuhan, P.R. China Wuhan, P.R. China
2
College of Mathematics & Computer Science E-mail: [email protected]
Hubei University
Wuhan, P.R. China
E-mail: [email protected]

Abstract—Compare to the traditional annotation is enriched with the adoption of semantic annotation to
technology for web service, a novel approach for the semantic improve the flexibility of the WSR and help users with
annotation to web service based on MFI4OR is proposed in solutions (services) that are close enough to what they
the paper. This proposal permits multi-ontology based would have liked to get (even if they do not fully match
annotation to web service in the same domain which will lead their expectations). Therefore, how to annotate these items
to the flexible annotation, and then the flexible publication becomes a key point in the issue of semantic annotation. In
and discovery of web service in Web Service Registry (WSR). recent years, a large number of research based on semantic
The result of experiment illustrates the efficiency of web techniques have been proposed, such as the efforts around
service discovery and the degree of satisfaction to users in
OWL for Services (OWL-S) [8], the Web Services
WSR is highly improved.
Modeling Ontology (WSMO) [9], WSDL-S [10] and
Keywords-Semantic Annotation, Web Service, Domain METEOR-S [12].
Ontology, MFI4OR In this paper, a novel approach for the semantic
annotation for web service based on ISO SC32 19763-3:
I. INTRODUCTION MFI4OR (Metamodel Framework for Interoperability:
With XML based standards like UDDI, WSDL and Metamodel for Ontology Registration) [4] is proposed,
SOAP, web services play a key role in universal which will make semantic annotation more accurate and
connectivity and interoperability of applications and efficient. The remainder of the paper is organized as
services [1]. With the growing popularity of Web services, follows: Section 2 presents the traditional model of
there arise issues of finding relevant services, especially semantic annotation. In Section 3, we provide some
with the possibility of the existence of thousands of Web improved methods of semantic annotation by use of domain
services. Web Service Registry (WSR), which acts as an ontology. Section 4 proposes the MFI4OR-based approach
information mediation role in the triangular SOA, is to of semantic annotation. The related work is discussed in
solve this problem. In WSR, service providers can publish Section 5, followed by the conclusion and future work in
service description information more effectively and Section 6.
service consumers can discover appropriate services more
accurately and easily. However, currently web services are II. TRADITIONAL MODEL OF SEMANTIC ANNOTATION
described as WSDL descriptions, which provide WSR is supported by domain ontology for semantic
operational information such as Inputs and Outputs etc. It is publication and discovery of the Web services. The
a type of syntactic level description and its interface domain ontology is created from concepts and
parameters do not have meaningful names, which obstructs
terminologies that are likely to be used by Web services in
service discovery and composition. For example, one
a particular domain.
service A has an output item which use specialty as name to
describe what field an expert engages in, while another For example, in the electronic commerce domain, the
service B may use research field for describing such an domain ontology can consist of concepts from the ebXml
item. As we all know, specialty and research field represent Core Component Dictionary. Another example could be
the same information item, but they are different at domain ontology for the Traffic domain which may
syntactic level, and same at semantic level. Therefore, the include concepts like Car, CarPrice, ArrivalCity and
key to solve the limitation of keyword-based searches is to DepartureCity. As a result, before the semantic annotation
add semantic annotations to Web service specifications [2]. of web service, we must choose a concrete domain
ontology. After that, we can manually use the mapping
Domain ontology is used to add semantic information between these items in WSDL file (mainly refers to WSDL
on WSDL interface parameters to overcome the limitations
interface parameters of Web Service such as Inputs,
of a syntax-based search [3]. The discovery of web services

615
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_107, © Springer Science+Business Media B.V. 2010
616 ZENG ET AL.

Outputs and Operations) and the selected domain ontology includes terms related to a given application domain. We
to capture the meaning implied by the Web service assume that this ontology can be built by domain experts
provider in that domain. Finally, the mappings are written who analyze the terms included in the Web services
to the WSDL file as annotations. The annotated WSDL file published in the registry. The latter includes all the possible
along with the original WSDL file is shown in figure 1. terms (at this stage we use Wordnet).
However the manual process of annotation can not assure We assume that programmers of Web Services have a
the accuracy of semantic annotation. Section 3 is ready to general understanding about the domain which Web
overcome this problem. Services belong to, such as the Electronic Commerce
domain or the Traffic domain. Therefore, when defining the
names of operations and parameters in Web Service, they
may use the concept in the domain, which can avoid the
naming at will. When annotating the names of operations
and parameters, our algorithm will realize the following
two steps: (1) Making a match between these names and
the concept items in the domain ontology, and producing
the candidate concepts list from the domain ontology to
annotate. (2) Sorting the candidate concepts by the priority
of (Exact, Plug-in, Subsume, Fail) according to the
algorithm introduced in the [7]. The sorted candidate
concepts list is shown in Figure 2. Then the publisher can
select the most satisfying concept from candidate concepts
to annotate the names used in Web Service.
Figure 1． Fragment of original and annotated WSDL files However, it is possible that the names of operations and
parameters may not exist in the domain ontology, or reflect
the naming convention usually adopted by programmers:
III. IMPROVED MODEL OF SEMANTIC ANNOTATION UserName or currencyExhange are more frequent than the
In this section, a semi-automatically mapping simple names directly included in the ontology. For this
algorithm is used in the process of semantic annotation reason, if the names are composite words, we will tokenize
which will improve the accuracy of semantic annotation. the word and return the similar word to replace it; if the
Figure 2 shows the improved annotation process of names are not directly included in the ontology, our
mapping WSDL concepts to the nodes in Domain during algorithm will make use of the general-purpose ontology
semantic annotation. Different from traditional model, we (i.e. WordNet) to build the synset list of the missing name.
have developed an algorithm to semi-automatically map And then each word in the synset will be made matching in
each individual concept in the WSDL description (i.e. the same way introduced above. The sorted candidate
Inputs, Outputs, and Operations) to an ontological concept. concepts list is also created for the provider to annotate.

Phone

Mobile
Phone

Mobile
Nokia Moto Battery

subClassOf partOf

Figure 3ˊ Fragment of Ontology

In addition, our algorithm considers the other match-

making relationship: partOf and siblingOf. In [7], the
Candidate Concepts List algorithm only proposes the four degree of matching: Exact,
Figure 2． Improved Model of Semantic Annotation Plug-in, Subsume and Fail, which has some disadvantages
in matchmaking. For example, Figure 3 shows an
To achieve this goal, the algorithm relies on two kinds fragement of phone ontology. If one service has an output
of ontologies: the domain ontology which is discussed in named as Nokia, the concept Mobile phone and phone will
section 1 and the general-purpose ontology. The first be selected as candidate concepts according to the
 A MFI4OR-BASED APPROACH OF SEMANTIC ANNOTATION FOR WEB SERVICE 617

algorithm in [7]. In fact, the concept Moto should also be concepts, attributes, properties, constraints and so on.
considered as the candidate concept because it has a Theoretically, the three basic rules can cover all the
siblingOf relationship with the concept Nokia. For another possible modifications on RO and LO because any
case, if one service has an output which uses Mobile battery complex evolution can be decomposed, viewed as a
as name, there is no matching concept using the algorithm sequence consisting of these three operations.
in [7]. Actually, the concept Mobile Phone, which has a
partOf relationship with Mobile battery, has a possible B. MFI4OR-based Model of Semantic Annotation
opportunity to be the candidate concept. Our algorithm has As mentioned in section 2 and section 3, the process of
considered these possible matching relationships between semantic annotation to web service is based on the single
the signatures of Web Service and the concept items in the domain ontology. However there also exist some cases that
domain ontology when annotating, which ensures to different ontologies in the same domain are used to
achieve the satisfying annotation at most. annotate web service, which will lead to the issue of the
Whatever traditional model or improved model interoperability of multi-ontology in the same Domain.
mentioned in section 2 and 3, they are both focus on the As we all know, Domain Ontology is always
single domain ontology to annotate. In section 4, the constructed and standardized by Domain Experts which use
method of multi-ontology based annotation to web service common approved concepts and relationship among them,
is proposed which will lead to the flexible annotation in but sometimes service providers maybe define their own
WSR. concepts to semantically annotate their web services for
meeting their requirements. We regard the domain ontology
IV. MFI4OR-BASED MODEL OF SEMANTIC ANNOTATION defined by experts as Domain Reference Ontology (DRO)
and the domain ontology defined by users as Domain
A. What is MFI4OR? Local Ontology (DLO). However the process of defining
the new concepts in DLO is not at will, but evolved from
Ontology provides a basic technology to support the
DRO according to three ontology evolution rules (SameAs
semantic interoperability of different systems over a
Rule, Enhancement Rule, and Contraction Rule) defined in
distributed and heterogeneous environment. If the MFI4OR. That is to say, DRO and DLO are not isolated but
ontologies have the semantic relation, they could interrelated to each other based on these evolution rules.
understand each other to some extent, which is the Figure 4 shows that DRO and DLO can both be adopted to
precondition of semantic interoperability. MFI4OR is add semantic information to Web Service. It can extend the
mainly used for providing a common framework to scope of semantic annotation in the Domain, which will
register ontology and its evolution information for enhance the flexibility during the process of service
semantic interoperation between ontologies or web-based publication and discovery.
applications based on them. It provides the solutions by
defining “Reference Ontology” (RO) and “Local
Ontology” (LO) which are both registered by a common
structure formed as Ontology-Ontology Component-
Ontology Atomic Construct [4]. The Ontology
Components defined in RO can be reused by the other
ontologies, while that defined by LO can only for its own
definition purpose. RO is always constructed by domain
experts, while LO is evolved from RO. The LOs and RO
can interoperate with each other, which provide a solid
foundation for interoperability between the different
ontologies. Besides RO and LO, the ontology evolution
rules between them are the core part in MFI4OR which
can help record the detailed changes when LOs evolve
from RO. These rules can ensure that the different
ontologies used to annotate web service will not hamper
the interoperability among them.
General speaking, there always exist three rules:
SameAs Rule, Enhancement Rule, and Contraction Rule [5, Figure.4. Multi-Ontology annotation Based on MFI4OR
6]. SameAs Rule can set up the equivalent mapping
between instances of Ontology, Ontology Component or
Ontology Atomic Construct. Enhancement Rule is special As mentioned above, DLO is evolved from DRO, and
for adding some new elements to the original ontology the evolution information between DRO and DLO should
when versioning and evolving. Contraction Rule is often be registered and recorded, which will improve the
adopted in ontology evolution management to delete interoperability between them in WSR. Although there
618 ZENG ET AL.

exist three evolution rules between DRO and DLO, we are

ready to adopt SameAs rule because this rule is the part of Person Person
the first version of MFI4OR which has already become
International Standard since 2008, and the other two
evolution rules will be in the second version of MFI4OR in Businessman Businessman

the future.
C. Realization of the MFI4OR-based Model Consumer Seller Buyer Seller

Figure 5 shows the User Interface (UI) of annotation subClassOf

subClassOf
with DRO or DLO. Once service provider selects the
Domain that web service belongs to, such as logistics or (a) DRO (b) DLO
transport, DRO and DLO will be imported to this platform
for the use of annotation to web service. The evolution Fig.6. Fragment of DRO and DLO
information between DRO and DLO will lead to the
interoperability between them.
D. Evaluation
To evaluate our approach, we take precision, the
average time and the average number of web services as
the evaluation metrics. Precision is defined as the ratio of
correct web services retrieved to the total number of result
retrieved. The performance is measured on a workstation
of 1GB RAM, 2.6GHz CPU with Microsoft Windows XP.
Evaluation is performed on TMSA (Traditional Model
of Semantic Annotation), IMSA (Improved Model of
Figure.5. UI of annotation with DRO or DLO Semantic Annotation), and MMSA (MFI4OR-based
Model of Semantic Annotation). Figure 7 shows the
For example, there is a fragment of DRO shown in relationship of retrieving time and the number of web
Figure 6 (a). When evolving from DRO to DLO, Service service in different models of semantic annotation. The
Providers can reuse the concepts of Person, businessman average retrieving time is increased with the increasing
and seller in DRO, and more important can define the new number of web services. Meanwhile in MMSA, the
concept of buyer for their requirement shown in Figure 6 average time of retrieving web service from is also
(b). Meanwhile we adopt SameAs rule to record the increased compared to two other models because of the
evolution information between consumer in DRO and time-consuming on evolution operation between DRO and
buyer in DLO shown in table 1. It shows that DLO is not DLO in the same domain. Although the average retrieving
separated from DRO but evolved from DRO through the time is spent more in MMSA, the high precision can be
evolution information. If service provider use the concepts achieved because it can return more satisfied web services
in DLO to annotate web service, when querying the web to be chosen during retrieving time. See Figure 8.
service whose output is buyer, the returned candidate list of
web service will include the web service whose output is
2
consumer according to evolution information besides the
Average Time : s

1.8
web services whose output is businessman or person
according to traditional semantic model. That is to say, this 1.6
model of semantic annotation will return the more web 1.4
services to satisfy the users. 1.2
TMSA
1
IMSA
Table 1. Evolution information of DLO 0.8
0.6 MMSA
Evolution information 0.4
attribute of LO Value 0.2
Name Buyer 0
Number of web service
URI URI_ buyer 100 200 300
modelType OWL Figure 7. The relationship of retrieving time and the number of web
beforeChange Consumer service in TMSA, IMSA, MMSA
afterChange Buyer
transformationRule SameAs Rule
consistentChange True
 A MFI4OR-BASED APPROACH OF SEMANTIC ANNOTATION FOR WEB SERVICE 619

1 when annotation which will improve the efficiency of

precision 0.9 semantic annotation.
0.8
0.7 As future work, we plan to expand the Ontology
0.6 Evolution Rules which make the evolution information to
0.5
MMSA
be recorded comprehensively and clearly which will
achieve more satisfied annotation results to fulfill user’s
IMSA
0.4
TMSA
0.3 requirements.
0.2

ACKNOWLEDGMENTS
0.1
0
Number of web service
100 200 300 This work is supported by National Basic Research
Figure 8. The relationship of precision and the number of web service in Program (973) of China under grant No. 2007CB310801,
TMSA, IMSA, MMSA National High Technology Research and Development
Program (863) of China under grant No. 2006AA04Z156,
National Natural Science Foundation of China under grant
V. RELATED WORKS No. 60873083,60803025, 60970017 and 60903034, Natural
Science Foundation of Hubei Province for Distinguished
Our work presents an approach for adding semantics to
Young Scholars under grant No. 2008CDB351, Natural
Web services based on MFI4OR. In fact, much work has
Science foundation of Hubei Province under grant No.
been done on semantic annotation to Web services. We
2008ABA379, Research Fund for the Doctoral Program of
discuss two representatives of them. A classification
Higher Education of China under grant No. 20070486065
method is described in [11]. It uses machine learning
and 20090141120022, China Postdoctoral Science
algorithms to help users simplify the search for relevant
Foundation under grant No. 20090460107, and the
ontology. Unfortunately, the method considers just names
Research Project funded by Wuhan University under grant
of WSDL concepts, ignoring the structure of them.
No. 6082005 and 6082007, Scientific and Technological
Moreover, it uses vocabularies, but does not use ontology
research projects of Hubei Province Education Department
which is more descriptive and capture domain more
under Grant No.D20091012.
accurately for classification. MWSAF [12] uses a media
structure called SchemaGraph to facilitate the matching
between XML schema and ontology to find the relevant REFERENCES
ontology. However, since the matching results are not good [1] R. Chinnici, J. Moreau, A. Ryman, and S. Weerawarana, Eds.,
if WSDL files do not have good structure or the ontology “Web Services Description Language (WSDL) Version 2.0 Part 1:
becomes much comprehensive, it is also not suitable for our Core Language”, W3C Candidate Recommendation 27 March
2006; http://www.w3.org/TR/2006/CRwsdl2020060327
context. Moreover, [13] proposes a template-based markup
[2] K. Verma and A Sheth, “Semantically Annotating a Web Service,”
tool, which can improve annotation accuracy significantly. IEEE Internet Computing 11 (2),March–April 2007, pp. 83–85
But as used for semantic web content annotation, the [3] Jian Wang, Keqing He, Wei Liu, Rong Peng, Bing Li, “A Service-
template is over complex, and has a poor reusability Oriented Domain Modeling Framework for Networked Software”,
considering the various web contents. Dynamics of Continuous Discrete and Impulsive Systems - Series
B, Special Issue on Software Engineering and Complex Networks,
The common factor in these approaches is only relating December 2007, pp. 246-252
signatures of Web service to SINGLE domain ontology, but [4] ISO/IEC 19763-3:2007. Keqing He.
our approach can use DIFFERENT ontologies in the same http://www.iso.org/iso/iso_catalogue/catalogue_tc/catalogue_detail.
domain, which are not isolated but interrelated to each other htm?csnumber=38637
based on ontology evolution rules defined in MFI4OR, to [5] Chong Wang, Keqing He, Yangfan He. MFI4Onto: Towards
annotate web service, which will extend the scope of Ontology Registration on the Semantic Web [A]. In Proceedings of
6th International Conference on Computer and Information
semantic annotation in the Domain and greatly enhance the Technology (CIT 2006) [C], Seoul, 2006, pp.40-46
flexibility during the process of service publication and [6] Cheng Zeng, Keqing He, Zhitao Tao. Towards Improving web
discovery. service Registry and Repository Model Through Ontology-Based
Semantic Interoperability, GCC2008, 2008, pp.747-752
VI. CONCLUSION AND FUTURE WORK [7] Paolucci, M, Kawamura, T, Payne, T. R, and Sycara,K, Semantic
Matching of Web Services Capabilities, In Proceedings of the 1st
At present, few academic models of semantic International Semantic Web Conference (ISWC2002).
annotation extend beyond simple subsumption relationship [8] D. Martin, M. Burstein, J. Hobbs,OWL-S: Semantic Markup for
based input and output matching in single domain ontology. Web Services”, W3C Member Submission, Nov.2004;
In this paper, in IMSA we have considered all possible http://www.w3.org/Submission/2004/SUBM-OWL-S-20041122/.
matching relationships between the signatures of Web [9] H. Lausen, A. Polleres, and D. Roman, “Web Service Modeling
Service and the concept items in the domain ontology when Ontology (WSMO)”, W3C Member Submission,
2005;http://www.w3.org/Submission/WSMO/.
annotating. More important, in MMSA we are not limited
[10] R. Akkiraju, J. Farrell, J. Miller, M. Nagarajan, M. T. Schmidt, A.
to single domain ontology to annotate. Different ontologies Sheth and K.Verma, “Web Service Semantics - WSDL-S”, W3C
in the same domain based on MFI4OR can be adopted
620 ZENG ET AL.

Member Submission, 7 November 2005; [12] Patil A, Oundhakar S, Sheth A, Verma K. METEOR-S Web service
http://www.w3.org/Submission/WSDL-S/. annotation framework. In Proceedings of the 13th Int'l World Wide
[11] A. Hess and N. Kushmerick. Automatically attaching semantic WebConference, New York, ACM Press, 2004, pp.553-562.
metadata to Web services. Proceedings of the 2nd International [13] Brian Kettler, James Starz. A Template-Based Markup Tool for
Semantic Web Conference (ISWC 2003). Semantic Web Content. Proceedings of ISWC2005, LNCS 3729,
2005.
 Ethernet Node for Serial Communications
A.S. Popa1, M. Popa2, T. Slavici3, D. Darvasi4
1,2,3
“Politehnica” University from Timisoara, Romania
4
”Ioan Slavici” University from Timisoara, Romania

Abstract—Several protocols exist in serial communications, • MISO and MOSI: data lines; for a master device MISO is
such as SPI, USB, RS232, RS485, CAN etc. In order to obtain a data input and MOSI is a data output and for a slave device
flexible and fast communication, serial communication nodes the roles of the two lines are inversed; all the MOSI lines are
were developed. They are generally based on microcontrollers
and can be seen as protocol converters. Their capabilities were
connected together and all the MISO lines are connected
enriched by adding the remote command possibility, through an together; there is only one master which will command the
Ethernet interface. This paper presents a serial communication SCLK line and will generate data on its MOSI line to the
node based on the powerful 32 bit MPC5553 microcontroller. It MOSI inputs of the slaves; only the selected slave will
implements a bridge between two SPI, two RS485 and one generate data on its MISO output to the MISO input of the
Ethernet interface. It is used for collecting data from sensors and master;
sending them over the Internet to a testing station.
• /SS: the significance of this line differs; if the device is a
slave one, then /SS is an input for its validation and the master
I. INTRODUCTION must generate so many /SS lines how many slaves are
Several protocols exist in serial communications, such as therefore for a system with n slaves there are n + 3 lines; if the
I2C, SPI, USB, RS232, RS485, LIN, CAN etc. They are more device is a master one, /SS may be output or input, its
or less appropriate for certain applications. For example, the significance being software selected; if it is an output it will
CAN protocol is very used in automotive applications, due to validate a slave and if it is an input it can be used for error
its reliability, RS232 protocol is used in industrial and PC detection or to receive an external request from other device
based applications, due to its simplicity and low cost and so for the master position.
on. RS485 is a protocol for serial asynchronous communication
In order to obtain flexible and fast communication, serial between more devices directly connected to the same lines.
communication nodes were developed. They are generally The connection is single master many slaves type. The logical
based on microcontrollers and can be seen as protocol levels are represented by differential voltages which increase
converters. Their capabilities were enriched by adding the the number of lines but the protection to noises too leading to
remote command possibility, through an Ethernet interface. a maximum distance between the devices of 1200 m.
Such microcontroller based serial nodes exist, with
different performances. In reference [1] a node for wireless III. THE PROPOSED SOLUTION
sensor networks is presented. It is based on an Atmel ATmega
128L microcontroller with two UART, one SPI and one I2C A. The Hardware
interfaces and with input ports for sensors. Reference [2] Fig. 1. presents the bloc diagram of the proposed serial
describes a network node based on an Atmel ATmega 128L node. It is based on the powerful MPC5553 microcontroller,
microcontroller, which has an I2C, two UART and one [3], based on the PowerPC Book E architecture. It collects
Bluetooth interfaces. In reference [4] a powerful node is data through two internal SPI interfaces. Each such interface
presented. It is based on two microcontrollers, an 8 bit Atmel is connected to SPI lines and to RS485 lines. Both interfaces
ATmega 128L and a 16/32 bit Atmel AT91FR4081 ARM were connected in order to ensure compatibility with newer
THUMB and includes I2C, RS232, RS485, SPI and RF sensors which transfer data through the SPI bus, and with
interfaces. older sensors which transfer data through the RS485 bus. A
This paper presents a serial communication node based on differential driver was foreseen on the SPI lines because of
the powerful 32 bit MPC5553 microcontroller. It implements the high distance between the sensors and the microcontroller.
a bridge between two SPI, two RS485 and one Ethernet The RS485 lines are connected to the SPI lines of the
interface. It is used for collecting data from sensors and microcontroller through a RS485 – SPI adapter. The Ethernet
sending them over the Internet to a testing station. The next interface was implemented with a LAN8187i transceiver
section shortly describes the SPI and RS485 protocols, the connected to the internal MII interface. The MII interface
third section presents the new solution and the fourth section ensures the connection to the internal specific FEC (Fast
outlines the conclusions and future development directions. Ethernet Controller) peripheral. FEC corresponds to the MAC
layer and requires an external transceiver for being connected
II. THE SPI AND RS485 SERIAL PROTOCOLS to the physical interface. The CAN interface will be used to
connect the module to other blocks from the vehicle,
The SPI protocol describes the communication via a full
connected to the CAN bus and the SPI3 interface will be used
duplex, master – slave interface. The interface lines are:
for writing parameters into the EEPROM internal memory
• SCLK: common clock for all the devices: it is an output through an USB interface.
for a master device and an input for a slave device;

621
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_108, © Springer Science+Business Media B.V. 2010
622 POPA ET AL.

Fig.1. The bloc diagram of the serial node

The MPC5500 microcontroller has the e200z6 core which • data reception: the bytes received from the SPI interface
implements a version of the PowerPC Book E architecture. are buffered until the reception of the whole multiframe and
This is appropriate for high performance embedded systems • data transmission: the data bytes are sent over the SPI
applications. The e200z6 core integrates an arithmetic interface.
execution unit, a branch control unit, units for fetching
instructions and for reading/ writing data and a multiport The Ethernet driver has the following roles:
register able to support up to three read operations or two • driver initialization: establishment of a connection to the
write operations in the same clock period. Most of the Ethernet and establishment of the MAC address,
instructions are executed in a single clock period. E200z6 is a • Ethernet frames reception: frames are buffered,
scalar core with 64 general purpose registers. Floating point • Ethernet frames transmission: an Ethernet frame is sent
instructions are not supported by the hardware but are over the MII interface to the external transceiver which will
emulated, through the traps. It exists also a signal processing send it on the Ethernet and the maximum length of an
extension for supporting floating point operations in real time. Ethernet frame is 1518 bytes.
Other features of the e200z6 core are: powerful interrupts/
exceptions handling unit, memory management unit with full The communication protocol between the sensors, the node
associative 32 inputs TLB, 2 – way set – associative 8 Kbytes and the Ethernet achieves the following operations:
cache, delay and watchdog functions and low consume • the frames received from the Ethernet are unpacked and
modes. the command contained in the data field is sent on the SPI
The PWM signals are obtained by program. Internal bus;
counters are used and their content (C) is continuously • the Ethernet frames with a destination address different
compared with programmed values (PW). If C < PW, 0 logic than the MAC address of the serial node are ignored;
will be sent on the PWM output and if C • PW, 1 logic will be • the frame received from Ethernet is not verified; it is
sent on the PWM output. supposed as being correct;
B. The Software • the bytes collected from the sensors are packed in one or
several frames and are sent on the Ethernet;
The application was conceived in Visual C++ based on
MFC (Microsoft Foundation Class) library which permits easy • the destination address for the frames is broadcast type;
implementation of the Windows based programs. The • in case of errors at an Ethernet frame transmission, for
application has a graphical interface with dialog windows example because collisions are detected, a retransmission is
specific to the Microsoft Visual C++ environment. tried and after 10 such tries the transfer is aborted and error
The SPI driver has the following roles: situation is indicated through a led.
• initialization of the attributes of the transfer: speed, The software was conceived in two layers:
number of bits/frame, polarity and phase of the clock signal,
initial mode,
 ETHERNET NODE FOR SERIAL COMMUNICATIONS 623

Fig. 2. The hierarchy of packets and classes

• the Hardware Abstraction Layer: it consists of the classes The SPI_Driver class uses the
corresponding to the microcontroller interfaces; it contains the SPI_Driver_Interface interface for implementing the
following classes: SPI_Driver, ETH_Driver, callback mechanism. There are two methods for enabling,
DMA_Driver and TimerDriver; (b_AttachCallbackHandler(SPI_Driver_Interf
• the Application Layer: contains the application, more ace*CallbackHandler)), and disabling,
exactly the protocol between the sensors, the serial node and (v_DetachCallbackHandler()), this mechanism. It
the Ethernet; it contains the SPI_EthHandler and informs the client that the transmission was successfully or
Handler_ETHDriver classes. that a new SPI frame was received and it was sent to the
This structure is implemented by a hierarchy of packets and client. Thus, the client will be informed automatically about
classes. Fig. 2. presents this hierarchy and fig. 3. presents the the occurrence of an event without the need of continuously
UML diagram of the software architecture. interrogating the class SPI_Driver about that event. The
SPI_Driver_Interface specifies the methods
Fig. 4. presents the UML diagram of the SPI_Driver
v_RxCompleted(T_UWORD received_data) and
class. The singleton design pattern was used for implementing
this class. The reason is that the microcontroller has three SPI v_TxCompleted() for the two events above mentioned.
interfaces, therefore the class must be instantiated three times. Fig. 5. presents the UML diagram of the ETH_Driver
The spi_GetInstance static method returns the instance class. It implements the singleton pattern and defines the
of the SPI channel specified by the e_spi_channel interface corresponding to the FEC peripheral. It contains
parameter. methods for initializing the FEC peripheral, the LAN8187i
transceiver, for transmitting and receiving Ethernet frames
The SPI_Driver class contains methods for initializing and for other configurations. The return of an instance of the
the internal DSPI peripheral, for transmitting and receiving class is done through the eth_GetInstance()method.
SPI frames and for configuration.
The method ub_Init() configures the pins
The method v_Init() initializes the SPI internal
corresponding to the MII interface and initializes the internal
peripheral. The terminals and the attributes of the transfers are
registers through which the origins of the memory zones
configured. The method T_VOID v_Send(T_UWORD allotted to the circular lists of the transmission and reception
transmitData, T_BOOL b_EOQFlag) has the role of descriptors and the dimension of the reception buffers are
sending commands and data on the SPI interface. specified.
624 POPA ET AL.

Fig. 3. The UML diagram of the software architecture

 ETHERNET NODE FOR SERIAL COMMUNICATIONS 625

The method ub_Open(T_UWORD uwMode, const The parameters of the connection can be by established
T_UBYTE* pubMacAdress, T_UWORD uwPhyType) by configuration or through the auto – negotiation process.
achieves the operations necessary to the FEC peripheral to be This is a mechanism for exchanging information between two
ready for transmitting and receiving. These operations are: interconnected nodes and in which the operation mode with
initialization of the descriptor lists, specification of the MAC higher performance is selected among the operation modes
address, configuration of the control registers for transmitting supported by the two nodes. The auto – negotiation process
and receiving and initialization of the PHY external can be started automatically when the connection beween the
transceiver. For initializing the descriptor lists, two private two nodes is lost, by setting a specific bit or can be started by
methods were foreseen, v_RxBDInit() and software. The configuration information sent by the PHY
v_TxBDInit(), called from the ub_Open() method. transceiver to the partner during the auto – negotation process,
are saved in specific registers. At the end, one register
contains the partner’s parameters and an other one contains
the connection’s parameters. The results of the auto –
negotiation operation can be used by the FEC peripheral too,
for example it can set the communication type as being half –
duplex or full – duplex.

Fig. 4. The UML diagram of the SPI_Driver class

The initialization of the PHY transceiver implies the

configuration of its internal registers. This is done through
reading and writing management frames. The operations are
implemented in two dedicated methods:
ub_MiiWrite(T_UBYTE ubPhyAddr, T_UBYTE
ubRegAddr, T_UWORD uwData) and
ub_MiiRead(T_UBYTE ubPhyAddr, T_UBYTE
ubRegAddr, T_UWORD* puwRetVal). Their
parameters specify the address of the transceiver PHY, the
address of the register and the value which will be written in,
respectively a pointer to the memory location where the read Fig. 5. The UML diagram of the ETH_Driver class
value will be stored. The PHY address is estableshid
according to the hardware configuration of certain pins of the In the transmission, a frame will be divided in multiple
circuit. buffers. For each occupied buffer, a flag will be set. The last
buffer containing the frame will have a specific flag activated.
626 POPA ET AL.

There is no a predetermined relation between a frame and an IV. CONCLUSIONS

Ethernet header, as a consequence one can have a descriptor The paper has presented a serial node based on the
corresponding to the data buffer and an other one MPC5553 microcontroller. It is used for collecting data from
corresponding to the buffer containing the Ethernet headers. sensors, through two SPI and two RS485 interfaces and
The two transmission flags, one for the data and the other for sending them over the Internet.
the header, must be set in a specific order: first the one
corresponding to the data and second the one corresponding Future development directions are:
to the header. Otherwise, an error will be indicated. There are • The extension of the performances of the existing
frames with the same or with different Ethernet headers and interfaces, by increasing the communication speed and by
different methods will be used for sending them. The allowing to connect multiple sensors on the same interface;
transmission is done through DMA channels. Aftre reading a • Adding new interfaces; the focused interfaces are CAN,
buffer, the flag indicating its state will be reset, thus the USB and for analogue inputs/outputs.
software will know which buffers are free for an other
incoming frame. REFERENCES
At the reception, the length of the frames is not previously [1] H. Abrach, J. Carlson, H. Dai, J. Rose, A. Sheth, A, B. Shuker. and R.
known. It is necessary to set a variable for defining the length Han, “System Support for MultimodAINetworks of In-situ Sensors”,
of all the reception buffers. A frame may occupy several Technical report CU – CS – 950 – 03, April 2003, Available:
buffers and the state of the buffers will be indicated by http://iguanaworks.net/~shucker/papers/CU-CS-950-03.pdf
specific flags. The descriptor corresponding to the last [2] J. Beutel, F. Mattern, K. Romer, F. Siegermund and L. Thiele,
occupied buffer will indicate the length of the frame. “Prototyping Wireless Sensor Network Applications with BT nodes”,
Lecture Notes in Computer Science, Vol. 2920, 2004, Springer, pp. 323
The processing of the received Ethernet frames is achieved – 338
either through the callback mechanism or by repetidevely [3] R. Dees and R. Marshall, MPC5500 family overview, Freescale
interogating the ETH_Driver class. The callback Semiconductor Engineering Bulletin, No. 11, 2006, Available:
mechanism is similar to that of the SPI driver and supposes http://www.freescale.com/files/32bit/doc/eng_bulletin/EB659.pdf
the implementation of the interface [4] A. Savvides and M. B. Srivastara, “Distributed Computation Platform
ETH_Driver_Interface, by the clients of the for Wireless Embedded Sensing”, in Proceedings of IEEE International
Conference on Computer Design, Freiburg, Germany, September 2002,
ETH_Driver class, and their registration through the pp. 220 – 225
method [5] *** SPI description – Available :
b_AttachCallbackHandler(ETH_Driver_Interf http://www.rpi.edu/dept/ecse/mpsd/SPI.pdf
ace* CallbackHandler). [6] *** RS485 description – Available : http://www.rs485.com
 EvalTool for Evaluate the Partitioning
Scheme of Distributed Databases
Adrian Runceanu Marian Popescu
University Constantin Brâncu i Târgu-Jiu, Romania
[email protected] [email protected]

Abstract. In this paper we present one tool named accessing during the transaction process. If the
EvalTool for evaluate the partitioning scheme of fragments are closer as possible to the needs of the
distributed database. This tool could be use for transactions set, then the transactions processing cost
calculate the best partition scheme both for horizontal can be reduced.
and vertical partitioning, and then alocate the
fragments in the right places over the network. We use
For the distributed databases design, the
an algorithm studied before, which could calculate transaction cost is reduced by increasing the local
some cost for access data to the local site and the other transactions (on a site) and at the same time by
cost for access data from the remote site. The minimum reducing the amount of data views which are not
value of this costs give the best fragmentation scheme local. The aim of vertical partitioning technique (and
for each relation. in general, partitioning techniques) is to find a
1. ABOUT FRAGMENTATION OF DISTRIBUTED
partitioning scheme to meet the objective outlined
DATABASES above.
Note that the partitioning problem can be
The distributed databases design is an tackled on different levels of detail considering some
optimization process that requires to obtain solutions additional information.
to several interpenetrating problems, namely data In paper [1], we have to consider only the
fragmentation, data allocation and local optimization. information about transactions as input data to
Each problem can be solved through different manage the partitioning problem effectively. In fact,
approaches, and therefore the design of distributed the global optimization problem (which includes a
databases becomes a very difficult task. Usually the large number of parameters and a metric complex) is
design process is heuristical defined. divided into several smaller optimization problems in
The distributed databases design derived order to reduce the search space and to reduce the
from the non-distributed databases design only in complexity of each problem separately.
terms of distribution. The design involves data In speciality literature are proposed several
acquisition, database partitioning, the allocation and vertical partitioning algorithms, so it can be
replication of partitions and local optimization. The measured the affinity between pairs of attributes and
database partitioning can be performed in several try to group attributes together under the affinity
different ways: vertical, horizontal and hybrid (also between, using the algorithm BEA (bond energy
called mixed) partitioning. This aspect of a database algorithm). In an article we use an heuristical cost
design will be highlighted in this thesis and namely estimator for design a file to obtain a partitioning
the developing of an algorithm for checking some scheme “bottom-to-top”. In another article extends
various algorithms proposed in literature for the approach proposed by BEA and algorithm has
distributed databases partitioning (fragmentation). two phases for vertical partitioning. The partitioning
Basically, it will be considered the problem algorithms mentioned above use several heuristical
of vertical data partitioning (fragmentation), also methods for a relationship fragmentation.
known as the attributes partitioning. This technique is As input date for these algorithms it is used
used for the databases design to improve the the attributes matrix (AUM). This matrix has the
performance of transactions. In vertical partitioning, attributes as columns, the transactions as lines and
the attributes of a relation R are together in groups the transactions frequency access are values in the
that do not overlap and the relationship R is designed matrix. The most previous algorithms used for the
on relations fragments in accordance with these data fragmentation use of affinity matrix of attributes
attributes groups. (AAM) which is derived from AUM. AAM is a
In the distributed databases, these fragments matrix in which for each pair of attributes it is stored
are allocated on different sites. Here comes the aim the total frequency access of the transactions that
of vertical partitioning to create vertical fragments of access the pair of attributes. The results of different
a relationship to minimize the cost of the data algorithms are often different, even if it is the same

627
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_109, © Springer Science+Business Media B.V. 2010
628 RUNCEANU AND POPESCU

affinity matrix of the attributes, so it indicates that allocate the calculated fragments. It is also possible
the objective functions used are different. Most to load an external fragmentation and use it for the
vertical partitioning algorithms do not have an allocation, this might be useful as horizontal
objective basis for assessing the correctness of fragmentation.
partitions that are obtained through the application of For both the fragmentation and allocation,
those algorithms. Also, there is a common criterion using the implemented algorithms, the tool reports all
or objective function to evaluate the results of these the possible solutions and the one considered the
algorithms for vertical partitioning. best.
The fragmentation and allocation algorithms
Bellatreche et al. [3] presents an horizontal are mainly based on the book [10]. We modified the
fragmentation method and an analytical cost model to cost analysis for the allocation, the model used is
evaluate the query execution time in horizontally described in the next section.
fragmented databases. The fragmentation schema The algorithms implemented are:
with the best performance according to the cost - Bound Energy Algorithm
model is achieved through a hill-climbing algorithm, - VF algorithm
by selecting a subset of classes for primary horizontal - Best-Fit heuristic
fragmentation. - Cost Analysis (based on the model
The work by Ezeife and Barker [6] presents described in the next section)
a set of algorithms for the horizontal fragmentation The tool works with text files as input, at
of all classes in a database schema. It takes least it needs:
relationships between classes into account to propose - one file for general information (sites names,
primary and derived horizontal fragmentation. applications frequencies, network values, and other
However, this approach works at the instance level, information)
where the class instances already exist in the - one file for the information about a relationship
database to proceed with the fragmentation process. (attributes names, useMatrix, and other information)
It also assumes a storage structure for the objects Once the 2 text files are loaded it is possible
hierarchy in the database class in which an instance to vertical fragment the relationship. The tool works
of a subclass physically contains a pointer to the with the following steps:
instance of its super-class that is logically part of it. - It is calculated the affinity matrix
This assumption leads to considering inheritance - Clustering affinity matrix is calculated by the Bond
links in the horizontal fragmentation process. Energy Algorithm (in the relationship text file is also
Ceri et al. [4] proposes a methodology for possible to specify the column insertion order)
the horizontal fragmentation of all classes in a - all the solutions for splitting CA matrix into two
database schema. The choice between primary and fragments are evaluated with the VF algorithm
derived horizontal fragmentation on each class - if the best solution is to fragment the relationship,
considers its relationships, which are defined by VF algorithm is iterated for each of the fragments
analyzing only the method calls between classes in calculated
the schema. The work does not present an algorithm The tool reports the CA matrix and all the
to support the methodology. steps of both the Bound Energy Algorithm and VF
Navathe et al. [8, 9] proposes a complete algorithm. Is also possible to store this information in
fragmentation methodology for OODBMS, which is a text file.
divided in three phases. First, there is an Analysis The allocation uses:
Phase to assist distribution designers in defining the - the best fit heuristic
most adequate fragmentation algorithm (horizontal, - a cost analysis (the model used is described in the
vertical, or both) to be applied in each class of the next section)
database schema. The Analysis Phase also considers EvalTool uses 2 policies of allocation:
the case in which no fragmentation of a class is the - “one for each site” (the default one)
best alternative. Second, they present an algorithm to - “one for all sites” (can be activated selecting the
perform Vertical Fragmentation in a class. Finally, relative option button)
the authors present an algorithm to perform An example will help to understand:
Horizontal Fragmentation on the whole class or on a consider that you are designing a distributed database
vertical fragment of a class, which may result in for your university.
mixed fragmentation. The university has one central site and five
peripheral sites managing all the didactics.
2. MAIN CARACTERISTICS Each peripheral site is independent from the
others: just use the information in the database about
In few words EvalTool considers one its students, its exams, its teachers, and so one, it is
relationship for time, once loaded is possible to easy that a few of the relationships of the database
fragment it (in the first time only vertically) and
 EVALTOOL FOR EVALUATE THE PARTITIONING SCHEME 629

will be fragmented in one fragment for each QPCi = PCi + TCi (4)
peripheral site (e.g. the student relationship) and, Processing component consists of three cost
because of the “independence” of the sites, the only factors: access cost (CA), the integrity constraint (IC)
solutions for allocation with sense will be to allocate and control competitor (CC):
the fragment on its site or on the central one: none of PCi = ACi + ICi + CCi. (5)
the other sites will use it; is easy that a lot of Detailed specification of these three cost
relationship will be fragmented in this way, think just factors depend on the algorithms used for these tasks.
to the students, the exams and the teachers For example, AC has the form:
relationships, each one fragmented in 5 fragments:
not considering others fragmentation 15 fragments ¦STC ≤storagecapacityat
∀Fj ∈F
jk
site Sk, ∀S∈S. (6)
are created.
Using the “one for each site” (a more correct The first two terms from the previous
name would be “one fragment for each peripheral formula calculates the number of accesses to the
site”) is possible to reduce the number of the query qi the fragment Fj and UR + RR value
fragments that the user has to manage: the input for provides the total number of accesses that executes
the allocation won't be 3 text files with 5 fragments work | retrieve and update them. Local cost of
each one, but 3 text files with 1 fragment each one. processing them is assumed to be identical, given
But there's also the possibility that a LPCk multiplying the total cost of access to station
relationship is not fragmented in one fragment for Sk and multiplying xjk show that will select only
each peripheral site (because is not possible to relate those values of the cost for stations where parts are
a tuple to a single peripheral site or simply because stored.
there's no need). In this case there aren't no-sense Access cost function requires that each
solutions for the allocation of the relationship (or the application can be divided into subqueries that run on
fragments obtained from other fragmentations), it can each fragment stored on a station, the results are
be allocating in all the sites central or peripherals; in returned to the station from which the request was
this case must be used the “one for all sites” policy sent. This is a simplistic view that ignores the
(one fragment for all sites). This policy can also be complexity of database processing. For example, it
used if the user in the example before wants to create was revealed the cost of execution join unlor or cost
3 text file with 5 fragments each one. competitiveness and integrity.
EvalTool was developed as part of a study The cost of transmission can be made
for a grant research. Maybe the tool “fits too much” analogous to the cost of access. The queries that are
our problem so that there could be problems for using updating is required to inform all stations where the
it in a more general one, but we considered useful to replicas, while the queries that are restored, it is
offer our solution to the community. sufficient to access only one of the descendants, in
addition, after the application update, no data
3. COST MODEL FOR ALLOCATION transmission back to state of origin (except a
confirmation message), while the queries that are
We present an allocation model that restored may result in transmission of data.
minimizes the total cost of storage and processing, Component update transmission function has the
trying to comply with certain restrictions imposed on form:
response time.
We consider a decision xkj variable, defined as: TCU
i
= ¦ ¦ uij *xjk *go(i), k + ¦ ¦ uij *xjk *gk, o(i),
∀Sk∈S ∀Fj ∈F ∀Sk ∈S ∀Fj ∈F
(7)
­1, if fragmentFk is placed on site Sj
xkj = ® (1) where the first term is to send a message to update
¯ 0, in the other cases
the initial station 0 (0 demand all copies of fragments
Total cost function has two components: that must be updated and the second term is to
processing applications and storage. This can be confirm transmission). Component retrieve can be
expressed as: specified as follows:

TOC = ¦ QPC i + ¦ ¦ STC jk , (2) seli(Fj )

where
QPCi is
∀ q i ∈Q ∀ S k ∈S ∀ F j ∈ F
the cost of
TCRi =
∀Fj∈F
¦
min (r * xjk * go(i),k + rij * xjk *
Sk∈S ij fsize
* gk,o(i)).
processing the application qi and STCjk fragment Fj is (8)
the cost of storage per stationSk.
We can define the cost of storage with the formula: The first term in TCR is the cost of
STCjk = USCk*size(Fj)*xjk (3) forwarding the application to recover all stations
The cost of processing applications (QPC) contain copies of the fragments to be accessed, and
can be separated into two components: the actual the second term is the cost of transmission results
processing cost (PC) and the cost of transmission from these stations to the original station.
(TQ).
630 RUNCEANU AND POPESCU

The cost of transmission for qt query can be matrix. In [5], Chakaravarthy et al. computes the
described as: processing costs of distributed fragments using an
TCi = TCUi + TCRi (9) application/attribute matrix that counts the number of
Constraint functions can be analyzed times applications access attributes at distributed
similarly. For example, response time constraint can sites.
be described thus: execution time of the call qi ≤ This paper computes the partition evaluator
maximum response time of applications qi, ∀qi ∈ Q , value (PE) with this formula:
and storage constraints may be described by the PE = EL2 + ER2 (11)
relation: where
M T ª § S ·º
ACi = ¦ ¦(u * UR +r * RR )* x * LPC , (10)
∀Sk∈S ∀Fj∈F
ij ij ij ij jk k E L2 = ¦¦ «qt2 ∗ S it ¨¨1 − it ¸¸» (12)
«
i =1 t =1 ¬ © ni ¹»¼
The heuristic methods must be sought that n - Total number of attributes in a relation that is
produce sub-optimal solutions and must test the being partitioned.
closeness of the results of heuristic algorithm for T - Total number of transactions that are under
optimal allocation. Unfortunately there is insufficient consideration.
qt - frequency of transaction t for t = 1,2, …, T.
information to determine the “near” optimal
solution. M - Total number of fragments of a partition.
ni - number of attributes in fragment i.
Sit - set of relevant attributes in fragment i that the
4. PARTITION EVALUATION FUNCTION transaction t
accesses; it is empty if t does not need fragment i.
The partition evaluation function is an |Sit| - number of relevant attributes in fragment i that
objective function used to evaluate the "goodness" of the transaction t accesses.
a particular partitioning scheme. It can be used to
measure the performance of a fragmentation system ª º
( ) ∗ nR
T
E R2 = ¦ min iM=1 ¦ «¦ f tpkk
2 itk
whenever a major change in access pattern
remote » (13)
information or global conceptual schema occurs. t =1 k ≠i ¬ Pk itk ¼
However, the defining of the suitable function for the
vertical fragments and the building of EvalTool for
ni – number of attributes in fragment i.
vertical fragmentation are the main contributions of remote
this work. [1] nitk - Total number of attributes that are in
fragment k accessed remotely with respect to
5. THE PARTITIONING EVALUATION FUNCTION fragment i by transaction t.
Attributes of relations are the members of f tji - Frequency of transaction t accessing attribute j
relational vertical fragments. On the other hand, the in fragment i
methods and attributes of object classes are the Ritk - Set of relevant attributes in fragment k
member of object vertical fragments. The instances
are members of the horizontal fragments object. The accessed remotely with respect to fragment i by
set of instance objects in fragments created by transaction t; these are attributes not in fragment i but
different fragmentation schemes may be different. It needed by t
is necessary to develop a modified partition evaluator Ritk - Number of relevant attributes in fragment k
to deal with object horizontal fragmentation schemes. accessed remotely with respect to fragment i by
This partition evaluator counts the number of local transaction t
irrelevant accesses made to fragments by queries as The algorithm implements the objective
well as the numbers of remote accesses made to function defined above. For measuring the
fragments for each set of horizontal fragments performance of a partition scheme, algorithm needs
created with a scheme. two inputs. One is every horizontal fragment from
Therefore, in order to measure the system the partition scheme; the other is the application
performance each fragment is assumed placed at one object instance sets. Then produces the PE value of
of the distributed sites, while the PE function is used this partition scheme. Each PE value corresponds to
to measure the partition evaluation value (the PE the total penalty cost incurred by each scheme
value) for the system. The higher the PE value of the through both local irrelevant access costs and remote
system, the worse the system performance. relevant access costs. A lower PE value means less
Computing the PE value of horizontal fragments performance penalty and thus, a better performance.
requires an input matrix that shows for each The algorithm is below. This algorithm accepts the
application the number of times an instance object is application object instance sets of al1 applications,
accessed. This is called the application instance the set of horizontal fragments allocated to
 EVALTOOL FOR EVALUATE THE PARTITIONING SCHEME 631

distributed sites and application access frequencies at fragment and that fragment haven’t one single
distributed sites. It then computes the local irrelevant attribute accessed by that transaction, then
access cost of each class as the sum of local transaction not be run on that fragment. We consider
irrelevant access cost of al1 applications for each 3 cases of matrix of attributes.
fragment of the class. Similarly, it computes the For the first test we used a matrix of
remote relevant access cost as the sum of remote attributes use with ten attributes accesed by eight
access cost made by al1 applications to fragments of transactions.
this class at remote sites. The sum of these two costs Tranzactions \ Attributes 1 2 3 4 5 6 7 8 9 10
makes up the PE value of the class. T1 25 0 0 0 25 0 25 0 0 0
We implemented the algorithm based on the T2 0 50 50 0 0 0 0 50 50 0
formula from equation (11) calculates the value of T3 0 0 0 25 0 25 0 0 0 25
PE, for a given fragmentation scheme so we used an T 4 0 35 0 0 0 0 35 35 0 0
T5 25 25 25 0 25 0 25 25 25 0
entrance date: the matrix used for attributes - A; the
T6 25 0 0 0 25 0 0 0 0 0
lots of fragments on which it calculated the value of
T7 0 0 25 0 0 0 0 0 25 0
PE, the relation -R.
T8 0 0 15 15 0 15 0 0 15 15
Algorithm
Input: A = attribute usage matrix;
R = Relation ; F = fragments set We present in Figure 1 values for each
Output: E2L: irrelevant local attribute cost; 2 2
number of fragments and values for EM , E R and
E2R: relevant remote attribute cost;
EP : partition evaluator value EP .
Begin Total number of fragments evaluates was
2 115975. Optimal value (minimum) is obtained for 3
E L=0
for i from 1 to M do fragments – fragment I (1,5,7), fragment II (2,3,8,9)
ei=0 and fragment III (4,6,10).
begin The program to generate all the
for j from 1 to T do combinations of ten attributes accessed by eight
ei=ei+qt2*|Sit| * (1 - |Sit| / ni ) transactions offers three solutions (for five
end_for fragments) and two solutions (for eight fragments),
E2M = E2M + ei having the same value for EP. However, the project
end_for of distributed database can choose which scheme of
E2R=0 partition wishes to use it.
for t from 1 to t do
begin
Number
of Partition scheme EM2 E R2 EP
fragments values values values
minim =maxint
for i from 1 to M do 1 (1,2,3,4,5,6) 24895 0 24895
for k from 1 to M do 2 (1,4) (2,3,5,6) 7565 55 7620
3 (1) (2,3,5,6) (4) 7565 276 7841
begin 4 (1) (2) (3,5,6) (4) 5063 11336 16399
if k ≠ i then 5 (1) (2) (3,6) (5) (4) 0 22492 22492
begin 6 (1) (2) (3) (4) (5) (6) 0 40913 40913
if exist attribute in matrix A who is from k Fig. 1 Results for test
fragment then
E2R= E2R+ (f tk)2 * | Rit | * | Rit | / 7. CONCLUSIONS
remote
n itk)
end_if In this paper is presented a general approach
2
if E R min < minim then of the fragmentation issue of the dates from a
minim = E2R min distributed database. Using this tool for obtained the
end_if best partitioning scheme with implementation of
end_for several clasic algorithms is one solution in designing
E2R = E2R + E2R min phase of a distributed database.
end_for
REFERENCES
end_for
2 2
PE = E L + E R
End. [1] Runceanu A. Fragmentation in distributed
databases, (2008), International Joint
6. EXPERIMENTAL RESULTS Conferences on Computer, Information, and
Systems Sciences, and Engineering (CISSE
For the execution of one transaction, we 2007) Conference, Conference Proceedings
know that if a transaction could be run at one book, December 3-12, 2007, University of
632 RUNCEANU AND POPESCU

Brigeport, USA, publish in Innovations and system” International Journal of Distributed

Advanced Techniques in Systems, Computing and Parallel Databases, vol. 3, no. 3,
Sciences and Software Engineering, pp. 247–272, 1995.
Springer Science, ISBN 978-1-4020-8734-9 [7] W. Kim, Introduction to Object-Oriented
(Print) 978-1-4020-8735-6 (Online), DOI Databses, MIT Press, 1990.
10.1007/978-1-4020-8735-6_12. [8] S.B. Navathe, K. Karlapalem, and M. Ra,
[2] J. Banerjee, W. Kim, and K.C. Kim, “Amixed partitioning methodology for
“Queries in object oriented databases,” in distributed database design,” Journal of
Proceedings of the IEEE Data Engineering Computer and Software Engineering, vol. 3,
Conference (ICDE’88), February 1988, pp. no. 4, pp. 395–426, 1995.
31–38. [9] S.B. Navathe and M. Ra, “Vertical
[3] L. Bellatreche, K. Karlapalem, and A. partitioning for database design: A graphical
Simonet, “Horizontal class partitioning in algorithm,” in Proceeding of the International
object-oriented databases,” in proceedings Conference on Management of Data, ACM-
of the 8th International Conference on SIGMOD’89, 1989, pp. 440–450.
Database and Expert Systems Applications [10] Tamer O. and Valduriez P. (1999)
(DEXA’97), September 1997, pp. 58–67, Principles of Distributed Database Systems.
Lecture Notes in Computer Science 1308. Prentice Hall Englewood Cliffs, Second
[4] S. Ceri, M. Negri, and G. Pelagatti, Edition, New Jersey 07362.
“Horizontal data partitioning in database [11] Grofu F., Popescu L. and Cercel C. (2008)
design,” in Proceedings of the ACM Acquisition System for Monitoring
SIGMOD International Conference on Vibrations, International Joint Conferences
Management of Data, 1982, pp. 128–136. on Computer, Information, and Systems
[5] Chakravarthy S., Muthuraj R., Varadarajan Sciences, and Engineering (CISSE 2007)
R., and Navathe S. (1994) An objective Conference, Conference Proceedings book,
function for vertically partitioning relations December 3-12, 2007, University of
in distributed databases and its analysis. In Brigeport, USA, publish in Novel Algorithms
Distributed and parallel databases, pages and Techniques In Telecommunications,
183-207, Kluwer Academic Publishers. Automation and Industrial Electronics, ISBN
[6] C.I. Ezeife and K. Barker, “A 978-1-4020-8736-3 (Print) 978-1-4020-8737-
comprehensive approach to horizontal class 0 (Online), pg. 89-92, DOI 10.1007/978-1-
fragmentation in distributed object based 4020-8737-0_16.
 Automatic Creation of Daily News Digests from
Internet News Sites
David B. Bracewell
General Electric Global Research
Niskayuna, NY 12309
[email protected]

Abstract—With the advent of 24-hour news channels and Internet However, there seems to be little research done on
news sites, the ease of finding news sources has become easier automatically creating digests.
than ever. However, with the increased number of sources
covering a greater variety of topics it is becoming increasingly This paper presents techniques for gathering and classifying
difficult for users to weed through the important and not so news articles and for creating simple digests. The system is
important news of the day. This paper presents a system that based on a keyword extraction algorithm that is applicable to
automatically creates a daily digest of the important news topics multiple languages. Currently, we are focused on creating an
of the day. News articles are mined from Internet news sites and English and Japanese version of the system.
the articles are stored for later analysis. At the end of the day, or
specified period, the articles are analyzed to determine the This paper will continue as follows. In section 2 a brief
important news topics and to create a listing of the top stories of overview of the proposed system will be given. Then, in
the day. The digest shows the top stories with their title, section 3 the news gathering and classification module will be
keywords, link and summary information. Through human discussed. In section 4 an overview of creating daily digests
judged evaluation it is shown that the created digests satisfy from the classified news articles is given. Then, in section 5 a
users. look at the current implementation details will be given.
Section 6 shows evaluation results. Finally, in section 7
Keywords-component; Digest Creation, Online News, concluding remarks will be made and future work discussed.
Personalization, Synopsis Creation
II. SYSTEM OVERVIEW
I. INTRODUCTION The system is made up of two modules: the news
In the past 10 years the number of online news sources has monitoring and article processing module and the digest
grown tremendously. With the increase in sources have also creation module. The news monitoring and article processing
come increases in the number of news articles, topics covered, module monitors a list of user defined RSS feeds and news site
etc. that are available. Keeping adequately informed, in an era pages for new news and the downloads, extracts and processes
of information overload, requires constant attention to Internet the articles. Processing the articles includes keyword
news sites, the sites' RSS feeds or backtracking at the end of extraction, category classification and summary generation.
the day to determine the important events. The digest creation module takes the downloaded and
processed articles at the end of the day and creates a digest of
Online news has become a hot topic in the information
the news by determining which topics and articles were the
retrieval world. From workshops on topic detection and
most important for the day. An overview can be seen in
tracking to various research projects, the information retrieval
of online news is of growing importance. There has been work figure 1.
done to automatically create RSS feeds of news archives [4] to
make past news more easily available. There have also been
efforts to combine news, such as the Pan-European news
distribution network [8]. However, keeping track of important
and popular stories and topics is still a problem.
In order to help the user find popular stories that may be of
interest or may be important, some online news sites offer most
recommended and most viewed sections. This allows for users
to quickly scan articles that others have found interesting. In
addition, news sites are making good use of RSS feeds that
allow users to view only specific content. Another way of Figure 1. System Overview
helping users is to offer daily digests. A digest summarizes the
most important topics and stories of the day. Some sites are The system is designed to be automatic and does not
offering daily digests, which are typically manually created. require any intervention by end users. The only requirement
from the end user is for them to set the time at which the news
monitoring and article processing module runs. For example,
This research was originally done while the author was at The University
of Tokushima.

633
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2_110, © Springer Science+Business Media B.V. 2010
634 BRACEWELL

the user could set the module to run from 6am to 5pm not been processed then the link is used to download the page
everyday. At completion, the digest creation module would containing the article. The article is then extracted from the
take over and create a digest for the time span given by the HTML web page using a rule-based extraction process.
user.
The digest creation module outputs in HTML, which could B. Article Extraction
be easily used as a digest, web page served on the Internet. The After examining many news sites we found a certain pattern
main output of the system is the digest, which is made up of the that described the prototypical article. The pattern is made up
top articles of the day with their category or genre (sports, of a title, then the article body and the footer. Certain sites,
business, politics, etc.), top keywords and a summary. however, utilize handcrafted rules in the form of regular
Additional output includes ranked lists of keywords and expressions that allow for very precise extraction. However,
categories for the day. this poses a possible bottleneck for the addition of new news
sites.
III. NEWS MONITORING AND ARTICLE PROCESSING The first step in article extraction is to find the title in the
News monitoring and article processing handles body of the HTML. This is done using the title extracted from
monitoring, downloading and processing news articles during the RSS feed or web page. Since these titles sometimes have
an assigned time span. An overview of the module can be seen extraneous information, such as web site name or provider,
in figure 2. The module monitors a list of RSS feeds and news they cannot be fully matched to a line in the HTML. Instead,
web sites for new news. After articles are downloaded and each line is examined to see if it is a substring of the title. If it
extracted they are processed by extracting keywords, assigning is, then that line is marked as the article's title.
categories and generating summaries. The next step is to find the article footer. This is done by
examining each line that comes after the article's title to see if it
contains one of the defined footer elements. Some of the
currently used footer patterns are Copyright, ©, (C), Email this,
Related Stories, etc. If no footer pattern is found then the
bottom of the HTML body is used.
After the footer is found the article can be extracted. The
article body is the text that falls between the title and the footer.
The article is then cleansed. The cleansing process removes
multiple white space and unimportant multiple punctuation
marks, such as --, **, ..., etc.

C. Article Processing
Article processing consists of keyword extraction, category
classification and summarization. Keyword extraction extracts
the most important words and phrases for describing the article
and its content. Category classification assigns to each article
one or more predefined news categories. Summarization gives
the important points of the article in a few sentences.
1) Keyword Extraction
For keyword extraction, the method proposed by Bracewell
et al. [1] was used. It is capable of extracting keywords from a
single document without the need for training data or corpus
statistics. A brief description will be given in the next couple of
Figure 2. Gathering and Classification Overview paragraphs.
The algorithm uses linguistic information in the form of
A. News Monitoring noun phrases to extract keywords. Phrases can make better
The news monitoring part of the module watches a set of keywords than single words as they can keep concepts
RSS feeds and news site web pages for new news articles. The portrayed in compound words together. For example ``New
sites are processed one at a time in sequential order. Depending York'' would be a better keyword than having ``New'' and
on the site either an RSS feed or HTML page is parsed and the ``York'' separately.
links and article titles are extracted. The algorithm is made up of three modules: morphological
After the articles are collected they are processed one at a analysis, NP extraction and scoring, and NP clustering and
time. First, the article's link and title are checked for in the scoring. Morphological analysis takes care of word
article database, which stores the article and its processing segmentation, part-of-speech tagging, and word stemming. For
results. If the link or the title is in the database then the article English, Porter's stemmer [7] and Brill's tagger [3] were used.
is treated as already processed and is ignored. If the article has For Japanese, Chasen [6] was used.
 AUTOMATIC CREATION OF DAILY NEWS DIGESTS FROM INTERNET NEWS SITES 635

NP extraction and scoring extracts simple noun phrases and IV. DIGEST CREATION
gives them a score based on their frequency in the article and After the current day's news has been downloaded and
the frequency of the words that make it up in the article, see [1] processed, a digest can be created that will highlight the
for details on the score. NP clustering and scoring brings important news of the day. The digest creation module is made
together noun phrases that share some commonality and is an up of two parts: article set determination and synopsis creation.
attempt to prevent redundancy in the final set of keywords. For Article set determination uses clustering in an attempt to
example, “bank bailout” and “bailout” would be clustered determine the broad topics covered in the day's articles.
together. The clusters are scored based on the scores of the Synopsis creation takes the top topics and determines the
noun phrases in them. From each of the clusters a representative article from them.
representative is chosen and becomes a keyword for the article.
2) Category Classification A. Article Set Determination
For category classification, we used the method proposed in The first step in creating the digest is to determine which
[2], which is based on the previously mentioned keyword set of articles were the most important of the day. This is done
extraction algorithm. It is a keyword based classification by clustering the articles and ranking the resulting clusters. The
method that only requires positive examples. It has the added overview of this can be seen in figure 3.
ability that it is capable of updating the training information in
an online manner as well as easily adding new categories
without a complete retraining. It also had an F-Measure of 93%
on the test set in that paper. A brief description will be given in
the next couple of paragraphs.
Figure 3. Article Set Determination Overview
The algorithm is made up of N independent classifiers,
where N is the number of categories. Training involves First, a feature vector is created for each article. The feature
building a model from a set of articles that are in the desired vector is simply a vector space model representation of the
category. A category model is made up of a category name, article using the keywords that were extracted in the new
total number of documents counter and a keyword list. The monitoring and article processing module. The next step
keywords list contains the keywords and their raw document clusters the articles in an attempt to group articles of similar
frequency for the category. topic together.
The classifiers output the likelihood that an article is in the K-Means clustering [5] is used and the distance between
category. The likelihoods from all the categories are used to two articles is calculated as one minus the cosine similarity,
determine which of the categories should be assigned to the shown in equation 2. The value of K is automatically estimated
article. The categories that have a likelihood greater than the in the following manner. First, the range of possible K is
mean plus one standard deviation are assigned to the article. determined by using the lower bound of N/2 where N is the
The assumption is that these categories stand out among the number of articles downloaded and an upper bound N.
group and are the best choices for the article.
3) Summarization
Ai • A j
For summarizing the article, the extracted keywords are Distance( Ai , A j ) = 1 − (2)
used to assign a score to each of the sentences. The top N Ai × A j
sentences are then extracted and become the article's summary.
In this research N = 4. The idea behind the summarization
Five random, but unique, values are chosen from the range.
technique is that the sentences with a high number of keywords
Each of these values are chosen to represent K and the K-
in them are more important than sentences that have a lower
Means algorithm is allowed to run for a few steps. After
number or no keywords.
completion, the average intra-cluster and extra-cluster distance
Summarization is done in three steps. The first step is to are calculated. The value with the best combination of low
break the article up into individual sentences. This is done intra-cluster and high extra-cluster distance is chosen and the
using a rule-based approach. The second step is to score each K-Means algorithm is run until completion or until 100 steps
of the sentences using equation 1, where k1, ... ,kn are the have been completed.
keywords in sentence S and KScore(ki) is the keyword score of
ki. The final step is to create the summary. For doing this, the
sentences are first sorted by score and the top four extracted. CScore(Cluster ) = P(Category ) × Cluster (3)
Then the four sentences are sorted by position in the article and
used as the summary. After the articles are clustered and have a category
N
assigned, they are ranked. The category chosen for the cluster
is the most abundant one among the articles in the cluster. For
Score(k1  k n ∈ S ) = ¦ log(1 +KScore(k ))
i =1
i (1)
ranking each cluster is first scored using equation 3. The
probability of a category, P(Category), is estimated using the
frequencies of the classified categories for the day’s articles
and as such, will vary day to day.
636 BRACEWELL

B. Synopsis Creation
After the clusters are ranked by cluster score, the top
clusters can be chosen as the most important ones. Currently,
the top 10 clusters are used, but the user could easily change
this. For each of these clusters an article is chosen to represent
the cluster. To choose the representative, article the importance
of each article is calculated using equation 4, where A is the set
of keywords that define the article, DF(ki) is the number of
articles the keyword appeared in, and KScore(ki) is the
keyword score of keyword ki.

N
Importance( A) = ¦ DF (k ) × KScore(k )
i =1
i i (4)

For the top clusters, a brief synopsis is created. The

synopsis consists of the cluster's category, the representative
article's title, summary and link to the full article, and the top 5
highest scoring keywords in the cluster.
Figure 4. Digest Example
V. IMPLEMENTATION
Politics 0.43
The system has been developed in C# and MySQL is used
for storing the articles and their classification results. Currently, Business 0.21
English and Japanese news is being monitored and gathered Science & Nature 0.15
using a set of 12 RSS feeds from the news sites listed below. Crime & Misfortune 0.14
Entertainment 0.13
• Yahoo! Top Stories (English) Sports 0.12
• Yahoo! Most Viewed (English) Health 0.09

• Google News (English) Figure 5. Category Ranking Example

• Washington Post Top News (English)

minut 0.20
• Nikkeibp.jp News (Japanese) recent 0.18
violenc 0.18
• Asahi (Japanese) tool 0.18
• Livedoor Main Topics (Japanese) video 0.17
warner 0.17
Eight categories, business, crime and misfortune, politics, read 0.17
sports, health, entertainment, science and nature, and lllp 0.17
technology, are being used for category classification. Each
reserv 0.17
category was trained using a collection of 1,000 documents
cabl network 0.17
representing the topic. Each category has a Japanese and an
English version. Figure 6. Keyword Ranking Example
The news monitoring and article processing module
processes the feeds every hour. This is, currently, resulting in
acquiring and processing about 1,200 articles a day. Because of
the large amount of articles that are being collected, the news VI. EVALUATION
digest creation algorithm is taking upwards of an hour to The proposed system was evaluated by comparing the
complete. Figures 4-6 show examples of output from the digests to human chosen important topics. The topics were
system. taken from titles of articles appearing during the day and
manually cleaned up to be more generic. Examples of topics
are “Colts win Superbowl” and “Abe elected prime minister.”
Forty topics were created for each day over a one-week period
starting March 13th, 2007 taken from the 10 stories chosen by
the proposed system and 30 other frequent topics found. In
addition to these topics, judges were told to write-in their own
topics if they did not appear in the list.
 AUTOMATIC CREATION OF DAILY NEWS DIGESTS FROM INTERNET NEWS SITES 637

Evaluation was done for English and Japanese looking at In the future, we hope to configure the system to work on
the agreement for the top 10 stories of the day. Each language the web and add more news sources. We would also like to add
had two judges. news tracking, which would track stories over time. It would
also be interesting to add named entity recognition to find the
English results can be seen in Table 1. Judge 1 and Judge 2 top people and places in the news for the day. We would like to
had very different agreements. Judge 2 leaned more in favor of also create a Chinese version of the system. Finally, we want to
political articles, while judge 1 more world events. The use of explore ways to improve the system. Personalizing the digests
personalization techniques could be used to improve the per user based on user defined topics of interest and feedback
individual results. could improve the agreement. We hope to examine a large
number of human created digests to mine features of digest-
TABLE I. ENGLISH EVALUATION RESULTS worthy articles.
Judge Agreement
1 26%
REFERENCES
2 78%
[1] D. B. Bracewell, F. Ren, and S. Kuroiwa. Multilingual single document
1∩2 83% keyword extraction for information retrieval. In Proceedings of the 2005
IEEE International Conference on Natural Language Processing and
Knowledge Engineering, Wuhan, China, November 2005.
Japanese results can be seen in Table 2. Judge 1 and Judge [2] D. B. Bracewell, F. Ren, and S. Kuroiwa. Category classification and
2 had a greater similarity in agreement than judges for English. topic discovery of news articles. In Proceedings of Information-MFCSIT
Both judges leaned towards political and business articles. 2006, Cork, Ireland, August 2006.
[3] E. Brill. A simple rule-based part-of-speech tagger. In Proceedings of
3rd Applied Natural Language Processing, pages 152–155, 1992.
TABLE II. JAPANESE EVALUATION RESULTS [4] S. Debnath. Creating rss for news articles and beyond. In Proceedings of
the 19th FLAIRS Conference, pages 275–280, 2006.
Judge Agreement
[5] J. B. MacQueen. Some methods for classification and analysis of
1 43% multivariate observations. In Proceedings of the 5th Berkeley Symposium
2 41% on Mathematical Statistics and Probability, pages 281–297, 1967.
[6] Y. Matsumoto, A. Kitauchi, T. Yamashita, Y. Hirano, H. Matsuda, K.
1∩2 49%
Takaoka, and M. Asahara. Morphological analysis system chasen
version 2.2.9 manual. Technical report, Nara Institute of Science and
Technology, 2002.
[7] M. Porter. An algorithm for suffix stripping. Program, 14:130–137,
VII. CONCLUSION 1980.
In this paper a system that creates daily digests of news [8] M. Schranz, S. Dustdar, and C. Platzer. Building an integrated pan-
articles was presented. The system monitors and gathers news european news distribution network. In Proceedings of the 6th IFIP
to download and classify. Classification is made up of keyword Working Conference on Virtual Enterprises, PRO-VE 2005, pages
26–28, 2005.
extraction, category classification and summarization. Using all
of this information a digest can be created that highlights the
important news of the day as well as the top categories and
keywords. Currently, the system works on English and
Japanese news.
 Index

A Cloud model, 2
Cluster
Academic advising, 25–29 ensemble, 305–308, 311–312
Academic record, 151–153 head, 539–542, 549–552
Academic social network, 151–156 Clustering algorithm, 306, 312, 542, 549–553
Acquisition system, 243–245, 253, 396 CMOS, 339, 405, 416
Adaptive assessments, 19–23 Coarse-grained PEDT/PSS, 263, 267
Adjacency relationship, 519–522 Coaxial outputs, 357
Agile software, 603–606 Collaborative learning, 83–87, 122, 134, 171
All Share Price Index (ASPI), 269–273 Collinear and triangle equation, 351–354
Analogical-digital converter, 373 Colombo Stock Exchange, 269
Analytical Hierarchy Process (AHP), 77, 82 Command circuit, 373
Architectural education, 163–166 Common-wavelength, 499, 501
Artificial intelligence, 44, 46, 73, 75, 147, 227, Communication, 45, 55, 113, 145, 149, 243,
361, 421–422, 424 461–465, 570
Artificial neural networks, 225 Communication integration, 461–465
Assessment, engineering education, 59 Component, 33, 244, 617, 629
Augmented reality, 215–216 Computational modeling, 193–195
Automated advising, 25 Computer graphics, 114, 181, 215, 316
Automation, 73, 109, 111–112, 163, 175–179, Computer Numerical Control (CNC),
187–192, 250, 327, 345–349, 398 163–166, 169
Autonomic computing, 421, 424 Computer vision, 182, 253–255, 278, 282, 312,
Autonomous system, 517–520, 522 353, 361, 433, 437, 439
Congestion control, 493–496, 503, 579–580
B Congestion window, 493–496, 498
Construccionism, 71
Ballistics system, 327 Constructivism, 71, 73, 133–134, 137
Bandwidth allocation, 479–481, 483 Content-addressable memory (CAM), 405–410
Binary sensor network, 381–386 Context, 12, 20, 22, 43–47, 73, 79–80, 134,
Biometric identification, 303–307 139–143, 161–162, 172, 559, 574–575
Bioorganic compounds, 377–378 Conventional proportional-integral control, 203
Blow-CAST-Fish, 523–528 Converter, 243–246, 257, 259, 288, 373, 376, 393
Bologna process, 151–156 Cooperative maritime network, 449–454
Both side differential transmissions, 357 Coremaking plant, 247–251
Break-even point, 105, 107 Course scheduling, 25
Browser, 27–28, 47, 79–81, 129, 134, 154–155, Covert channel, 503–506
160, 328, 461–465, 557 Critical thinking, 55–58, 86, 121–126, 134
Curriculum management system, 89–94
C
D
Calibration techniques, 185
Cardiac cavity, 351–354 Daemon, 443, 445
CAT instrument, 55–58 Data hiding, 367
Cellular systems, 562 Data protection, 455, 458–459, 460, 560
Chaotic, 367–371 Decomposition method, 427–432
Chromaticity, 362–365 Deconvolution method, 393–398
Chromatic modulation, 361–365 Design of SOHO, 555–560
Circuit analysis, 59–60, 193 Diameters of carbon nanotubes, 275–279
Clamped joints, 417–419 Didactic application, 443, 448
Cloud Migration Strategy, 2 Diesel fuel, 377–378

639
K. Elleithy et al. (eds.), Technological Developments in Networking, Education and Automation,
DOI 10.1007/978-90-481-9151-2, © Springer Science+Business Media B.V. 2010
640 INDEX

Differential cryptanalysis, 523–528 Fractional pore area, 234

Differential evolution, 209–211 Fragmentation scheme, 627, 630–631
Differential geometry, 387 Frequency allocation, 479
Diffraction, 507 Friction torque, 417–419
Digest creation, 633–636 Fuzzy logic, 496, 591–595
Digital citizenship, 49–53
Digital content delivery, 37–40 G
Digital fabrication, 163–164, 166
Digital logic laboratory, 109–114 Generalized Minimum Variance Controller
Digital system description, 127–132 (GMVC), 291–296
Discriminant analysis method, 275 Genetic algorithms, 269–273, 422–423, 425, 511
Discussion forums, 89–90, 121–126 Geocast, 567–572
Dissolved gas analysis, 225–229 Greedy algorithm, 479, 481–482
Distributed databases, 627–631 Grid site RO-14-ITIM, 609–612
Domain ontology, 98–100, 152, 158–161, 576,
615–617 H
Double pipe heat exchanger, 387–391
Dual-mode micro-mirror, 411–415 Half-wave, 287–289
Dynamic allocation, 479 Handover, 533–537, 561–566
Dynamic change window, 399 Hardware description languages, 127
Harmonic gear, 357–360
E Henon map, 367–368, 371
High boost filter, 351–353
E-cheating, 115–117 Higher education, 1–5, 25, 37, 41, 72, 133–137,
Edge detection, 183, 297–298, 303, 307, 322, 351, 139–140, 143, 151–152, 156, 169
368 Holistic enterprise management, 455
Educational robotics, 71–73 Hologram design, 507–509
E-education, 43, 103–108 Honeytokens, 449–454
E-learning, 19–20, 43–47, 77–82, 89, 93, HPC, 609
103–104, 107, 115–118, 127–128, 133 HTTP, 12, 33, 129, 461–465, 560
Electrode, 364, 415, 507, 609 Human learning, 31–32, 137
Electronic curriculum, 89–90, 92–94 Humidity elimination, 377–378
Electroporation, 231–234 Hydrophilic polymer antifouling action, 377
Energy dissipation, 539–542, 571 Hypermedia application, 65, 67
Energy-saving sealing unit, 193–195
Enzyme kinetics parameters, 13–18 I
E-speed start, 493–497
Ethernet node, 621–626 IAX, 585, 588
EvalTool, 627–631 Internet Control Message Protocol (ICMP), 443,
Evaluation method, 77–78, 82 503–506, 557–559
Execution time, 288, 423, 425, 628, 630 IEEE21, 533–537, 802
Expectation-Minimization, 303–304 IIR digital filter, 219–223
Extreme programing, 603–606 Image authentication, 367
Image binarization, 275, 322, 333
F Image processing, 66, 182, 253–254, 275–279,
299, 303, 305, 312, 327–328, 367
Feedback model, 579–584 INCDTIM, 609–612
Feistel cipher, 523 Individualized learning, 83–84, 87
Firearm identification, 327–331 Industrial PID controller, 291, 295
Fixed-MobileSPADE, 597–601 Informatics curricula, 145
Flooding attack, 555, 558–559 Information quality, 139–143
Fourth generation, 561 Information security, 347–348, 450, 455–460,
Foveated images, 281–285 556, 573
 INDEX 641

Information systems, 83–87, 134, 139–140, Minimum data buffer, 473

145–146, 148, 151, 345 Mining mobile, 597–601
Integrated web-based educational application, MobileSPADE, 597–601
65–70 Mobility sensory platform, 433–441
Intelligent plasma, 361–365 Modular design, 433–441
Interactive applications, 215–216 Moment invariant, 333–335, 337–338
Interdisciplinary automation, 175–179 Monoalphabets, 529–531
IP-PSTN integration, 585–589 Monte Carlo simulations, 381, 385
Ipv6, 443–448, 504 Morphology, 263, 297–299, 351–352
ISDN, 563, 585, 587–589 MPC5553, 621
Isolated thermal power system, 203–206 Multi-agent learning, 421
Isolation gate bipolar transistor, 257 Multi-agent system, 421–422
Istopolis, 65–70 Multicore system, 339–344
IT change management tool, 399–403 Multimedia security, 367
Multiobjective optimization, 209–210
K Multiplicative decrease network, 493–498
Multiprocessor System-on-Chip (MPSOC), 339
Keyless polyalphabetic cipher, 529–532 Multi-sink wireless sensor network, 591–595
Knowledge sharing mechanism, 31–32 Multiuser, 479–480
Knowledge states, 7–12 Multi-walled CNTs, 275

L N

Ladder-diagram, 178, 192 Nature Inspired Optimization Technique (NIT),

Lathe, 169 421–423, 425
LDR, 197–199 Negative Laplacian, 351–353
Learning management system, 21, 89, 93, 122 Network on Chip (NOC), 339–340
Level diagrams, 209–214 Networks, 87, 99, 104, 160, 219, 228, 269, 340,
Life long learning, 43, 171 451, 484, 497, 500, 503–504, 533, 539, 555,
Linear regression, 13, 16, 184–185 560, 562, 567–568, 579–581, 586, 592, 601,
Liquid crystal, 507–509 621
Locating targets, 381–386 Network security, 556–560, 592
Lost in cyberspace, 65, 67, 69 Neural network, 147, 203–206, 219–225, 227,
Low power VLSI, 405 269–272, 282, 338, 511–515
NGN architecture, 585–589
M Nonlinear observer, 387–391
Normalized difference vegetation index, 181, 186
Managing congestion, 579–584 Numerical model, 231, 234, 427–428
Markov chain, 427, 429–430
Markov random field, 303–304, 309–312, O
315–317
Mathematical morphology, 297–299, 351 Object detection, 281–285
MATLAB, 16, 59–63, 221, 228, 233, 254–255, Object identification, 309–315
272, 284, 394, 474–475, 580, 584 Online news, 633
MesoDyn simulation code, 263 Ontology-based intrusion detection, 467–472
Mesoscale simulation, 263–267 OPCUA automation protocol, 345–350
MFI4OR, 615–619 Open standards, 19–24
Michaelis-Menten equations, 13 Optical networks, 499
Microcontroller, 109–113, 199, 243–244, Optical sensors, 113, 361
287–290, 376, 393–394, 621–623 Optimize scheduling, 399–403
Microelectromechanical Systems (MEMS), 411, Orthogonal complementary codes, 487–492,
413, 415, 507 543–544
MIH, 533–534 OSI-network model, 345
642 INDEX

P Responsibility, 17–18, 49–54, 66, 87, 121, 140,

459, 550, 586, 605
Packet switched, 340, 562–563, 579–584 Retraining, 103, 511–515, 635
Panels, 175, 184, 197–201 RFID, 440, 467–472
Pareto front, 209–214 RISCbot ІІ, 433–434, 439–441
Partitioning scheme, 627–631 Risk management, 455
Peer-to-peer (P2P), 449–454 Robotic swarm, 71, 73–74
Personalization, 43–47, 97, 633, 637 Rotary transducer, 373–376
Personalized ontology, 97, 100 Router advertisement, 443–446
P-Hub median problem, 237–241 Routing, 339–344, 405–406, 447, 499–500,
Piezoelectric sensor, 393 517–522, 549–550, 558–559, 562, 567–571,
Piezoelectric technology, 411 591–594, 597
Pilot, 4, 37–38, 40, 89–90, 92, 141–143, 197–200, Routing policy, 517–518, 522
387–388 RS232 interface, 111, 243–244
Pipeline, 80, 405–410 RS245, 394, 485, 621
Piston micro-mirror, 411
Plaintext, 523–524, 526–531 S
Plant health, 181–186
Polymeric PEDT/PSS, 263–268 Samoan students, 83
Pore radii, 231, 233 Scale-invariant feature transform (SIFT), 253–256
Power simulation, 405, 407, 409–410 SCRUM, 446, 603, 605–607
Power transformers, 225–229 Seal identification, 333–334
Prediction, 4, 269, 271, 273, 511–515, 597, Sealing lifetime, 193
600–601 Seal segmentation, 333–338
Presence, 49, 87, 145, 166, 215, 259, 294–295, Security measures, 503, 555–556, 592
377, 395, 461–466, 511, 513, 568, 570 Segmentation, 186, 297, 309–320, 322, 330,
private key, 347–348, 523 333–338, 351–354, 563
Proactive advising, 25 Selecting patterns, 77–82
Probability, 8, 38, 44, 249, 263–264, 304, Self management distributed approach, 151
309–310, 429–432, 496, 523–528, 568, 583, Self-optimization, 421–426
597, 635 Semantic annotation, 152, 160, 615–619
Programmable logic controllers (PLC), 109, Semi-controlled rectifier, 287–290
175–179, 187, 192 Serial communications, 621–626
Programmable logic devices, 8, 38, 44, 97–98, Service, 1–2, 43–47, 139, 152, 154, 159, 249, 345,
249, 263–264, 304, 309–310, 317, 382, 347–348, 401, 464, 517, 522, 533–534,
429–432, 452, 496, 512, 523–528, 539, 549, 561–564, 580, 597, 615–619
568, 581, 583–584, 597, 635 Simulation game, 31–34
Simulation model, 247–251, 561, 563
Q Simulation results, 129, 203, 205–206, 293, 339,
405, 409–410, 453, 483–484, 500–502, 512,
Quality management, 455–460 514, 534, 541–542, 549, 552–553
Quasiorthogonal complementary pairs, 543 SIP, 465–466, 585, 587–589
Query expansion, 97–99, 160 Sliding mode control, 209–210
Queuing system, 248, 427–432, 580 Social components, 157–162
Society and computing, 71
R Software agents, 97–98, 100
Solar disk, 297–299, 321, 323
RC filters, 473–477 SPADE, 597–599, 601
Real-time control system, 169 Spare resources, 499–502
Recurrent neural network, 219–224 Spatial light modulator, 507–510
Region filter, 351–353 Sri Lankan Stock Market, 269–273
Regression analysis, 377–379 Stateless autoconfiguration, 443–448
Resource management, 455–458, 479–484, 609 State services, 43–44
 INDEX 643

Steady state probabilities, 427–428 Unsupervised segmentation, 297, 301, 316,

Stereo spectral imaging, 181–186 321–323
Stochastic model, 303–320 Unsupervised stochastic model, 315–320
Strategy, 1–2, 4–5, 37, 40, 73, 84–85, 87, 93, 121,
139–140, 146, 148–149, 151, 161, 171, 228, 399, V
407, 431–432, 456, 459, 479, 579, 587, 597
Subcarrier, 479–484, 488, 544 Vertical handover, 533–537
Sunspots, 297–301, 321–324 Vibration signal processing, 393–398
Sunspots groups, 297, 299–301, 321–324 Virtual reality, 215
Support vector machines, 225, 227, 321–324 Visual attention, 281–283, 285
Surveillance region, 381–386 Voice gateway, 585–589
Switching power devices, 257–261
Synopsis creation, 633, 635–636 W
Synthesis of signals, 487, 543 Watermarking, 47, 367–371
System on a Programmable Chip (SOPC), Wavelengths, 181–184, 361, 363, 365, 499–502
339–344 WDM networks, 499–502
Weak orderings, 7–12
T Web, 1, 5, 7–8, 12, 19, 26, 49, 51–53, 65–70,
78–79, 90, 94, 97–101, 127, 133–135, 137, 141,
Telephony, 5, 461–462, 562, 585, 587, 589 148, 152, 154, 156, 159, 346, 439, 462, 464,
Textured color, 309–320 466, 470–471, 557, 559–560, 615–619
Threshold, 183, 185, 198, 263–264, 275–277, 279, Web 2.0, 49, 133–135, 137, 148, 152, 154, 156
284, 297–298, 303, 308, 322, 330, 334, Web repositories, 97–101
336–338, 342, 351–352, 381–386, 437–438, 452, Web service, 140, 151–152, 154, 159–160,
469, 494, 512, 514, 534–537, 539, 598–601 345–346, 348, 439, 462–464, 560, 615–619
Topology, 74, 263, 266, 340, 499–501, 504, Wide Area Networks (WAN), 579–584
517–522, 534–535, 549–550, 564–567, WiFi, 533–537
570–571, 580, 587–589, 591, 594 Wiki, 133–137
Torsional micro-mirror, 411 Wikibook, 133–137
Transition network, 187–192 WiMAX, 533–537
Transmission Control Protocol (TCP), 348, Windows, 3, 69, 84, 110–111, 128, 181, 254, 271,
462–465, 493–497, 503–506, 557–559 279, 345, 399, 401, 403, 440, 443, 445–447,
Triple Bottom Line (TBL), 37 498, 543–544, 556, 560, 618, 622
TTL, 109, 114 Wireless cellular networks, 597–601
Typology of costs, 103 Wireless Sensor Network (WSN), 539–542,
549–553, 567–568, 591–595, 621
U WLCG, 609–613

Universal Mobile Telephone system (UMTS), X

484, 561–566
Unmanned aircraft, 237–240 X-ray images, 303–308