Real Time Location System using Passive UHF RFID

Dean Brennan Dr. Jan Kolaja

Applied Research and Innovation Services Department of Applied Cybernetics
Southern Alberta Institute of Technology Technical University of Liberec
Calgary, Alberta, Canada Liberec, Czech Republic
[email protected] [email protected]

Abstract—RFID technology is becoming more and more

popular these days and on its top stands passive UHF RFID II. BACKGROUND AND RELATED WORK
technology for it's economical as well as practical benefits. Even Several attempts [1]-[6] have been made by several vendors
though this technology is intended for inventory, several attempts in creating a real-time location tracking system using passive
have been made by several vendors in creating a real-time RFID solutions in the past. Most of these have met with
location tracking system using passive RFID solutions.
limited success at best. The downfalls these solutions face is
The scope of this paper is description of experiments when
either the need for custom specialized equipment, expensive
developing such a RTLS with 9 RFID readers spread into a equipment or too much of both. Not one system has yet to of
rectangular grid at the area of 64 to 100 meters square. Each nailed the ideal cost to coverage ratio desired by industry.
reader connected to 4 antennas facing specified direction. This proposal of the project which is described in this paper
Location algorithm combines signal strength values with number is to create a real-time location tracking system using passive
of reads to obtain tags location probability. Experiments with UHF RFID tags and readers which are present at the RAD Lab
this settings were run in October and November 2013 at the
so that no additional RFID equipment is needed. This should
Southern Alberta Institute of Technology, Calgary, Canada in
be a home grown system created in-house that uses different
cooperation with Dean Brennan of the RFID Application
Development (RAD) Lab Team and visiting professor from
metrics to calculate the location of objects.
Technical University of Liberec, Czech Republic (Jan Kolaja). Location tracking for passive RFID tags works much
differently as those of the active variety. For active tags, the
The most interesting result of these space consuming tag sends out a signal every predetermined interval. Then it is
experiments is self-calibrating algorithm. Although the system picked up by various antennas in range that record a time of
has not been sufficiently finished yet as it is limited by having not
arrival notification for each. These can be then converted into
enough powerful antennas, the experiments have earned many
distance metrics using simple physics and spherical geometry
beneficial
knowledge.
calculates the location of the tag. Unfortunately passive tags
and their antennas don't work this way.
Keywords—RFID; tracking system; Alien RFID; UHF passive As a result all current location tracking models using
tags passive RFID tags fall into one of two types. It is either a
custom hardware design that doesn’t use typical off the shelf
I. INTRODUCTION TO RFID readers and tags, or it is a probabilistic model for calculating
Radio Frequency Identification (RFID) is a set of location. The former requires custom and typically expensive
technologies that allow contact-less reading or writing data equipment to implement, the latter has its own pitfalls that
sources within some area. The reach area differs with used must be overcome.
frequency as well as used data sources – the tags. Tags may be Common probabilistic models can be generalized into two
active or passive depending on where they get the energy to main categories.
transmit their data. Active tags have their own source (battery),
passive tags are energized by electromagnetic field which is at A. Models using RSSI value
the same time necessary for data transmission. Generally
speaking the higher frequency means better results in reach and The first uses Received Signal Strength Indication (RSSI)
speed of communication on the other hand it requires much values from antennas reading the tags, and where all antennas
more precise equipment. can read the zone in which the tag could be in. This model
suffers from a few key flaws, first is the very limited usefulness
The reason why passive UHF tags are so popular is its that this could be deployed. These models typically require 4
lower prize in compare with active tags and better results of antennas to have overlapping zones of reading and as such
Ultra High Frequency (865 MHz to 868 MHz for Europe) in require them to be in a very close area (Fig. 1). For most
compare with lower frequencies. From the economic point of readers and antenna in the market today that would be no more
view it is better to buy one expensive reader which can read than a 5 meter square with antennas at each corner. This
millions of cheap tags than to buy one cheaper reader to read limited range makes any practical deployment of such a model
thousands of expensive tags. unfeasible due to the costs involved.

58
978-1-4799-3528-4/14/$31.00 ©2014 IEEE
Fig. 1. Typical layout of RTLS models using RSSI value

B. Models Using Read Rate

The other category of probabilistic model leverages read Fig. 2. RTLS layout with reference tags and omni directional antennas
rates by the readers receiving the tag reads, and a series of
known reference tags to calibrate the system to calculate the
tracking tag locations. These systems enjoy greater coverage
area but come at other costs. First all reference tags must be
deployed in advance and their geographic locations must be a
known quantity. This has a very lengthy deployment time.
Second the processing power needed to calibrate the system
isn’t small.
The typical example of such a system (Fig. 2) has a large
number of reference tags that are used to generate a statistical
regression model to allow for the calculation of the tracking
tag. This typically uses a Log Likelihood regression model.
The reference tags are used frequently to update the functions
needed to track objects. Finally the model must assume to have
precise control over the interrogation cycles per second of each
antenna and all antennas must have the same interrogation Fig. 3. Initial setup of the experiment
cycle.
In either case both models only use one of the two possible III. A NEW RTLS ALGORITHM
metrics we can record from a passive tag read, either the
The algorithm used in this experiment attempted to use the
probability of tag reads, or the RSSI values received. The two
best features of both models to allow for larger area, easier set
models only leverage one of them. We wish to continue
up time, and still use off the shelf equipment. The equipment
experimenting with a model that uses both metrics and would
used in this experiment were nine Alien 9900+ UHF RFID
forgot the need for reference tags being deployed. However
Readers. Each reader was connected to with a number of 60
before we arrive there we must first show the strengths and
degree arc directional RFID antennas and placed in an 8 meter
weaknesses of existing models and show improvements upon
by 8 meter grid for testing. The tags used in this experiment
them. As such it is our hope to develop and deploy two
were Frog 3D UHF RFID tags. The only known information
different models for larger scale testing.
needed to be included in the system was the geographical
location of each antenna on each reader. The geographical
information of all tags is unknown and will be calculated by
the new algorithm. The readers used in the experiment were
arranged initially as shown at Fig. 3:

A. Basic Algorithm
The first model discussed using RSSI values is unreliable in
real world applications. The RSSI can be misleading at the
best of times. Backscatter and signal bounce can register the
same RSSI values at 4 meters as it could at 15 meters. The
only difference is that the number of times a tag is read at 4
meters is vastly greater than a tag read at 15 meters. As such
RSSI alone is insufficient in determining time of arrival
information for location tracking.

2014 15th International Carpathian Control Conference (ICCC) 59

 The second model has several issues involving the
interrogation cycle for each antenna. In any larger system with
multiple readers and differing numbers of antennas on each,
synchronizing the interrogation cycle of each reader can prove
difficult if not impossible. As such the algorithm treats each
antenna (Ai) separately and calculates for each antenna its
maximum (Rmaxi) and minimum (Rmini) RSSI values over a
specified time interval. During this time interval the tag read
information and RSSI information is logged. For each antenna
the tag with the most reads will become the expected value of
tag reads for that antenna (Ni). As such that tag will have a
probability of read of 1. For each tag its number of reads for
tag k by antenna I is denoted as Tki. This allows us to calculate Fig. 4. New asymetric layout of the experiment
the probability of tag k being read by antenna i as
Pki = Tki / Ni. (1)
Next the average RSSI value for each tag is calculated to
each antenna and denoted as Ravgki. To calculate the effective
RSSI value for each tag k for antenna i is defined as:
Rki = (Pki * Ravgki – Rmini) / Rmaxi . (2)
Once the effective RSSI for each tag to antenna
combination is calculated the location of each tag can be
determined. For each tag find the set of antennas that have
reads during the interrogation time interval. Since each
antenna’s geographical information is known (as x & y
coordinates) the centroid point for that tag Ck can be found.
The centroid is the point that minimizes the distance between
each antenna Ai. This is found by averaging the x and y
coordinate values Fig. 5. Common speaker stand equipped with Alien reader and 4 antennas

Ck = 1/n * ™Ai. (3)

Lastly we vector shift from the centroid towards or away
from each antenna that has read the tag based on its effective
RSSI values. This arrives at the final position of the tag Tk
OTk = OCk – ™Rki * CkAi. (4)

IV. EXPERIMENTATION RESULTS

The initial experiments tried using a symmetric distribution
of the readers and antennas. This arrangement however was
actually discovered entirely by accident to be the less optimal
deployment for the model. When one of the readers used in the
experiment suddenly stopped reporting its tag reads it had to be
removed from the experiment. The grid was reduced to a 16
by 10 meter grid now instead of its initial 16 by 16. The
readers were deployed in the following arrangement (Fig. 4)
and the accuracy of tag location approximations increased
Fig. 6. First experimenting
substantially.
This new smaller and less symmetric setup allowed for
better overlapping of antenna reading zones. Further
experimentation revealed that the required minimum number of
tags needed for reference for each antenna was three. So
placing tags inside the system to act as reference tags to allow
the system to properly calibrate itself requires tags placed in
locations that have at least two antennas high probability read
zones overlapping its location. The exact tag location isn’t
needed to be known as the algorithm calculates it.

2014 15th International Carpathian Control Conference (ICCC) 60

 V. EXPERIMENTATION BACKGROUNDS badges in a conference hall, once enough people have entered
Stands on which antennas and readers were located are the system area the system would automatically be calibrated.
common three leg aluminum stands used by musicians for
speakers or lamps (Fig. 5). VII. CONCLUSION AND FUTURE WORK
Experiments to prove presented algorithm were run at The work presented in this paper shows it is possible to
several room of SAIT Polytechnic. The first experimentation build such a RTLS system using off the shelf equipment. Self
room (Fig. 6) exposed the problem of impossible read of tag calibrating algorithm enables placement of this RTLS in the
placed directly on the ground. The most probable reason for shortest possible time - still it is needed to transport these large
this was the conductivity of flooring. The solution to this but not heavy stands.
problem is simply mounting tags on paper stem. From the economical point of view initial cost of this model
Final experiments were run in a gym (Fig. 7). The is pretty high. The system is expensive when measured in
attractiveness of such a place enabled testing only for few dollar per square meter of served area. On the other hand the
hours so the time saving self calibrating algorithm was more tags are in the area which means the more subjects are
appropriate. tracked the better is the ratio dollar per tag (dollar per tracked
subject or dollar per person). Hand in hand with this
economical benefit goes previously declared statement that the
more tags are in the system the better the self adjusting
algorithm works.
From the practical point of view use of such a RTLS
system is at the moment imaginable at crowded places where it
is beneficial to know each person's position like conference
halls, restaurants, hospitals etc.
Algorithm might also be useful for RFID sport timing
system to precise results when more athletes appear near the
finish line at the same time. This functionality is going to be a
subject of future work.
Future work will also concern on transferring and testing
the model on Linux platform. (At the moment the algorithm is
proceed by .NET enviroment.) Then the team will logically
work on system's improvement to achieve better precision or
easier set up.
Fig. 7. Experimenting at the largest area possible
ACKNOWLEDGMENT
It must also be said that the positioning algorithm was Results presented in this paper were obtained while Jan
changed several times while experimenting. The main parts Kolaja's visit at SAIT in October and November 2014 which
which were changed were impacts of RSSI and Read Rate was possible thanks to Mobility Fund of Technical University
values. As these impacts were tested several times with several of Liberec.
layouts it is highly probable that final algorithm should work
well in general layout. Three of readers used for this work have been lent by Alien
Company.
VI. FUTURE USE
While other models require reference tags to have known
locations this model does not. The tags merely need to be in at REFERENCES
least two different antennas read zones when placed. In each [1] A. Bekkali, H. Sanson, and M. Matsumoto, "RFID Indoor Positioning
antennas read zones a total of at least three tags must be present based on Probabilistic RFID Map and Kalman Filtering", Proc. of the
to ensure proper calibration of the model. Using tag filtering IEEE International Conference on Wireless and Mobile Computing,
Networking and Communications, New York, 2007, pp. 21-27.
the tags placed as such can be classified as reference tags then
not shown on output displays and tags that one wishes to track [2] Chuan-Chin Pu, Chuan-Hsian Pu and Hoon-Jae Lee (2011). Indoor
Location Tracking Using Received Signal Strength Indicator, Emerging
such as people wearing UHF RFID ID badges, will be the only Communications for Wireless Sensor Networks, Anna Foerster and
tags shown. Alexander Foerster (Ed.), ISBN: 978-953-307-082-7, InTech, DOI:
10.5772/10518. Available from:
The reference tags and the tracking tags have identical http://www.intechopen.com/books/emerging-communications-for-
functionality, the final system output is the only difference wireless-sensor-networks/indoor-location-tracking-using-received-
here. It must be pointed out that a sufficiently large number of signal-strength-indicator
tracking tags in the reading zones would eventually remove the [3] Da Yan, Zhou Zhao, Wilfred Ng, “Leveraging read rates of passive
need for the reference tags all together. So if tracking ID RFID tags for real-time indoor location tracking”, Proceedings of the
21st ACM international conference on Information and knowledge

2014 15th International Carpathian Control Conference (ICCC) 61

 management, Pages 375-384, ACM New York, NY, USA 2012, ISBN:
978-1-4503-1156-4
[4] S. Sarkka , V. Viikari , M. Huusko and K. Jaakkola "Phase-based UHF
RFID tracking with nonlinear Kalman filtering and smoothing", IEEE
Sens. J., vol. 12, no. 5, pp.904 -910 2012
[5] Wu, Jiaqing, "Three-Dimensional Indoor RFID Localization System"
(2012). Industrial and Management Systems Engineering --
Dissertations and Student Research. Paper 36. Available from:
http://digitalcommons.unl.edu/imsediss/36
[6] J. L. Brchan, Z. Lianlin, J. Wu, R. E. Williams, and L. C. Perez, "A
Real-Time RFID Localization Experiment Using Propagation Models",
Proc. of the IEEE International Conference on RFID, Florida, 2012, pp.
141-148.

2014 15th International Carpathian Control Conference (ICCC) 62