03014_Cover.

pdf

10/20/11

11:30 AM

DIGITAL COMMUNICATIONS
PHYSICAL LAYER EXPLORATION LAB USING THE NI USRP PLATFORM

Robert W. Heath Jr., Ph.D., P.E.

i
book2 2011/10/28 14:05 page i #1

Digital Wireless
Communication
Physical Layer Exploration Lab
Using the NI USRP
Student Lab Manual

Robert W. Heath Jr.

i
i

i
book2 2011/10/28 14:05 page ii #2

ISBN-10: 1-934891-18-5
ISBN-13: 978-1-934891-18-6

10

Publisher: Tom Robbins

General Manager: Erik Luther
Technical Oversight: Sam Shearman
Marketing Manager: Amee Christian
Development Editor: Catherine Peacock
Compositor: Paul Mailhot, PreTeX Inc.
c
2012
National Technology and Science Press.
All rights reserved. Neither this book, nor any portion of it, may be copied or reproduced in any
form or by any means without written permission of the publisher.
NTS Press respects the intellectual property of others, and we ask our readers to do the same.
This book is protected by copyright and other intellectual property laws. Where the software
referred to in this book may be used to reproduce software or other materials belonging to others,
you should use such software only to reproduce materials that you may reproduce in accordance
with the terms of any applicable license or other legal restriction.
LabVIEW, USRP, Universal Software Radio Peripheral, and National Instruments are trademarks
of National Instruments.
All other trademarks or product names are the property of their respective owners.
Additional Disclaimers:
The reader assumes all risk of use of this book and of all information, theories, and programs
contained or described in it. This book may contain technical inaccuracies, typographical
errors, other errors and omissions, and out-of-date information. Neither the author nor the
publisher assumes any responsibility or liability for any errors or omissions of any kind, to
update any information, or for any infringement of any patent or other intellectual property
right.
Neither the author nor the publisher makes any warranties of any kind, including without
limitation any warranty as to the sufficiency of the book or of any information, theories,
or programs contained or described in it, and any warranty that use of any information,
theories, or programs contained or described in the book will not infringe any patent or other
intellectual property right. THIS BOOK IS PROVIDED AS IS. ALL WARRANTIES,
EITHER EXPRESS OR IMPLIED, INCLUDING, BUT NOT LIMITED TO, ANY AND ALL
IMPLIED WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR
PURPOSE, AND NON-INFRINGEMENT OF INTELLECTUAL PROPERTY RIGHTS,
ARE DISCLAIMED.
No right or license is granted by publisher or author under any patent or other intellectual
property right, expressly, or by implication or estoppel.
IN NO EVENT SHALL THE PUBLISHER OR THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, SPECIAL, INCIDENTAL, COVER, ECONOMIC, OR CONSEQUENTIAL DAMAGES ARISING OUT OF THIS BOOK OR ANY INFORMATION, THEORIES,
OR PROGRAMS CONTAINED OR DESCRIBED IN IT, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGES, AND EVEN IF CAUSED OR CONTRIBUTED TO BY
THE NEGLIGENCE OF THE PUBLISHER, THE AUTHOR, OR OTHERS. Applicable law
may not allow the exclusion or limitation of incidental or consequential damages, so the above
limitation or exclusion may not apply to you.

i
i

i
book2 2011/10/28 14:05 page iii #3

Dedication
I dedicate this to Garima, Pia, and Rohan for their love and support,
to my parents Bob and Judy Heath for their encouragement
and to Dr. Mary Bosworth for her passion in the pursuit of higher education.
R. W. H.

i
i

i
book2 2011/10/28 14:05 page iv #4

i
i

i
book2 2011/10/28 14:05 page v #5

Contents
Preface

vii

About the Author

xi

Lab 1: Part 1 Introduction to NI LabVIEW

Lab 1: Part 2 Introduction to NI RF Hardware

10

Lab 2: Part 1 Modulation and Detection

22

Lab 2: Part 2 Pulse Shaping and Matched Filtering

35

Lab 3: Synchronization

51

Lab 4: Channel Estimation & Equalization

63

Lab 5: Frame Detection & Frequency Offset Correction

82

Lab 6: OFDM Modulation & Frequency Domain Equalization

99

Lab 7: Synchronization in OFDM Systems

115

Lab 8: Channel Coding in OFDM Systems

130

Appendix A: Reference for Common LabVIEW VIs

139

Bibliography

141

v
i

i
i

i
book2 2011/10/28 14:05 page vi #6

i
i

i
book2 2011/10/28 14:05 page vii #7

Preface
Wireless communication is fundamentally the art of communicating information without wires. In principle, wireless communication encompasses
any number of techniques including underwater acoustic communication,
semaphores, smoke signals, radio communication, and satellite communication, among others. The term was coined in the early days of radio, fell out
of fashion for about fifty years, and was rediscovered during the cellular telephony revolution. Wireless now implies communication using electromagnetic wavesplacing it squarely within the domain of electrical engineering.
Wireless communication techniques can be classified as either analog or
digital. The first commercial systems were analog including AM radio, FM
radio, television, and first generation cellular systems. Analog communication is rapidly being replaced with digital communication. The fundamental
difference between the two is that in digital communication, the source is
assumed to be digital. Modern applications of digital communication include cellular communication, wireless local area networking, personal area
networking, and high-definition television.
Digital communication is widely taught in electrical engineering programs, at both the undergraduate and the graduate levels. Many digital communication courses take an abstract approach that emphasizes theory. This is
no surprisethe mathematical theory of digital communication is beautiful
and elegant. There has been a growing separation, however, in the abstract
content of a typical course and what is implemented in practice. This difference is particularly evident in wireless communication systems, which have
their own set of specific challenges. As courses are becoming more abstract,
there is also increasing reliance on simulation to validate the theory. Typical
simulations, though, are constructed under certain simplifying assumptions,
which leaves important practical issues undiscovered. As a byproduct of the
growing rift between theory and practice, and the increasing reliance on simulation, many students are not prepared to apply their knowledge immediately
to a real digital wireless communication system.
vii
i

i
i

i
book2 2011/10/28 14:05 page viii #8

viii

Preface

This book is inspired by my own experiences as a graduate student and

working at a startup company. My first summer project at Stanford University
working withArogyaswami Paulraj was to implement the physical layer of the
receiver for a GSM system. The idea was to implement different space-time
processing algorithms developed in Prof. Paulrajs Smart Antennas Research
Group and to show practically how multiple antennas can improve system
performance using real transmitted waveforms. I found the implementation
of the space-time processing algorithms to be straightforward (not easy but
for the most part following the theory). Most of my time was spent, however,
dealing with more practical issues like channel estimation, frame synchronization, and carrier frequency synchronization. I found that the algorithms
for dealing with these practical impairments were vitally important yet I had
never encountered them in multiple courses on digital communication and
wireless communication systems.
I used what I had learned later at a startup company then called Gigabit
Wireless and later Iospan Wireless, to develop a prototype of a three transmit
and three receive antenna MIMO (multiple input multiple output) communication system. This prototype was key to demonstrating the promise of
MIMO communication, which in 1998 was not yet an established hot research area. Coincidentally this prototype used multichannel DAC and ADC
products from National Instruments.
As a new Assistant Professor at The University of Texas at Austin, I longed
to teach wireless communication from a practical perspective. I quickly found
one reason why these concepts were not taught in the typical communication
courseit was not included in most textbooks. As a result, I developed course
notes that include these concepts and will be published in a forthcoming
textbook [3]; the key concepts are included in this lab manual.
I did not just want to teach wireless communication including some theory
about dealing with practical impairments. I wanted to teach a class where
students would send and receive actual communication waveforms. Unfortunately, when I started developing the course materials in 2003 there
were few viable solutions for the radio frequency components to create
a repeatable and reliable system. All of this changed with the introduction of hardware platforms with flexible RF upconverters and downconverters, matched to high speed DACs and ADCs that facilitate the concept
of software defined radio. Finally there was a way to create and process
waveforms with high quality and reproduciabilty. Using such equipment,
the students could focus on the algorithms and yet be exposed to practical wireless system engineering, without having to worry with the challenging analog front end and RF circuit design issues. The present book
makes use of National Instruments Universal Software Radio Peripheral
(USRP) [13]. This low-cost hardware allows Universities to set up mul-

i
i

i
book2 2011/10/28 14:05 page ix #9

Preface

ix

tiple workstations with a moderate budget. I am thrilled that this option

has become available.
This book is a result of five years of development to make a complete
laboratory course that teaches the principles of wireless digital communication. It consists of a series of labs with three components: a pre-lab, a
laboratory experiment, and a lab report. The idea of the pre-lab is to write
the essential LabVIEW code and getting it to work in a simulator before
doing the laboratory experiment. This models conventional design practices
where communication algorithms are tested in simulation before being tested
over a wireless link. It is important to complete the pre-lab prior to starting
the laboratory experiment. The laboratory experiments were designed to be
completed in a time period of up to three hours. In the lab, the code from
the pre-lab is run over the wireless link, some experiments are performed,
and the results are recorded. The lab report is the final component of the lab.
The purpose of the lab report is to discuss what was observed in the lab and
to answer several questions related to wireless communication engineering.
The lab report is an opportunity to synthesize what was learned.
Digital signal processing (DSP) is at the heart of the approach taken in
this lab manual. No background in digital communication is assumed, though
it would be helpful. The utility of a DSP approach is due to the bandlimited nature of wireless systems. Consequently with a high enough sampling
rate, thanks to Nyquists theorem, it is possible to represent the bandlimited
continuous-time wireless channel from its samples. This allows the transmitted signal to be represented as a discrete-time sequence, the channel as
a discrete-time linear time-invariant system, and the received signal as a
discrete-time sequence.
The labs explore both single carrier and multicarrier transmission. Single carrier transmission uses quadrature amplitude modulation (QAM) and
raised-cosine pulse-shaping. Over the course of the labs, complexity is added
to the receiver design including functions like detection, channel estimation,
equalization, frame synchronization, and carrier synchronization. In later
labs, the system is extended to incorporate multicarrier modulation in the
form of orthogonal frequency division multiplexing (OFDM) with coding.
This lab manual could not have been possible with the support from many
graduate teaching assistants at The University of Texas at Austin including
Roopsha Samanta, Sachin Dasnurkar, Ketan Mandke, Hoojin Lee, Josh Harguess, Caleb Lo, Robert Grant, Ramya Bhagavatula, Omar El Ayache, Harish
Ganapathy, Zheng Li, and Tom Novlan. Some specific contributions deserve
special mention. Roopsha Samanta developed the initial set of laboratory
drafts and software in addition to being the first TA for the course. Ketan
Mandke reworked the software to have a complete system implemented and
a channel simulator, where components could be removed and replaced with

i
i

i
book2 2011/10/28 14:05 page x #10

Preface

software from the student. He also revised the lab manual to reflect the new
system implementation. Hoojin Lee undertook the job of creating figures
for the early course material. Josh Harguess created a nice introduction to
LabVIEW presentation. Robert Grant spent time seeing how to convert to
an early version of the USRP platform, when it was just released. Caleb Lo,
Ramya Bhagavatula, Omar El Ayache, and Tom Novlan all made important
revisions to the notes and software to keep it up-to-date with current versions
of LabVIEW. All the teaching assistants ran the labs, worked with the students, collected feedback, and make suggestions and revisions. I sincerely
thank them for their support.
The development of this book and course was supported by several groups.
National Instruments made the course possible through its establishment of
the Truchard Wireless Lab at The University of Texas at Austin. NI provided summer support that funded the endeavor, and contributed countless
hours of employee time. Several groups from University of Texas at Austin
including the Department of Electrical Engineering and the Cockrell School
of Engineering provided teaching assistant support to develop and maintain
the course materials. Working on the labs has inspired my research. I am
pleased to acknowledge research support from the National Science Foundation, DARPA, the Office of Naval Research, the Army Research Labs,
Huawei, National Instruments, Samsung, Semiconductor Research Corporation, Freescale, Andrew, Intel, and Cisco. Their support has allowed me
to train many graduate students who share my appreciation of theory and
practice.
I hope you enjoy this book and doing the laboratory experiments.
Robert W. Heath Jr.
Austin, Texas

i
i

i
book2 2011/10/28 14:05 page xi #11

About the Author

Robert W. Heath Jr. is an Associate Professor in the Department of Electrical and Computer Engineering at The University of Texas at Austin and a
member of the Wireless Networking and Communications Group. He is also
President and CEO of MIMO Wireless Inc. and VP of Innovation at Kuma
Signals LLC. He received his B.S. and M.S. degrees from the University of
Virginia, Charlottesville, VA, in 1996 and 1997 respectively, and the Ph.D.
from Stanford University, Stanford, CA, in 2002, all in electrical engineering.
Prior to joining The University of Texas, he was with Iospan Wireless Inc,
San Jose, CA where he worked on the design and implementation of the physical and link layers of the first commercial MIMO-OFDM communication
system.
His current research interests include several aspects of wireless communication and signal processing including limited feedback techniques, multihop
networking, multiuser and multicell MIMO, interference alignment, adaptive
video transmission, manifold signal processing, and 60 GHz communication
techniques. He has published some 300 papers in archival journals and conference proceedings on these topics, and has more than 25 U.S. patents.
He has been an Editor for the IEEE Transactions on Communication, an
Associate Editor for the IEEE Transactions on Vehicular Technology, lead
guest editor for an IEEE Journal on Selected Areas in Communications special issue on limited feedback communication, and lead guest editor for an
IEEE Journal on Selected Topics in Signal Processing special issue on Heterogenous Networks. He currently serves on the steering committee for the
IEEE Transactions on Wireless Communications. He was a member of the
Signal Processing for Communications Technical Committee in the IEEE
Signal Processing Society. Currently he is the Chair of the IEEE COMSOC Communications Technical Theory Committee. He was a technical
co-chair for the 2007 Fall Vehicular Technology Conference, general chair
of the 2008 Communication Theory Workshop, general co-chair, technical
co-chair and co-organizer of the 2009 IEEE Signal Processing for Wireless
Communications Workshop, local co-organizer for the 2009 IEEE CAMSAP
xi
i

i
i

i
book2 2011/10/28 14:05 page xii #12

xii

About the Author

Conference, technical co-chair for the 2010 IEEE International Symposium

on Information Theory, technical chair for the 2011 Asilomar Conference
on Signals, Systems, and Computers, and is technical co-chair for the 2014
IEEE GLOBECOM conference.
He was a co-author of best student paper awards at IEEE VTC 2006 Spring,
WPMC 2006, IEEE GLOBECOM 2006, IEEE VTC 2007 Spring, and IEEE
RWS 2009. He is co-recipient of the Grand Prize in the 2008 WinTech
WinCool Demo Contest and is co-recipient of the 2010 EURASIP Journal on
Wireless Communications and Networking best paper award. He was a 2003
Frontiers in Education New Faculty Fellow. He is the recipient of the David
and Doris Lybarger Endowed Faculty Fellowship in Engineering. He is a
licensed Amateur Radio Operator and is a registered Professional Engineer
in Texas. He is a Fellow of the IEEE.

i
i

i
book2 2011/10/28 14:05 page 1 #13

Lab 1: Introduction to Wireless Communications

Lab
Part 1: Introduction to NI LabVIEW

Summary
A software defined radio (SDR) is a communication radio that uses software
to implement the algorithms necessary for digital communication. In this lab,
you will be designing and implementing an SDR using National Instruments
hardware and software (i.e., National Instruments LabVIEW). Through the
implementation of this SDR you will investigate practical design issues in
wireless digital communication.
This introductory lab is divided into two parts. In the first part you will
learn about the software used in this lab; in the second part you will learn
about the RF hardware to be used in lab. This part of the lab will answer the
following questions:
What software will I need for this lab?
What is NI LabVIEW and why do we use it in this lab?
What are some of the advanced LabVIEW tools used in this lab?
Is there such a thing as good code style in LabVIEW?
Also, in this part of the lab, you will build some of the basic components
of the SDR that you will be designing throughout the labs. This lab is meant
to introduce LabVIEW as a tool for the design of digital communications
systems in this course, and to give you a chance to get comfortable with basic
LabVIEW programming concepts.

Software and Other Materials for the

Course

Throughout the course you will be required to do some portion of the lab
work (e.g., pre-labs) outside of the lab. This means that you will need to get
access to the appropriate software tools and packages outside of class. The
NI software below will be used in the course.
1
i

i
i

i
book2 2011/10/28 14:05 page 2 #14

Lab 1: Part 1 Introduction to NI LabVIEW

NI LabVIEW 2010 or later
NI LabVIEW Modulation Toolkit 4.3.1 or later

Note: Your instructor will provide information about how to access LabVIEW
at your University.

Introduction to National Instruments

LabVIEW

2.1 What Is LabVIEW?

LabVIEW is a graphical programming language developed by National Instruments. The basic building block of LabVIEW is the virtual instrument
(VI). Conceptually, a VI is analogous to a procedure or function in conventional programming languages. Each VI consists of a block diagram and a
front panel. The block diagram describes the functionality of the VI, while
the front panel is a top level interface to the VI. The construct of the VI provides two important virtues of LabVIEW: code reuse and modularity. The
graphical nature of LabVIEW provides another virtue: it allows developers
to easily visualize the flow of data in their designs. NI calls this Graphical
System Design. Also, since LabVIEW is a mature data flow programming
language, it has a wealth of existing documentation, toolkits, and examples
which can be leveraged in development.
In this course you will use National Instruments RF hardware. LabVIEW
provides a simple interface for configuring and operating various external
I/O, including the NI RF hardware used in lab. This is the main reason why
you will use LabVIEW as the programming language to build an SDR in this
course. You should realize that the algorithms considered here could also be
programmed in optimized C/C++, assembly, or VHDL and implemented on
a DSP, microcontroller, or an FPGA. The choice of hardware and software in
this lab is mostly a matter of convenience.
In future labs you will need to be familiar with LabVIEW and the documentation/help available to you. This is the only lab in this course which will
give you the opportunity to learn and practice LabVIEW programming; so
it is important that you take this opportunity to ask the instructor any questions you might have about LabVIEW programming. The following tutorials
and reference material will help guide you through the process of learning
LabVIEW:
LabVIEW 101 [11],
LabVIEW Fundamentals from National Instruments [12],

i
i

i
book2 2011/10/28 14:05 page 3 #15

Lab 1: Part 1 Introduction to NI LabVIEW

and online LabVIEW tutorials from NI [9, 10].

New LabVIEW programmers should carefully review all of the material in
[11] and [9]. Please remember to refer to [12] and [10] often as they are
excellent references for all basic LabVIEW questions. In order to test your
familiarity with LabVIEW, you will be asked at the end of this lab to build
some simple VIs in LabVIEW and integrate them into a simulator. Bring
any questions or concerns regarding LabVIEW or these tutorials to your
instructors attention. For the remainder of this lab you should be familiar
with the basics of LabVIEW programming and where to look for help.

2.2 What Advanced LabVIEW Tools Will I Use?

Many of the algorithms you will implement in the lab (and in digital communication in general) use linear algebra. LabVIEW provides a variety of
support for matrix and vector manipulation and linear algebra. Some of the
functions you will use regularly include matrix inversion (Inverse Matrix.vi),
matrix multiplication (A x B.vi), and reshaping arrays (Reshape Array.vi).
LabVIEW also has many built-in signal processing functions which will be
used in the lab, such as the fast Fourier transform (FFT.vi), inverse FFT
(Inverse FFT.vi), and convolution (Convolution.vi). The Modulation Toolkit
is a toolset of common digital communication algorithms which will also be
leveraged in this lab. Note that many of the functions you will implement in
this lab are already available in the Modulation Toolkit in some form. The
objective of this course is to understand the principles of wireless digital communication by implementing the physical layer in as much detail as possible.
Once you are familiar with these concepts, you will decide when to use existing VIs and when to write your own. It is recommended that you explore
the following tool palettes in order to get some exposure to the VIs that you
will likely use in this course:
Signal Processing palette
Digital palette (part of the Modulation Toolkit palette)
Structures palette
Complex palette (part of the Numeric palette)
and Array palette
Appendix A enumerates some common VIs to be used in this course and how
to access them.

i
i

i
book2 2011/10/28 14:05 page 4 #16

Lab 1: Part 1 Introduction to NI LabVIEW

2.3

Code Style in LabVIEW

Googling the phrase labview code style will yield a number of sources
for general code style in LabVIEW. Whatever style you choose to follow,
remember that your code style should accomplish the following:
1. Consistency - All of your code should follow the same code style.
2. Readability - Your code should be user-friendly and easy to maintain.
3. Documented - Documentation should not be few and far between.
Good documentation not only comes in the form of free text labels
(see the Decorations palette, which is part of the Structures palette),
but it also means good variable and function names.
When incrementally designing a moderately complicated system, it can
be tempting to continually add hacks until you end up with spaghetti code
(see Figure 1).
The design of a digital communication system is modular and sequential
in nature. Thus it is natural to adopt a code style which makes use of a
great deal of hierarchy and modularity. Since many algorithms in digital
communications can often be boiled down to a recipe (e.g., A then B then C),
it is critical to use sub-VIs whenever possible.

Pre-Lab

In this pre-lab you will build two sub-VIs: source.vi and error detect.vi. You
will then plug your code into a simulator provided to you (shown in Figure 2).
The goals of this pre-lab are (1) to understand how to build and use sub-VIs
and (2) to practice using good code style.
The details of each sub-VI that you must construct are given in Tables 1
and 2. In all future labs you will be provided with an appropriate simulator.
You will typically reimplement certain blocks within the simulator and use it to
perform some kind of measurement. This incremental process will continue
throughout the course as you learn about each part of a digital communication
system. The general simulator you will use in future labs is composed of many
files. Figure 3 shows an overview of the hierarchy of the various parts of the
simulator.
It should be noted that the simulator you will be provided with will have
full functionality; however, the block diagram of the functional blocks of the
simulator will be locked. This means that although you will be able to see
how the simulator is supposed to function and look at the front panel of all the
VIs, you will not be able to see the block diagram (i.e., the implementation)
of the digital communication sub-VIs.

i
i

i
book2 2011/10/28 14:05 page 5 #17

Figure 1: Spaghetti code design can be messy.

Lab 1: Part 1 Introduction to NI LabVIEW

i
i

i
book2 2011/10/28 14:05 page 6 #18

Lab 1: Part 1 Introduction to NI LabVIEW

Figure 2: Digital communications simulator. Block diagram of awgn simple sim.vi.

i
i

i
book2 2011/10/28 14:05 page 7 #19

Lab 1: Part 1 Introduction to NI LabVIEW

Figure 3: Overview of simulator hierarchy in LabVIEW.

Table 1: Description of source.vi

source.vi - generates a random bit sequence1
INPUTS

OUTPUTS

length

I32 (32 bit

integer)

length of bit sequence

output bit
sequence

array of U8
(unsigned
bytes)

random bit sequence (array of

unsigned bytes taking value 0
or 1)

Once you have constructed the VIs for this lab (source.vi and error detect.vi), be sure to appropriately edit the icon of each VI and connect all of the
relevant terminals. Now you are ready to replace the corresponding blocks
inside the simulator provided to you (shown in Figure 2). After placing your
code in the simulator and verifying that it works correctly, observe what
happens to the bit-error rate as you decrease the noise power in the simulator.
(Note: If your subVI appears greyed out after placing it in the simulator, right
click the subVI and choose "relink subVI" to fix this issue)
1 Hint: There are a number of ways to generate a random bit sequence in LabVIEW. For
instance, you could do it using the random number generator provided in the Numeric palette,
or you could just use the Generate Bits.vi module located in the Digital palette (see Modulation
Toolkit palette).

i
i

i
book2 2011/10/28 14:05 page 8 #20

Lab 1: Part 1 Introduction to NI LabVIEW

Table 2: Description of error detect.vi

error detect.vi - computes the bit error rate of a bit sequence

INPUTS

OUTPUTS

estimated
bit
sequence

array of U8

estimated bit sequence

key bit
sequence

array of U8

correct bit sequence

bit-error
rate

DBL
(double
precision)

Hamming distance2 between

estimated sequence and key
sequence divided by length of
bit sequence

The simulator will allow you to run multiple iterations and collect statistics
from these runs. In this case, you will only be concerned with the bit-error
rate (BER) statistics (i.e., the average BER).
Plot a curve of the average bit-error rate versus 1/N0 over a number of
trials, where N0 is the noise power (realize N0 is given in dB in the simulator).
It is sufficient to vary noise power from 10 dB to 0 dB in 2 dB steps. Use
BPSK modulation and choose an input length of 200 bits. Note that in order
to observe a nonzero error for low noise powers you may need to run in excess
of 104 iterations.


linear
HINT: (X/Y ) in dB is equal to 10 log X
Ylinear , where XdB
= 10 log(Xlinear ) and YdB = 10log(Ylinear ). The ( )linear is a linear scale
value, not dB.

Pre-Lab Turn In
Submit your implementations of source.vi and error detect.vi. Turn in your
plots of average bit-error rate (BER) versus 1/N0 .
A typical example of a bit-error-rate plot is shown in Figure 4. You may
use LabVIEW, Excel, or any other graphing utility to generate these plots
(i.e., you only need to use LabVIEW to generate the data, not necessarily to
generate the plots).
2 The Hamming distance of two bit sequences is equal to the number of positions in which
the sequences differ.

i
i

i
book2 2011/10/28 14:05 page 9 #21

Lab 1: Part 1 Introduction to NI LabVIEW

Figure 4: BER plot for BPSK and QPSK modulation in AWGN channel.

Lab Experiment

There is no in-lab component for this lab.

Lab Report

After most labs you will write a lab report. You will only turn in one lab
report per lab. This means that if a lab has multiple parts (e.g., Parts 1 and 2
of Lab 1), you will only be required to turn in one lab report. The contents of
this report will be explicitly stated in the lab draft. Note: This means that you
will not be submitting a lab report for this part of Lab 1. Upon completion
of Part 2 of Lab 1, you will be required to generate a lab report.

i
i

i
book2 2011/10/28 14:05 page 10 #22

Lab 1: Introduction to Wireless Communications

Lab
Part 2: Introduction to NI RF Hardware

Summary
In this part of the introductory lab you will learn about the NI RF hardware
to be used in the Wireless Communications Lab. You will acquaint yourself
with the relevant documentation and references available to you with regard
to the RF hardware. You will build some simple example VIs from the NI
documentation and use some of the existing LabVIEW tools (e.g., RF Signal
Analyzer Demo Panel). In the final part of this lab you will explore some of
the VIs that you will use in future labs to control the RF hardware. This part
of Lab 1 introduces NI RF hardware as a tool for digital communication, and
gets you familiar with the help/documentation available for the hardware you
will be working with throughout the course.
In this lab, for the pre-lab, you are required to turn in only the questions
given in Section 2. There are no pre-lab VIs for this lab. As a part of the
lab session, you will construct two VIs by following the instructions given in
the NI documentation. The questions given in Section 3 need to be answered
during the lab only. Please do not submit the answers to these questions with
your pre-lab. At the end of this lab, you will submit a lab report, the details
of which are given at the end of this document.

RF Hardware in Wireless
Communications Lab

In this course you will use software recofigurable RF hardware from National
Instruments to build a digital communication system. This hardware can be
easily configured using LabVIEW. The RF Hardware used in the lab is the
National Instruments USRP (Universal Software Radio Peripheral). Figure 1
shows a USRP connected to a PC (running LabVIEW). This PC controls the
USRP through the gigabit ethernet cable connecting the two together. In this
section you will learn more about these modules.
10
i

i
i

i
book2 2011/10/28 14:05 page 11 #23

Lab 1: Part 2 Introduction to NI RF Hardware

11

Figure 1: Hardware setup in Wireless Communications Lab.

1.1

NI USRP (Transmit)

Lets begin by taking a look at the NI USRP when acting as a transmitter.

The USRP receives a waveform from the host PC with 16 bits of resolution
sampled at up to 25 MSamples/second. This signal is upconverted to a radio
frequency (RF) before being sent to an amplifier and then transmitted over
the air. For more information on the NI USRP, please refer to the help files
associated with the USRP (HelpNI-USRP HelpDevicesNI USRP292x Specifications)

1.2

NI USRP (Receive)

The NI USRP is also capable of receiving, and does so in the following manner. The received signal is mixed with a desired carrier frequency in order
to down-convert it to a complex IQ baseband signal sampled at 100 MSamples/second. The digital signal is then downsampled to a rate specified by
the user and passed to the host PC for processing.
When the ADC of the NI USRP samples at the full digitizer rate
(100 MSamp/sec), it can acquire a bandwidth of 20 MHz. Sampling at
such a high rate with high resolution (14 bits) produces a large amount of
data. However, to acquire a signal with smaller bandwidth (a narrowband
signal), it is sufficient to sample at a rate twice that of the signal bandwidth1 .
1 This sampling rate is known as the Nyquist rate of the system. It is the minimum sampling
rate required to reproduce a bandlimited signal from discrete samples.

i
i

i
book2 2011/10/28 14:05 page 12 #24

12

Lab 1: Part 2 Introduction to NI RF Hardware

Decimation is the process of reducing the sampling rate of a discrete-time

signal. In practice, this entails applying an anti-aliasing filter and throwing
away some samples. The NI USRP features a digital down converter (DDC),
which allows it to perform decimation in hardware instead of in software
(which can be considerably slower).
Using the DDC, the USRP can acquire narrowband signals at sample
rates much less than the full digitizer rate (100 MSamp/sec), thus reducing
the memory required to store a waveform. For example, a waveform sampled at 100 MSamp/sec for 1 second produces 100 times more data than the
same waveform sampled at 1 MSamp/sec. The effective sample rate of the
USRP (including DDC) ranges from a maximum of 25 MSamp/sec down to
200 kSamp/sec and is directly configurable by the user. For more information
on the NI USRP, please refer to the associated help files.

Pre-Lab

Review the documentation for the USRP (HelpNI-USRP Help). You will
find that the help files found under DevicesNI USRP-292x Specifications
and Devices NI USRP-2920 contain the most comprehensive information
about the USRP. For a deeper understanding of the specifications, please refer
to the following documents:
Bandwidth, Sample Rate and Nyquist Theorem [8]
Understanding RF & Microwave Specifications - Part I [14]
Understanding RF & Microwave Specifications - Part II [15]
Answer the following questions about the USRP after reading through the
help files.
1. What is the range of allowable carrier/center frequency supported by
the NI-USRP?
2. What is the maximum allowable bandwidth supported by the NI-USRP?
3. What is the maximum sampling rate of the NI-USRP?
4. Why do you think the DDC is implemented? What is its main benefit?
Answer the following general concept questions using [8], [14] and [15].
1. In your own words, describe what the bandwidth of an instrument is.

i
i

i
book2 2011/10/28 14:05 page 13 #25

Lab 1: Part 2 Introduction to NI RF Hardware

13

2. What is meant by the sampling rate of an instrument?

3. Why are these specifications important for designing a transmitter and
receiver in a wireless communications system?

Pre-Lab Turn In
Submit your answers for all of the above pre-lab questions. As a reminder,
this is all that you need to turn in for the pre-lab. There are no pre-lab VIs to
be submitted for this lab.

Lab Experiment

3.1

Building a Basic Sine Generation VI

For the first task in this lab you will build a basic sine generation VI described below. After you have completed building this example VI, verify
that the block diagram of your VI resembles Figure 2, and that the front panel
resembles Figure 3.
Use the following procedure to begin using the NI USRP to generate a
signal:
1. Select StartProgramsNational InstrumentsLabVIEW
2010LabVIEW to launch LabVIEW. The LabVIEW dialog appears.
2. Click New, choose Blank VI, and click OK to create a blank VI.
3. Display the block diagram by clicking it or selecting WindowShow
Block Diagram.
4. Navigate to the NI USRP VIs on the FunctionsInstrument I/O
Instrument DriversNI-USRPTX palette. With LabVIEW running, click here to find the NI USRP TX LabVIEW palette.
5. Create the block diagram shown in Figure 2 by placing the four core
VIs on the block diagram in the order they appear in the NI USRP RX
functions palette.
6. Hover the cursor over the device name terminal on the niUSRP Open
TX Session VI and right-click. Select CreateControl to create a
front panel field where you specify the NI USRP device name.
7. Hover the mouse tool over the IQ rate, carrier frequency, and gain
terminals of the niUSRP Configure Signal VI.

i
i

i
book2 2011/10/28 14:05 page 14 #26

Lab 1: Part 2 Introduction to NI RF Hardware

Figure 2: Block diagram of Basic Sine Wave Generation VI.

14

i
i

i
book2 2011/10/28 14:05 page 15 #27

Lab 1: Part 2 Introduction to NI RF Hardware

15

Figure 3: Front panel of Basic Sine Wave Generation VI.

8. Right-click each terminal and select CreateControl from the shortcut menu to create frequency, IQ rate, and gain controls.
9. Display the front panel by clicking it or selecting WindowShow
Front Panel. Fields are displayed in which you can specify a frequency, IQ rate, and gain.
Continuous waveform generation is controlled by means of a STOP button.
A STOP button is typically used within a While Loop. This example places a
While Loop around the Write TX Data VI so that signal generation continues
until you click STOP.
Build a While Loop and STOP button by completing the following steps:
1. Display the block diagram by clicking it or selecting WindowShow
Block Diagram.
2. Select the While Loop on the All FunctionsStructures palette. With
LabVIEW running, click here to find the While Loop on the Structures
palette.
3. Enclose the niUSRP Write TX Data (Poly) VI in the While Loop.
4. Hover the mouse tool over the Loop Condition terminal of the While
Loop.
5. Right-click the Loop Condition terminal and select Create Control
from the shortcut menu to create a STOP button on the VI front panel.

i
i

i
book2 2011/10/28 14:05 page 16 #28

16

Lab 1: Part 2 Introduction to NI RF Hardware

6. Create an error indicator by right-clicking on the error out terminal of
the niUSRP Close Session VI and selecting CreateIndicator.

Now we need to produce a waveform to be transmitted. We will be transmitting a sinusoid at a carrier frequency that is specified by the carrier frequency control. This means that the baseband signal we will be passing to
the NI USRP is simply a constant signal. To generate this signal, complete
the following steps:
1. Using the quick-drop functionality described in Lab 1, Part 1, place the
Initialize Array function down on the block diagram.
2. Right-click the element input of the Initialize Array function and select CreateConstant. Double-click on the constant that was created
and change it to 1.
3. Right-click on the dimension size input and select CreateControl.
4. Wire the output of the Initialize Array function block to the data input
of the niUSRP Write Tx Data (poly) VI.
5. Finally, wire the STOP button that was created for the While Loop to
the end of data? input of the niUSRP Write Tx Data (poly) VI and
select CDB from that VIs drop down menu.
Remember that in order to operate your sine generation VI, you must
properly configure the USRP IP Address for the USRP (an input to niUSRP
Initialize.vi). You can find the IP addresses for all of the USRP devices
connected to your PC by using the NI-USRP Configuration Utility. Once
that is properly configured, set the rest of the parameters as follows and run
the VI:
carrier frequency = 2.4 GHz
IQ rate (samples/s) = 400k
gain (dB) = 20
waveform size = 10000
active antenna = TX1
You will now test your VI using the niUSRP EX Spectral Monitoring
(Interactive) VI, which can be found in the Functions palette. This spectrum
analyzer tool will allow you to observe the frequency domain response of your
basic sine wave transmitted over a wireless link. The following directions
describe the basic operation of the Spectral Monitoring VI.

i
i

i
book2 2011/10/28 14:05 page 17 #29

Lab 1: Part 2 Introduction to NI RF Hardware

17

1. First, open the Spectral Monitoring VI. This can be accessed from
Functions PaletteInstrument I/OInstrument Drivers
NI-USRPExamplesniUSRP EX Spectral Monitoring (Interactive).vi.
2. Verify that the USRP IP Address match the values specified in the
NI-USRP Configuration Utility.
3. After opening the example VI, you will want to modify the hardware
settings on the front panel. Use the parameters listed below for this
experiment.
Center Frequency = 2.4 GHz
IQ Sampling Rate [s/sec] = 400k
Acq Duration [sec] = 50m

Questions
Answer the following questions regarding the NI USRP.
1. What is the IP Address for your NI USRP?
2. Describe what happens to the spectral response of the transmitted signal
when you place an obstruction between the antenna of transmitter and
that of the receiver (i.e., blocking the line of sight component of the
waveform).

Turn In
Demonstrate to the instructor that your basic sine generation VI is working
properly. You will need to start the Spectral Monitoring VI and show that
your basic sine generation VI produces a carrier signal at 2.4 GHz.
Submit your answers to the above questions in your lab report. More
details about the requirements for the lab report are discussed near the end of
this document.

3.2

Programming the RF Signal Analyzer

In the second part of this lab experiment you will build a Continuous IQ
AcquisitionVI and it should look like the Continuous IQAcquisitionVI shown
in Figure 4. Add all of the appropriate controls and indicators necessary to
control the sub-VIs in your code.

i
i

i
book2 2011/10/28 14:05 page 18 #30

Lab 1: Part 2 Introduction to NI RF Hardware

Figure 4: Continuous IQ Acquisition VI.

18

i
i

i
book2 2011/10/28 14:05 page 19 #31

Lab 1: Part 2 Introduction to NI RF Hardware

19

To test your acquisition VI, you will graph the real and complex parts of
the signal acquired by niUSRP Fetch Rx Data (poly).vi. To do this:
1. Place two waveform graphs on the front panel of your VI. These can be
accessed from the All Controls tool palette (All ControlsGraph
Waveform Graph.vi).
2. Retrieve the real and imaginary waveform components from the IQ
Waveform output of niUSRP Fetch Rx Data (poly).vi. Use Get Complex
IQ Component.vi to retrieve these components from IQ Waveform. This
VI can be accessed from the All Functions tool palette on the block diagram (All FunctionsModulationAnalogUtilitiesGet Complex IQ Component.vi).
3. On the block diagram, connect the terminal for each waveform graph
to its corresponding waveform for the real and imaginary parts of the
acquired signal (inside of the While-Loop structure).
You will use the basic sine generation VI created in the previous section
to test your acquisition VI. In order to do so, you must first set up the parameters for the receiving USRP. Set the input parameters for niUSRP Configure
Signal.vi listed below. Notice the carrier signal generated by your basic sine
generation VI will be offset from the center frequency of your acquisition VI,
effectively creating a 100 kHz single-sideband (SSB) modulated waveform
at baseband.
Carrier Frequency = {2.4001 GHz @ transmitter} and {2.4 GHz @
receiver}
IQ rate = 800k
gain = 20 dB
number of samples = 10000
active antenna = RX1
You should now be ready to continuously acquire a signal. After setting
the parameters described above, run both VIs and observe the in-phase and
quadrature-phase waveforms (i.e., real and imaginary parts respectively) of
the acquired signal.

Questions
Answer the following question about the NI USRP.
1. What are the three elements in the data cluster output from niUSRP
Fetch Rx Data (poly).vi? Give the name and data type of each element.

i
i

i
book2 2011/10/28 14:05 page 20 #32

20

Lab 1: Part 2 Introduction to NI RF Hardware

Turn In
Demonstrate to the instructor that your acquisition VI is working properly. It
must correctly display both I and Q phases of the acquired signal.
Submit your answers to the questions in your lab report.

3.3

Looking Ahead

For the rest of this course, VIs for RF hardware configuration and operation
will be provided to you. This will allow you to focus on the details of the
digital communications algorithms you will be implementing. Even though
you will be provided with the VIs necessary for controlling the RF hardware,
you will be expected to understand what these VIs do and how they work.
You will be using the VIs listed below in lab to configure and operate the
RF hardware. Explore the front panels and block diagrams of these VIs.
TXRF init.vi - initializes a USRP transmit session with the appropriate
RF parameters.
TXRF send.vi - executes a transmit operation (i.e., sends a complex
waveform), and closes the USRP session.
RXRF init.vi - initializes an USRP receive session with the appropriate
RF parameters.
RXRF recv.vi - acquires a complex waveform, and closes the USRP
session.

Questions
Answer the following questions about the VIs you will be using in lab. Hint:
The help files for VIs can be accessed from the block diagram. Right-click on
the VI you wish to find out more about and select Help.
1. In RXRF config.vi the IQ rate parameter is used by niUSRP Configure
Signal.vi. Can the USRP support any arbitrary IQ rate? If not, what will
the USRP do when an unsupported bandwidth parameter is requested?

Turn In
Submit all of your answers in your lab report.

i
i

i
book2 2011/10/28 14:05 page 21 #33

Lab 1: Part 2 Introduction to NI RF Hardware

21

DIRECTIONS FOR SUBMISSION OF THE LAB

REPORT
Your lab report should accomplish the following tasks.
1. Answer all of the questions asked in lab (i.e., do not reanswer pre-lab
questions).
2. Discuss problems you encountered in Lab 1, Part 1 and how you overcame these obstacles.
3. In 2-3 sentences describe the basic operation of the NI USRP when
transmitting (e.g., what are the inputs/outputs for the USRP?).
4. In 2-3 sentences describe the basic operation of the NI USRP when
receiving (e.g., what are the inputs/outputs for the USRP?).
5. Discuss problems you encountered in this part of Lab 1 and how you
overcame these obstacles.

i
i

i
book2 2011/10/28 14:05 page 22 #34

Lab 2: Baseband QAM Modem

Part 1: Modulation and Detection

Summary
In Lab 2, you will build a baseband digital modem that can use binary phaseshift keying (BPSK) or quadrature phase-shift keying (QPSK) modulation.
The complexity of the transmitter and the receiver will grow in subsequent
labs as more kinds of receiver impairments are considered. The main concepts in this lab include constellation mapping, oversampling, pulse shaping,
matched filtering, maximum-likelihood detection, and probability of error
calculation.
The system considered in this lab is illustrated in Figure 1. This modem
transmits symbols and applies a detection algorithm assuming only an additive white Gaussian noise (AWGN) channel. The transmitter maps bits to
elements of a symbol constellation. The sequence of symbols is upsampled
then filtered in discrete-time by a transmit pulse shape. The filtered sequence

Transmitter

bits

Symbol
Mapping

s[n]

gTX [n]

D/C

x(t)

x(t )

Tx

v(t)

AWGN Channel

Ex

z(t) =x(t)+v(t)

Receiver

z (t )

C/D

g RX [n]

M
M

Detection

y[n]

Inverse
Symbol Map

bits

Tz

Figure 1: The system under consideration in this lab. The transmitter uses digital
pulse-shaping and upsampling to create the transmit waveform. The baseband signal
encounters additive white Gaussian noise during transmission. The receiver uses
digital pulse-shaping and downsampling, followed by detection, to find a good guess
of the transmitted symbols.

22
i

i
i

i
book2 2011/10/28 14:05 page 23 #35

Lab 2: Part 1 Modulation and Detection

23

is then passed to the discrete-to-continuous converter. The receiver samples

the received signal and filters it with the receiver pulse shaping filter. The
filtered symbols are passed to a detection block which determines the most
likely transmitted symbol for that observation. The detected symbols are
passed to an inverse symbol mapping block to produce a good guess of the
transmitted bits.
This lab is broken up into two parts. In this first part, you will build the
modulation and decoding blocks of the modem in LabVIEW. After verifying
the correctness of your design using a simulator, you will implement your
code in lab using RF Hardware to see how your code works over a real
wireless link. The goal of this part of Lab 2 is to introduce the basic concepts
of modulation and detection. In the second part, you will implement a pulseshaping filter (on the transmitter side) and the matched filter (on the receiver
side).
You will be required to turn in two VIs (modulate.vi and demodulate.vi),
plot the BER curves for QPSK and BPSK, and answer the questions in Section 2, as a part of the pre-lab submission. The lab report for this part of the
lab will need to be submitted with the lab report in Part 2 of Lab 2 (i.e., there
is no lab report required for this part of the lab).

Background

This section provides some review of the concepts of digital modulation

and maximum likelihood detection. The emphasis is on BPSK and QPSK
modulation, with maximum likelihood detection assuming AWGN.

1.1

Modulation

Digital modulation consists of mapping a sequence of bits to a (finite) set

of analog waveforms which are suitable for transmission over a particular
channel. There are many different kinds modulation. This and subsequent
labs focus on what is known as complex pulse amplitude modulation. With
this modulation technique, symbols are modulated onto pulse-shapes.
The source for digital modulation is a sequence of bits {b[n]}. The bit
sequence is passed to the physical layer of a radio by higher layers.
The sequence of bits is processed by the symbol mapping block to produce
a sequence of symbols {s[n]}. Essentially, each value s[n] is a complex
number that comes from a finite set of symbols called the constellation and
written as
C = {s0 , ..., sM2 , sM1 } .

i
i

i
book2 2011/10/28 14:05 page 24 #36

24

Lab 2: Part 1 Modulation and Detection

The entries of the constellation are different (possibly complex) values. The
size of the constellation, or cardinality, is denoted |C| = M where M is
the number of symbols in the constellation. For practical implementation
M = 2b where b is the number of bits per symbol so that a group of b input
bits can be mapped to one symbol.
This lab considers two of the most common modulations found in commercial wireless systems.
Binary Phase Shift Keying (BPSK) has
C = {+1, 1}.
This is arguably the simplest constellation in use. The bit labeling is
provided in Table 1. BPSK is a real constellation since it does not
contain an imaginary component.
Quadrature phase shift keying (QPSK) is a complex generalization of
BPSK with
C = {1 + j, 1 + j, 1 j, 1 j, }.
Essentially QPSK uses BPSK for the real and BPSK for the imaginary component. QPSK is also known as 4-QAM. The bit labeling is
provided in Table 1.
For implementation and analysis, it is convenient to normalize the constellation to have unit energy. This means scaling the constellation such that
M1
1 
|sm |2 = 1.
M
m=0

The normalized QPSK constellation is illustrated

in Figure 2.
The symbols are pulse-shaped and scaled by Ex to produce the complex
baseband signal
x(t) =



Ex
s[n]gtx (t nT ).

(1)

n=

The scaling factor Ex is used to model the effect of adding energy or power
to x(t). The scaling factor will be added typically in the RF by controlling
the gain of the transmit power amplifier. The pulse-shape is given by the
function gtx (t). To preserve energy, the pulse-shaping function is assumed

to be normalized such that |g(t)|2 dt = 1. The symbol period is given

by T . The symbol rate is 1/T .

i
i

i
book2 2011/10/28 14:05 page 25 #37

25

Lab 2: Part 1 Modulation and Detection

Im
(00)

(01)

Re

(10)

(11)
4-QAM

Figure 2: QPSK or 4-QAM constellation with generally accepted bit to symbol

mappings based on Gray labeling.
Input Bit
0
1

Symbol
1
1

(a)
Input Bits
00
10
11
01

Symbol
1+j
1 + j
1 j
1j

(b)
Table 1: Symbol Constellation Mappings for (a) BPSK and (b) QPSK

We refer to x(t) as a complex pulse amplitude signal because complex

symbols modulate a succession of pulses given by the function gtx (t). Effectively, symbol s[n] rides the pulse gtx (t nT ). The symbol rate corresponding
to x(t) in Eq. (1) is 1/T . The units are symbols per second. Typically in wireless systems this is measured in kilo-symbols per second or mega-symbols
per second. Be careful to note that 1/T is not necessarily the bandwidth of
x(t), which depends on gtx (t). The bit rate is b/T (where recall there are
2b symbols in the constellation) and is measured in bits per second. The bit
rate is a measure of the number of bits per second the baseband waveform
x(t) carries.
Complex baseband signals are used for notational convenience. The signal x(t) is complex because the resulting wireless signal is upconverted to
ride on a carrier frequency. After upconversion, the resulting passband

i
i

i
book2 2011/10/28 14:05 page 26 #38

26

Lab 2: Part 1 Modulation and Detection

signal will be xp (t) = Re{x(t)} cos(2fc t) Im{x(t)} sin(2fc t). Upconversion is performed in the analog hardware. The signal xp (t) will be
launched from the transmit antennas into the propagation environment.
For practical implementation, the pulse-shaping is performed in discretetime. One way to implement the pulse-shaping is illustrated in Figure 1.
The idea is to use a discrete-to-continuous converter operating with a sample
period of Tx = T /L where L > 0 is an integer that denotes the oversampling
factor. The symbol sequence is upsampled by L and then filtered by the
oversampled pulse-shaping filter defined by gtx [n] := gtx (nTx ).

1.2

Detection at the Receiver

When a signal propagates through the wireless channel, it will be affected by

noise and other types of channel distortion. They type of distortion impacts
the architecture of the receiver and the design of the signal processing algorithms. In this lab, it is assumed that the transmitted signal is only impaired
by AWGN. This assumption will gradually be relaxed in future labs.
The AWGN communication channel is a good model for the impairments
due to thermal noise present in any wireless communication system. Mathematically
z(t) = x(t) + v(t),
where x(t) is the complex baseband signal, v(t) is the AWGN, and z(t) is the
observation. The assumption that v(t) is AWGN means the noise is an independent and identically distributed (i.i.d.) complex Gaussian random variable. In the presence of thermal noise, the total variance is v2 = No = kTe ,
where k is Boltzmans constant k = 1.38 1023 J/K and the effective noise
temperature of the device is Te in Kelvins. The effective noise temperature
is a function of the ambient temperature, the type of antennas, as well as the
material properties of the analog front end.
Under the assumed transmitted structure, the optimum receiver structure
involves matched filtering, sampling at the symbol rate, and detection. The
structure that will be implemented in this lab is illustrated in Figure 1. The
received signal z(t), which is assumed to be bandlimited, is sampled with
a sampling rate of 1/Tz . The sampling rate should be such that Nyquist is
satisfied. In this lab we assume that Tz = T /M where M is the integer
oversampling factor. The received signal is filtered by grx [n] = grx (nTz )
where grx (t) = gtx (t) is the matched filter. After the digital filtering, the
signal is downsampled by M to produce the symbol-rate sampled signal given
by y[n]. The
 digital filtering is performed such that y[n] is the equivalent
to y(t) = t grx ( )z(t )d sampled at nT .

i
i

i
book2 2011/10/28 14:05 page 27 #39

27

Lab 2: Part 1 Modulation and Detection

Under assumptions about the transmit and receive pulse-shapes, described

in more detail in Part 2 of Lab 2, a model for the received signal is

Ex s[n] + v[n],
(2)
y[n] =
where v[n] is i.i.d. complex Gaussian noise with Nc (0, v2 ) where v2 = No
for complex thermal AWGN. More background on matched filters and pulseshapes is provided in Part 2 of Lab 2.
A symbol detector is an algorithm that given the observation y[n] produces
the best s [n] C according to some criterion. In this lab we consider the
maximum likelihood (ML) criterion which solves
s [n] = arg max fy|s (y[n]|s[n] = s),
sC

(3)

where fy|s () is the conditional distribution of y[n] given s[n] and is known
as the likelihood function. For additive white Gaussian noise, the conditional
probability has a simple form. This allows the detection problem in Eq. (3)
to be simplified to

s [n] = arg min ln fy|s (y[n]|s[n] = s) = arg min |y[n] Ex s|2 . (4)
sC

sC

Essentially the algorithm works as follows. Given an

observation y[n], it
determines the transmitted symbol s C, scaled by Ex , that is
closest to
y[n] in terms of the euclidean distance in the error term y[n] Ex s. The
algorithm in Eq. (4) can be simplified by exploiting structure and symmetry
in the constellation.
2 + x 2 , where x
Note: If x is a complex symbol, then x2 = xreal
real
imag
and ximag are the real and imaginary parts of x respectively.

Pre-Lab

To prepare for this part of the lab you will be required to build the modulation
and decoding blocks of Figure 1. Tables 2 and 3 describe the details of the
VIs you must implement. Plug your code into transmitter.vi and receiver.vi
and run Simulator.vi (see the VI hierarchy in Figure 3) to verify functionality
and correctness. Do not use the Modulation Toolkit (MT) VIs to implement
these two VIs (i.e., do not just create a wrapper around an MT VI).
Note: From this point forward, please uniquely name your VIs so that
they do not exactly match those of the VIs in the simulator (i.e., instead of
modulate.vi, use yourname modulate.vi).
Note that both modulate.vi and decode.vi do not have an input to specify
modulation type. Modulation type is a parameter that can be found bundled
in the Modulation Parameters cluster, which has already been inserted into
the template VIs for you.

i
i

i
book2 2011/10/28 14:05 page 28 #40

28

Lab 2: Part 1 Modulation and Detection

Table 2: Description of modulate.vi

modulate.vi - maps bit sequence to symbols using appropriate constellation;
modifies symbol energy of output symbol sequence
input bit
sequence

INPUTS

OUTPUTS

modulation
type

array of U8
(unsigned
bytes)
string

Symbol
Energy

double

output
symbols

array of CDB
(complex
doubles)

input bit sequence (each array

element takes on value 0 or 1)
determines type of
constellation map (i.e., BPSK
or QPSK)
determines the per-symbol
energy of the output sequence
(default = 1)
appropriately modulated
output symbols

Table 3: Description of decode.vi

decode.vi - performs ML detection on a given symbol sequence
INPUTS

OUTPUTS

input
symbols

array of CDB
(complex
doubles)

input symbols to be decoded

estimated
bit
sequence

array of U8
(unsigned
bytes)

decoded output bit sequence

You have been provided with templates for the VIs you need to create for
this lab that already have all the inputs and outputs wired for you. What
is required of you is to finish constructing the block diagram to provide the
functionality of the VIs.
Throughout the course, several of the VIs you will create will have a
modulation parameters cluster passed into and out of them. The modulation
parameters in contains many of the parameters needed by your VIs and will
be unbundled from them. This has already been done for you in the template
VIs if necessary. Some VIs will also have modulation parameters out, so that
the cluster can be passed on through to the VIs that follow. Please ensure that
these clusters remain wired the way they are in the template VIs, as changing
how they are wired will cause VIs further down the line to break.

i
i

i
book2 2011/10/28 14:05 page 29 #41

Lab 2: Part 1 Modulation and Detection

29

Figure 3: Simulator hierarchy.

Error clusters will allow you to propagate error messages through your
code, which will help in debugging your LabVIEW code. If a subVI in your
code requires an error cluster as an input, wire error in to the appropriate
subVI before passing it on to error out; otherwise, directly connect these two
terminals together.
After you have created your subVIs properly (i.e., wiring all inputs and
outputs and editing the icon appropriately), insert them into the simulator
provided to you for this lab. To do this, first open up transmitter.vi and
receiver.vi and replace the modulate and decode subVIs with your own (refer
to Figures 5 and 6 for where to find these subVIs in the block diagrams).
Then you can open up Simulator.vi and, using the default parameters, run
the simulator to verify that your VIs operate according to the specifications
provided in Tables 2 and 3.
Using the simulator and the code you have written, plot average bit-error
rate (BER) versus signal-to-noise ratio as in Part 1 of Lab 1, but make sure to
obtain data for both BPSK and QPSK modulation1 . Remember that if signal
power is held constant, then decreasing noise power, N0 , is equivalent to
increasing SNR [i.e., you can assume unit power, giving SNR = N0 (in dB)].
Also, to observe errors at high SNR (or low N0 ) you need to run significantly
more iterations in the simulator.
1 Please ensure that the y-axis is in a logarithmic scale.

i
i

i
book2 2011/10/28 14:05 page 30 #42

30

Lab 2: Part 1 Modulation and Detection

Answer the following questions about PSK modulation:
1. PSK is a constant envelope modulation scheme. This means that all
symbols have equal energy after modulation. What is the energy of the
BPSK and QPSK modulated symbols shown in Tables 1(a) and 1(b)
respectively?
2. In modulate.vi, what should be the default value for the Symbol Energy
of the modulated sequence?
3. Consider a uniformly distributed random bit stream {bn } (i.e.,
Prob{bn = 1} = Prob{bn = 0}). Suppose {bn } is BPSK modulated
such that each symbol has energy Es . In an AWGN channel, what is
the average probability of bit error as a function of SNR when using
ML estimation? You may assume that noise is a real-valued Gaussian
random variable with variance N20 , [i.e., N (0, N20 )].
HINT: you may leave the answer in terms of the Q-function, which is
derived from the N (0, 1) distribution.

Pre-Lab Turn In
1. Submit your implementation of modulate.vi and decode.vi. Remember,
you will be penalized if you do not wire the error cluster inputs and
outputs in your code. Note: If you need to generate any additional
subVIs for this or any future labs, please submit these subVIs along
with your other pre-lab VIs.
2. Turn in a plot which compares the BER curves for BPSK and QPSK
modulation. You will need to collect average BER statistics using your
code for SNR values of 0 to 10 dB (in 2 dB increments). Remember to
use a logarithmic scale for BER and a dB scale for SNR in your plot2 .
You must use your LabVIEW code to generate the data for your plots.
3. Submit your answers to all of the questions in the pre-lab section.

Lab Experiment

In this lab you will run your implementation of modulate.vi and decode.vi
over a real wireless link. Remember to use your version of source.vi and
error detect.vi, which you have already built in Part 1 of Lab 1. Using your
ES
2 HINT: If X is a real number, then X
dB = 10 log X. Thus, SN RdB = 10 log( N ), where

ES is the average signal power (or symbol energy) and N0 is average noise power.

i
i

i
book2 2011/10/28 14:05 page 31 #43

Lab 2: Part 1 Modulation and Detection

31

code from Part 1 of Lab 1 in conjunction with the new blocks you have created,
you will complete the framework for the transmitter and receiver blocks in
lab. This framework will be provided in lab. The task of plugging your code
into this framework may seem trivial at this point, but as the course progresses
you will incrementally re-implement each block in the framework.
Figure 4 describes the hierarchy of files you will be using in lab. Rounded
rectangles in the block diagram represent files which you will be able to
alter. The files top tx.vi and top rx.vi are the top level of the transmitter and
receiver respectively. This level connects the digital communications blocks
of the transmitter and receiver (in transmitter.vi and receiver.vi) with the
VIs needed to control the NI RF hardware. You explored the RF hardware
and these control VIs in Lab 1, and will need to recall how to modify the
parameters/settings for the RF hardware using these VIs.

Figure 4: Hierarchy of code dramework used in lab.

3.1

Insert Your Code

The block diagram of transmitter.vi is shown in Figure 5. Replace source.vi

and modulate.vi with your code by right-clicking on the appropriate VI and
selecting Replace Select a VI.

i
i

i
book2 2011/10/28 14:05 page 32 #44

32

Lab 2: Part 1 Modulation and Detection

Figure 5: Block diagram of transmitter.vi.

Similarly, you will need to replace decode.vi and error detect.vi in receiver.vi with your code. The block diagram of receiver.vi is shown in Figure 6. Add error cluster appropriately.

3.2

Communication Over a Wireless Link

Before you begin testing your code over a real wireless link, set the following
baseband parameters on the front panel of top tx.vi and top rx.vi:
Packet length = 100 bits
Modulation type = QPSK
Leave all other modulation parameters set to their default value. The
following RF parameters have already been set up in the HW parameters tab
of top tx.vi and top rx.vi for you:
Carrier frequency
TX Gain
RX Gain
Trigger Level
Capture Time
The wireless digital communication system used in lab generates a small
burst of data, transmits it over a wireless link, and decodes the received signal.
To run the system:
1. Start the top level of your transmitter (i.e., run top tx.vi).
2. Wait until the Transmitting indicator is lit.
3. Start the top level of your receiver (i.e., run top rx.vi).

i
i

i
book2 2011/10/28 14:05 page 33 #45

33

Figure 6: Block diagram of receiver.vi.

Lab 2: Part 1 Modulation and Detection

i
i

i
book2 2011/10/28 14:05 page 34 #46

34

Lab 2: Part 1 Modulation and Detection

Open the front panel of top rx.vi to see the constellation of the received
signal. Note that a queue has been used to pass the key bit sequence to the
receiver for error detection. Also notice, on the front panel of top tx.vi there
is a control for the channel model parameters used by transmitter.vi. These
parameters are used to apply a desired channel model to the transmitted signal.
The real wireless channel will convolve with this artificially imposed channel
model to create a kind of equivalent channel.
Answer the following questions about the digital communication system
in lab:
1. What is the name of the queue used to pass the generated bit sequence
from top tx.vi to top rx.vi (i.e., the bit sequence needed to perform error
detection)?
2. Describe what happens to the received constellation as you increase
noise power in the top level of the transmitter, (i.e., top tx.vi).
3. Based on your observation of the received signal constellation, explain
which modulation scheme (BPSK or QPSK) performs better (on average) in noisy channels, and why. Assume that both schemes are using
the same average symbol energy.

Lab Turn In
Demonstrate to the instructor that your code is working properly. You will
need to show the instructor that both BPSK and QPSK modulation work in
your system. Please answer the questions above. You will be required to
submit these answers in a future lab report (i.e., in Lab 2, Part 2).

i
i

i
book2 2011/10/28 14:05 page 35 #47

Lab 2: Baseband QAM Modem

Part 2: Pulse Shaping and Matched Filtering
Summary
In Part 1 of Lab 2, you implemented the digital modulation and detection
parts of a baseband QAM modem. This part of the lab will deal with pulse
shaping in digital communication systems. You will implement the pulse
shaping and matched filtering blocks of the baseband QAM modem shown
in Figure 1.
Transmitter

bits

Symbol
Mapping

s[n]

gTX [n]

D/C

x(t)

x(t )

Tx

v(t)

AWGN Channel

Ex

z(t) =x(t)+v(t)

Receiver

z (t )

C/D

g RX [n]

M
M

Detection

y[n]

Inverse
Symbol Map

bits

Tz

Figure 1: The system under consideration in this lab. The transmitter uses digital
pulse-shaping and upsampling to create the transmit waveform. The baseband signal
encounters additive white Gaussian noise at the receiver. The receiver uses digital
pulse-shaping and downsampling, followed by detection, to find a good guess of the
transmitted symbols. Symbol synchronization could be included as well but is not
illustrated in the figure.

We begin by discussing the motivation and mathematical model for pulse

shaping. You will then build the pulse shaping, up-sampling, and matched
filtering blocks of the modem in LabVIEW. After verifying the correctness
of your design using a simulator, you will implement your code in lab using
the NI-USRP to see how your code works over a real wireless link. The goal
of this part of Lab 2 is to understand the motivation for pulse shaping and
observe the properties of various pulse shaping filters.
35
i

i
i

i
book2 2011/10/28 14:05 page 36 #48

36

Lab 2: Part 2 Pulse Shaping and Matched Filtering

In this lab, you will have to turn in twoVIs (pulse shaping.vi and matched filtering.vi), and answer the questions given in Section 2, as part of the pre-lab
submission. The lab reports for Parts 1 and 2 of Lab 2 have to be submitted
as a single lab report at the end of this lab.

Pulse Shaping and Matched Filtering

The objective of the course is to construct the digital signal processing blocks
necessary to operate a wireless digital communication link. A digital signal
(from the digital source, constructed in Part 1 of Lab 1) can not be transmitted
over a communication channel. The digital signal needs to be mapped to an
analog waveform that can easily be transmitted over the channel.
The model for transmission and reception considered in Lab 2 is illustrated
in Figure 1. The transmitter creates a complex pulse amplitude modulated
signal of the following form:
x(t)

Ex

s[n]gtx (t nT ).

(1)

n=

Conceptually, the mapping of bits to waveforms can be viewed in two steps.

1. Source bits to symbolsThis involves mapping the bits emitted by
the (digital) source to (possibly complex) symbols, s on a constellation,
C determined by the modulation scheme. This was accomplished in
Part 1 of Lab 2.
2. Symbols to pulsesThis involves creating an analog pulse train to
be transmitted over the wireless channel, using the complex symbols
generated in the first step. This is done using a pulse-shaping filter
gtx (t) at the transmit side.
The receiver processing depends on the type of distortion in the communication channel. Assuming an additive white Gaussian noise (AWGN)
channel, the receiver performs analogous operations as at the transmitter.
1. Pulses to symbolsIn this step, the symbols are recovered from the
received pulses using a matched filter at the receive side.
2. Symbols to bitsThis involves mapping the symbols received to bits
(depending on the modulation scheme chosen), using some form of a
detection technique. This was accomplished in Part 1 of Lab 2 using
the maximum likelihood detection algorithm.

i
i

i
book2 2011/10/28 14:05 page 37 #49

Lab 2: Part 2 Pulse Shaping and Matched Filtering

37

To implement the maximum likelihood detector as described in Part 1 of

Lab 2 using single-shot detection, the pulse shaping filters gtx (t) and grx (t)
have to satisfy some properties.

The transmit pulse shape is normalized such that |gtx (t)|2 dt = 1.
This is primarily for convenience in our notation to avoid additional
scaling factors.
(t), which simplifies to g (t) =
The receive filter grx (t) = gtx
rx

gtx (t) since the pulse-shaping function is usually real. This means
that the receive filter is a matched filter. Effectively the filtering by grx (t)
becomes a correlation with gtx (t) thanks to the relationship between
convolution and correlation. The choice of a matched filter maximizes
the received signal-to-noise ratio and thus gives the maximum likelihood detector its best performance.

The composite filter g(t) = grx ( )gtx (t )dt is a Nyquist pulse
shape. This means that it satisfies g(nT ) = [n]. The Nyquist pulseshape property eliminates intersymbol interference, allowing the detector to take a very simple form as described in Part 1 of Lab 2.

Defining the bandwidth of the transmitted signal x(t) requires some mathematical subtleties. From the perspective of measuring the power spectrum
of a signal, the following definition of power spectrum for the transmitted
signal is sufficient:
(2)
Px (f ) Ex |Gtx (f )|2
where it is assumed that the transmit constellation is normalized to unit energy.
It is clear that |Gtx (f )|2 determines the bandwidth of x(t), where Gtx (f ) is
the frequencyresponse of the pulse-shape gtx (t). By normalizing the transmit
pulse shape, f Px (f )df = Ex where Ex is known as the symbol energy
and Ex /T as the power.
A simple choice for gtx (t) is the rectangular pulse-shape. Unfortunately,
such a pulse-shape would have poor spectral properties, since the Fourier
transform of a rectangular function is a sinc function. This means that a lot
of bandwidth would be required to transmit the signal.
Another choice for gtx (t) is the sinc function. A sinc pulse has ideal
)
spectral properties. For example, if gtx (t) = sin(t/T
then Gtx (f ) = T
t/T
for f [1/2T , 1/2T ] giving an absolute bandwidth of 1/2T . This is
the smallest bandwidth a complex pulse amplitude modulated signal could
have with symbol rate 1/T . Unfortunately the sinc pulse shaping filter has a
number of problems in practice. Ideal implementations of g(t) do not exist in
analog. Digital implementations require truncating the pulse shape, which is a
problem since it decays in time with 1/t. Further the sinc function is sensitive

i
i

i
book2 2011/10/28 14:05 page 38 #50

38

Lab 2: Part 2 Pulse Shaping and Matched Filtering

to sampling errors (not sampling at exactly the right point). For these reasons
it is of interest to consider pulse shapes that have excess bandwidth, which
means the double sided bandwidth of the pulse-shape is greater than 1/T .

1.1

A Common Pulse Shape

The most common Nyquist pulse, besides the sinc pulse, is the raised cosine.
The raised cosine pulse-shape has Fourier spectrum

T,



T
1 + cos T
|f |
Grc (f ) =
2

0,

1
2T

0 |f | (1 )/2T
,

1
2T

|f | >

1+
2T
1+
2T

and transform
grc =

sin t/T cos( t)

.
t/T 1 4 2 t 2 /T 2

The parameter is the rolloff factor, 0 1 (sometimes is used

instead of ) and is often expressed as percentage excess bandwidth. In
other words, 50% excess bandwidth would correspond to = 0.5. At large
values of the tails of the time-domain pulse decay quickly, but the spectrum
roll off is smoother (i.e., it requires more excess bandwidth). Reducing the
time-domain tails of the pulse shape reduces the amount of ISI introduced by
sampling error. Thus rolloff factor can be used to tradeoff excess bandwidth
for resistance to ISI caused by sampling errors. The raised cosine pulse is
illustrated in Figure 2.
The transmitter does not use the raised cosine pulse directly. The reason is
that the raised cosine pulse shape corresponds to g(t) while
 we use gtx (t) at the
transmitter and grx (t) at the receiver. Recall the g(t) = grx ( )gtx (t )d
and that grx (t) = gtx (t). In the frequency domain this means that G(f ) =
Gtx (f )Gtx (f ). Consequently we choose gtx (t) to be a square root of the
raised cosine. Such a pulse shape is known as a square root raised cosine, or
a root raised cosine and is written
gsqrc (t) =

sin[(1)t/T ]
4 cos [(1 + ) t/T ] +
4t/T
.

1 (4t/T )2
T

Of particular interest note that gsqrc (t) is even, thus if gtx (t) = gsqrc (t) then
grx (t) = gtx (t) = gtx (t) and the transmit pulse shape and receive pulse
shape are identical. The square root raised cosine is its own matched filter!

i
i

i
book2 2011/10/28 14:05 page 39 #51

Lab 2: Part 2 Pulse Shaping and Matched Filtering

39

Figure 2: Spectral response of the raised cosine pulse shaping function g(t).

1.2

Practical Implementation of Pulse Shaping

Recall that the Nyquist theorem states it is sufficient to sample a given signal at
a sampling frequency is at least twice the largest frequency in the signal, to be
able to reconstruct the signal effectively. Sampling at a rate higher than what
is required by the Nyquist theorem is known as oversampling. Oversampling
at the analog front ends of a digital communication system (DAC and ADC)
will make the design of all the filters much simpler. The increased bandwidth,
due to a larger sampling frequency, will relax the sharp cutoff requirements on
the filters needed to remove aliasing effects. Oversampling in digital systems
is also important for efficient synchronization.
The upsampling and downsampling processes are critical in order to reap
the benefits of oversampling. Upsampling a sequence {s[n]} by a factor L is
equivalent to inserting L 1 zeros between every sample of the sequence.
Downsampling a sequence {y[n]} by a factor M is equivalent to throwing
away M 1 out of every M symbols (i.e., ydownsamp [k] = y[kM]). Upsampling and downsampling are used to facilitate the digital implementation
of gtx (t) and grx (t).

i
i

i
book2 2011/10/28 14:05 page 40 #52

40

Lab 2: Part 2 Pulse Shaping and Matched Filtering

Simple explanations of transmit and receive pulse shaping using the concepts of upsampling and downsampling are provided in this lab. More efficient implementation computationally is possible using filter banks but that
is not required to complete the assignment.
Transmit Pulse Shaping
The challenge with implementing transmit pulse shaping is that the symbol
rate may be less than the sample rate required for pulse shapes with excess
bandwidth. For example, for a square root raised cosine pulse shape, the
Nyquist rate is (1 + )/T while the symbol rate is only 1/T .
Let Tx be some sampling period such that 1/Tx is greater than twice the
maximum frequency of gtx (t). For simplicity we take Tx = T /L. Other
choices of Tx would require a resampling operation. The continuous-time
complex baseband signal is

x(t) =



Ex
s[n]gtx (t nT ).

(3)

n=

Since x(t) is bandlimited by virtue of the bandlimited pulse shape gtx (t),
there exists a sequence {c[n]} such that
x(t) =



sin((t nTx )/Tx )
Ex
c[n]
,
(t nTx )/Tx
n=

where c[n] = x(nTx ). Substituting Eq. (3) gives

c[n]

=
=


m=

s[m]gtx (nTx mT )
s[m]gtx [n mL],

(4)

m=

where gtx [n] = gtx (nTx ). Using multi-rate signal processing identities, it
can be shown that Eq. (4) is the result of upsampling s[m] by L, then filtering
with gtx [n].
Receiver Pulse Shaping
It is easier to justify the digital implementation of receiver pulse shaping.
Let z(t) denote the complex baseband input to the continuous-to-discrete
converter. Assume that z(t) has already been bandlimited by the RF in the

i
i

i
book2 2011/10/28 14:05 page 41 #53

Lab 2: Part 2 Pulse Shaping and Matched Filtering

41

analog front end. Let Tz = T /M for some integer M such that 1/Tz is
greater than the Nyquist rate of the signal. This is known as oversampling.
Let z[n] := z(nTz ). Using results on filtering bandlimited signals, it can be
shown that



y[n] = grx ( )z(nT ) = Tz
z[m]grx ((nM m)Tz ) (5)
m=

z[m]grx [nM m],

(6)

m=

where grx [n] = Tz grx (nTz ). Using multi-rate signal processing identities, it
can be shown that Eq. (6) is the result of downsampling the filtered oversampled signal z[n] by grx [n].

1.3

Pulse Shaping in LabVIEW

LabVIEW provides many useful digital signal processing tools. In this

lab, you will use MT Generate Filter Coefficients.vi, a Modulation Toolkit
VI found in the Modulation palette (FunctionAddonsModulation
DigitalUtilitiesMT Generate Filter Coefficients.vi). This Modulation Toolkit VI allows you to generate a pulse shaping filter (e.g., root raised
cosine) and its associated matched filter based on a set of input parameters.
You will need to use this VI to build the pulse shaping and matched filtering
blocks of Figure 1. The inputs to this VI [modulation type, pulse shaping
filter, filter parameter and filter length (symbols)] can be obtained from the
pulse shaping parameters available in the modulation parameters cluster.
As a note about code style in LabVIEW, when building a VI requiring
many inputs/outputs (i.e., about five or more of each), you should bundle
related inputs/outputs into clusters. Clusters, found in the Cluster palette,
allow you to group together terminals of different datatypes. Please refer
to [12] for more information on how to build a cluster or access the elements
of a cluster. For an example of a cluster you will be using in the lab, please
look at the pulse shaping parameters terminal, which can be accessed from
the front panel of top tx.vi.

Pre-Lab

To prepare for this part of the lab you will be required to build the pulse
shaping and matched filtering blocks of Figure 1. Tables 1 and 2 describe
the details of the VIs you must implement. You will plug your code into
receiver.vi and transmitter.vi and run the simulator provided to you in order

i
i

i
book2 2011/10/28 14:05 page 42 #54

42

Lab 2: Part 2 Pulse Shaping and Matched Filtering

Table 1: Description of pulse shaping.vi

pulse shaping.vi - oversample input and apply transmit pulse shaping filter
input

array of CDB
(complex
doubles)

input symbol sequence

output

array of CDB

oversampled input convolved

with pulse shaping filter

INPUTS

OUTPUTS

Table 2: Description of matched filtering.vi

matched filtering.vi - apply matched filter corresponding to pulse shaping parameters

INPUTS

OUTPUTS

input
complex
waveform

IQ Waveform
cluster1

received sequence

output
complex
waveform

IQ Waveform
cluster

received sequence convolved

with matched filter

to verify functionality and correctness. Please name your VIs as per the
convention yourname VIname.vi.
You have been provided with templates for the VIs you need to create for
this lab that already have all the inputs and outputs wired for you. What
is required of you is to finish constructing the block diagram to provide the
functionality of the VIs.
Throughout the course, several of the VIs you will create will have a
modulation parameters cluster passed into and out of them. The modulation
parameters in contains many of the parameters needed by your VIs and will
be unbundled from them. This has already been done for you in the template
VIs if necessary. Some VIs will also have modulation parameters out, so that
the cluster can be passed on through to the VIs that follow. Please ensure that
these clusters remain wired the way they are in the template VIs, as changing
how they are wired will cause VIs further down the line to break.
Figure 3 shows the dependencies between the files that make up the simulator. The top level of this simulator, simulator.vi, connects top tx.vi to
1 IQ Waveform clusters contain three elements: t0, dt, and Y. Please recall from Part 2 of
Lab 1 what these elements are and what their data types are.

i
i

i
book2 2011/10/28 14:05 page 43 #55

Lab 2: Part 2 Pulse Shaping and Matched Filtering

43

Figure 3: Hierarchy of code framework for new simulator.

top rx.vi and provides each with the appropriate inputs. The parts of the
simulator you will be modifying are located in transmitter.vi and receiver.vi
shown in Figures 4 and 5 respectively.
You will be putting your VIs into transmitter.vi and receiver.vi, replacing
the locked versions that are already there. After doing this, you will then
open up simulator.vi, that you will use to confirm your VIs operate correctly
before implementing them on the NI-USRP.
Notice that pulse shaping.vi and matched filtering.vi do not take any parameters as inputs. All of the pulse shaping and oversampling parameters you
need to use for these VIs can be accessed from the modulation parameters in
cluster. After building these VIs, replace the existing code in the simulator
with your code. Replace a subVI in the transmitter or receiver with your code

Figure 4: Block diagram of transmitter.vi.

i
i

i
book2 2011/10/28 14:05 page 44 #56

Lab 2: Part 2 Pulse Shaping and Matched Filtering

Figure 5: Block diagram of receiver.vi.

44

i
i

i
book2 2011/10/28 14:05 page 45 #57

Lab 2: Part 2 Pulse Shaping and Matched Filtering

45

by right-clicking on the appropriate VI and selecting ReplaceSelect a VI.

Remember to wire error clusters where appropriate.
As described in the previous section, use MT Generate Filter Coefficients.vi
from the Modulation Toolkit to generate your pulse shaping and matched
filters. After generating each filter, you can apply the filter by convolving it
with the complex array component of the appropriate input waveform. Do
not use a Modulation Toolkit VI to apply either the pulse shaping filter or
matched filter.
In LabVIEW, Convolution.vi, which is located in the Signal Processing tool
palette
(FunctionsSignal
ProcessingSignal
Operation
Convolution.vi), supports complex convolution. You can use this VI when
creating your subVIs for this lab and others.
In summary:
Use MT Generate Filter Coefficients.vi to generate the pulse shaping
filter and its associated matched filters. Do not use Modulation Toolkit
VIs to apply these filters. Instead, use convolution, and remember to
make any necessary changes needed for complex convolution.
The modulation parameters cluster contains many of the parameters
needed for pulse shaping, including pulse shaping parameters, TX oversample factor, and RX oversample factor.
After building pulse shaping.vi and matched filtering.vi, insert them
into the simulator provided to you to verify functionality and correctness.

Using an Eye Diagram as an Analytical Tool

In this lab you will use an eye diagram to make some qualitative statements
about the received signal. An eye diagram is constructed by taking a time
domain signal and overlapping traces of a certain number of symbol times.
For example, if Ts = 10 sec is the symbol time for a given waveform, an
eye diagram for an eye length of two would be generated by overlaying traces
of duration 2Ts starting at an ideal sampling point. In the absence of noise,
these overlapped signals might look something like Figure 6.
The eye diagram of Figure 6 shows the overlapped traces of a BPSK
modulated bit stream using a sinc pulse for an eye length of two. This eye
diagram can give us some qualitative insight about our received signal. For
example, the opening of the eye indicates the period of time over which
we can successfully sample the waveform (i.e., without suffering from ISI).
Along with indicating the amount of ISI in the system, the distortion of the
signal in the diagram also indicates various qualitative parameters, such as

i
i

i
book2 2011/10/28 14:05 page 46 #58

46

Lab 2: Part 2 Pulse Shaping and Matched Filtering

Figure 6: Eye diagram for sinc pulse shape in the absence of noise.

the variation of zero crossings or distortion related to SNR. Please refer to

[6] and [1] for more detailed discussions about the use of the eye diagram as
a tool for qualitative analysis in digital communications.

Questions
Answer the following questions, and submit the answers as part of your prelab.
1. MT Generate Filter Coefficients.vi requires an input for the number
of pulse shaping samples per symbol. This tells the VI the oversample factor being used to design the filter. When using this VI in
pulse shaping.vi, where do you obtain this parameter from? Give any
relevant cluster names and/or variable names if appropriate.
2. According to [1], impairments visible in the eye diagram can occur
at many places along the communication path (i.e., from source to
sink). What are some of the places where impairments visible in the
eye diagram might occur (name at least three)?
3. When using inconsistent sampling rates at the transmitter and receiver,
it is important to remember that the symbol rate (1/Ts ) must remain
constant.
(a) In terms of the sampling rates at the transmitter and receiver
(1/TM and 1/TN respectively), what is the relation between the
oversample factor at the transmitter (M) and receiver (N )?
T
(b) What can you say about the type of number the ratio of M
TN
(i.e., the ratio of the sampling rate of the receiver to that of the
transmitter) must be?
Hint: Note that M and N must be positive integers.

i
i

i
book2 2011/10/28 14:05 page 47 #59

Lab 2: Part 2 Pulse Shaping and Matched Filtering

47

Pre-Lab Turn In
1. Submit your implementation of pulse shaping.vi and matched filtering.vi. Remember, you will be penalized if you do not wire the error
cluster inputs and outputs in your code.
Note: If you need to generate additional subVIs for this or any future
labs, please submit these subVIs along with your other pre-lab VIs.
2. Submit your answers to all of the questions in the pre-lab section.

Lab Experiment

In this lab you will run your implementation of pulse shaping.vi and matched
filtering.vi over a real wireless link. Using your code from the prior labs in
conjunction with the new blocks you have created, you will complete the
framework for the transmitter and receiver blocks in this lab. This framework
will be provided to you as in the Part 1 of Lab 2.
Figure 7 describes the hierarchy of files you will use in this lab. Rounded
rectangles in the block diagram represent files you will be able to alter. The
files top tx.vi and top rx.vi are the top level of the transmitter and receiver
respectively. This level connects the digital communications blocks of the
transmitter and receiver (in transmitter.vi and receiver.vi) with the VIs needed
to control the NI-USRP. You explored the USRP and these control VIs in
Lab 1, and will need to recall how to modify the parameters/settings for the
USRP using these RF hardware VIs.

Figure 7: Hierarchy of code framework used in lab.

i
i

i
book2 2011/10/28 14:05 page 48 #60

48

Lab 2: Part 2 Pulse Shaping and Matched Filtering

3.1

Communication Over a Wireless Link

Before you begin testing your code over a real wireless link, set the following
baseband parameters on the front panels of top tx.vi and top rx.vi:
Packet length = 200 bits
Modulation type = QPSK
Pulse shaping filter = Root Raised Cosine
Filter parameter = 0.5
Filter length = 8
Also make sure that the following RF parameter is set up appropriately
(using the HW parameters accessible on the front panel of top rx.vi):
Capture time = 2 msec
Let us digress now to clarify the operation of top rx.vi, the top level of
the receiver. The receiver has two states: (1) listen and (2) receive. Figure 8
shows a finite state diagram of the operation of the receiver. The RXRF recv.vi
features a digital triggering subVI. When this feature is enabled, the digitizer
will continually listen to the received signal until the energy of the incoming
signal crosses a predefined trigger level. Once this trigger occurs, the NIUSRP captures a window of samples, which it then passes to receiver.vi.
After processing the captured signal, the receiver returns to the listening
state. The receiver will periodically timeout while in the listen state to update
RF configuration parameters.
Now you are ready to test your code over a real wireless link. The wireless
digital communication system used in the lab generates a small burst of data,

Figure 8: State diagram for receiver implemented by top rx.vi.

i
i

i
book2 2011/10/28 14:05 page 49 #61

Lab 2: Part 2 Pulse Shaping and Matched Filtering

49

transmits it over a wireless link, and decodes the received signal. To run the
system:
1. Start the top level of your transmitter (i.e., run top tx.vi).
2. Wait until the Transmitting indicator is lit.
3. Start the top level of your receiver (i.e., run top rx.vi).
4. Open the front panel of top rx.vi to see the constellation and eye diagram of the received signal.
Note that a queue has been used to pass the key bit sequence to the receiver
for error detection. Also notice on the front panel of top tx.vi there is a control
for the channel model parameters used by transmitter.vi. These parameters
are used to apply a desired channel model to the transmitted signal. The real
wireless channel will convolve with this artificially imposed channel model
to create a kind of equivalent channel.
After you have verified the correct operation of your code using the parameters above, vary the input parameters to the system and observe how these
changes affect the eye diagram of the received signal. Some of the parameters
you might vary include noise power, pulse shape, filter parameter, or channel
model (e.g., you may enter your own channel impulse response using the ISI
channel model).

3.2

Questions

Answer the following questions about the digital communication system:

1. Using the pulse shaping parameters control on the front panels of
top tx.vi and top rx.vi, vary the filter parameter (i.e., the rolloff factor)
of a root raised cosine pulse from 0 to 1 in increments of 0.20. Observe
how the eye diagram (on the front panel of top rx.vi) changes as you
vary this excess bandwidth parameter.
(a) Describe what happens to the opening of the eye as you increase
the excess bandwidth parameter from 0 to 1.
(b) Explain what impairment is being added or removed from the
received signal as you vary this parameter.
2. Based on what you have learned from [1], and from varying the pulse
shaping parameters, what is the relation between the excess bandwidth
parameter of the root raised cosine pulse and inter-symbol interference?
Does ISI increase or decrease with increasing and why.

i
i

i
book2 2011/10/28 14:05 page 50 #62

50

Lab 2: Part 2 Pulse Shaping and Matched Filtering

Lab Turn In
Demonstrate that your code is working properly. You will need to show that
your pulse shaping VIs can work with a variety of pulse shaping parameters.
Also answer the questions above. You will be required to submit these answers
in your lab report.

DIRECTIONS FOR LAB SUBMISSION

Your lab report should accomplish the following tasks.
1. Answer all questions from Parts 1 and 2 of Lab 2 (not including pre-lab
questions). Note that this also includes the questions that you answered
during the lab session.
2. Discuss any problems you might have encountered in either Part 1 or 2
of Lab 2 and how you overcame these obstacles.
3. Describe how the code you have developed in lab for modulate.vi and
decode.vi might be extended or generalized to be used with an arbitrary
constellation (i.e., if you were given some arbitrary constellation C
as a 1-D array of complex doubles in LabVIEW, explain how you
would modify your code for modulate.vi and decode.vi to use this new
constellation). Please only explain details of your code that pertain
to your implementation of algorithms for symbol mapping, energy
normalization, and maximum-likelihood detection. You do not have to
construct the VI, only an explanation is needed.
4. The background section of this lab stated that two main goals for pulse
shaping are:
To reduce the energy of spectral sidelobes (i.e., excess bandwidth
requirements).
To reduce the time-domain pulse tails which can lead to intersymbol interference.
Are these conflicting or complimentary goals and why?

i
i

i
book2 2011/10/28 14:05 page 51 #63

Lab 3: Synchronization:
Symbol Timing Recovery in Narrowband
Channels

Summary
In this lab you will consider the problem of symbol timing recovery also
known as symbol synchronization. Timing recovery is one of several synchronization tasks; others will be considered in future labs.
The wireless communication channel is not well modeled by simple additive white Gaussian noise. A more realistic channel model also includes
attenuation, phase shifts, and propagation delays. Perhaps the simplest channel model is known as the frequency flat channel. The frequency flat channel
creates the received signal
z(t)

ej x(t d ) + v(t),

(1)

where is an attenuation, is a phase shift, and d is the delay.

The objective of this lab is to correct for the delay caused by d in discretetime. The approach will be to determine an amount of delay k and then to
delay the filtered received signal by k prior to downsampling. This will
modified the receiver processing as illustrated in Figure 1.
Two algorithms will be implemented for symbol synchronization in this
lab: the maximum energy method and the Early Late gate algorithm. The
maximum energy method attempts to find the sample point that maximizes
the average received energy. The earlylate gate algorithm implements a
discrete-time version of a continuous-time optimization to maximize a certain

z (t )

C/D
Tz =

zk

g RX [n]

M
k

T
M

Symbol
Sync

Figure 1: Receiver with symbol synchronization after the digital matched filtering.

51
i

i
i

i
book2 2011/10/28 14:05 page 52 #64

52

Lab 3: Synchronization

cost function. For the best performance, both algorithms require that there is
a lot of oversampling (i.e., M is large).
You will build two VIs, one for each of the two algorithms. After verifying
the correctness of your design using a simulator, you will implement your
code in lab using the NI-USRP hardware to see how your code works over a
real wireless link. The goal of this lab is to understand the need for symbol
timing recovery and the performance of two common symbol timing recovery
algorithms.
In this lab, you will have to turn in two VIs (align MaxEnergy.vi and
align ELgate.vi) and submit the answers to the questions in Section 2, along
with the plot depicting timing error statistics, as part of the pre-lab submission.

Background

This section provides some background on the timing recovery problem and
reviews the two algorithms that will be implemented in this lab. Remember
that you will be implementing these synchronization methods using a purely
digital approach.

1.1

Introduction to Symbol Timing Recovery

In the presence of frequency-flat fading as described in Eq. (1), the input to

the receiver has the form
 
z(t) = ej Ex
s[m]gtx (t mT d ) + v(t).
(2)
m

Now let g(t) be the convolution of gtx (t) and grx (t). With the model in Eq. (2)
after matched filtering and sampling, the received signal is


s[m]g((n m)T d ) + v[m].
y[n] = Ex ej
m

Several different impairments are possible depending on the value of

the d . Unless otherwise specified, Nyquist pulse shapes are assumed.
First consider the case where d is a fraction of a symbol period thus
0 < d < T . This can model the effect of sample timing error, or not
sampling at the right point. Under this assumption



y[n] = Ex ej s[n]g(d ) + Ex ej
s[m]g((n m)T d ) + v[n] .




m=n
desired

 noise

ISI

i
i

i
book2 2011/10/28 14:05 page 53 #65

53

Lab 3: Synchronization

Intersymbol interference is created when the Nyquist pulse shape is not sampled exactly at nT. You will need symbol synchronization or equalization to
mitigate the effect of sample timing error.
Second, suppose that d = dT for some integer d. This is the case where
symbol timing has been resolved but an unknown propagation delay, which
is a multiple of the symbol period, remains. Under this assumption


y[n] =
Ex ej
s[m]g((n m)T d ) + v[m]
=
=




Ex e

s[m]g((n m)T dT ) + v[m]

Ex ej s[n d] + v[m].

Essentially integer offsets create a mismatch between the indices of the transmitted and receive symbols. You will need frame synchronization to find the
beginning of the data.
Third, suppose that the time error has been removed from the channel;
equivalently d = 0. Then

y[n] = Ex ej s[n] + v[m].
Sampling and removing delay leaves distortion due to and , which are
unknown to the receiver. The amplitude and phase shift will be estimated and
equalized in Lab 4 (the focus of this lab is on the unknown delay d ).

1.2 The Maximum Output Energy Solution

Let y(t) be the matched filtered version of z(t) in continuous-time. Given an
advance , notice that the output energy function defined as

J ( ) = E |y (nT + )|2 = Ex
|g (mT + d ) |2 + v2
m

Ex |g(0)|2 + v2 .

The maximum of the function J ( ) occurs when d is an integer multiple

of the symbol rate. This value of is known as the maximum output energy
solution. The resulting value of by that maximizes J ( ) is known as the
maximum output energy solution. In this lab we implement two algorithms
for finding a maximum output energy solution in discrete-time. Other criteria
for performing symbol synchronization are possible, for example maximum
likelihood solutions. The approach described here is robust to additive noise,
certain kinds of fading channels, and small carrier frequency offsets.

i
i

i
book2 2011/10/28 14:05 page 54 #66

54

Lab 3: Synchronization

You will implement a digital version of the maximization of J ( ). This

means that you will perform the estimation and correction after the digital
matched filter at the receiver, prior to downsampling.

1.3

Direct Maximization of the Output Energy in

Discrete Time

Let
r[n] =

z[m]gtx [n m]

be the output of the matched receiver filter. The discrete-time output energy
can be calculated as
J [k] = E |r (nMT + k)|2 ,
where k is the sample offset between 0, 1, . . . , M 1 corresponding to an
estimate of = kT /M. The expectation can be replaced with a time average
over P symbols to create the function
Japprox [k] =

P 1
1 
|r (pMT + k)|2 .
P

(3)

p=0

The direct maximum output energy solution is given by k = max Japprox [k].
Because the receiver is only performing symbol synchronization, it is sufficient to evaluate Japprox [k] for values k = 0, 1, . . . , M 1. The resulting
value of k would be implemented in the advance operation in Figure 1 given

in the Z-domain as zk . Instead of an advance, a delay could be used instead

with value zMk . Larger values of P generally give better performance.

1.4

Indirect Maximization of the Output Energy in

Discrete Time

An alternative approach, useful for adaptive implementations, involves differentiating the cost function and discretizing the result. The approach here
follows the description from [6]. Differentiating the expectation and assuming they can be interchanged


d
d
2
|y (nTs + )|
Jopt ( )  E
d
d





= E y(nT + ) y (nT + ) +E y (nT + ) y(nT + )

 


= 2Re E y(nT + ) y (nT + ) .

i
i

i
book2 2011/10/28 14:05 page 55 #67

Lab 3: Synchronization

55

Using a first-order difference approximation

y (nT + )  y (nTs + + ) y (nTs + ) .

Now as before replacing the expectation with a time average

P 1



d
1 
Jopt ( ) 
2Re y (nTs + ) y (nTs + +)y (nTs + ) .
d
P
n=1

Choose to be a multiple of T /M. Using r[n] as defined before let

J [k] =

P
1





2Re r [nP + k] r [nP + k + ] r [nP + k ] . (4)

n=0

What we call the indirect maximum output energy solution is given by k =

maxk=0,1,...,M1 J [k]. A typical value of is 1.

Pre-Lab

Please answer the following questions about symbol timing recovery:

1. Show that in the absence of noise, and in Eq. (2) do not have any
impact on the maximum output energy solution.
2. What are the two critical assumptions used to formulate the indirect
maximization of the output energy? Consider how the presence of the
flat fading channel AWGN can impact this method. Specifically, using
at least one of the critical assumptions, explain how you might mitigate
the impact of these impairments by suitable selection of parameters.
3. After downsampling a sequence originally sampled at rate 1/Tz by a
factor M, what is the sample period of the resulting signal?

Programming
In this pre-lab you will generate LabVIEW code to estimate the direct maximization of the output energy in discrete time and the earlylate gate algorithm. Tables 1 and 2 describe the details of the VIs you must implement.
Please name your VIs as per the convention yourname VIname.vi.

i
i

i
book2 2011/10/28 14:05 page 56 #68

56

Lab 3: Synchronization

Table 1: Description of align MaxEnergy.vi

align MaxEnergy.vi - compute sampling offset by maximizing Japprox [k] and
correct for offset

INPUTS

OUTPUTS

input
complex
waveform

IQ Waveform
cluster

received sequence after

matched filtering

output
complex
waveform

IQ Waveform
cluster

received sequence after

symbol timing recovery

offset

I32 (integer)

computed offset

Table 2: Description of align ELgate.vi

align ELgate.vi - estimate timing offset by minimizing J [k] and correct for offset

INPUTS

OUTPUTS

input
complex
waveform

IQ Waveform
cluster

received sequence after

matched filtering

output
complex
waveform

IQ Waveform
cluster

received sequence after

symbol timing recovery

offset

I32 (integer)

computed offset

You have been provided with templates for the VIs you need to create for
this lab that already have all the inputs and outputs wired for you. What
is required of you is to finish constructing the block diagram to provide the
functionality of the VIs.
Throughout the course, several of the VIs you will create will have a
modulation parameters cluster passed into and out of them. The modulation
parameters in contains many of the parameters needed by your VIs and will
be unbundled from them. This has already been done for you in the template
VIs if necessary. Some VIs will also have modulation parameters out, so that
the cluster can be passed on through to the VIs that follow. Please ensure that
these clusters remain wired the way they are in the template VIs, as changing
how they are wired will cause VIs further down the line to break.

i
i

i
book2 2011/10/28 14:05 page 57 #69

Lab 3: Synchronization

57

Figure 2: Hierarchy of code needed for symbol timing recovery in lab.

You will plug your code into the baseband simulator provided to you in
order to verify functionality and correctness. You will not be replacing one of
the main blocks inside receiver.vi as you have done in previous labs. Instead,
you will need to insert your code into symbol timing.vi, a subVI in the digital
communications library built for this course. Figure 2 describes the hierarchy
of the files needed for symbol timing recovery in this lab.
Figure 2 shows the block diagram of synchronize.vi, where you can find
symbol timing.vi. Notice there are a number of synchronization methods
available in this VI, however you will only be dealing with Timing Estimation (the option for performing symbol timing recovery independent of
frame synchronization or channel estimation).
After replacing align ELgate.vi and align MaxEnergy.vi in symbol timing.vi, you should test your code by adding a small delay to your channel.
Delay in the channel model can be modified using the channel model parameters control on the front panel of Simulator.vi. Note that delay in the channel
model is bounded in channel.vi (i.e., you will not be able to add an arbitrarily
large delay to the channel). You can use the front panel controls to modify
the symbol timing recovery method used in the simulator. After choosing the
STR method and modifying the delay, you should notice the estimated offset
generated by the receiver changes as you vary delay. This integer offset value
includes delay introduced by pulse shaping and matched filtering.
The simulator outputs the estimated delay of the channel (i.e., after correcting for any offset caused by filtering) as an error statistic. The VI simulator.vi

i
i

i
book2 2011/10/28 14:05 page 58 #70

Lab 3: Synchronization

Figure 3: Block diagram of synchronize.vi.

58

i
i

i
book2 2011/10/28 14:05 page 59 #71

59

Lab 3: Synchronization

compares this estimated delay with the actual delay and computes the mean
square error static, which has been normalized by symbol period Ts

2 
[N ] = E ( (N ) d )/Ts  ,
(5)
where N is the oversample factor at the receiver, (N ) is the estimated delay
of the channel computed at the receiver1 , and d is the actual delay of the
channel.
Plot a graph of the timing error statistic (N ) versus oversample factor N ,
using the parameters below. Since your channel will have no noise, you will
only need to run a single iteration of the simulator to gather the timing error
statistic. Note when modifying oversample factor N , symbol period Ts must
remain constant (i.e., since Ts = N TN , you must also modify the sampling
rate at the receiver appropriately).
Delay d = 0.34Ts
Oversample factor N = 2, 4, 6, . . . , 20
Channel Model: h = 0.75ej /4 (i.e., narrowband channel)
Noise Power = dB
In summary:
After building align ELgate.vi and align MaxEnergy.vi, place these VIs
into your simulator (i.e., inside symbol timing.vi).
Modify the symbol timing recovery method and the delay introduced
by the channel from the front panel of Simulator.vi.
After verifying the functionality of your code, observe how these algorithms perform in the presence of a narrowband channel with AWGN.
In the presence of these impairments, observe how the packet length
affects the performance of the algorithms you have implemented.
Using delay d = 0.34Ts and the parameters listed above, plot the
timing error statistic (N), as defined in Eq. (5), for oversample factors
N = 2, 4, 6, . . . , 20 (i.e., even integers up to 20). Use a logarithmic
scale for timing error and a linear scale for N .

1 Estimated channel delay (N ) = kTs , where k is the integer offset estimated at the receiver

(after correcting for delay due to filtering), Ts is the symbol period, and N is the oversample
factor at the receiver

i
i

i
book2 2011/10/28 14:05 page 60 #72

60

Lab 3: Synchronization

Pre-Lab Turn In
1. Submit your implementation of align Elgate.vi and align MaxEnergy.vi.
Note: If you need to generate additional subVIs for this or any future
labs, please submit these subVIs along with your other pre-lab VIs.
2. Submit your answers to all of the questions in the pre-lab section along
with your plot of timing error (N ) versus oversample factor N .

Lab Experiment

In this lab you will run your implementations of align ELgate.vi and align
MaxEnergy.vi over a real wireless link. Using your code from the prior labs
in conjunction with the new blocks you have created, you will complete the
framework for the transmitter and receiver blocks in lab. This framework
will be provided to you as in previous labs.

Figure 4: Hierarchy of code framework used in lab.

Insert your code into this framework as you did in the pre-lab. In the
pre-lab you observed how timing error varies with the oversampling factor
N at the receiver. Now you will observe this relationship over a real wireless
link.
Set up the following parameters in your baseband system.
Packet length = 500 bits
Modulation type = QPSK

i
i

i
book2 2011/10/28 14:05 page 61 #73

Lab 3: Synchronization

61

Pulse shaping filter = Root Raised Cosine

Filter parameter = 0.5
Filter length = 8
Channel model: h = 0.75ej /4 (i.e., narrowband channel)
Noise power = Inf dB
TX sample rate = 20 MSamp/sec
TX oversample factor = 20
RX sample rate = 2 MSample/sec
RX oversample factor = 2
Symbol timing recovery method = Max Energy
Also make sure the following RF parameter is set up appropriately (using
the HW parameters control accessible from the front panel of top rx.vi):
Capture time = 400 sec

Questions
Answer the following questions about the digital communication system:
1. What is the symbol rate of the system based on the parameters above?
2. As you did in the pre-lab, vary the oversample factor N at the receiver.
Observe the general shape of the received constellation for a number of
packets (i.e., to eliminate any variance that might be caused by errors
in frequency offset correction or channel estimation). Observe how the
constellation changes when N = 2, 4, 10, and 20.
Based on what youve observed and what you learned in the pre-lab,
describe how the relationship between sampling error and oversample
factor N manifests itself on the signal constellation of top rx.vi.
3. For each of the oversample factors in the previous question (i.e., N =
2, 4, 10, 20), specify what value you set the RX sample rate to on the
front panel of top rx.vi.

i
i

i
book2 2011/10/28 14:05 page 62 #74

62

Lab 3: Synchronization

Lab Turn In
Demonstrate to the instructor that your code is working properly. You will
be required to identify that a sampling error has occurred and qualitatively
support your claim based on the visualization tools available in top rx.vi. You
will then need to show the instructor how the performance of your system
changes as you vary oversample factor N at the receiver. Also answer all of
the questions above. You will be required to submit these answers in a lab
report.

DIRECTIONS FOR LAB SUBMISSION

Your lab report should accomplish the following tasks.
1. Answer all questions from the lab experiment (i.e., do not reanswer
pre-lab questions).
2. Discuss any problems you might have encountered and how you overcame these obstacles.
3. What underlying assumption about the noise in the system has been
made, but not stated explicitly in this lab? Explain why this assumption
is important for the maximum output energy maximization problem.
Hint: This is a property of AWGN.
4. Air any grievances about the structure, content, or workload for this
lab.

i
i

i
book2 2011/10/28 14:05 page 63 #75

Lab 4: Channel Estimation & Equalization

Summary
In Lab 4 you considered a communication system using a narrowband channel model. This scalar channel model does not reflect the frequency selective
nature of most realistic wireless channels. This lab will introduce digital
transmission and reception in a frequency selective channel, where multiple
paths in the propagation environment create distortions in the transmitted.
Correcting for distortions introduced by the channel requires more complicated receiver processing. In this lab you will implement additional features
in the transmitter and receiver to mitigate channel distortions. The new transmitter, channel, and receiver structure are illustrated in Figure 1.
The main concepts explored in this lab are linear least squares estimation
and linear equalization. After reviewing these concepts you will build the
channel estimation and equalization blocks of the model in Figure 1. You
will verify the functionality and correctness of your code in a simulator.
Then, you will implement your code in lab using the USRP hardware to see
how your code works over a real wireless link.
The goal of this lab is to understand the need for channel equalization and to
understand basic algorithms for channel estimation and channel equalization.
For the pre-lab submission, you have to turn in the three VIs described in
Section 2 (LLSE.vi, toeplitz.vi, and direct equalizer.vi). Furthermore, you
have to submit the answers to questions in Section 2.

Background

In wireless communication systems, the signal encounters a effects in the

propagation environment including reflections, scattering, diffraction, and
path-loss. The different mechanisms of propagation create multiple propagation paths between the transmitter and receiver. For example, if there are
two propagation paths, the receiver may observe the signal
z(t)

= 0 ej 0 x(t 0 ) + 1 ej 1 x(t 1 ) + v(t),

(1)

which consists of two different delayed, attenuated, and phase shifted versions
of the signal x(t). The delays 0 and 1 are determined by the propagation
time of the paths, which would generally be different. If the difference 1 0
is significant (at least a fraction of a symbol period T ) then the received signal
will experience intersymbol interference. This can not be corrected using the
63
i

i
i

i
s[n]
Insert
Training

Tz = T/M

C/D

g RX [n]
M
M
Extract
Training

y[n]

Compute
Equalizer

gTX [n]
Tx

D/C
Ex

Detection

x(t )

Inverse
Symbol Map

bits

Figure 1: The system under consideration in this lab. The transmitter uses digital pulse-shaping and upsampling to create the transmit
waveform. A known training sequence is inserted into the transmitted signal. The baseband signal encounters a frequency selective channel
and additive white Gaussian noise at the receiver. The receiver uses digital pulse-shaping and downsampling, followed by linear equalization
and detection, to find a good guess of the transmitted symbols. The linear equalizer is computed using the known training sequence in the
received signal. An advance is applied after the equalization to correct for delays.

z (t )

v(t)

64

Receiver

x(t)

AWGN Frequency Selective Channel

training symbols

Symbol
Mapping

bits

Transmitter

i
i

book2 2011/10/28 14:05 page 64 #76

i

Lab 4: Channel Estimation & Equalization

i
book2 2011/10/28 14:05 page 65 #77

65

Lab 4: Channel Estimation & Equalization

symbol synchronization techniques of Lab 3 since there is a sum of two

different received signals.
In general, there may be many propagation paths between the transmitter
and the receiver. A good generalization of Eq. (1) is

he ( )x(t )d + v(t),
(2)
z(t) =

where he ( ) is the baseband frequency selective channel. For complex pulse

amplitude modulations, the channel distorts the pulse-shaping function. Consider the received signal after matched
filtering and down-sampling. Using

for convolution define h(t) := Ex he (t)

gtx (t)
grx (t)
x(t) and let
h[n] = T h(nT ) be the sampled version of the channel. Then

s[m]h[n m] + v[n]
y[n] =
m

h[0]s[n] +


m=0

s[m]h[n m] +v[n].


(3)

intersymbol interference

Unless h(t) is a Nyquist pulse-shape, the second term in Eq. (3) will not be
zero.
There are two main challenges associated with the received signal in
Eq. (3).
1. The channel coefficients {h[]} create intersymbol interference. The
solution we pursue in this lab is to augment the detection procedure
with a linear equalization step.
2. The channel coefficients {h[]} are unknown to the receiver. The solution we pursue in this lab is to estimate the unknown channel coefficients and use them to determine the linear equalizer, or to estimate
the coefficients of the linear equalizer directly.
The steps of estimation and equalization benefit from making some additional assumptions about the propagation channel. It is reasonable to assume
that the propagation channel is causal and FIR (finite impulse response). It
is causal because, naturally, the propagation channel can not predict the future. It is FIR because (i) there are no perfectly reflecting environments, and
(ii) the signal energy decays as a function of distance between the transmitter
and receiver. Essentially every time the signal reflects some of the energy
passes through the reflector thus it looses energy. Additionally as the signal
is propagating, it loses power as it spreads in the environment. Multipaths
that are weak will fall below the noise threshold. With the FIR assumption, it

i
i

i
book2 2011/10/28 14:05 page 66 #78

66

Lab 4: Channel Estimation & Equalization

is common to assume that the composite channel h(t) is also FIR thus Eq. (3)
becomes
y[n]

L


s[m]h[n ] + v[n].

(4)

=0

The unknown channel coefficients are {h[]}L

=0 where L is the order of the
filter.
In this lab we will employ a linear equalizer. The goal of a linear equalizer

Lf
is to find a filter that removes the effects of the channel. Let fnd [l] l=0
be
an FIR equalizer. The equalizer will be applied to the received signal so that

r[n]

Lf


fnd []y[n ] s [n nd ],

(5)

=0

where nd is the equalizer delay and is generally a design parameter. Generally

allowing nd > 0 improves performance. The best equalizers consider several
values of nd and choose the best one.

1.1

Linear Least Squares

The main mathematical technique that will be used in this lab is known as
linear least squares, which will be used both for estimating the channel and
for computing the equalizer.
Let A denote a N M matrix. The number of rows is given by N and
the number of columns by M. If N = M we say that the matrix is square.
If N > M we call the matrix tall and if M > N we say that the matrix is
fat. We use the notation AT to denote the transpose of a matrix, A to denote
the Hermitian or conjugate transpose, and Ac to denote the conjugate of the
entries of A. Let bbe a N 1 dimension vector. The 2-norm of the vector
N
2
is given by a =
n=1 |bn | .
Consider a collection of vectors {xn }K
n=1 . We say that the vectors are linearly
independent if there does not exist a set of nonzero weights {n } such
that n n xn = 0. In CN and RN, at most N vectors can be linearly independent. We say that a square matrix A is invertible if the columns of A (or
equivalently the rows) are linearly independent. If A is invertible, then a matrix inverse A1 exists and satisfies the properties AA1 = A1 A = I. If A
is tall, we say that A is full rank if the columns of A are linearly independent.
Similarly, if A is a fat matrix, we say that it is full rank if the rows are linearly
independent.

i
i

i
book2 2011/10/28 14:05 page 67 #79

Lab 4: Channel Estimation & Equalization

67

Consider a system of linear equations written in matrix form

Ax

(6)

where A is the known matrix of coefficients sometimes called the data, x is

a vector of unknowns, and b is a known vector often called the observation
vector. We suppose that A is full rank. First consider the case where N = M.
Then the solution to Eq. (6) is x = A1 b.
Now suppose that N > M. In this case A is tall and the system overdetermined in general. This means there are N equations but M unknowns, thus
it is unlikely (except in special cases) that an exact solution exists. In this
case we pursue an approximate solution known as least squares. Instead of
solving Eq. (6) directly we instead propose to find the solution to the squared
error
min Ax b2 .

(7)

Using matrix calculus, it can be shown that the solution to this problem
assuming that A is full-rank is
xLS

(A A)1 A b.

(8)

Note that A A is a square matrix and invertible because of the full-rank

assumption. We refer to xLS as the linear least square error (LLSE) solution
to Eq. (6).
We use the squared error achieved by xLS to measure the quality of the
solution. With some calculus and simplifying, it can be shown that the least
squares error achieved is given by
J (xLS )

=
=

AxLS b2 = x A (Ax b) b (Ax b)

b b b Ax.

(9)
(10)

Now you will use the least-squares solution for channel estimation and equalization.

1.2

Channel Estimation

There are several different criteria for designing an estimator including the
maximum likelihood criterion, minimum mean squared error, and least squares.
Of these, perhaps the least squares technique is the simplest thus we will discuss it here. The added advantage of the least squares estimator is that in
AWGN the least squares estimator is also the maximum likelihood estimator.
t
Suppose that {t[n]}N
n=0 is a known training sequence. This is a set of
symbols that is known a priori. Suppose that the known training symbols

i
i

i
book2 2011/10/28 14:05 page 68 #80

68

Lab 4: Channel Estimation & Equalization

are inserted into the transmitted sequence such that s[n] = t[n] for n =
0, 1, . . . , Nt . To solve for the estimate of the channel using the LLSE solution
in Eq. (8), it is necessary to form a system of linear equations. Consider
y[n]

L


h[l]s[n l] + v[n],

l=0

where s[n] = t[n] for n = 0, 1, . . . , Nt . Since s[n] is unknown for n Nt

(that is the unknown data), we need to write the squared error only in terms
of the unknown data. With this in mind the least squares problem is to find
the channel coefficients that minimize the squared error

2
N
L
t 1 





{h[0],
h[1],
. . . , h[L]}
= arg
min
a[l]t[n l] .
y[n]


a[0],a[1],...,a[L]
n=L

l=0

The summation starts with n = L to ensure unknown data is not included. The
least squares estimator is simply a generalization of the narrowband estimator.
A straightforward way to solve this problem is to differentiate with respect
to a [m], build a set of linear equations, and solve. An alternative approach
is to build a suitable set of linear equations. This is a powerful approach to
solve a large class of least squares problems.
First write the observed data as a function of unknown in matrix form

t[L]

t[0]
a[0]
y[L]

..
..
y[L + 1]
a[1]
.
t[L + 1]
.

.
,
=

..
.
.
.

..
..
.
.

a[L]
y[Nt 1]


  t[Nt 1]  t[Nt 1 L]    
y

where we refer to T as the training matrix. If T is square or tall and full

rank, then T T is an invertible square matrix, where stands for conjugate
transpose (Hermitian). The tall assumption (square with equality) requires
that
Nt L L + 1,
or, equivalently
Nt 2L + 1.
Generally choosing Nt much larger than L+1 (the length of the channel) gives
better performance. The full rank condition can be guaranteed by ensuring
that the training sequence is persistently exciting. Basically this means that it
looks random enough. Random training sequences perform well while the all
constant training sequence fails. Training sequences with good correlation
properties generally satisfy this requirement.

i
i

i
book2 2011/10/28 14:05 page 69 #81

69

Lab 4: Channel Estimation & Equalization

1.3

Channel Equalizer Calculation

L in hand, the next task is to remove

With an estimate of the channel {h[l]}
l=0
the effects of this channel. In this lab we consider linear equalization that
performs the operation in Eq. (5). There are several approaches for designing
a linear equalizer. In this lab we compute the least-squares equalizer. The
objective is to find a filter such that
Lf


l] [n nd ].
f [l]h[n

(11)

l=0

Except in trivial channels Eq. (11) can not be satisfied exactly. The reason
is that an FIR filter requires an IIR filter. The parameter nd in Eq. (11) is
the equalizer delay and is generally a design parameter. Generally allowing
nd > 0 improves performance. The best equalizers consider several values
of nd and choose the best one.
A straightforward approach is the least-squares equalizer. The key idea
is to write a set of linear equations and solve for the filter coefficients that
ensure that the Eq. (11) minimizes the squared error. Writing Eq. (11) in
matrix form

f [0]
0

h[0]
0

h[1]

h[0]
0

f [1] ..
.

.
..
..
.

1
.
.

nd + 1

.

= ..
..
h[L]

.
.

.
..
h[L]

..
.
..
.
f [Lf ]
0




   

end

is a type of Toeplitz matrix, sometimes called a filtering matrix.

The matrix H
is full rank, which is guaranteed if any of the channel
Assuming that H
coefficients are nonzero, the linear least squares solution is

1
H

end .
fnd = H
H
f end 2 . The squared error
The squared error is measured as Jf [nd ] = H
can be minimized further by choosing nd such that Jf [nd ] is minimized. This
is known as optimizing the equalizer delay.
The equalizer order is Lf . The choice of Lf is a design decision that
depends on L. The parameter L is the extent of the multipath in the channel

i
i

i
book2 2011/10/28 14:05 page 70 #82

70

Lab 4: Channel Estimation & Equalization

and is determined by the bandwidth of the signal as well as the maximum delay
spread derived from propagation channel measurements. The equalizer is an
FIR inverse of an FIR filter. As a consequence the results will improve if Lf
is large. The complexity required per symbol, however, also grows with Lf .
Thus there is a tradeoff between choosing large Lf to have better equalizer
performance and smaller Lf to have more efficient receiver implementation.

1.4

Direct Least-Squares Equalizer

Using the channel estimation and equalization methods from the previous
sections requires solving two least squares estimation problems. This can be
computationally expensive. Designing the equalizer estimate directly from
the received sequence can be more efficient as this method only requires
formulating a single least-squares estimation problem. The least-squares
equalizer requires solving two least-squares problems. The first is to estimate the channel coefficients using a least squares approach; the second is
to estimate the least-squares equalizer. An alternative approach is a direct
solution. This means that the equalizer is found directly from the observed
training data. Such an approach is somewhat more robust to noise.
Consider the received signal after linear equalization with delay nd
s [n nd ] =

Lf


fnd [l]y[n l].

(12)

l=0

Suppose that s[n] = t[n] for n = 0, 1, . . . , Nt is the known training data.

Then s [n nd ] = t[n nd ] for n = nd , nd + 1, . . . , nd + Nt . Rewriting
Eq. (12) with knowledge of the training data
t[n] =

Lf


fnd [l]y[n + nd l]

l=0

for n = 0, 1, . . . , Nt .
Now build a linear equation

y[nd Lf ]
y[nd ]
t[0]
fnd [0]

..
..
t[1]
fnd [1]
.
y[nd + 1]
.

.
..
..

.
.

.
.
.
.

.
.
t[Nt 1]
fn [Lf ]

  y[nd + Nt 1]  s[nd + Nt Lf ]   d


t

Ynd

fnd

(13)

i
i

i
book2 2011/10/28 14:05 page 71 #83

71

Lab 4: Channel Estimation & Equalization

Solving under the assumption that Y is full rank, which is reasonable in the
presence of noise, the least squares solution is
fnd

Ynd Ynd

Ynd t.

(14)

nd fnd 2 . The squared

The squared error is measured as Jf [nd ] = t Y
error can again be minimized further by choosing nd such that Jf [nd ] is
minimized.
Note that to ensure that Y is square or tall requires Lf Nt 1. Thus
the length of the training determines the length of the equalizer. This is a
major difference between the direct and indirect methods. With the indirect
method, an equalizer of any order Lf can be designed. The direct method,
on the other hand, avoids the error propagation where the estimated channel
is used to compute the estimated equalizer. Note that with a small amount
of training the indirect method may perform better since a larger Lf can be
chosen while a direct method may be more efficient when Nt is larger.

Pre-Lab

You have learned about two equalizer estimation methods, indirect and direct. In this lab you will be implementing the direct least-squares equalizer
described in the last section. To do this you will need to build a number of
Table 1: Description of LLSE.vi
LLSE.vi - for linear system Ax = b, as in Eq. (6),
compute linear least squares estimate of parameter x
A
INPUTS
b

x.estimate

OUTPUTS

least
squared
error
error
code

2-D array of
CDB
(complex
doubles)
1-D array of
CDB
1-D array of
CDB
DBL (double)

I32 (integer)

m x n matrix

length m vector
linear least squares estimate of
parameter x, as defined by
Eq. (8)
squared error of estimate,
Ax b2
error code generated by
Inverse Matrix.vi

i
i

i
book2 2011/10/28 14:05 page 72 #84

72

Lab 4: Channel Estimation & Equalization

Table 2: Description of toeplitz.vi

toeplitz.vi - generate an m-by-n Toeplitz structured matrix from

the first row (length n) and column(length m) of the matrix
row

INPUTS
column

OUTPUTS

Toeplitz
array

1-D array of
CDB
(complex
doubles)
1-D array of
CDB

first row (length n) of the

matrix

first column (length m) of the

matrix

2-D array of
CDB

m-by-n Toeplitz structured

array

auxiliary subVIs (i.e., LLSE.vi and toeplitz.vi described in Tables 1 and 2).
You will then use these VIs in your implementation of the direct least-squares
equalizer in direct equalizer.vi. After building all of your VIs, you will revise
the equalizer.vi called by receiver.vi and use the simulator provided to you
and verify your code as you have done in the previous labs.
You can use the error code output of LLSE.vi to find out if there were any
errors in computing the LLSE solution. For example, error code can be used
to indicate an error when the matrix (A A) is singular.
You have been provided with templates for the VIs you need to create for
this lab that already have all the inputs and outputs wired for you. What
is required of you is to finish constructing the block diagram to provide the
functionality of the VIs.
Throughout the course, several of the VIs you will create will have a
modulation parameters cluster passed into and out of them. The modulation
parameters in contains many of the parameters needed by your VIs and will
be unbundled from them. This has already been done for you in the template
VIs if necessary. Some VIs will also have modulation parameters out, so that
the cluster can be passed on through to the VIs that follow. Please ensure that
these clusters remain wired the way they are in the template VIs, as changing
how they are wired will cause VIs further down the line to break.
After building these VIs, it is important that you verify their functionality
before proceeding as you will need to use them in your implementation of
direct equalizer.vi. To verify the functionality of LLSE.vi, do the following:

Let A be a 3x5 matrix of all zeros except for its major diagonal, which

i
i

i
book2 2011/10/28 14:05 page 73 #85

73

Lab 4: Channel Estimation & Equalization

is all ones:

A=

1
0
0
0
0

0
1
0
0
0

0
0
1
0
0

Let b be any vector of length 5, such that the last two positions of b
are zero (i.e., b = [b1 b2 b3 0 0]T contains three nonzero values in its
first three positions).
Verify that LLSE.vi computes x.estimate to be the first three elements
of b.
Also verify that your VI returns an error when A is not full rank. To test
this, replace the last column of A with zeros and repeat the above test (i.e.,
checking to see that an error has occurred). Next, verify your implementation
of toeplitz.vi is working by visual inspection of Toeplitz array. This matrix
should have constant negative-sloping diagonals.
To receive full credit you must correctly implement both LLSE.vi and
toeplitz.vi. However, if you are unable to do so, you may use the equivalent VIs from digital comm.llb to implement direct least-squares equalizer
estimation.
Next, you will build direct equalizer.vi, which will be inserted inside
equalizer.vi as shown in Figure 2. This block diagram shows how equalizer.vi
implements the delay optimization for both direct and indirect equalizer estimation.
To build direct equalizer.vi, implement the direct least-squares estimation
solution in Eq. (14). You should use your implementations of toeplitz.vi
and LLSE.vi to formulate and solve the LLSE problem of Eq. (14) in direct equalizer.vi. These subVIs should simplify your code significantly. Be
careful when generating the matrices of Eq. (14) as small errors in the formulation can lead to critical problems.
Table 3 describes the details of direct equalizer.vi. Notice the VI takes an
explicit equalizer delay input. Do not confuse this value with the equalizer
delay parameter in the modulation parameters in cluster. Use the equalizer
delay input to form the received signal matrix of Eq. (13) (i.e., do not use the
delay parameter from the modulation parameters in cluster).
To build the training vector t of Eq. (14) use Training Sequence, a 1-D array
of CDB, from the modulation parameters in cluster. You will only use the
first half of Training Sequence to generate t, since the training sequence from
the modulation parameters in cluster is composed of two repeated sequences
(i.e., first half of Training Sequence is the same as the second half). This

i
i

i
book2 2011/10/28 14:05 page 74 #86

Lab 4: Channel Estimation & Equalization

Figure 2: Block diagram of equalizer.vi.

74

i
i

i
book2 2011/10/28 14:05 page 75 #87

75

Lab 4: Channel Estimation & Equalization

Table 3: Description of direct equalizer.vi

direct equalizer.vi - solves for the least-squares solution of the equalizer
directly from the received sequence as in Eq. (14)
input
INPUTS

OUTPUTS

1-D array of
CDB
I32 (integer)

received sequence after frame

detection
filter delay parameter nd

filter
estimate
mean
square
error

1-D array of
CDB

solution to LLSE of Eq. (14)

DBL

mean square error of estimate,

Ynd fnd t2

actual
equalizer
delay

I32 (integer)

delay parameter modulo

Lf + 1

equalizer
delay

means Nt , the length of t, will be one-half the size of the training sequence
array from the modulation parameters in cluster.
In summary:
Build LLSE.vi using the linear algebra tools available in the Linear
Algebra tool palette. Be aware that these polymorphic VIs operate in
a number of modes and may require you to specify the use of complex inputs. The Linear Algebra tool palette can be accessed from
MathematicsLinear Algebra.
Build toeplitz.vi using the array tools available in the Array palette.
Be sure to verify your code before using this VI in direct equalizer.vi
as small errors in formulation can lead to critical problems.
After building direct equalizer.vi using the previously mentioned VIs,
insert your code into equalizer.vi, which can be accessed from the block
diagram of receiver.vi

Questions
Answer the following questions, and submit the answers as part of your prelab.
1. In your implementation of toeplitz.vi you were required to build a
Toeplitz matrix given the initial row and column of the matrix. Notice

i
i

i
book2 2011/10/28 14:05 page 76 #88

76

Lab 4: Channel Estimation & Equalization

the first element of row and column should be equal. What will your
VI do if the initial element of each array is different?
2. Test your channel equalizer algorithm using the channel h[0] = 1,
h[1] = 0.35ej /4 . You can modify the length of the equalizer from
the front panel of the simulator. In the absence of noise, what happens
to the received signal constellation when you set the equalizer length
to one? Describe what happens to the constellation as you vary the
equalizer length from one to six.
3. Using the same channel, observe how the bit-error rate performance
of your equalizer varies with SNR for various equalizer lengths. Plot
average BER as a function of SNR for Lf + 1 = 1 and Lf + 1 = 6.
Vary SNR from 0 dB to 14 dB in increments of 2 dB. Use the default
value for any parameter not specified below.
Modulation type = QPSK
Packet length (bits) = 500
Equalization method = Direct
Equalizer length (Lf + 1) = 1,6
ISI Channel = {h[0] = 1, h[1] = 0.35ej /4 }
Remember if signal power is held constant, then decreasing noise
power, N0 , is equivalent to increasing SNR. To observe errors at high
SNR you will need to run significantly more iterations in the simulator.
Use a logarithmic scale for BER and a dB scale for SNR in your plot.
Also, on the same graph, plot BER as a function of SNR for Lf +1 = 1
in an AWGN channel for the same range of SNR values. For this curve,
you will only need to plot error rates in excess of 106 .

Pre-Lab Turn In
1. Submit your implementation of LLSE.vi, toeplitz.vi, and direct equalizer.vi. Remember, you will be penalized if you do not wire the error
cluster inputs and outputs in your implementation of direct equalizer.vi.
Note: If you need to generate additional subVIs for this or any future
labs, please submit these subVIs along with your other pre-lab VIs.
2. Submit your answers to all of the questions in the pre-lab section.

i
i

i
book2 2011/10/28 14:05 page 77 #89

Lab 4: Channel Estimation & Equalization

77

Lab Experiment

In this lab you will run your implementation of direct equalizer.vi over a
real wireless link. Using your code from the prior labs in conjunction with
the new blocks you have created, you will complete the framework for the
transmitter and receiver blocks in lab. This framework will be provided to
you as in previous labs.

Figure 3: Hierarchy of code framework used in lab.

Insert your code into this framework as you did in the pre-lab. In the prelab you observed the BER performance of the direct least-squares equalizer
over a multipath channel. Now you will observe how a multipath channel
impacts the constellation of the received signal. You will also learn about the
power-delay profile of a multipath channel.

3.1

Narrowband Channel

In this part of the lab experiment you will observe how the channel of a
real wireless link impacts the received QAM constellation. To begin, setup
the following parameters in your system using the appropriate controls in
top rx.vi and top tx.vi:
Packet length = 500 bits
TX sample rate = 20 MSamp/sec
TX oversample factor = 200

i
i

i
book2 2011/10/28 14:05 page 78 #90

78

Lab 4: Channel Estimation & Equalization

RX sample rate = 2 MSamp/sec
RX oversample factor = 20
Equalizer length = 1
Capture time = 3.5 msec

Use the default values for any parameters not listed above. Next, in order
to observe the impact of the wireless channel on your received QAM constellation you will need to rewire part of the block diagram in receiver.vi.
As shown in Figure 4, rewire the wire labeled recovered symbols so it is
connected to the output of synchronize.vi and not strip control.vi. This will
allow you to observe the received constellation prior to equalization. Notice
that since this modification bypasses strip control.vi it will plot the entire
received sequence (i.e., including training data and any additional symbols
or zeros at the end of the array).

Questions
After setting up the appropriate parameters and modifying receiver.vi as previously described, run your system and observe how the narrowband channel
alters your received constellation.
1. What is the symbol rate of your system?
2. What is the passband bandwidth of your system?
3. Based on your observations, describe the impairments imparted to the
received constellation. You may need to auto-scale the axes of your
constellation in receiver.vi in order to observe the effects of the narrowband wireless channel.
Submit your answers to these questions as part of your lab report.

3.2 Wideband Channel

In this part of the lab experiment you will learn about the power-delay profile
of the channel. The power-delay profile of a continuous-time channel h( )
is


P ( ) = E |h( )|2 ,
(15)
where E{ } is the expectation operator. The discrete-time equivalent of the

power-delay profile can be computed from the estimated channel h[n]

(i.e.,
2

P [n] = E{|h[n]| }). In this part of the lab experiment you will study the
power-delay profile of a wideband system.

i
i

i
book2 2011/10/28 14:05 page 79 #91

79

Figure 4: Modification of receiver.vi needed to observe constellation prior to equalization.

Lab 4: Channel Estimation & Equalization

i
i

i
book2 2011/10/28 14:05 page 80 #92

80

Lab 4: Channel Estimation & Equalization

Set up the following parameters in your system:
Packet length = 500 bits
TX sample rate = 20 MSamp/sec
TX oversample factor = 4
RX sample rate = 10 MSamp/sec
RX oversample factor = 2
Channel estimate length = 6
Equalizer length = 6
Capture time = 80 sec

Use the default values for any parameters not listed above. You will now
start the transmitter of your system (i.e., top tx.vi). Place your antennas at an
elevated height so they will be free of obstructions and able to capture a large
portion of the multipath energy in your wireless environment. Begin receiving
for at least 5 of the
packets using top rx.vi. Record the estimated channel h[n]
transmitted packets. Discard any measurements in which the bit-error rate of
the received packet exceeds 0.10; typically BER exceeds 101 when errors
in frame detection occur. You should also disable frequency offset correction
from the front panel of top rx.vi, as this can lead to additional unwanted
errors. Also, be sure to rewire the wire labeled recovered symbols so it is
connected to the output of strip control.vi and not synchronize.vi (in contrast
to the previous section).

Questions
Answer the following questions about this part of the lab experiment.
1. What is the symbol rate of your system?
2. What is the passband bandwidth of your system?
3. Based on the data you have collected in this part of the experiment,
make a plot of the power-delay profile of the wideband wireless link
in lab. Discard any outlying observations that might have been caused
by errors in frame detection or noise. Average the power-delay profile
over the remaining observations.
Submit your answers to these questions along with your plot of powerdelay profile as part of your lab report.

i
i

i
book2 2011/10/28 14:05 page 81 #93

Lab 4: Channel Estimation & Equalization

3.3

81

Lab Turn In

Demonstrate that your code is working properly. Show that your code works
for a variety of equalizer lengths (Lf + 1 = 1 and Lf + 1 = 6).

DIRECTIONS FOR LAB SUBMISSION

Your lab report should accomplish the following tasks.
1. Answer all questions from the lab experiment (i.e., do not reanswer
pre-lab questions).
2. Discuss any problems you might have encountered and how you overcame these obstacles.
3. In this lab you observed how your code operates over a narrowband
channel and a wideband channel. The following questions will explore
the differences between these two channels.
(a) What is the symbol period of the narrowband system in Section 3.1?
(b) What is the symbol period of the wideband system in Section 3.2?
(c) If the continuous-time channel h( ) has nonzero support for
[0, 2.5 sec), then how many nonzero taps will h[n], the
equivalent discrete-time channel, have in the narrowband and
wideband system respectively?
In other words, if Tnb and Twb are the symbol periods of the narrowband and wideband systems respectively, find the cardinality
(i.e., number of elements) of the set A = {n|h[n]  = 0} for each
system. See the hint below for more on h[n].
(d) Which of the two systems can resolve the channel response
h( ) more accurately?
HINT: The equivalent discrete-time channel is h[n] = h(nTs ), where
Ts is the symbol period of the system.

i
i

i
book2 2011/10/28 14:05 page 82 #94

Lab 5: Frame Detection & Frequency Offset

Correction

Summary
The algorithms for channel estimation and equalization discussed in Lab 4
were formulated under the assumption that the location of the training data
was known, which we call the start of the frame. In practice, due to propagation and signal processing delays, the location of the beginning of the
frame is unknown. This makes it difficult to apply the channel estimation
and equalization algorithms from Lab 4. Up to this point it has also been assumed that the carrier frequency of the transmitter and receiver were locked
together. Even small differences, however, lead to a distortion known as carrier frequency offset. In this lab you will implement a particular method for
frequency offset correction, called the Moose algorithm, and a method for
frame detection based on correlation.
The frequency offset correction and frame detection algorithms you will
implement in this lab will be embedded within the synchronization block of
the receiver. After building the VIs needed to implement these algorithms,
you will verify the functionality and correctness of your code in a simulator.
Then, you will implement your code in lab using RF Hardware to see how
your code works over a real wireless link.
Note that after this lab you will have considered symbol timing synchronization, frame detection, and channel estimation/equalization all as independent problems. In actuality, any number of these problems could be solved
jointly (e.g., jointly solve for the symbol timing offset and channel estimate
which minimizes mean-squared error). The task of joint optimization will
not be explored in this lab, but you should be aware that the receiver you have
been implementing in lab is not necessarily the optimal one. In particular,
it is common that frame synchronization is performed jointly with carrier
frequency offset estimation.
For the pre-lab submission, you have to turn in the two VIs described in
Section 2 (Moose.vi and sliding correlator.vi). Further, you have to submit
the answers to the questions in Section 2.

Background

In this section we provide some background on training sequences then review

algorithms for frame synchronization and frequency offset synchronization.
82
i

i
i

i
g RX [n]
M
M

y[n]

Frame
Synch

Frequency
Synch

Equalize

Detection

Inverse
Symbol Map

bits

Figure 1: The receiver considered in this lab. The new synchronization functionality is introduced after downsampling and before the
equalization. The first step is determining the location of the training sequence, known as frame synchronization. The second step is
estimation and correction of the carrier frequency offset, known as carrier frequency synchronization. The linear equalizer block includes the
channel estimation and equalization functions from Lab 4.

Tz = T/M

C/D

z (t )

Receiver

i
i

book2 2011/10/28 14:05 page 83 #95

Lab 5: Frame Detection & Frequency Offset Correction

83

i
book2 2011/10/28 14:05 page 84 #96

84

Lab 5: Frame Detection & Frequency Offset Correction

1.1 Training Sequences

Up to this point we have assumed that a training sequence has been provided
at both the transmitter and receiver. Not all training sequences are created
equal. Without a good training sequence, the estimation techniques in the labs
would perform poorly. This section discusses some desirable properties of
training sequences and introduces one family of sequences known as Barker
codes. Many other sequences are used in research literature and in commercial wireless systems as well (e.g., Frank sequences, Zadoff-Chu sequences,
or Gold sequences).
Training sequences serve as a known reference signals, which can be used
for many purposes at the receiver. In these labs, we use training sequences
for two main purposes: (1) synchronization and (2) channel estimation. In
general, good training sequences for synchronization possess strong autocorrelation properties. In Lab 4, the training sequence needed to be suitably
designed in order to provide sufficient rank properties. Training sequences
with strong autocorrelation properties can satisfy these rank properties. In
this lab, you will use such a training sequence to create a correlation based
detector that finds the beginning of a frame.
Barker codes are one set of sequences with good aperiodic autocorrelation
t
and rank properties. A length Nt Barker sequence {ak }N
k=1 is a sequence of
values, 1, such that
%N k
%
t
%
%
%
%
ai ai+k % 1,
%
%
%

(1)

i=1

for all 1 k Nt . Because of this strong autocorrelation property, Barker

sequences are commonly used in digital communication systems that employ
spread spectrum techniques. They have been used for example in IEEE
802.11 and IEEE 802.11b. Table 1 has a list of known Barker codes1 . The
complete set of Barker sequences is the closure of this set of codes under
reversal and negation.
The known binary sequences that satisfy Eq. (1) are listed in Table 1. It
may happen that longer training sequences are required in a given application. Remember that longer training sequences give better estimation performance. One approach is to relax the condition in Eq. (1) to allow for large
cross-correlation values or the binary restriction and thus look for another sequence family. Another approach is to concatenate several Barker sequences
together. This will give higher cross-correlation peaks, but provides some
other properties that will be exploited for frequency offset estimation.
1 Shorthand notation is used in this table. and + represent 1 and +1 respectively.

i
i

i
book2 2011/10/28 14:05 page 85 #97

Lab 5: Frame Detection & Frequency Offset Correction

Code Length
2
3
4
5
7
11
13

85

Barker Sequence
[+, ]
[ +]
[ + , + ++]
[ + ]
[ + + +]
[ + + + + + +]
[ + + + + ]

Table 1: Barker Sequences

In the lab we will use a training sequence that consists of four length 11
Barker sequences prepended to the data segment of the packet. This means
that you may treat the training sequence as a concatenation of two length 22
training sub-sequences (i.e., the complete training sequence is 44 symbols,
and the first and second half are the same).

1.2

Frame Synchronization

The purpose of frame synchronization is to resolve multiple symbol period

delays, determining what we call the beginning of the frame. Essentially
the objective is to find a known reference in the transmitted signal so that
subsequence receiver operations can proceed like channel estimation and
equalization. There is a hugh theory surrounding frame synchronization. We
consider one approach based on correlation properties of the training signal.
To develop the frame synchronization algorithm we neglect the presence of
carrier frequency offset in this section.
Consider a flat-fading channel where symbol synchronization has already
been employed to give
y[n] = hs[n d] + v[n]
after the matched filter and down-sampling at the receiver where h is an
unknown complex channel coefficient and d is an unknown frame offset.
Suppose that data is transmitted in frames consisting of a length Nt training
t
signal followed by P Nt data symbols. Suppose that {t[n]}N
n=0 is the
training sequence known at the receiver. Since the training sequence has
good correlation properties, we correlate the received signal with the training
signal to compute
%2
%N 1
t
%
%
%
%
t [k]y[n + k]%
(2)
R[n] = %
%
%
k=0

i
i

i
book2 2011/10/28 14:05 page 86 #98

86

Lab 5: Frame Detection & Frequency Offset Correction

and solve for the frame offset as

d = max R[n].

(3)

The correlation in Eq. (2) results in two terms

%N 1
%2
N
t
t 1
%
%
%
%

= %
t [k]hs[n d] +
t [k]v[n]% .
%
%

R[n]

k=0

k=0

Because the noise is zero mean, the second term should be approximately
zero if the training sequence is large enough. This leads to
R[n]

%N 1
%N 1
%2
%2
t
t
%
%
%
%
%
%
%

2%

t [k]hs[n + k d]% = |h| %

t [k]s[n d]% . (4)
%
%
%
%
%
k=0

k=0

Assuming that the data is different than the training sequence, the correlation
peak should occur at the location of the training data, assuming that the
training data is contained in the window of observation samples.
Now suppose that the channel is frequency selective and
y[n] =

L


h[]s[n  d] + v[n] = h[]
s[n d] + v[n],

=0

where we use
to denote convolution. In this case, the received correlation
R[n] neglecting noise is
%
%2
R[n] %h[]
t [n]
s[n d]% .
(5)
The effect of the channel is to smear the correlation peak. Correlation based
frame detection still works, but it becomes less accurate as the peak may shift
due to the channel. In practice the peak value of R[n] may be shifted by an
offset d = maxn R[n]  for some  > 0 typically  < L/4. Equalization
takes care of the residual offset.

1.3

Frequency Offset Estimation

Passband communication signals are used in wireless communication systems. A passband signal has energy that is nonzero for a frequency band concentrated about some carrier frequency fc . At the transmitter the baseband
signal is upconverted to a passband signal while at the receiver a passband
signal is downcoverted to a baseband signal. In most systems the carrier
frequency used for upconversion at the transmitter does not match the car-

i
i

i
book2 2011/10/28 14:05 page 87 #99

Lab 5: Frame Detection & Frequency Offset Correction

87

RF Downconversion

RF Upconversion
propagation
channel

Figure 2: Frequency offset between transmitter and receiver.

rier frequency at the receiver, as illustrated in Figure 2. This creates what is

known as a carrier frequency offset.
From a baseband signal perspective, the problem of carrier frequency offset
can be visualized using the signal model in Figure 3 where the carrier offset
is fo = fc fc . The effect of the offset is to rotate the received signal by
exp(j 2fo t). The larger the offset, the faster the received signal rotates. The
carrier offset complicates most of the receiver processing functions including
frame synchronization, channel estimation, and equalization.
If the carrier offset is small and sufficient sampling is employed, then the
discrete-time signal at the receiver after matched filtering and downsampling
can be written as
y[n]

ej 2 n

L


h[l]s[n l] + v[n],

l=0

where = fo T . The objective of carrier frequency offset synchronization

is to estimate the parameter . Given an estimate , removal of the offset
amounts to performing
y[n]

= ej 2 n y[n].

(6)

To estimate the carrier frequency offset, we use a data aided approach

proposed by Moose [7]. This method relies on a periodic training sequence
and is applicable to many commercial wireless systems (e.g., IEEE 802.11a/g
systems). It a self-referenced synchronization algorithm as it employs the
periodic structure in the training but not the correlation properties. Consider
v(t)

Figure 3: Frequency offset between transmitter and receiver.

i
i

i
book2 2011/10/28 14:05 page 88 #100

88

Lab 5: Frame Detection & Frequency Offset Correction

a complex pulse-amplitude modulation where there is a repetition of two

training sequences followed by the data. Let the training sequence start at
n = 0. Thus
s[n] = s[n + Nt ] = t[n]
for n = 0, 1, . . . , Nt 1. Note that symbols s[n] for n < 0 and n Nt are
unknown (they are either zero or correspond to the unknown portions of the
data). For L n Nt 1
y[n]

j 2 n

L


h[l]s[n l] + v[n]

l=0

y[n + Nt ]

j 2 (n+Nt )

L


h[l]s[n + Nt l] + v[n + Nt ].

l=0

Using the fact that s[n + Nt ] = s[n] = t[n] for n = 0, 1, . . . , Nt 1

y[n + Nt ]

ej 2 Nt ej 2 n

L


h[l]t[n l] + v[n + Nt ]

l=0

ej 2 Nt y[n].

(7)

To see the significance of this result, remember that the channel coefficients
{h[l]}L
l=0 are unknown!
One way to solve the frequency offset estimation problem is to formulate
and solve a least-squares problem. Because appears in the exponent, we
solve a modified least squares problem. Consider the squared error
J (a) =

N
t 1

y[l + Nt ] ay[l]2 .

l=L

Using the concept of liner least squares from Lab 4, the coefficient a is
Nt 1
y[l + Nt ]y [l]
a = l=L
.
Nt 1
2
l=L |y[l]|
Since only the phase of a is of interest, there is no need to compute the
denominator. A simple estimate of the frequency offset is then
Nt 1
phase l=L
y[l + Nt ]y [l]
=
(8)
2 Nt
or
Nt 1
phase l=L
y[l + Nt ]y [l]

,
(9)
fe =
2 T Nt

i
i

i
book2 2011/10/28 14:05 page 89 #101

Lab 5: Frame Detection & Frequency Offset Correction

89

where phase denotes the principle phase of the argument.

Thanks to the periodicity of the discrete-time exponential, the estimate of
will only be accurate for | Nt | 21 or equivalently
| |

1
2Nt

or

1
.
2T Nt
Choosing larger Nt improves the estimate since it results in more noise averaging but reduces the range of offsets that can be corrected. A way of solving
this problem is to use multiple repetitions of a short training sequence, but
this will not be exploited in this lab. An enhanced carrier frequency offset
correction range can be obtained by using properties of the training sequence.
|fe |

1.4

Combining Frame Detection and Frequency

Offset Estimation

The frame synchronization algorithm in this lab was developed assuming

there was no carrier frequency offset. The presence of frequency offset reduces the accuracy of the correlation based frame detection method. For small
offsets this effect can be negligible while for larger offsets, the autocorrelation
properties of the training sequence can fall apart.
The carrier frequency offset estimation algorithm considered in this lab
assumed that frame synchronization was already performed. The periodic
correlation properties though could be used to perform frame synchronization
as well. The observation is that the correlation peak should occur when the
pair of training sequences is encountered at the receiver. A robust approach
involves evaluating the received correlation in a window and solving
N
t 1

y[n + d + Nt ]y [n + d]
d = arg max &
.
&
N
N
t 1
t 1
2
2
|y[n + d]|
|y[n + d + Nt ]|
n=L

n=L

(10)

n=L

Frame synchronization with the Moose method is robust to intersymbol interference.

In the lab we will employ the training based correlation for frame synchronization followed by the Moose method for carrier frequency offset estimation. It should be clear though that other approaches are possible, including
algorithms that jointly perform frame synchronization and carrier frequency
offset estimation.

i
i

i
book2 2011/10/28 14:05 page 90 #102

90

Lab 5: Frame Detection & Frequency Offset Correction

Table 2: Description of Moose.vi

Moose.vi - implements the Moose algorithm described by Eq. (9) and performs
correction per Eq. (6).
input

INPUTS

1-D array of
CDB
(complex
doubles)
DBL (double)

received sequence after frame

detection

channel
estimate
length

I32 (integer)

Lh + 1

output

1-D array of
CDB

frequency
offset

DBL

received sequence after

correcting for estimated
frequency offset, i.e., y[n]

as
described by Eq. (6)
frequency offset estimate
foffset = /T

symbol
rate

OUTPUTS

symbol rate of the system

Pre-Lab

In this lab you will implement the correlation based frame detector and the
Moose algorithm. You will first build Moose.vi, that will implement the
Moose algorithm. You will then build sliding correlator.vi, which will implement the frame detection algorithm. This VI will also use Moose.vi to
estimate/correct for the frequency offset. These VIs are described in Tables 2
and 3. After building your VIs, you will insert your code into frame detect.vi
in the simulator provided to you and verify your code using the simulator.
Please name your VIs in the same manner as your previous submissions (i.e.,
follow the yourname VIname.vi format).
After building these VIs, insert your code into the simulator provided to
you. As shown in Figure 4, you will insert sliding correlator.vi into the
appropriate case structure in the block diagram of frame detect.vi.
You can find frame detect.vi in the block diagram of synchronize.vi. The
hierarchy of these files is shown in Figure 5. After inserting your code appropriately, if you find that the simulator does not compile check to make sure
that you have used the same connector topology described by moose.vi in the
digital communication library.
You have been provided with templates for the VIs you need to create for
this lab that already have all the inputs and outputs wired for you. What

i
i

i
book2 2011/10/28 14:05 page 91 #103

91

Figure 4: Block diagram of frame detect.vi.

Lab 5: Frame Detection & Frequency Offset Correction

i
i

i
book2 2011/10/28 14:05 page 92 #104

92

Lab 5: Frame Detection & Frequency Offset Correction

Table 3: Description of sliding correlator.vi

sliding correlator.vi - implements the correlation based detector of Eqs. (2)(3) ;
also estimates/corrects for frequency offset using Moose.vi.

INPUTS

OUTPUTS

input
complex
waveform

IQ waveform
cluster

received sequence after

matched filter and symbol
timing recovery

correct
frequency
offset?

Boolean

determines whether or not

your VI should correct for the
frequency offset estimated by
Moose.vi

output

1-D array of
CDB

received sequence after

correcting for estimated delay
d and frequency offset (when
appropriate)

frame
offset

I32

estimated delay d of Eq. (3)

frequency
offset

DBL

frequency offset estimate

computed by Moose.vi

Figure 5: Hierarchy of files associated with frame detect.vi.

is required of you is to finish constructing the block diagram to provide the

functionality of the VIs.
Throughout the course, several of the VIs you will create will have a
modulation parameters cluster passed into and out of them. The modulation

i
i

i
book2 2011/10/28 14:05 page 93 #105

Lab 5: Frame Detection & Frequency Offset Correction

93

parameters in contains many of the parameters needed by your VIs and will
be unbundled from them. This has already been done for you in the template
VIs if necessary. Some VIs will also have modulation parameters out, so that
the cluster can be passed on through to the VIs that follow. Please ensure that
these clusters remain wired the way they are in the template VIs, as changing
how they are wired will cause VIs further down the line to break.
After building these VIs and inserting them into the simulator as you have
done in previous labs, it is important that you verify their functionality. First,
verify that your correlation based detector is working by doing the following:
1. Set the Correct Frequency Offset control to FALSE on the front panel
of the simulator. Setting this constant to FALSE tells frame detect.vi
not to correct for the frequency offset estimated by Moose.vi.
2. Set the following parameters in your simulator.
TX oversample factor = RX oversample factor = 10
TX sample rate = RX sample rate = 10 MHz
Packet length (bits) = 500
Synchronization Method = Timing Estimation
ISI Channel = {h[0] = 0.35ej /4 }
Use the default value for any parameters not specified above.
3. Set the delay of the channel to 10Ts , where Ts is the symbol period of
the system.
After running the top level of your simulator, verify that your receiver
has estimated the correct delay in the channel. You can find the measured
delay in the Measured channel impairments cluster on the front panel of the
simulator. Additionally, you should verify that you have no bit errors. After
you have verified that your correlation based frame detection algorithm has
been implemented properly, you can now check that your implementation of
the Moose algorithm is working properly by doing the following:
1. Set the Correct Frequency Offset control to TRUE on the front panel
of the simulator Setting this constant to TRUE tells frame detect.vi to
correct for the frequency offset estimated by Moose.vi.
2. Use the same parameters to test your implementation of Moose.vi.
3. Set the delay of the channel to 10Ts , where Ts is the symbol period of
the system.

i
i

i
book2 2011/10/28 14:05 page 94 #106

94

Lab 5: Frame Detection & Frequency Offset Correction

4. Set the frequency offset of the channel to 201 Hz.

Verify that the estimated frequency offset displayed in the Measured channel impairments cluster reflects the correct frequency offset. Also, you should
have no bit errors. Finally, set the Correct Frequency Offset control back to
FALSE. Verify that your code estimates the frequency offset, but does not
correct for it when this Boolean control is set to FALSE.
In summary:
After building Moose.vi and sliding correlator.vi, insert your code into
the simulator (i.e., inside frame detect.vi).
Note: You will use the unbundled parameter Training Sequence from
the modulation parameters in cluster for the training sequence in this
lab.
Verify that your implementation of the correlation based frame detector
and the Moose algorithm are working properly.

Questions
Answer the following questions, and submit the answers as part of your prelab.
1. Set up your simulator according to the steps used for verifying your
correlation based frame detector. In particular, make sure that the
control Correct Frequency Offset is set to FALSE. Set the noise power
of your channel to 5 dB. Also, use AWGN channel for this question
instead of ISI. Describe what happens to the error rate of your system
when the estimate for the delay in the channel is off by more than one
symbol time (i.e., when d  = d).
Note: You may need to modify your VIs to observe this effect, or you may
guess the impact from the structure of the frame detection equations.
2. Set up your simulator according to the steps used for verifying your
implementation of the Moose algorithm. Set the Correct Frequency
Offset control to FALSE. Describe what happens to the received constellation if you do not correct for the 201 Hz frequency offset.
3. Based on the system parameters described in the pre-lab, what are
the range of frequency offsets that you can estimate/correct using the
Moose algorithm? Since the Moose algorithm cannot estimate/correct
for arbitrary frequency offsets it is often considered to perform what is
known as fine frequency synchronization.

i
i

i
book2 2011/10/28 14:05 page 95 #107

Lab 5: Frame Detection & Frequency Offset Correction

95

Pre-Lab Turn In
1. Submit your implementation of Moose.vi and sliding correlator.vi. Remember, you will be penalized if you do not wire the error cluster inputs
and outputs in sliding correlator.vi and Moose.vi.
Note: If you need to generate additional subVIs for this or any future
labs, please submit these subVIs along with your other pre-lab VIs.
2. Submit your answers to all of the questions in the pre-lab section.

Lab Experiment

In this lab you will run your implementation of sliding correlator.vi over a
real wireless link. Using your code from the prior labs in conjunction with
the new blocks you have created, you will complete the framework for the
transmitter and receiver blocks in lab as in previous labs. Insert your code
into this framework as you did in the pre-lab. Next, set up the following
parameters in your system.
TX sample rate = 4 MSamp/sec
TX oversample factor = 20
RX sample rate = 4 MSamp/sec
RX oversample factor = 20
Channel estimate length = 2
Equalizer length = 4
Capture time = 2 msec
Leave any unspecified parameters set to their default values. Also make
sure that the Correct Frequency Offset control on top rx is set to TRUE. You
will now perform two experiments to explore frequency offsets and frame
detection over a real wireless link. Before you begin, estimate the order of
magnitude of the frequency offset between the transmitter and receiver of
your system. Take note of this value as you may be asked about it later.

3.1

Correlation Based Frame Detection Under Large

Frequency Offsets

The correlation based detector discussed in this lab can be impacted by a

number of impairments, including frequency offset, as discussed previously.

i
i

i
book2 2011/10/28 14:05 page 96 #108

96

Lab 5: Frame Detection & Frequency Offset Correction

Answer the following questions to explore the relationship between frequency

offset and your correlation based detector.
1. Before changing any of the settings, calculate an average value (using five runs or so) of the inherent offset between the transmitter and
receiver.
2. Based on the system parameters, what is the range of frequency offsets
that can be estimated by your frequency offset estimation algorithm?
3. Let fM be the maximum correctable frequency offset of your system
(i.e., as calculated in the previous question). Modify the carrier frequency of your transmitter so that you will cause an effective offset of
0.80fM at the receiver. What is the new value of the carrier frequency
at your transmitter?
4. How does this frequency offset impact the performance of your correlation based frame detector (i.e., does your correlation based detector
still find the start of the frame)? Describe what happens to the error
rate and constellation of your system.
You will include your answers to these questions in your lab report. Note
that the frequency offset 0.80fM is two orders of magnitude greater than the
inherent offset between the transmitter and receiver.
The correlation based detector discussed in this lab can be impacted by a
number of impairments, including frequency offset, as discussed previously.
Answer the following questions to explore the relationship between frequency
offset and your correlation based detector.
1. Based on the system parameters, what is the range of frequency offsets
that can be estimated by your frequency offset estimation algorithm?
2. Let fM be the maximum correctable frequency offset of your system
(i.e., as calculated in the previous question). Modify the carrier frequency of your transmitter so that you will cause an effective offset of
0.80fM at the receiver. What is the new value of the carrier frequency
at your transmitter?
3. How does this frequency offset impact the performance of your correlation based frame detector (i.e., does your correlation based detector
still find the start of the frame)? Describe what happens to the error
rate and constellation of your system.
You will include your answers to these questions in your lab report.

i
i

i
book2 2011/10/28 14:05 page 97 #109

Lab 5: Frame Detection & Frequency Offset Correction

3.2

97

Self referenced frame synchronization

In Section 3.1, you saw how the correlation based detector can break under large frequency offsets. You will now explore another frame detection
algorithm which is more robust to frequency offsets. As discussed in class, a
self referenced frame synchronization method can be derived from the leastsquares expression of the Moose algorithm. This frame detection performs
the calculation in Eq. (10). This frame detection method has already been
implemented in digital comm3.2b.llb. Change the frame detection method
of your system by modifying the appropriate parameter on the front panel
of top rx.vi. Observe how the Self Reference Frame Synchronization method
performs under the large frequency offset computed in the last section.
Consider a simple flat fading channel model with frequency offset foffset =
/T (T is the symbol period of the system) such that the received sequence
is given by
y[n] = ej 2 n t[n D],
(11)
where t[n] is the transmitted training sequence, is a complex scalar, and D
is the delay introduced by the channel. Include the answer to the following
question in your lab report:
Question: Using the simple channel model of Eq. (11), explain why the self
referenced frame detector of Eq. (10) works better than the correlation
based detector under large frequency offsets. You may also need to
consider the mathematical description of the correlation based detector
given in Eq. (2). Hint: Notice that the training sequence is periodic
(i.e., t[n + W ] = t[n] for all n {0, 1, . . . , W 1}). Also you should
notice that the output of the frame detection algorithm should be D for
this particular channel model.

3.3

Lab Turn In

Demonstrate to the instructor that your code is working properly. Show

the instructor that your correlation based detector breaks under a large
frequency offset (as discussed in Section 3.1). Then show the instructor
that under this same frequency offset, the self referenced frame synchronizer
works. Also answer all of the questions above. You will be required to submit
these answers in a lab report.

DIRECTIONS FOR LAB SUBMISSION

Your lab report should include the following.
1. Answer all questions from the lab experiment (i.e., do not reanswer
pre-lab questions).

i
i

i
book2 2011/10/28 14:05 page 98 #110

98

Lab 5: Frame Detection & Frequency Offset Correction

2. Discuss any problems you might have encountered and how you overcame these obstacles.
3. In this lab you observed how your code operates over a real wireless
link. Answer the following questions about your experience in lab.
(a) Name three impairments in a real wireless system which impact
the accuracy of the correlation based detection method.
(b) Under large frequency offsets, the correlation based detector
breaks. What property of the training sequence breaks down
under large frequency offsets? Hint: You may need to consider
your answer to the question in Section 3.2.
(c) In the self referenced frame synchronization method, is it still
critical that the periodic training sequence have good autocorrelation properties? If so, why? Use the simple channel model of
Eq. (11) to help answer this question, if necessary.

i
i

i
book2 2011/10/28 14:05 page 99 #111

Lab 6: OFDM Modulation & Frequency Domain

Equalization

Summary
In this lab you will implement the key features of the orthogonal frequency
division multiplexing (OFDM) multicarrier modulation technique. OFDM is
a transmission technique with a special structure that permits low complexity linear equalization at the receiver. Several commercial wireless systems
have adopted OFDM modulation including wireless LAN standards like IEEE
802.11g, IEEE 802.11a, and IEEE 802.11n; broadband wireless access including IEEE 802.16 (WiFi); mobile broadband wireless known as IEEE
802.20; digital video broadcasting DVB (used in Europe); as well as several
releases of the 3GPP cellular standards. The OFDM system considered in
this lab is illustrated in Figure 1.
Although the design of an OFDM system has several important differences from the single carrier system considered in previous labs, it must still
perform many of the same functions (e.g., channel estimation, equalization,
and frequency offset estimation). In this lab you will still perform all functions related to synchronization and channel estimation in the time domain
using the same training sequence discussed in Lab 5. The framing structure
considered in this lab is illustrated in Figure 2.
For the pre-lab submission, you have to turn in the three VIs described in
Section 2 (OFDM modulate.vi, OFDM demodulate.vi, and FEQ.vi). Further,
you have to submit the answers to questions in Section 2.

1
1.1

Background
OFDM Modulation

Figure 1 shows a block diagram of the basic OFDM modulator you will be
implementing in this lab. OFDM is a type of digital modulation where information is modulated into discrete-time sinusoids. With OFDM, the symbols
after the constellation mapping are considered to start in the frequency domain. OFDM operates on groups of symbols, the resulting group of symbols
being called an OFDM symbol. To begin the explanation, we assume K = 0
and that there are N frequency-domain symbols in one OFDM symbol. The
number of subcarriers is given by N , typically a power of 2. The transmitter
operates as follows. Given {s[n]}N1
n=0 and cyclic prefix of length Lc , the
99
i

i
i

i
Symbol
Mapping

Tz = T/M

C/D

Insert
Training

g RX [n]

s[n]

M
M

...

Extract
Training

y[n]

Serial
to
Parallel
1:N-K

...

Estimate
Channel

Serial
to
Parallel
( N+L c ):1

Insert
K Null
Tones

...

FFT
N

FFT

discard

IFFT
N

Add
CP

...
Remove
K Null
Tones

Parallel
to
Serial
(N+L c ):1

Parallel
to
Serial
N-K:1

Detection

gTX [n]

Inverse
Symbol Map

Tx

D/C

Ex

bits

x(t )

Figure 1: The OFDM system considered in this lab. The transmitter sends groups of N K symbols where N is the number of subcarriers and
K the number of null subcarriers. Null symbols are inserted into groups of N K transmit symbols, which are transformed from frequency to
time domain, then prepended with a cyclic prefix of length Lc at the transmitter. The receiver processes blocks of length N + Lc , discarding
the first Lc samples of each block. Single tap equalization is performed in the frequency domain. The operations of carrier frequency
synchronization and frame synchronization are omitted for simplicity.

z (t )

OFDM Receiver

bits

100

...

...

OFDM Transmitter

i
i

book2 2011/10/28 14:05 page 100 #112

i

Lab 6: OFDM Modulation & Frequency Domain Equalization

i
book2 2011/10/28 14:05 page 101 #113

Lab 6: OFDM Modulation & Frequency Domain Equalization

OFDM symbol
train

train

CP1 OFDM 1

preamble

CP2 OFDM 2

101

...

OFDM payload

Figure 2: The framing structure considered in this lab. The preamble consists of two
repeated training sequences from Lab 5 (a total of four Barker codes). The preamble
is sent using the usual complex pulse amplitude modulation. Subsequent data is sent
using OFDM modulation, which is a block-based modulation technique.

transmitter produces the sequence

N1
m(nLc )
1 
w[n] =
s[m]ej 2 N n = 0, ..., N + Lc 1
N
n=0

that is passed to the transmit pulseshaping filter. The samples from n =

Lc , Lc + 1, . . . , N + Lc 1 are the output of the inverse discrete Fourier
transform (DFT) of the input symbols {s[m]}N1
n=0 , which could be implemented using the fast Fourier transform (FFT).
Now observe that
w[n] = w[n + N ]
for n = 0, 1, . . . , Lc 1. This means that the first Lc samples of w[n] are
the same as the last Lc samples. The first Lc samples are known as a cyclic
prefix. The length of the cyclic prefix Lc should be at least as long as Lh .
This will ensure that there is no intersymbol interference between adjacent
OFDM symbols (i.e., the CP guards against ISI). The cyclic prefix serves
another important purpose. It helps convert (part of) the linear convolution
into a circular convolution.
In OFDM rectangular pulse-shaping is common; windowing of the symbols is used to shape the frequency spectrum. This is not considered in this
lab.
Assuming carrier frequency synchronization and frame synchronization
have been accomplished, the received observes the following signal after
matched filtering, symbol timing, frame synchronization, and downsampling
y[n] =

L


h[]w[n ] + v[n].

=0

The receiver operates on blocks of data.

i
i

i
book2 2011/10/28 14:05 page 102 #114

102

Lab 6: OFDM Modulation & Frequency Domain Equalization

Note: This exposition implies that to recover multiple blocks of symbols

(i.e., the input to the OFDM modulator), you will need to process multiple
OFDM symbols, each of length N + Lc . Assuming that the indexing starts
at the first OFDM symbol, the receiver discards the first Lc samples to form
for n = 0, 1, . . . , N 1 (neglecting noise) y[n]

= y[n + Lc ]. With some

mathematical manipulations it can be shown that
y[n]

L


h[l]w[n + Lc l]n = 0, ..., N 1

l=0

=
=
=

L
N1

m(n+Lc Lc l)
1 
N
h[l]
s[m]ej 2
N

1
N
1
N

l=0

m=0

L


N1


h[l]

l=0
N1


'

m=0

s[m]ej 2

m=0
L


h[l]ej 2

mn
N

ej 2 N

ml

(
ml
N

s[m]ej 2

mn
N

l=0

The receiver then takes the DFT (or FFT) of these samples to produce
(after some simplification)
Y [k]

=
=

DF T [y[n]]

H [k]s[k] + V [k],

(1)

where
H [k] =

L


h[l]ej 2 N ,
kl

l=0

which is just the DFT of the zero padded channel. The frequency domain interpretation of OFDM comes from Eq. (1). Essentially information is sent on
discrete-time sinusoids, or carriers. The information on the k th discrete-time
sinusoid experiences the channel response determined by H [k]. Equalization simply requires dividing Y [k] by H [k]. We refer to this as the frequency
domain equalizer (FEQ).
The following parameters are useful when discussing OFDM.
T is sample period.
T (N + Lc ) is the OFDM symbol period.
The guard interval, or cyclic prefix duration, is Lc T .

i
i

i
book2 2011/10/28 14:05 page 103 #115

Lab 6: OFDM Modulation & Frequency Domain Equalization

103

The passband bandwidth is 1/T assuming use of a sinc pulseshaping

filter.
The subcarrier spacing c =

BW
N

1
NT

The length of the cyclic prefix Lc should be the same as or exceed the
order of the channel response Lh . Recall that the length of a channel of
order Lh is Lh + 1. This will ensure that there is no intersymbol interference
between adjacent OFDM symbols (i.e., the CP guards against ISI). The
guard interval is a form of overhead. This reduces the effective data rate that
is transmitted (i.e., for an OFDM system with OFDM sample rate Rs and M
bits/symbol, the effective data rate Rd < Rs ).
The guard interval serves to separate different OFDM symbols thus the
name. In practice the guard interval is determined by the maximum delay
spread. As the bandwidth increases, Lc must increase to compensate. The
subcarrier spacing refers to the spacing between adjacent subcarriers as measured on a spectrum analyzer. The larger the N , the smaller the subcarrier
spacing. The subcarrier spacing determines the sensitivity to Doppler and
residual carrier frequency offset.
The subcarrier spacing is inversely proportional to the number of subcarriers N . This means that as N increases you will get higher spectral efficiency,
but at the cost of increased sensitivity to frequency offsets.
It is common for not all subcarriers to be used in an OFDM system. Some
subcarriers are usually nulled or zeroed out in the frequency domain
(i.e., the null tones). The zero frequency or DC is commonly nulled due to
RF distortion at DC. Also, guard bands (i.e., frequencies at the edges of the
frequency response corresponding to those around s[N/2]) are commonly
nulled to prevent interference with signals in adjacent frequency bands. In
the lab we assume that K subcarriers are zeroed with the zero locations being
specified separately.

Pre-Lab

In this lab you will implement the OFDM modulator and demodulator as
described in the background section. You are required to build two VIs:
OFDM modulate.vi and OFDM demodulate.vi. These VIs will implement
the appropriate transmit and receive operations from Figure 1.
Before discussing the details of these VIs, you will learn about a set of
helper VIs which have been provided to you in OFDM comm1.0.llb, a new
digital communications library for processing related to OFDM. The VIs
described in Table 1 are located in OFDM comm1.0.llb and may be helpful
in your implementations of OFDM modulation and demodulation.

i
i

i
book2 2011/10/28 14:05 page 104 #116

104

Lab 6: OFDM Modulation & Frequency Domain Equalization

Name of VI
S2P.vi

P2S.vi

OFDM insert null tones.vi

OFDM add CP.vi
OFDM remove null tones.vi
OFDM remove CP.vi
OFDM FEQ.vi

Description
converts serial input stream (1-D array) to parallel blocks structure (2-D array in which each
row represents M consecutive elements from the
input stream)
converts parallel input stream (2-D array) to a
serial output stream (1-D array which contains
successive rows of input concatenated together)
insert columns of zeros into 2-D input array at
specified locations
prepends last C columns of 2-D input array to
start of array
remove specified columns from 2-D input array
remove first Lc columns from 2-D input array
computes frequency domain equalizer based on
channel estimate and equalizes parallel input
stream

Table 1: Helper VIs located in OFDM comm1.0.llb.

To efficiently process parallel inputs, these VIs use a parallel block structure represented in a 2-D array. In this structure, each row of the 2-D array
corresponds to a single block of data. Thus, the number of rows corresponds
to the number of blocks that are being processed.
Note: For-Loops in LabVIEW provide an auto-indexing feature which can
be used to extract the i th element of a 1-D array. Auto-indexing can also be
used on a 2-D array to extract the i th row of the 2-D array.
Tables 2 and 3 describe the VIs that you must construct. Many of the
OFDM related parameters you will need to use (such N , Lc , and null tones)
Table 2: Description of OFDM modulate.vi
OFDM modulate.vi - implements the OFDM modulator described in Section 1.1

INPUTS

OUTPUTS

input
symbols

1-D array of
CDB
(complex
doubles)

input symbol stream to be

modulated using OFDM

output
samples

1-D array of
CDB

output sample stream after

OFDM modulation; one
OFDM symbol corresponds to
(N + Lc ) samples

i
i

i
book2 2011/10/28 14:05 page 105 #117

Lab 6: OFDM Modulation & Frequency Domain Equalization

105

Table 3: Description of OFDM demodulate.vi

OFDM demodulate.vi - implements the OFDM demodulator described
in Section 1.2
input
samples

1-D array of
CDB

channel
estimate

1-D array of
CDB

number
of data
symbols

I32

equalize
channel?

Boolean

determines whether or not you

should apply FEQ to OFDM
demodulated symbols; set this
value to TRUE by default

demodulated
symbols
FD
channel
estimate

1-D array of
CDB

stream of output symbols after

OFDM modulation and FEQ
(when appropriate)

1-D array of
CDB

frequency domain response of

channel estimate computed
taking DFT of channel
estimate

INPUTS

OUTPUTS

input sample stream to be

demodulated using OFDM;
one OFDM symbol
corresponds to (N + Lc )
samples
channel estimate computed
prior to demodulation (used
for FEQ)
number of data symbols to be
recovered1

can be found unbundled from the OFDM parameters in cluster. After building
your VIs, insert your code into OFDM transmitter.vi and OFDM receiver.vi
as shown in Figure 3 and 4 respectively.
You have been provided with templates for the VIs you need to create for
this lab that already have all the inputs and outputs wired for you. What
is required of you is to finish constructing the block diagram to provide the
functionality of the VIs.
Throughout the course, several of the VIs you will create will have a
modulation parameters cluster passed into and out of them. The modulation
parameters in contains many of the parameters needed by your VIs and will
be unbundled from them. This has already been done for you in the template
VIs if necessary. Some VIs will also have modulation parameters out, so that
the cluster can be passed on through to the VIs that follow. Please ensure
1 Figure 1 implies that the output of the OFDM demodulator produces multiples of N symbols.
You should truncate the array to the appropriate size (as dictated by number of data symbols).

i
i

i
book2 2011/10/28 14:05 page 106 #118

Lab 6: OFDM Modulation & Frequency Domain Equalization

Figure 3: Block diagram for OFDM transmitter.vi.

106

i
i

i
book2 2011/10/28 14:05 page 107 #119

107

Figure 4: Block diagram for OFDM receiver.vi.

Lab 6: OFDM Modulation & Frequency Domain Equalization

i
i

i
book2 2011/10/28 14:05 page 108 #120

108

Lab 6: OFDM Modulation & Frequency Domain Equalization

that these clusters remain wired the way they are in the template VIs, as
changing how they are wired will cause VIs further down the line to break.
In addition, you will now have access to the OFDM parameters cluster for
OFDM-specific parameters you will need for you VIs. These are accessed in
the same way that parameters from the modulation parameters cluster are.
After inserting your code into the simulator provided to you, verify that
your code is working properly by observing that the transmitted and received
constellations reflect the appropriate PSK modulation scheme you have chosen.

Pre-Lab Turn In
1. Submit your implementation of OFDM modulate.vi and OFDM demodulate.vi. Remember, you will be penalized if you do not wire error
cluster inputs and outputs. Note: If you need to generate additional
subVIs for this or any future labs, please submit these subVIs along
with your other pre-lab VIs.
2. EXTRA CREDIT (20pts) - implement FEQ.vi (i.e., your own version
of OFDM FEQ.vi). Table 4 describes the details of this VI. Submit
your implementation of FEQ.vi with your other VIs. Remember to
wire error clusters.
Table 4: Description of FEQ.vi
FEQ.vi - performs frequency domain equalization as described in Section 1.3
input

2-D array of
CDB

channel
estimate

1-D array of
CDB

equalized
output

2-D array of
CDB

equalized symbols; each row

corresponds to a block of N
N1
)
symbols (i.e., {X[k]}
k=0
) = Y [k]/H
*[k])
where X[k]

FD
channel
estimate

1-D array of
CDB

frequency domain response of

channel estimate computed by
taking DFT of channel
estimate (i.e.,
*[k] = DF T {h[l]})
H

INPUTS

OUTPUTS

parallel block structure of

symbols (i.e., output of DFT);
each row corresponds to a
block of N symbols (i.e.,
{Y [k]}N1
k=0 )
channel estimate computed
prior to demodulation

i
i

i
book2 2011/10/28 14:05 page 109 #121

Lab 6: OFDM Modulation & Frequency Domain Equalization

109

Lab Experiment

A single carrier system operating with symbol rate 1/T would transmit one
symbol every symbol period T . Thus in time N T , the single carrier system would transmit N symbols; maintaining orthogonality (or separation)
between symbols through time division multiplexing (or TDM). As you have
learned in previous labs, a frequency selective channel can undo this orthogonality, making some form of equalization necessary.
Multicarrier systems use a cyclic prefix to enable digital frequency domain
equalization. An interpretation is that OFDM divides a frequency selective
channel into N flat fading subchannels. Thus OFDM can be seen as N
parallel flat fading channels, (each of bandwidth 1/N T ), multiplexed in the
frequency domain. Hence, a simple zero-forcing equalizer can be applied to
each subchannel in order to equalize channel impairments. In this lab you
will examine the frequency selectivity of wireless channels and explore when
it might be advantageous to use OFDM to forgo the complexities of linear
equalization.
Along with the advantage of low complexity equalization, OFDM systems
also provide a framework which allows for many advanced digital communication techniques such as adaptive modulation and power control. Additionally, the frequency domain interpretation of OFDM (i.e., as a set of N parallel
flat fading subchannels) allows for many clever techniques with regard to error control coding by taking advantage of frequency diversity (topics that are
outside of the scope of this course). All of these advantages do not come for
free. There are a number of tradeoffs associated with single and multicarrier
systems. In this lab you will explore one particular tradeoff, the sensitivity
of OFDM systems to frequency offsets.
In this lab you will run your implementation of OFDM modulate.vi and
OFDM demodulate.vi over a real wireless link. As in previous labs, you will
complete the framework for the transmitter and receiver blocks in lab using
your code from this and previous labs. Insert your code into this framework
as you did in the pre-lab. You will then perform two experiments to explore
the previously mentioned topics: 1) the frequency selectivity of wireless
channels and 2) the sensitivity of OFDM systems to frequency offsets.

3.1

Insert Your Code

Figure 5 depicts the dependencies between files in the new OFDM framework
you will be using in lab. Note that this new framework still leverages many of
the blocks in the original digital communications library (digital comm.llb),
but also includes some of the elements from the new OFDM digital communications library (OFDM comm1.0.llb).

i
i

i
book2 2011/10/28 14:05 page 110 #122

110

Lab 6: OFDM Modulation & Frequency Domain Equalization

Figure 5: Hierarchy of files for OFDM framework.

Once you have inserted your code for OFDM modulate.vi and OFDM demodulate.vi into OFDM transmitter.vi and OFDM receiver.vi respectively,
you will need to set up the parameters listed below. These parameters are
located on the front panels of top tx.vi and top rx.vi. Leave any unspecified
parameters set to their default value.
Packet length = 500 bits
Modulation type = QPSK
Channel estimate length = 4
FFT size (N ) = 64
Length of CP (Lc ) = 8
Null tones = {0, 31, 32, 33}

3.2

Frequency Selectivity of Wireless Channels

In this first experiment you will observe the frequency response of narrowband
and wideband channels. Set up the following parameters for a narrowband
system.
TX sample rate = 4 MSamp/sec

i
i

i
book2 2011/10/28 14:05 page 111 #123

Lab 6: OFDM Modulation & Frequency Domain Equalization

111

TX oversample factor = 20
RX sample rate = 4 MSamp/sec
RX oversample factor = 20
Capture time = 2.4 msec
After transmitting a packet successfully, observe the frequency response
of the narrowband channel, using the Channel Response graph located on
the front panel of OFDM receiver.vi. Also examine the instantaneous powerdelay profile using the appropriate graph on the front panel of the same VI.
Take note of the effective length of the channel response (i.e., the number of
nonzero taps in the channel).
Next, you will observe the frequency response of a wideband channel.
To set up a wideband system, set the following parameters in your OFDM
system.
TX sample rate = 20 MSamp/sec
TX oversample factor = 4
RX sample rate = 10 MSamp/sec
RX oversample factor = 2
Capture time = 100 sec
For this part of the experiment it is critical that you place your antennas at an elevated height so that all reflected paths in the lab environment reach the antennas. After transmitting a packet successfully observe the frequency response
and power-delay profile of the wideband channel in OFDM receiver.vi. Also,
take note of the effective length of the channel response.
Questions
Answer the following questions regarding the frequency selectivity of wireless channels.
1. According to the parameters above, what is the respective OFDM symbol rate in the narrowband and wideband systems you have set up (i.e.,
the reciprocal of the OFDM symbol period as previously defined)?
2. What are the effective lengths of the narrowband and wideband channels respectively?
3. Describe the frequency responses of each channel. In particular, are
the frequency responses of these channels frequency selective or flat?

i
i

i
book2 2011/10/28 14:05 page 112 #124

112

Lab 6: OFDM Modulation & Frequency Domain Equalization

4. Consider the multipath channel model in the absence of noise

y[n] =

Lh


h[l]x[n l],

(2)

l=0

where h[n] = 0, n
/ (0, 1, . . . , Lh ).
Show that in an OFDM system, when Lh = 0 the frequency
response of the channel is necessarily flat fading. In other words,
show that the frequency response of all subchannels is the same
(i.e., H [n] = H [m], n, m, where H [k] = DF T {h[l]}).
Show that when Lh > 0 the frequency response of the channel
is frequency selective (i.e., show that Lh > 0 H [n]  = H [m]
for at least one different n and m).
Submit your answers to these questions as part of your lab report.

3.3

Sensitivity to Frequency Offsets

In this experiment you will observe how the performance of an OFDM system degrades in the presence of a frequency offset. Set up the following
parameters in your system.
TX sample rate = 20 MSamp/sec
TX oversample factor = 20
RX sample rate = 4 MSamp/sec
RX oversample factor = 4
Capture time = 500 sec
Frequency offset (Hz) = 200 Hz
To observe how a frequency offset impacts your system, you will need to
disable the frequency offset correction algorithm implemented in the synchronization block. To do this, simply set Correct Frequency Offset on the
front panel of top ofdm rx.vi to FALSE. Now you will consider two OFDM
systems which use different values for N . First, consider a system which
uses N = 64 subcarriers. Set up the following OFDM parameters.
FFT size (N ) = 64
Length of CP (Lc ) = 8

i
i

i
book2 2011/10/28 14:05 page 113 #125

Lab 6: OFDM Modulation & Frequency Domain Equalization

113

Null tones = {0, 31, 32, 33}

Observe how a frequency offset impacts the received signal constellation.
Also make note of the impact of a frequency offset on BER performance.
Increase the amount of frequency offset in your system in increments of 200
Hz and observe how the signal constellation and BER performance vary. Now
consider a system which uses N = 1024 subcarriers.
FFT size (N ) = 1024
Length of CP (Lc ) = 32
Null tones = {0, 511, 512, 513}
Again, observe how a frequency offset of just 200 Hz impacts the received
constellation and BER performance of your system. To understand how
a frequency offset impacts OFDM systems, it is convenient to think of a
frequency offset as a shift in the frequency domain. This shifting can cause
what is known as inter-carrier interference (or ICI).
Questions
Answer the following questions about the sensitivity of OFDM systems to
frequency offsets.
1. Recall from the Lab 5 that in a single carrier system, a frequency
offset causes a time varying phase offset which effectively smears the
received constellation as it rotates it. How is the impact of a frequency
offset in OFDM systems different from that of single carrier systems?
In particular, how is the impact on the signal constellation different?
2. What is the subcarrier spacing c of your system when N = 1024 and
64 respectively?
3. Which of the systems (i.e., N = 1024 or 64) is more severely impacted
by a 200 Hz frequency offset? Why?
4. As mentioned, frequency offsets can be interpreted as shifts in the
frequency domain. There is a duality between the effect of this shifting
in a multicarrier system and symbol timing error in a single carrier
system (discussed in Part 2 of Lab 2).
Discuss this relationship between the effect of symbol timing error
in single carrier systems and frequency offset in OFDM systems. Be
sure to include in your discussion the effect of each impairment on the
received signal constellation. Hint: Think about the impact of each
type of error on the orthogonality (or separation) between symbols.
Submit your answers to these questions as part of your lab report.

i
i

i
book2 2011/10/28 14:05 page 114 #126

114

3.4

Lab 6: OFDM Modulation & Frequency Domain Equalization

Lab Turn In

Show the instructor that you have replaced the OFDM modulator and demodulator in lab with your own code. Demonstrate to the instructor that your
code is working properly. Also show the instructor that your OFDM system
breaks down when you have frequency offsets greater than the subcarrier
spacing that are not corrected for (i.e., when foffset > c ). Answer all of the
questions above and submit your answers in your lab report.

DIRECTIONS FOR LAB SUBMISSION

Your lab report should include the following.
1. Answer all questions from the lab experiment (i.e., do not reanswer
pre-lab questions).
2. Discuss any problems you might have encountered and how you overcame these obstacles.
3. In this lab you observed how an OFDM system operates over a real
wireless link. Answer the following questions about your experience
in lab.
(a) OFDM systems have an inherent overhead due to the need for a
cyclic prefix. Assuming QPSK modulation, in terms of N , Lc ,
T , and the number of null tones K (all as previously defined),
what is the effective data rate of the OFDM system in lab?
(b) Based on your experience in lab, discuss why or why not you
might use an OFDM system in a wideband and narrowband system respectively. Include in your discussion some of the benefits/costs of each scenario.
(c) OFDM systems in general have a large degree of flexibility. Name
at least three parameters of OFDM systems which contribute to
this flexibility and comment on how they do so.

i
i

i
book2 2011/10/28 14:05 page 115 #127

Lab 7: Synchronization in OFDM Systems using

Schmidl and Cox Algorithm

Summary
In Lab 6 you implemented key features of the orthogonal frequency division multiplexing (OFDM) multicarrier modulation technique. In particular
you implemented the encoding from frequency domain to time domain at
the transmitter, and the low complexity frequency domain equalization at
the receiver. Systems that employ OFDM modulation, like those that employ
single carrier modulation, also require frame synchronization and carrier synchronization, not to mention channel estimation. In Lab 6, frame and carrier
synchronization was performed using the single carrier modulated training
sequence from previous labs. In this lab you will perform frame and frequency synchronization using an OFDM modulated training sequence, using
the short training sequence, as defined by the IEEE 802.11a standard, to perform synchronization. The OFDM system considered in this lab is illustrated
in Figure 1.
The emphasis of this lab is using an OFDM modulated training signal
to perform carrier and frequency synchronization. First you will review the
IEEE 802.11a preamble including short and long training sequences. Then
you will learn about the Schmidl and Cox Algorithm (SCA), which leverages
the periodic structure of the OFDM training sequence to perform frame and
frequency synchronization.
For the pre-lab submission, you have to turn in theVI described in Section 2
(SchmidlCox.vi). Further, you have to submit the answers to questions in
Section 2.

1
1.1

Background
Impairments in Frequency Selective Channel

In this section we review the baseband model for delay and frequency offset in
the presence of a multipath channel. Referring to Figure 1, if the carrier offset
is small and sufficient sampling is employed, then the discrete-time signal at
the receiver after matched filtering and downsampling can be written as
y[n]

j 2 n

L


h[l]w[n l D] + v[n],

l=0

115
i

i
i

i
Symbol
Mapping

z (t )

Insert
Training

g RX [n ]

s[n]

M
M

y[n]

...

Joint Freq
& Frame Sync

Serial
to
Parallel
1:N-K

...

Extract
Training

Insert
K Null
Tones

...

Add
CP

Estimate
Channel

Serial
to
Parallel
( N+L c ):1

IFFT
N

Parallel
to
Serial
(N+L c ):1

FFT
N

FFT

discard

...

Remove
K Null
Tones

Parallel
to
Serial
N-K:1

gTX [n]

Detection

Tx

Ex

Inverse
Symbol Map

D/C

x(t )

bits

Figure 1: The OFDM system considered in this lab including joint carrier frequency offset estimation and frame synchronization.

Tz = T/M

C/D

OFDM Receiver

bits

116

...

...

OFDM Transmitter

i
i

book2 2011/10/28 14:05 page 116 #128

i

Lab 7: Synchronization in OFDM Systems

i
book2 2011/10/28 14:05 page 117 #129

117

Lab 7: Synchronization in OFDM Systems

where is the discrete-time frequency offset, D is an unknown delay,

h
and {h[]}L
=0 are the coefficients of the unknown frequency selective channel. The frequency offset is estimated and removed during carrier frequency
offset synchronization. The delay D is estimated and removed as part of
frame synchronization. The channel coefficients are estimated and used to
design an equalizer.

1.2

IEEE 802.11a Short Training Sequence

In this lab you will use a carefully constructed training sequence to perform
frequency offset synchronization and frame synchronization. Figure 2 shows
the IEEE 802.11a training structure described in the Physical Layer Specification [4]. The training sequence is composed of a short and long
training sequence. The short training sequence is composed of 10 identical
subsequences and has a total length of Nt = 160. The long training sequence
is composed of two identical subsequences with an extra long cyclic prefix.
In this lab you will use the short training sequence. The frame structure will
be the same as the Lab 6 with the time domain training replaced by the short
training sequence.
The short training sequence is created from two OFDM symbols in a
very special way. We describe how it is performed mathematically here.
Consider the symbols {s[n]}N1
n=0 that will be used to generate the coefficients
for one OFDM symbol. Suppose that s[n] is generated by upsampling the
N/L
symbols {s [n]}n=0 where L is an integer. First consider the IDFT with zero
cyclic prefix. Then
N1
2 mn
1 
s[m]ej N .
N

w[n]

(1)

m=0

Remembering that s[Ln] = s [n] and that s[n] = 0 for values of n that are
not an integer multiple of L, it follows that

w[n]

N/L1
2 Lmn
1 
s[Lm]ej N
N
m=0

1
N

N/L1


s [m]e

mn
j 2
N/L

m=0

i
i

i
book2 2011/10/28 14:05 page 118 #130

Lab 7: Synchronization in OFDM Systems

Figure 2: IEEE 802.11a training structure.

118

i
i

i
book2 2011/10/28 14:05 page 119 #131

119

Lab 7: Synchronization in OFDM Systems

Because of the periodicity of the complex exponential, it can be verified that

for n = 0, 1, . . . , N/L 1
N/L1
1 
j 2 m(n+kN/L)
N/L
s [m]e
N

w[n + kN/L] =

m=0

1
N

N/L1


s [m]e

2 mkN/L
mn
j 2
N/L j
N/L

m=0

N/L1
1 
j 2 mn
s [m]e N/L
N

w[n].

m=0

This means that by zero every L 1 subcarriers, it is possible to make the

OFDM symbol look periodic with L periods.
To conclude the discussion now suppose there is a cyclic prefix of length Lc
and that N/L = Lc
w[n]

N1
m(nLc )
1 
s[m]ej 2 N n = 0, ..., N + Lc 1.
N

(2)

m=0

Remembering the zero subcarriers for n = 0, 1, . . . , N + Lc 1

w[n]

N/L1
c)
1 
j 2 m(nL
N/L
s [m]e
N
m=0

=
=

1
N
1
N

L
c 1

s [m]ej 2

m(nLc )
Lc

m=0
L
c 1

mn

s [m]ej 2 Lc ,

m=0

which satisfies the property that w[n+kLc ] = w[n] for n = 0, 1, . . . , Lc 1

and k = 0, 1, . . . , N/Lc . Therefore the cyclic prefix becomes an additional
period of repetition!
The IEEE 802.11a standard [4] specifies that N = 64 and L = 4 in the
short training sequence, thus yielding a training sequence which is periodic
with period Lc = 16, which is also the length of the cyclic prefix. Two
OFDM symbols concatenated together form a sequence with 160 samples,
that consists of 10 repetitions of the same sequence. You will use this training
sequence in the lab.

i
i

i
book2 2011/10/28 14:05 page 120 #132

120

Lab 7: Synchronization in OFDM Systems

1.3 The Schmidl and Cox Algorithm

The Schmidl and Cox Algorithm (SCA) [16] is a clever specialization of the
Moose algorithm [7] to OFDM waveforms. It uses periodicity created in the
OFDM symbol to perform the frequency offset correction and frame synchronization. A main difference over the Moose algorithm is that it consists
of two steps of frequency offset correction. The first step corrects for offsets
that are a fraction of 1/N, the subcarrier spacing. The second step corrects
for integer offsets that are multiples of the subcarrier spacing. The integer
offset is performed using a specially constructed second training symbol. In
this lab we focus on the fine frequency offset correction.
The training sequence considered in the lab has 10 repetitions of the same
sequence in the 160 samples. For maximum compatibility and reuse with
the Moose algorithm developed in previous labs, we treat this sequence as
two repetitions of length W = 80. There are algorithms that can exploit the
multiple periods of repetition, however, they will not be considered in this
lab.
The steps of the SCA that you will implement are summarized as follows.
1. The first step is to compute the correlation metric
R[d]

N1


y [n + d]y[n + d + W ],

(3)

n=0

where W is half the size of the periodic training sequence and N is the
number of subcarriers.
2. The second step is to compute the average received signal power
P [d]

N1


|y[n + d + W ]|2 .

(4)

n=0

%
%
% R[d] %2
% .
3. From P [d] and R[d], compute M[d] = %%
P [d] %
4. Approximate delay parameter D by solving for d = argmax M[d]
d

searching over a reasonable range of samples.

5. Estimate the frequency offset parameter as
= phase

P [D]
.
R[D]

(5)

i
i

i
book2 2011/10/28 14:05 page 121 #133

Lab 7: Synchronization in OFDM Systems

121

6. Correct for the frame offset and frequency offset as in the Moose algorithm.
The algorithm described above is an abbreviated version of the full SCA
discussed in [16]. The full version of this algorithm includes a method for
coarse frequency offset estimation/correction1 , which will not be discussed
in this lab.

Pre-Lab

Questions
Answer the following questions before implementing the Schmidl and Cox
algorithm.
1. In SCA, the delay estimate d is computed through an exhaustive search
over the set of valid delay values (i.e., the search window). Suppose
that the length of the captured sequence is C samples. In terms of C, N ,
Lc , and Nof dm (the number of OFDM symbols in the OFDM payload,
which is described in Lab 6), what is the set of valid delay values you
should search through to find d ?
2. For the IEEE 802.11a short training sequence being used in lab, what
is the range of frequency offsets that can be estimated/corrected using
SCA (or equivalently, the Moose algorithm)? You may assume an
OFDM sample rate of 1/T = 1MH z.
3. Given that the IEEE 802.11a short training sequence contains 10 periods of a length 16 sequence (as described above), propose a frequency
offset scheme that allows you to correct for a greater range of frequencies than computed in the last question. You may assume that the
frame detection problem has been solved, and that an aligned sequence
is available.

Programming
In this lab you will be implementing the Schmidl and Cox algorithm for
frame and frequency synchronization. You may use your previously constructed implementation of the Moose algorithm (or the one available in the
digital communications library provided to you) for frequency offset estimation/correction. This means that you will only need to implement the steps
1 Coarse frequency offset is the same as a modulo offset [i.e., an integer multiple of
1/(2 W T )].

i
i

i
book2 2011/10/28 14:05 page 122 #134

122

Lab 7: Synchronization in OFDM Systems

Figure 3: Hierarchy of files associated with OFDM frame detect.vi.

of SCA that pertain to frame detection (as described in Section 1.3). You
will build the VI SchmidlCox.vi, described below. After building this VI you
will insert it into OFDM frame detect.vi shown in Figure 4. You can find
OFDM frame detect.vi in the block diagram of OFDM synchronize.vi. The
hierarchy of these files is shown in Figure 3.
As stated in Table 1 and mentioned above, you do not need to reimplement
the Moose algorithm for frequency offset estimation/correction. Reuse your
previous implementation of this algorithm or the one provided to you in the
digital communication library.
You have been provided with templates for the VIs you need to create for
this lab that already have all the inputs and outputs wired for you. What
is required of you is to finish constructing the block diagram to provide the
functionality of the VIs.
Throughout the course, several of the VIs you will create will have a
modulation parameters cluster passed into and out of them. The modulation
parameters in contains many of the parameters needed by your VIs and will
be unbundled from them. This has already been done for you in the template
VIs if necessary. Some VIs will also have modulation parameters out, so that
the cluster can be passed on through to the VIs that follow. Please ensure
that these clusters remain wired the way they are in the template VIs, as
changing how they are wired will cause VIs further down the line to break.
In addition, you will now have access to the OFDM parameters cluster for
OFDM-specific parameters you will need for you VIs. These are accessed in
the same way that parameters from the modulation parameters cluster are.

i
i

i
book2 2011/10/28 14:05 page 123 #135

123

Figure 4: Block diagram of OFDM frame detect.vi.

Lab 7: Synchronization in OFDM Systems

i
i

i
book2 2011/10/28 14:05 page 124 #136

124

Lab 7: Synchronization in OFDM Systems

Table 1: Description of SchmidlCox.vi

SchmidlCox.vi - implements SCA for frame and frequency synchronization;
uses Moose algorithm to perform frequency synchronization.

INPUTS

OUTPUTS

input
complex
waveform
correct
frequency
offset

IQ waveform
cluster

output

1-D array of
CDB

received sequence after

correcting for estimated delay
d and frequency offset (when
appropriate)

frame offset

I32

frequency
offset

DBL

estimated delay d computed by

SCA
frequency offset estimate
computed by Moose algorithm

Boolean

received sequence after

matched filter and symbol
timing recovery
determines whether or not
your VI should correct for the
frequency offset estimated by
moose.vi

After inserting your code into the simulator, verify that it correctly finds
the start of the frame and corrects for frequency offset. You should test that
your code works for nontrivial delay and frequency offset up to fM .
Note: The suboptimal Schmidl and Cox algorithm presented here does not
work in the presence of noise or multipath; do not attempt to optimize the
algorithm to compensate for this shortcoming.
You will need to submit a plot (or screen capture of a plot2 ) of M[d] the
frame correlation metric. Generate the plot using dB of noise power
and a delay of 80 sec. Please leave all other parameters set to their default
value.

Pre-Lab Turn In
1. Submit your implementation of SchmidlCox.vi. Make sure you use the
parameters Training Sequence FFT Size and Training Sequence Length
of CP from the OFDM parameters in cluster in your implementation,
as these are parameters corresponding to the training sequence and not
the OFDM payload.
2You can plot a 1-D array of DBL in LabVIEW, simply by placing a waveform graph on the
front panel of your VI and connecting its terminal to the appropriate array on the block diagram
of your VI.

i
i

i
book2 2011/10/28 14:05 page 125 #137

Lab 7: Synchronization in OFDM Systems

125

Remember, you will be penalized if you do not wire error cluster inputs
and outputs.
Note: If you need to generate additional subVIs for this or any future
labs, please submit these subVIs along with your other pre-lab VIs.
2. Submit your answers to all of the questions in the pre-lab section.
3. Submit your plot of M[d]; generated using the parameters listed below.
Noise power = dB
Delay = 80 sec

Lab Experiment

In the pre-lab you implemented the Schmidl and Cox algorithm and observed
its performance in channels with delay and frequency offset. In this lab experiment you will learn about how noise impacts SCA and an optimization
to SCA which increases the robustness of this frame and frequency synchronization method for OFDM systems.

3.1

Impact of Noise on Schmidl and Cox Algorithm

As discussed in Section 1.3, SCA attempts to maximize the correlation between the first and second halves of the training sequence. Note that the
training sequence is composed of two identical OFDM symbols [including
their respective cyclic prefixes (i.e., Nt = 2(N + Lc ))]. Because of the cyclic
prefix (or CP), the objective function3 M[d] is the same for all delay values
d (D + Lh , . . . , D + Lc ) in the absence of noise, where D is the actual
delay of the channel. This phenomenon can be graphically interpreted as a
plateau of length Lc Lh , i.e., on which M[d] is maximized.
When noise is added to the system, the entire objective function is perturbed as shown in Figure 5. The once flat plateau of M[d] has now been
distorted. Thus, finding the argument which maximizes M[d] may give only
a rough, often erroneous, approximation of the delay (or frame offset).
Notice that qualitatively speaking the perturbed objective function M[d]
still retains an approximate plateau. Intuitively, this suggests that a better
approximation of delay might be found by finding the start of the plateau.
In the following sections, you will learn more about frame detection errors
and a formal definition of the intuitive method alluded to.
3 The term objective function refers to the function being optimized in a given optimization
problem.

i
i

i
book2 2011/10/28 14:05 page 126 #138

126

Lab 7: Synchronization in OFDM Systems

Figure 5: Objective function M[d] from SCA. Noise power No = 20 dB and delay
D = 80.

3.2

Good and Bad Frame Synchronization Error in

OFDM Systems

In Lab 6 that dealt with frame and frequency synchronization you learned how
a frame detection error can lead to severe bit error rates (i.e., a frame offset
estimation error of one can lead to a BER of 0.5). To understand why this
is not necessarily the case in OFDM systems, consider the impact of cyclic
shifts on the DFT of a periodic discrete sequence. Let x[n] be a periodic
sequence with period N and X[k] be the discrete Fourier transform of x[n].
If z[n] = x[(n S) modulo N ] (i.e., x[n] after a cyclic shift of S), then
Z[k]

N1


z[n]e

j 2

kn
N

(6)

n=0

X[k]e

j 2

kS
N ,

(7)

describes the DFT of the shifted sequence z[n]. This implies that a cyclic
th
shift of S imparts a kS
N phase offset to the k subcarrier. Therefore, in an
OFDM system, any frame detection error which leads to a cyclic shift of
the OFDM samples (i.e., after removing the cyclic prefix) will only impart a
constant phase offset to each subcarrier. This phase offset can be accounted
for by a properly formulated channel estimation algorithm.
Consider a frame detection error = d D, where D is the actual delay of
the resulting sequence
the channel. After correcting for the estimated offset d,
may not be aligned to the boundary of an OFDM symbol period. This means
that the start of the first OFDM symbol (i.e., the N samples not including the
CP) is located at Lc + .

i
i

i
book2 2011/10/28 14:05 page 127 #139

Lab 7: Synchronization in OFDM Systems

127

In the OFDM demodulation algorithm, the first Lc samples of each OFDM

symbol are removed before demodulation by a DFT. Thus, if (Lc Lh )
< 0, then the OFDM samples of each symbol undergo a cyclic shift of ||
(after removing the CP). As discussed earlier, such a frame detection error
leads to phase shifts in the DFT of each OFDM symbol, which can be accounted for by an appropriate least squares channel estimator. Thus this type
of frame detection error is considered to be acceptable.
Consider the scenario in which > 0. In this situation, each OFDM
symbol (after removing the CP) contains || samples from an adjacent symbol.
This means that every subchannel will be corrupted by data from an adjacent
OFDM symbol; this is an impairment which cannot be undone by a simple
frequency domain equalizer (FEQ). It is also interesting to note that this type
of impairment must be modeled as a non-causal system. For these reasons, a
frame detection error > 0 is not considered acceptable (i.e., a bad error).
In the next section you will learn about an optimization of SCA that tends
to produce a frame detection error < 0 (i.e., typically underestimates delay).
Questions
Include the answers to the following questions in your lab report.
1. Prove that the DFT of a cyclically shifted periodic sequence results in
a frequency dependent phase shift [i.e., given x[n] and S, as defined
above, prove that Eq. (7) is true].
2. Using the simulator provided to you (i.e., without inserting your own
code), you will observe how overestimating and underestimating delay
impacts bit error rate. Set the noise power No = 10 dB. Describe
what happens to the average bit error rate of the system when the
estimated delay exceeds the actual delay of the channel. You may have
to run the simulator multiple times before such a frame detection error
occurs.
3. Based on the discussion in this section and your simulation experience,
is it better to underestimate or overestimate the delay of the channel?
Justify your answer.

3.3

Optimizing the Schmidl and Cox Algorithm

The Schmidl and Cox algorithm, as described in Section 1.3, does not perform well in channels with even a moderate amount of noise. This poor
performance stems from the fact that SCA typically overestimates the delay
of the channel. You can use your simulator to verify this behavior.

i
i

i
book2 2011/10/28 14:05 page 128 #140

128

Lab 7: Synchronization in OFDM Systems

In [16] an optimization for SCA is proposed. You will implement a slight

variation of this optimization. Recall that even in the presence of noise, M[d]
has a plateau of approximate length Lc Lh . This optimization seeks to estimate the beginning of that plateau. The algorithm you will be implementing
is described by the steps below.
1. Given received sequence y[n], compute M[d] as described in SCA (see
Section 1.2).
2. Find Mmax = max M[d].
d

3. Let D = {d|M[d] 0.90(Mmax )}.

4. Estimate delay by solving for d = infD.
apply Moose algorithm for frequency
5. After correcting for offset d,
offset correction.
This algorithm will be referred to as optimized SCA. The set D contains an
optimistic4 estimate of the possible delay values which make up the plateau.
By taking the infimum of this set, optimized SCA estimates a lower bound
on the delay of the channel.
After applying this optimization to your implementation of SchmidlCox.vi,
you will test your code over a real wireless link. Insert your code for optimized
SCA into the framework for the receiver in lab. Leave all parameters in your
system set to their default values.
Questions
Include the answers to the following questions in your lab report.
1. Graph M[d], a 1-D array of CDB, on the front panel of SchmidlCox.vi
using a waveform graph. Discuss the relationship between the estimated frame offset computed by optimized SCA and the beginning of
the M[d] plateau (which you should be able to approximate from your
graph). Please include a screen capture of your graph of M[d] in your
lab report and refer to it in your discussion.
2. What assumption has been made regarding the quantity Lc Lh ? Why
is this assumption critical?
4 The set D is said to be optimistic because the set forms a superset of the values that actually
make up the length Lc Lh plateau of M[d].

i
i

i
book2 2011/10/28 14:05 page 129 #141

Lab 7: Synchronization in OFDM Systems

3.4

129

Lab Turn In

Demonstrate to the instructor that your code is working properly. Show the
instructor that your implementation of optimized SCA still breaks when
the algorithm overestimates the delay of the channel. Answer the questions
below.
Additional Questions
In this lab you learned about optimized SCA and how it operates over a real
wireless link. Answer the following questions about your experience in lab.
1. Compare and contrast the following synchronization methods you have
learned about in lab: (1) the self-referenced frame detection method
for a single carrier training sequence5 and (2) optimized SCA for an
OFDM training sequence6 .
2. Using the standard Moose algorithm, with no optimizations or modifications and the entire IEEE 802.11a short training sequence [i.e., where
(Nt = 160) (W = 80)], what is the range of correctable offsets for
IEEE 802.11a systems? Is this the best that you could do?
Hint: You may need to look at the IEEE 802.11a standard to find out
the OFDM sample rate or equivalently the system bandwidth.

DIRECTIONS FOR LAB SUBMISSION

Your lab report should accomplish the following tasks.
1. Answer all questions from Lab 7 (not including pre-lab questions).
2. Discuss any problems you might have encountered in Lab 7 and how
you overcame these obstacles.

5 Here single carrier training sequence refers to the sequence derived from the length 11
Barker code.
6 OFDM training sequence in this context refers to the IEEE 802.11a short training sequence
used in lab.

i
i

i
book2 2011/10/28 14:05 page 130 #142

Lab 8: Channel Coding in OFDM Systems

Summary
As you have learned from previous labs, one of the main advantages of OFDM
is the ability to forgo the complicated equalizer design1 needed by single
carrier systems to combat intersymbol interference (ISI). For OFDM systems,
to fully utilize their potential for high-data-rate communication in frequency
selective (ISI) channels they must take advantage of the diversity inherent to
frequency selective channels, making use of channel coding and interleaving.
In this lab you will explore the topic of channel coding. Figure 1 shows how
channel coding fits into the system model for our wireless communication
system. By adding structured redundancy to the transmitted sequence, channel coding provides added resilience to channel impairments and improves
overall throughput. This lab begins by introducing the common classes of
channel codes. You will then learn about a simple code, the repetition code,
which will reveal some of the basic benefits and costs of channel coding.
The repetition encoding and decoding blocks will be provided to you. All
that is required from you this lab is to compare the bit-error performance of
the OFDM system you have been developing with and without coding. You
will implement the code over a real wireless link to estimate coding gain (a
metric of performance gain between coded and uncoded systems).

1
1.1

Background
Structured Channel Codes

Coding theory is a field which encompasses both source and channel coding.
It is largely based on results from finite field algebra. The main idea behind
channel coding is that adding structured redundancy to the source bits for a
transmitted sequence can improve the resilience of transmission to channel
distortion. In other words, a properly designed channel code can improve
the bit-error rate (BER) performance of a system (i.e., reducing the transmit
power required to achieve a particular error performance). A rate (k/n)
code takes a length L bit sequence and generates nL
k bits. There are two
main classes of channel codes: (1) block codes and (2) convolutional codes.
Channel coding is sometimes referred to as forward error correction (FEC)
or error control coding, or just coding.
1 Through the use of guard intervals and an appropriately designed FEQ, OFDM systems can
forgo the complicated equalizer design needed by wideband single carrier systems.

130
i

i
i

Figure 1: A typical digital communication system model. The channel coding is applied before the demodulation and the channel decoding
is applied after the demodulation. In more sophisticated systems, the modulation and coding are jointly designed and decoded.

book2 2011/10/28 14:05 page 131 #143

Lab 8: Channel Coding in OFDM Systems

131

i
book2 2011/10/28 14:05 page 132 #144

132

Lab 8: Channel Coding in OFDM Systems

Channel Code
Hamming Code
ReedSolomon

Code Type
Block
Block

Punctured Codes
Turbo Codes
Low Density Parity Check
(LDPC)

Convolutional
Convolutional
Block

Technology/Standard

disk drives, CDs, DVDs,

WLAN
Hybrid-ARQ, IEEE 802.11a
3G mobile, satellite comm.

Table 1: Various Channel Codes.

In block coding, the encoder first segments the source sequence into blocks
of k bits. Thus each block represents any one of 2k distinct messages. This
block is in turn mapped to a length n bit sequence called a codeword. Each
codeword contains r = n k redundant bits. This kind of code is called
an (n, k) block code and can be used to detect and possibly correct errors.
The resilience of block codes to channel distortion largely depends on the
structure of the r redundant bits.
Convolutional codes are a class of channel codes which employ convolutional encoders. The encoded sequence is a convolution of the algebraic polynomial describing the source bit sequence with a polynomial which describes
the encoder. Equivalently, the i th codeword generated by a convolutional encoder is a weighted sum of the previous input messages to the encoder. This
is similar to the convolution used in linear time-invariant systems and shares
many of the same properties.
Convolutional codes generally have very efficient implementations which
employ shift registers (to retain state or memory) and exclusive-or (XOR)
gates. Block codes, which generally require more logic to implement, are
much easier to analyze. There are a number of tradeoffs between both classes
of codes, but neither class of codes fundamentally provides better error protection than the other. Table 1 lists a variety of new and old channel codes.

1.2

Repetition Codes

Repetition codes are a very simple type of block code. In the remainder of
this section you will
 learn
about the encoder, decoder, and probability of error
analysis for a rate n1 repetition code.
Encoder

A rate n1 repetition code converts each message bit into a length n codeword.
More precisely, let I(x) be a mapping from the space of message bits to length

i
i

i
book2 2011/10/28 14:05 page 133 #145

133

Lab 8: Channel Coding in OFDM Systems

n codewords such that

I(x) =

(0, 0, . . . , 0),
(1, 1, . . . , 1),

x = 0,
x = 1.

(1)


Thus a rate 13 repetition code has the following input-output relationship
0
1

(0, 0, 0),
(1, 1, 1).

Decoder

A simple (although not necessarily optimal) decoder for a rate n1 repetition

code is a majority decoder. For each length n codeword received (i.e., after
demodulation), the majority decoder estimates the transmitted bit sequence
by finding the majority consensus of the n received bits. For example, if n = 5
and the receiver estimates the bit sequence 11001 after demodulation, then
the majority decoder guesses that the message bit m = 1 was transmitted.
Typically n is chosen to be odd to avoid a race condition.
Probability of Error Analysis
Suppose that after modulation, bit errors at the receiver can be modeled as
an i.i.d. Bernoulli random process with parameter p, where p reflects the
average bit error rate of the uncoded system. This means that a bit error occurs
in the coded system only when n/2 or more errors occur in a codeword.
Therefore, probability of bit error is given by
+ ,
n

n
Pe =
p k (1 p)nk .
k

(2)

k=n/2

Note that the number of errors in each codeword is being modeled as a

binomial(p) random variable. Also note that p, the probability of bit error in the uncoded system, is actually a function of SNR [i.e., p p( ),
where represents SNR].

Pre-Lab

In this lab you will be implementing the encoder and majority decoder for
a rate 13 repetition code. Tables 2 and 3 describe the two VIs that you
are required to build. You will insert your code into the baseband simulator
provided to you in order to verify functionality and correctness.

i
i

i
book2 2011/10/28 14:05 page 134 #146

134

Lab 8: Channel Coding in OFDM Systems

Table 2: Description of repeat encoder.vi

repeat encoder.vi - implements rate (1/3) repetition code

INPUTS

OUTPUTS

information bit
sequence

1-D array of
U8

bit sequence to be repetition

encoded

enable
channel
coding

Boolean

determines whether or not

your VI should be using
repetition code

encoded
bit
sequence

1-D array of
U8

bit sequence after repetition

encoding

Table 3: Description of repeat decoder.vi

repeat decoder.vi - implements majority decoding to decode rate (1/3) repetition
code

INPUTS

OUTPUTS

estimated
bit
sequence

1-D array of
U8

estimated bit sequence (i.e.,

after demodulation)

enable
channel
coding

Boolean

determines whether or not

your VI should be using
repetition code

decoded
bit
sequence

1-D array of
U8

bit sequence after majority

decoding

You will have to insert the repeat encoder.vi and repeat decoder.vi into the
block diagram for OFDM transmitter.vi and OFDM receiver.vi respectively.
Once you have verified that your code is working correctly, you will then
need to set up the following parameters in order to complete the remainder of the pre-lab. These parameters may be located on the front panel of
simulator.vi.
Correct frequency offset? = FALSE
Equalize channel? = FALSE
Synchronization method = Fixed Offset
Channel model = AWGN

i
i

i
book2 2011/10/28 14:05 page 135 #147

Lab 8: Channel Coding in OFDM Systems

135

Leave all unspecified parameters set to their default values. Using this prescribed setup, you will compare the performance of a coded and uncoded
OFDM system over an AWGN channel. You will be required to plot BER as
a function of SNR for both the coded and uncoded systems.

Questions
Answer the following questions about channel coding.
1. According to the union bound for probability of error in QPSK systems,
what SNR is required to maintain a symbol error rate of 103 in an
uncoded OFDM system operating in an AWGN channel?
2. Consider an OFDM system with the following parameters:
(N K) subchannels carry QPSK modulated data,
a length Lc guard interval (or cyclic prefix) is used,
a rate nk channel code is used,
and 1/T is the OFDM sample rate.
In terms of these parameters, what is the effective data rate of this
coded OFDM (COFDM) system? You may ignore the overhead cost
of training in your answer.
3. On the same plot, graph BER as a function of SNR for the coded and
uncoded OFDM systems in an AWGN channel. Vary SNR from 0
to 10 dB in increments of 2 dB (i.e., vary noise power No from 0 down
to 10 dB). Collect average BER statistics using the simulator with
your code for repetition coding.
You must use your LabVIEW code to generate the data for your plots.
Use a logarithmic scale for BER and a linear scale for SNR in your
plot.
4. From the plot you created in the last question, what is the approximate
difference between the SNR required to obtain an error rate of 103 in
the coded and uncoded systems?

Pre-Lab Turn In
1. Submit your implementations of repeat encoder.vi and repeat decoder.vi electronically. Remember to include error cluster inputs and outputs.
Note: If you need to generate additional subVIs for this lab, please
electronically submit these subVIs along with your other pre-lab VIs.

i
i

i
book2 2011/10/28 14:05 page 136 #148

136

Lab 8: Channel Coding in OFDM Systems

2. Submit your answers to all of the questions in the pre-lab section.

3. Submit your plot of the BER curves comparing coded and uncoded
OFDM.

Lab Experiment

In this lab, you will evaluate the bit-error rate performance of a coded OFDM
system and show that it is better than that of an uncoded system. The gap
in the SNR required to achieve a particular bit-error probability between
the coded and uncoded OFDM systems is defined as the coding gain of the
channel code.
 1 In this lab you will empirically approximate the coding gain
of the rate 3 repetition code over a real wireless link.

3.1

Coding Gain

Coding gain is a well-established metric for gauging the performance of

channel codes. To formally define coding gain, let Pe,c ( ) and Pe,uc ( )
be functions of SNR which describe the probability of error performance
of a coded and uncoded system respectively. If we assume that both BER
curves are continuously differentiable and monotonically decreasing, then
1 (p) and P 1 (p) such that P (P 1 (p)) = p
there exist two functions Pe,c
e,c e,c
e,uc
1 (p)) = p (i.e., there exists an inverse mapping from the
and Pe,uc (Pe,uc
probability of error values to SNR values for each curve). Thus, let the
gap function G() be defined as

(3)
G() =
.
1 
Pe,c Pe,uc ()
The gap function G() describes the gap between the coded and uncoded
error curves. More precisely, it describes the ratio between the SNR required
to sustain Pe,uc () error rate in the uncoded and coded systems. Thus, coding
gain can be formally defined as
C = lim G(),

or in a decibel scale
CdB = lim

.
1
(Pe,uc ()) ) .
10 log() 10 log(Pe,c

(4)

(5)

Coding gain approaches a constant at high enough SNR. This means that
the gap between the coded and uncoded error curves remains approximately
constant at reasonably high SNR. Intuitively, coding gain can be interpreted
as the amount of additional transmit power required to achieve a particular Pe
in an uncoded system (versus a coded system).

i
i

i
book2 2011/10/28 14:05 page 137 #149

Lab 8: Channel Coding in OFDM Systems

3.2

137

Measuring Coding Gain

In this lab experiment, you will use the intuitive interpretation

of coding gain
1
to empirically estimate the coding gain of a rate 3 repetition code. The
questions below will guide you through the process of approximating the
coding gain of your repetition code. Set the following parameters in your
system. Leave all unspecified parameters set to their default value.
TX sample rate = 20 MSamp/sec
TX oversample factor = 8
RX sample rate = 10 MSamp/sec
RX oversample factor = 4
Packet length = 1000 bits
Capture time = 1 msec
There are several places where bit errors can occur in your wireless communication system. In the Lab 7 you learned how critical frame detection
errors can lead to an error rate of approximately 21 . You have also learned
about frequency synchronization errors in OFDM systems, and how large
frequency offsets can also lead to severe error rates. In this experiment, you
will only be interested in errors caused by additive noise. That is, you should
disregard any trials in which an error might have occurred from a critical error
in frame detection or a severe frequency offset (i.e., throw out trials in which
severe error rates are observed).
Questions
Be sure to answer the following questions in the lab itself, as this lab does
not require you to submit a lab report.
1. Estimate the minimum prewireless channel SNR required by the coded
system to maintain an error rate below 102 (i.e., maintain an error rate
on the order of 103 ). You will be doing this by changing the artificial
noise power added to your signal before being transmitted. This will
be done from the front panel of top tx.vi inside the channel model
parameters control cluster. You will need to transmit multiple packets
to verify that your estimated SNR reliably produces the desired error
rate.

i
i

i
book2 2011/10/28 14:05 page 138 #150

138

Lab 8: Channel Coding in OFDM Systems

2. Estimate the additional pre-wireless channel SNR required by the uncoded system to maintain a similar error rate. To do this, disable the
channel coding feature on the front panels of top tx.vi and top rx.vi.
Also set the Packet length parameter to 3000. You will need to transmit
multiple packets to verify that your estimated transmit power reliably
produces the desired error rate.
3. Based on the SNRs you have estimated in the previous two problems,
what is the coding gain (in dB) of the resulting system?
Hint: X dBm Y dBm = (X Y) dB.

3.3

Lab Turn In

Demonstrate to the instructor that your code is working properly. Show

the instructor that the uncoded system requires higher SNR to maintain a
comparable error rate (i.e., versus the coded system).

DIRECTIONS FOR LAB SUBMISSION

This lab does not require any lab report. Please be sure to answer all the lab
questions and show your results to the instructor.
Answer the following questions about your channel coding experiment.
1. Does the coding gain you measured in lab correspond to the coding
gain you observed through simulation in the pre-lab?
2. Using channel coding adds resilience to a digital communication system. Discuss the tradeoffs associated with using coding, specifically
that of rate and performance. If you could adaptively decide when
to use coding (i.e., depending on the received SNR), would you use
coding when the SNR is high, low, or always and why?

i
i

i
book2 2011/10/28 14:05 page 139 #151

Appendix A
Reference for Common LabVIEW VIs
Appendix A enumerates some of the common VIs you will have to use in
this lab and how to access them (i.e., the path to get to them from the All
Functions tool palette). Note you can also use LabVIEW QuickDrop to
search for VIs by name by clicking <CTRL> + <SPACE> when you are
on a LabVIEW diagram window. For additional information about each VI,
place the VI in a LabVIEW block diagram, right-click on the VI, and select
Help. This will bring you to a detailed description of the VIs behavior as well
as its inputs and outputs.
Insert Into Array.vi

ProgrammingArray

Interleave 1D Arrays.vi

ProgrammingArray

Max & Min.vi

ProgrammingArray

Unbundle By Name.vi

ProgrammingCluster & Variant

Tick Count (ms).vi

ProgrammingTiming

Random Number (0-1).vi

MathematicsNumeric

Complex To Polar.vi

ProgrammingNumericComplex

Complex To Re/Im.vi

ProgrammingNumericComplex

Case Structure.vi

ProgrammingStructures

Local Variable.vi

ProgrammingStructures

Convolution.vi

Signal ProcessingSignal Operation

FFT.vi

Signal ProcessingTransforms

Inverse FFT.vi

Signal ProcessingTransforms

Mean.vi

MathematicsProb & Stat

A x B.vi

MathematicsLinear Algebra

Inverse Matrix.vi

MathematicsLinear Algebra

Generate Bits.vi

ModulationDigital

Fractional Resample.vi

ModulationAnalogUtilities

Generate Filter Coefficients.vi

ModulationDigitalUtilities

Align To Ideal Symbols.vi

ModulationDigitalUtilities

139
i

i
i

i
book2 2011/10/28 14:05 page 140 #152

i
i

i
book2 2011/10/28 14:05 page 141 #153

Bibliography
[1] Breed, G., Analyzing Signals Using the Eye Diagram, High Frequency
Electronics, pp. 5053, November 2005.
[2] Heath Jr., R. W., EE 371C / EE 381V: Wireless Communications Lab,
http://www.profheath.org/teaching/ee-371c-ee-381v-wireless-communications-lab/, accessed September 2011.
[3] Heath Jr., R. W., Introduction to Wireless Digital Communication: A
Signal Processing Perspective, course notes for EE 371C taught at The
University of Texas at Austin.
[4] IEEE-SA Standards Board, IEEE Std 802.11a-1999 (R2003)Part 11:
Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY)
Specification. High Speed Physical Layer in the 5 GHz Band, June 12,
2003, from http://standards.ieee.org/
[5] National Instruments, LabVIEW Fundamentals, August 2005, retrieved
December 14, 2005 from http://www.ni.com/pdf/manuals/374029a.pdf.
[6] Johnson, Richard C. and Sethares, William A., Telecommunications
Breakdown: Concepts of Communication Transmitted via SoftwareDefined Radio, Prentice Hall, 2003
[7] Moose, P., A technique for orthogonal frequency division multiplexing
frequency offset correction, IEEE Trans. Commun., vol. 42, no. 10, pp.
29082914, October 1994.
[8] National Instruments, Bandwidth, Sample Rate, and Nyquist Theorem,
retrieved September 13, 2011, from http://zone.ni.com/devzone/cda/
tut/p/id/2709.
[9] National Instruments, Getting Started with NI LabVIEW Student Training, June 2010, retrieved September 13, 2011, from http://zone.ni.com/
devzone/cda/tut/p/id/7466.
141
i

i
i

i
book2 2011/10/28 14:05 page 142 #154

142

Bibliography

[10] National Instruments, How Can I Learn LabVIEW?, retrieved September 13, 2011, from http://www.ni.com/gettingstarted/labviewbasics.
[11] National Instruments LabVIEW 101, retrieved September 13, 2011,
from http://www.ni.com/lv101.
[12] National Instruments, LabVIEW Fundamentals, August 2007, retrieved
September 23, 2011, from http://www.ni.com/pdf/manuals/374029c.pdf.
[13] National Instruments NI USRP information, retrieved September 13,
2011 from http://www.ni.com/usrp.
[14] National Instruments, Understanding RF & Microwave Specifications
Part I, retrieved September 13, 2011, from http://zone.ni.com/devzone/
cda/tut/p/id/4446.
[15] National Instruments, Understanding RF & Microwave Specifications
Part II, retrieved September 13, 2011, from http://zone.ni.com/devzone/
cda/tut/p/id/4448.
[16] Schmidl, T. M. and Cox, D., Robust frequency and timing synchronization for OFDM, IEEE Trans. Commun., vol. 45, no. 12, pp. 16131621,
1997.
[17] Wikipedia Software-defined radio, retrieved September 13, 2011, from
http://en.wikipedia.org/wiki/Software radio.

i
i

375775A-01.pdf

9/12/11

2:53 PM

ISBN 9781934891186

90000 >

*375775A-01*
375775A-01

03014

9 781934 891186