This article has been accepted for publication in a future issue of this journal, but has not been

fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JRFID.2017.2751592, IEEE Journal
of Radio Frequency Identification
1

UHF RFID in Automatic Vehicle Identification:

Analysis and Simulation
Andrey Larionov, Roman Ivanov, and Vladimir Vishnevsky

Abstract—In this paper the application of UHF RFID in conditions, the bus was identified if any one tag mounted
automatic vehicles identification is studied by the means of a on it was successfully read. The identifiers were written
new simulation framework. The system under study consists of into EPC and TID memory, both values were needed for
the stationary readers with antennas mounted over the road lanes
and vehicles license plates equipped with passive EPC Class 1 the identification. The experiment was run for several winter
Generation 2 UHF RFID tags. The simulation includes protocol months and revealed 0.92 mean bus recognition rate [4].
settings, antennas parameters and propagation media conditions. To enhance the system performance, e.g. increase recog-
Propagation analysis considers Doppler shift to compute the path nition rate, all factors that affected the system substantially
loss of moving tags properly. The estimations of an inventory should be considered. The parameters making a great contri-
round duration, the number of rounds the tag participates
in and the tag identification probability are provided. The bution into the system functioning quality are the following:
paper presents vehicles identification probabilities computed for antenna systems and transceivers [2], [3], [5]–[9], propagation
different protocol settings and movement speeds. In particular, it media parameters [10], [11] and protocol settings [12], [13].
was shown that a vehicle could be identified with 0.9 probability Thus, it is necessary to take into account both protocol
by EPC values only when the vehicle speed is 120 km/h or even specifics and signal propagation properties to build an adequate
more. If the vehicle is identified with both EPC and TID banks
values, the same 0.9 probability is achievable when the vehicle model.
speed is under 90 km/h. The results of the simulation coincide In this paper a detailed simulation framework enabling
with the experiment results involving the city buses. precise UHF RFID vehicle identification system performance
Index Terms—UHF RFID, automatic vehicle identification, evaluation is presented. The framework is implemented in
radio propagation, Doppler shift, k-ray model, simulation C++ and Python languages using OMNeT++ [14] and PyONS
[15] simulators. The model simulates both high-level and
low-level protocol features to take into account the set of
I. I NTRODUCTION
protocol parameters, makes use of an analytic model of signal
UTOMATIC vehicle identification is widely used in
A traffic rules enforcement systems. Speed cameras as a
crucial part of most identification systems have high cost and
propagation to evaluate bit error probability (BER) and, finally,
provides collision probability and recognition rate estimations
for different protocol parameters.
rely on clear optical conditions to enable the license plate
reading.
II. R ELATED W ORK
An alternative solution for vehicle identification is assumed
to take advantage of UHF RFID under control of EPC Class1 Vehicle identification is one of the primary RFID applica-
Gen2 [1]. Radio identification equipment has essentially lower tions. First RFID-based tolling systems appeared in the USA
cost and is more reliable in terms of weather conditions. At the in 1991 [16]. RFID usages in entrance control systems, tolling
same time UHF RFID is able to read the tags at large enough and others are devoted a significant amount of researches [11],
distance to make the vehicle radio identification feasible. [17]–[21]. RFID can be applied to solve other problems, such
RFID system consists of a Reader, or an Interrogator, and as performance enhancement of the traffic law enforcement
Tags. Such systems can be active, if the tag has an inner systems making use of cameras to identify vehicle license
power source and can independently initiate message exchange plates [4], [22].
with the reader, or passive otherwise. Passive RFID [2], [3] Various approaches were proposed for UHF RFID sys-
generally provides less reading distance as the signal has to tems simulation [23]–[25]. In [23] a model written in Java
pass double distance from the reader to the tag and back and using JiST simulator was proposed. The paper observes the
demonstrates large attenuation along this path, but such tags simulation model design, the model uses Java language and
are significantly cheaper than the active ones and are energy- could be extended for various scenarios. Unfortunately, JiST
independent. simulator seems to be outdated nowadays and is not widely
Before implementing the system in real life, a proof-of- used. The paper [24] is focused on wideband signals, fading
concept experiment should be carried out. In order to show channels and other low-level features while lacking high-level
viability of the system and to study its performance the protocol details. In contrast the present paper uses discrete-
pilot project was launched in Kazan city (Tatarstan, Russia) event simulation approach and considers radio propagation
in 2014. Four identification points were deployed and near along with high-level protocol details. The paper [25] proposes
1000 buses were equipped with tags placed both onto license a combination of OMNeT++ simulator for high-level modeling
plates and under the windshields. According to the experiment while using MATLAB for physical layer modeling. It presents

2469-7281 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JRFID.2017.2751592, IEEE Journal
of Radio Frequency Identification
2

Fig. 2. The scheme of vehicle identification system and geometry parameters

for ray tracing in path loss computation.

v. The tag identification probability P{At } depends on an

inventory round duration, a response transmission probability
and a tags number in reading area:
P{At } = 1 − (1 − (1 − P{Ac })P{Ar })Nr , (2)
Fig. 1. A round and a TID reading slot message exchange.
where Ac is a collision occurrence, Ar is an event in which
the tag successfully replies to the reader and Nr is a round
a UHF RFID simulation as a case study. While this work number in which the tag participates.
provides an efficient way of simulating UHF RFID systems, Let us consider Nt be a number of tags in the reading area
it has high complexity due to MATLAB integration. For and Q be a parameter of the anti-collision protocol [1]. Then
the best of the authors knowledge, to date there are no the collision probability is given by the following formula:
existing tools using a combination of OMNeT++ and Python-
P{Ac } = 1 − (1 − 2−Q )|Nt |−1 . (3)
based simulators for both low-level and high-level UHF RFID
discrete-event modeling. The number of rounds can be estimated as:
In order to provide the proper simulation of the message Dr 1
exchange involving the possibility of packet loss due to unre- Nr ≥ · , (4)
v Tr
liable media conditions the paper suggests careful propagation
modeling. This case was studied in papers [26], [27]. Both where Dr is a total length of the road part with good tag
works propose propagation models making use of ray tracing. reading conditions (e.g. estimated by upper bound of bit error
The authors of the paper [26] suppose in-door environment probability for the reader to decode a tag response), Tr is
with multiple reflections, but focus on stationary objects and a maximum round duration under the given protocol settings
ignore Doppler shift resulted from moving objects to be and given tag number, v is a vehicle speed. In fact, the round
identified. In [27] the authors model propagation considering duration is a random variable depending on protocol settings,
moving vehicles, but miss any high-level protocol parameters tag number and tag choice of a time slot to response, errors,
and bound mostly on power evaluation in comparison with etc.
the experimental data. Paper [28] proposes careful link budget According to [5] the key contribution into a successful tag
computation with different antennas schemes and parameters. response probability P{Ar } in collision-free environment is
made by a bit error probability B (BER) for the reader to
III. V EHICLE I DENTIFICATION S YSTEM OVERVIEW decode a tag response. Assuming |r| to be a tag response
length in bits and R be a set of tag responses, P{Ar } is given
A vehicle identification system consists of a reader deployed by formula:
above the road and a tag placed either onto a license plate or Y
P{Ar } = (1 − B)|r| . (5)
under a windshield. The reader can be equipped with up to 4
r∈R
antennas.
The performance of the system is measured by the ratio of The set R necessarily contains responses to Query or
successfully identified vehicles. We assume that identification QueryRep (RN16) commands and a response to ACK
of at least one tag installed on the vehicle is sufficient for its (PC + EPC + CRC); if identification is performed by TID
identification, though identification of more than one tag can memory bank or others, it contains response to Req RN
enhance system reliability and security. Mathematically this and Read as well (see fig.1).
expression can be described as follows: To maximize the vehicle identification probability the round
Y number and the tag response probability should be increased
P{Av } = 1 − (1 − P{At }), (1) while the collision probability, on the contrary, decreased.
∀t∈Tv
However increasing the round number leads to unreliable pro-
where Av is a vehicle v identification event, At is a tag t tocol settings that reduces BER, thus, more detailed analysis
identification event, Tv is a set of the tags placed on the vehicle is required.

2469-7281 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JRFID.2017.2751592, IEEE Journal
of Radio Frequency Identification
3

TABLE I
L INK B UDGET C OMPUTATION PARAMETERS 20 Path loss
Received power at tag

power, dBm (atten ation, dB)

Reader transmission power, Pt
(r)
31.5 dBm 0 Transmitted power by tag
Reader antenna gain, G(r) 8 dBi Received power at reader
−20
Tag antenna gain, G(t) 2 dBi
(t)
Tag sensitivity, Ps -18 dBm −40
Polarization loss, Apol -3 dB
−60
Backscatter loss, Abs -10 dB
Cable loss, Abs -2 dB −80

−100
IV. C HANNEL E STIMATION 0 5 10 15 20 25 30
distance, m
A passive tag makes use of backscatter modulation and thus
cannot provide proper signal gain. Therefore the signal power Fig. 3. Link budget computation versus distance between a reader and a
at the reader side is dramatically lower than the power incident tag. All plots decrease nonmonotonic as the consequence of ground reflection
to the tag. Hence the transmission errors more likely take (slight oscillations of the plot) and antenna radiation pattern attenuation (that
have a great affect at small value of distance where angle between antenna
place at the reader side, the reader bit error probability (or axis and direction of radiating / receiving is large). As the tag transmits if the
BER) makes a dominant contribution into successful message received power is higher then its sensitivity the power transmitted by tag is
exchange and it should be studied carefully. To perform the nonzero just for some region.
analysis of BER let us estimate signal-to-noise ratio (SNR)
via link budget computation. The results of such estimation
polarization and, thus, different reflection coefficient. Also the
for a stationary reader and a tag, that is placed in a car
cable loss Ac of the reader antenna should be mentioned.
moving with a speed of 60 km/h (a typical speed for some road
In the table I the link budget computation parameters are
regulations), are illustrated on the figure 4. The horizontal axis
given. We consider a reader to have circular polarized antenna
of all subfigures represents a distance between a transmitter
while a tag antenna to have a linear one, thus polarization loss
and a receiver, while the vertical shows the time spent from
Apol equals to -3 dB.
the moment when the wave took the receiver. As it will be
In the figure 4b the result of computation of the power
shown latter (see formula (9)) the choice of a specific value
incident to a tag is given. The figure 3 illustrates the result
of speed determines how fast the attenuation at a specific point
of the path loss, the power received at the tag, the power
variates with the time.
transmitted by the tag and the power received at the reader
computation in dependence on the distance d (see figure 2)
A. Link Budget
(r)
between the tag and the road infrastructure object where the
Denote a reader transmission power by Pt and an an- reader is deployed.
tenna gain by G(r) , thus equivalent isotropic radiated power
(r)
EIRP = Pt G(r) . We use a superscript in parenthesis for B. Path Loss
power or gain to point whether reader (r) or tag (t) is assumed
and a subscript to point whether it is transmission of reception Automatic vehicle identification operating supposes a tag to
power. Over transmission from a reader to a tag the signal move relative to a reader with non-zero speed in general case.
(d) In that application of the RFID system a signal is affected by
experiences the attenuation resulted from the path loss Apl
the Doppler effect resulting in frequency scaling with a scaling
depending on a radio-channel state and mutual position of
factor
the reader and the tag. The antennas of the devices can have υ cos ψ υ
different polarization that leads to additional polarization loss α=1− ≈ cos ψ,
c + υ cos ψ c
Apol . A tag antenna has gain G(t) . Hence the power incident
where c is a speed of light, υ is a relative velocity of a receiver,
to the tag:
(r) (d) that can be both a reader or a tag (let us note that a relative
Pr(t) = Pt G(r) Apl Apol G(t) . (6)
velocity of a reader and a tag are opposite in sign and have the
(t) same absolute value), ψ is an angle between the wave vector
If this power is lower than the tag sensitivity Ps the tag
does not power up and will not be able to communicate with ~k and the direction of movement, fc is a carrier frequency.
the reader. Otherwise the tag modulates a backscattering signal For a narrow-band signal being used for communication
(t) (t) in RFID this scaled frequency can be approximated by an
of power Pt that is related to Pr via backscattering loss
Abs , caused by modulation loss, impedance mismatch, etc., so expression αf ≈ f − υc cos ψ · fc = f − ν, where
(t) (t)
Pt = Pr + Abs . Finally, the power of a tag response (i.e. υ 1
reader reception power) is expressed as follows: fc cos ψ = (~υ , ~k)
ν=
c 2π
(r) is referred to as a Doppler shift.
Pr(r) = Pr(t) G(t) Abs Apl Apol G(r) . (7)
In multipath propagation the media produces multiple dupli-
In general case losses on direct (reader to tag) and Adpl cates of the transmitted signal s(t), each of them experiences
(r)
reverse (tag to reader) Apl paths may differ due to different individual complex attenuation hi , delay τi and Doppler shift

2469-7281 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JRFID.2017.2751592, IEEE Journal
of Radio Frequency Identification
4

2.0 2.0 −16 2.0

−48
1.8 −50 1.8 1.8 0.45
−18
1.6 −52 1.6 1.6 0.40
1.4 −54 1.4 −20 1.4 0.35
1.2 −56 1.2 −22 1.2 0.30
time, s

time, s

time, s
1.0 −58 1.0 1.0 0.25
−24
−60
0.8 0.8 0.8 0.20
−62 −26
0.6 0.6 0.6 0.15
−64
0.4 0.4 −28 0.4 0.10
−66
0.2 0.2 0.2 0.05
−68 −30
0.0 0.0 0.0
2 4 6 8 10 2 4 6 8 10 2 4 6 8 10
distance, m distance, m distance, m

(a) Path loss (b) Received power at tag (c) Bit error probability
Fig. 4. Link budget and BER computation. The horizontal axes represent the distance between a reader mount point and the tag (see d variable on the fig.
2), while the vertical shows the time spent from the start of the signal receiving. The color of the bar depicts attenuation in dB (4a), received power at tag in
dBm (4b) and a probability (4c). Presence Doppler effect results to a channel being time-selective, i.e. the vehicles entering a reading area at different time
passed since the constant wave is started radiating have different signal attenuation.

−45 an attenuation factor of the i-th ray resulting from reflection,

scattering and diffraction losses (for the LoS component we
−50 (r) (t)
assume R0 = 1 to simplify expression), Γi = Γi Γi is a loss
−55
attenuation, dB

corresponding to a reader and a tag antenna radiation patterns

−60 properly.
−65 Let us now consider a constant wave of unit power as a
−70
transmitted signal. After combining the formulas described
Parallel polarization
Perpendicular polarization
above the path loss as an instantaneous power of the receiver
−75
Circular polarization antenna is obtained:
−80 2
0 5 10 15 20 25 30  2 X N
distance, m 2 λ Ri Γi −jk(di −υt cos ψi ) 
Apl = |r(t)| = e  (9)

4π di

 
i=0
Fig. 5. Path loss computed for parallel, perpendicular and circular wave The given formula supposes to use deterministic analysis of
polarization versus distance in meters. For circular one a complex-valued
reflection coefficient is a half-sum of the parallel and perpendicular (see fig. 6). wave propagation. The evaluation of Ri , Γi , di and ψi requires
the rays paths computation. For a given environment contain-
ing scattering objects of simple form the analytic analysis can
νi according to different paths lengths, angles of departure be applied for a precise evaluation of the parameters listed
and arrival, velocities of scattering objects. For a line-of-sight above. In general, a ray tracing should be performed for an
(LoS) component and N -scattered components, or rays, the arbitrary environment.
received signal r(t) is defined by the formula: In the given paper we consider two rays only: a LoS compo-
N
X nent and a ground reflected NLoS. For reflected component an
r(t) = hi s(t − τi )ejνi t , (8) attenuation factor R1 equals to the reflection coefficient of the
i=0 ground. We also suppose a reader to be deployed 5 m above the
Figure 7 and figure 8 illustrates the power of the signal road, tags – above 0.5 m in our geometrical computation. Other
received by the reader from the tag and a bit error probability possible rays are implicitly modeled as random components
of the transmission correspondingly. One can see that consid- that leads to the usage of Rayleigh distribution in BER
ering the Doppler effect makes radio channel time-selective: evaluation as will be shown below.
the areas with good signal power migrates to the reader in the Figure 5 shows the results of path loss computation consid-
chosen interval of time. The figure 4 shows selectivity of time ering parallel, perpendicular and circular wave polarization.
for 2-second time interval and the distance ranges from 1 m Different polarization results in different reflection coeffi-
to 10 m. cients, thus, the reflected ray affects the LoS ray differently.
Without loss of generality, let us consider zeroth ray to be
the LoS component of received signal, hence the others N C. Radiation Pattern
rays are scatted, or non line of sight (NLoS), components.
A presence of the Doppler effect makes a radio-channel time- A tag antenna is modeled with a dipole antenna, which
selective. More detailed analysis is presented in the book [29]. radiation pattern is
Attenuation hi can be found as hi = λ/(4πdi )Ri Γi , where  cos π2 sin θ 
 
λ is a wave length, di is a length of the i-th path, Ri is Γ(θ) = 
 , (10)
cos θ 

2469-7281 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JRFID.2017.2751592, IEEE Journal
of Radio Frequency Identification
5

Parallel Perpendicular Circular −60

1.0 0.10s 0.07s 0.04s 0.01s
εr = 3, σ = 0.0014 −65 0.09s 0.06s 0.03s 0.00s
reflection coefficient

0.8 εr = 15, σ = 0.03 0.08s 0.05s 0.02s

εr = 30, σ = 0.15 −70

bit error probability

0.6
0.4 −75

0.2 −80
0.0 −85
0 30 60 90 0 30 60 90 0 30 60 90
grazing angle −90

−95
Fig. 6. Reflection coefficient absolute value for parallel, perpendicular 2 4 6 8 10 12 14 16 18
and circular polarization with different pairs of relative permittivity and distance, m
conductivity versus a grazing angle.
Fig. 7. RX power incident to the reader in a radio channel affected by the
Doppler effect versus time.
where θ is an angle between an antenna direction and a
radiating direction. A reader antenna is modeled with patch
antenna of width W and length L and the width of dielectric 0.12 0.10s
substrate h: 0.09s
0.10 0.08s
0.07s

bit error probability

r
λ 2 λ
W = , L= √ − 2∆L, 0.08 0.06s
2 r + 1 2 ref f 0.05s
0.06 0.04s
where ref f is an effective dielectric permittivity that equals 0.03s
0.04 0.02s
to:  −1/2 0.01s
r + 1 r − 1 h 0.02
0.00s
ref f = + 1+ ,
2 2 W
0.00
and ∆L = lef f − L, Lef f is an effective length, can be 2 4 6 8 10 12 14
distance, m
computed via the formula that is as follows:
W
ref f + 0.3 h + 0.264 Fig. 8. Bit error probability (BER) in a radio channel affected by the Doppler
∆L = 0.412h W
, effect versus time.
ref f − 0.258 h + 0.8

Using the formulas described above the radiation pattern

can expressed as [30]: 0.12

0.10
bit error probability

 
 sin kW cos θ kL  0.08
Γ(θ, φ) = sin θ kW2 cos ( sin θ sin φ) , (11)
 
2 0.06
2 cos θ
 
FM0
0.04 M2
where φ is an azimuth angle, θ is a tilt angle and k is a wave
0.02 M4
number. M8
0.00
2 4 6 8 10 12 14 16
D. Reflection distance, m

The ground reflection coefficient is defined by the following

formula [11]: Fig. 9. BER for different tag modulations.
√
sin φ − C
R= √ (12)
sin φ + C E. Bit Error Probability
2
where φ is a grazing angle, C = η − cos φ for horizontally Let us consider a radio-channel with additive white Gaus-
2
polarized component and C = η−cos η2
φ
, for vertical one; η = sian noise (AWGN). Then bit error probability can be ex-
r (f ) − j60λσ(f ), where r (f ) is a relative permittivity of a pressed as:
ground surface at the frequency f , σ(f ) is a conductivity. Per = 2Q(γ́)[1 − Q(γ́)], (13)
Relative permittivity and conductivity vary widely de-
pending on the ground humidity. Conductivity ranges from where Q(·) is a Q-function; γ́ = mEs /N0 cos2 φs , m is a
0.00014 S/m for very dry ground to 5 S/m for sea water. Miller coding order, Es is a symbol energy, N0 /2 is a white
Relative permittivity ranges from 3 to 70. noise spectral density and φs is a difference between signal
In figure 6 the absolute value of reflection coefficient in phase and tracked phase. The ratio Es /N0 can be transformed
dependence on a grazing angle is depicted for different values into γTs B, where γ is a signal-to-noise ratio, Ts is a symbol
of r and σ. duration and B is a bandwidth.

2469-7281 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JRFID.2017.2751592, IEEE Journal
of Radio Frequency Identification
6

TABLE II
C OMMAND AND R ESPONSE D URATION VERSUS PROTOCOL SETTINGS .
Reading EPC only Reading EPC and TID
QUERY AND PC+EPC+CRC DURATIONS ARE GIVEN IN µs 104 Tari=6.25us Tari=6.25us
Tari=12.5us Tari=12.5us
DR Tari TRcal BLF M QUERY PC+EPC+CRC Tari=25us Tari=25us

Round Duration, ms
103
64/3 6.25 33.38 639.2 1 252.13 211.2
64/3 6.25 33.38 639.2 8 270.88 1739.67
64/3 12.5 66.75 319.6 1 491.75 422.4 102
64/3 25.0 133.5 159.8 1 971.0 844.8
64/3 25.0 225.0 94.81 1 1062.5 1423.83 101
8 6.25 33.38 239.7 1 245.8 563.2
8 25.0 225.0 33.56 8 1112.5 31275.0
100
0 2 4 6 8 10 12 14 0 2 4 6 8 10 12 14
Q parameter Q parameter
To set up an initial phase a receiver utilizes a preamble. The
error of the phase evaluation specifies parameter φs , that can Fig. 10. Maximum round duration when reading EPC only and EPC and TID
be modeled with normal distributed random variable with zero depending on Q value and Tari, M = FM0.
p
mean and variance σs = 1/ γTpr B, where Tpr is a preamble
duration. Reading EPC only Reading EPC and TID
The usage of Gaussian model may be too optimistic if not 35
Tari=6.25us Tari=6.25us
all components of the multi-ray signal are tracked. For that 30 Tari=12.5us Tari=12.5us
case we use another expression that is obtained via averaging Tari=25us Tari=25us
25
Number of rounds
Per in formula (13) over Rayleigh-distributed γ́:
q 20
1 1 2 arctan 1 + γ́2
Per = + q + . (14) 15
2 1+ 2 π 1 + γ́2
γ́ 10
In figure 4c the result of BER computation is given. Ac- 5
cording to extremely large growth of BER with the drop of 0
signal-to-noise ratio there are distinct areas where a tag can 2 4 6 8 10 12 14 2 4 6 8 10 12 14
Q parameter Q parameter
be read with sufficient probability.

V. P ROTOCOL PARAMETERS A NALYSIS Fig. 11. Number of rounds when reading EPC only and EPC and TID
depending on Q value and Tari, M=FMO.
According to formula (2) the tag identification probability
depends on the number of rounds in which a tag participates.
The number of rounds is defined by the round duration which increases the round duration as well as the tag flags presetting
depends on Q-parameter value, the number of tags, collisions, by the Select command. We limit our study to two scenarios:
responses transmission errors, commands and responses dura- reading EPC for tag identification (inventory only) and reading
tions. both EPC and TID banks (inventory and access without
encryption).
A. Symbol Duration and Tag Encoding Figure 10 shows the maximum round duration and figure
11 – the number of rounds per meter depending on different
A reader makes use of DSB-ASK, SSB-ASK or PR-ASK
values of Q for the case when reading only EPC.
modulation. Due to PIE (Pulse Interval Encoding) the com-
mand transmission time depends on its content since the dura-
tions of data 0 and data 1 differ in 1.5-2 times. All Query B. Session Polling and Antenna Switching
commands are transmitted after a preamble, other commands An actual number of rounds the tag involved in depends on
follow a shorter synchronization sequence. Table II shows symbols durations, a tag encoding scheme and Q parameter
the commands and responses durations under various protocol value as well as a the session selection algorithm and antenna
settings. Parameters Tari, TRcal, QUERY and PC+EPC+CRC switching policy. Since the probability to read both EPC and
are given in microseconds, BLF is in kHz, M is a number of TID in a single round is small, the number of rounds the
symbols per bit in a tag response. tag participates in appears to be one of the key performance
The choice of DR, Tari, M, TRcal sufficiently influence on characteristics.
commands and responses durations. Thus, Query duration When starting an inventory round the reader selects a
may vary from 200 us to 1.1 ms and duration of a response to session (S0, S1, S2, S3) and its value (A or B). The tag
ACK (PC+EPC+CRC transmission) may vary from 211 us to responds only if its session value matches the value provided
31.2 ms. in the Query command. After backscattering its EPC the tag
The identification process may use only EPC, both EPC has to invert the session value (A → B, B → A). The standard
and TID or involve user memory access. Encryption usage defines different persistence values for different sessions, in

2469-7281 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JRFID.2017.2751592, IEEE Journal
of Radio Frequency Identification
7

Fig. 12. Simulation model design.

the simplest case (session S0) the tag sets its value equal to session has persistence time. If the tag was powered-off for
A after each power-up. a short interval due to propagation conditions or antenna
If the reader uses a single antenna without periodical switching and the interval is shorter than persistence, its
reloading, the actual number of rounds the tag participates session will be the same as before the power-off.
in depends on the session value polling strategy only. For The number of rounds changes if the reader turns the field
instance, figure 13 shows the number of rounds when querying off due to the local regulatory requirements. All computations
A value only (the lowest one horizontal line) and altering in this paper assumes the reader goes down each 2 seconds
A and B each round (the highest one). In the first case the for a 100 ms interval.
number of rounds is defined by the number of tag power-ups
under the propagation conditions and reader periodic switching
off. Altering the target value allows to increase the number of VI. S YSTEM S IMULATION
rounds more than six time. Fluctuations on the plot are due to A UHF RFID discrete-event simulation framework was
the stochastic nature of the tag reading process and the round developed to analyze the system performance under the vari-
duration dependence on the number of busy and empty slots. ous protocol settings, environment conditions and road traffic
The number of rounds may also be tuned by using Select parameters [31]. The simulation framework allows to take into
command. account miscellaneous specifics of the signal propagation such
When several antennas are used and the reader switches as multi-ray path loss environment involving ground reflec-
between them periodically, the number of rounds depends tions and Doppler shift, antennas radiation patterns, etc. The
on the antenna switching interval. Figure 13 illustrates this framework comprises a detailed EPC Class1 Gen2 protocol
dependence when the antennas are deployed to read the tags implementation enabling the performance estimation under
placed both in back and front license plates. Altering the target comprehensive settings including symbols durations, tag re-
value doesn’t provide such a significant performance increase sponses encoding schemas, session selection polices, antenna
in this case, but still rises the number of rounds more than switching rules and tags reading scenarios. The simulation
four times. framework is written in Python and C++ languages, uses
Session S2 usage was also studied. The results were similar OMNeT++ [14] and PyONS [15] discrete-event simulators and
and even worth for constant Target scenario since the S2 is available under the GNU GPLv3 license.

2469-7281 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JRFID.2017.2751592, IEEE Journal
of Radio Frequency Identification
8

TABLE III
Number of inventory rounds per tag
session S0, M4, Tari=18.75us AVERAGE NUMBER OF TAGS PER ROUND . V EHICLE SPEED IS 60 KM / H .
24 2 antennas, target = A (1)
Tv , sec Nv Nt Nt
20 2 antennas, switch target A <-> B
Number of rounds

1 antenna, target = A 0.5 5.67 0.1000 1.0

16 1 antenna, switch target A <-> B 1.0 3.38 0.0357 1.0
12
8
4 tions and path loss computation is implemented in Medium
0
modules which can be associated with specific geographical
0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40 0.45 0.50 regions.
Antenna switching interval, ms
The detailed description of the simulation models is outside
of this paper scope and can be found in [31]. It is worth point-
Fig. 13. Number of rounds the tag participates in depending on the session ing out that the key problem in the discrete-event simulation of
polling strategy
the UHF RFID system is the model performance. For instance,
the results presented in this paper required over twelve hours
Front Plates Back Plates
1.0 of four quad-core Xeon servers total runtime to simulate 5000
vehicles for each model configuration and for each average
0.8
vehicle speed. Due to very short interframe intervals, small
frames durations and very frequent antennas positions updates
Tag Read Probability

0.6
the model should carefully use the event queue to keep the
simulation time acceptable. To speed up the simulation the
models use events for data transfer between the layers and
0.4
EPC, Tari=6.25us EPC, Tari=6.25us devices only, while using different techniques like publish-
TID, Tari=6.25us TID, Tari=6.25us
EPC, Tari=18.75us EPC, Tari=18.75us subscribe patterns for control and management information
0.2 TID, Tari=18.75us TID, Tari=18.75us
EPC, Tari=25.0us EPC, Tari=25.0us
transmission between the modules.
TID, Tari=25.0us TID, Tari=25.0us
0.0
60 80 100 120 60 80 100 120
Speed, kmph Speed, kmph B. Simulation Results
By the means of the simulation framework we performed the
Fig. 14. License plate tag identification probability, M = 4 simulation assuming the reader equipped with two antennas
deployed above one road lane in opposite directions. Such
antenna placement allows to read both front and back license
A. Simulation Model Design plates. In order to select the proper value of the Q parameter
The UHF RFID simulation framework includes two inde- a mean number of tags participating in an inventory round
pendent models. The first one is written in Python language was computed. Table III shows the mean number of vehicles
using PyONS simulator. In consequence of using Python, Nv near the reader and the mean number of tags Nt for
this model can be easily extended to add support for any different time intervals Tv between vehicles arrivals. The
(1)
standard or experimental protocol features. The second model column Nt shows the mean number of tags in the round
is implemented in C++ language using OMNeT++ simulator. where at least one tag is responding. While a large number of
While providing less features than the first one, this model vehicles are located in the reading area, only a small fraction
provides better performance and is generally more suitable of the tags are powered up due to the multi-path propagation.
at handling large-scale simulations. To unify access to both The propagation loss affected with the ground reflection and
models the framework provides the API written in Python. Doppler shift leads to significant oscillations of the power
This API provides the means for parameters description, incident to a tag with variation of distance between the tag
simulation using Python or C++ models and results analysis and the reader. It can be concluded that Q = 2 value provides
and visualization. good enough collision probability mitigation.
Both models share the common design shown in figure 12. Figures 14 and 15 illustrate the estimated tag identification
At the top level each model consists of the devices (readers probability for different Tari and M values. It was assumed
and tags), things (e.g. vehicles) and management modules. that reader antennas gain was 8 dBi and cable loss was -2 dB,
Each device comprises logical, physical and radio layers: the tags used extended preambles in the replies and divide ratio
logical layer implements commands and replies generation, the was equal to 8. The reader worked with session S0, inverted its
physical layer simulates frames encoding and modulation as Target value in each Query command and switched antennas
well as link timing, and the radio layer simulates commands every 100 ms. It can be seen that the identification probability
and replies transmission over the air. The radio layer consists decreases along with the vehicle speed growth. While EPC
of antennas, transceivers and antenna switching units. Partic- identification rate is large enough for fast moving tags, TID
ularly, each transceiver includes a channel model for BER read probability decreases significantly. It should be noted that
computation. Besides that, each device includes power control choosing large Tari value doesn’t mandatorily lead to a better
modules and several management units. Environment condi- identification rate as well as choosing large M values. This

2469-7281 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JRFID.2017.2751592, IEEE Journal
of Radio Frequency Identification
9

Front Plates Back Plates Vehicle identification probability

1.0 1.0

0.8 0.9
Tag Read Probability

0.6 0.8

Probability
EPC, M2, Tari=12.5us
0.7 TID, M2, Tari=12.5us
0.4 EPC, M4, Tari=12.5us
EPC-M2 EPC-M2
TID-M2 TID-M2 TID, M4, Tari=12.5us
EPC, M8, Tari=12.5us
EPC-M4 EPC-M4 0.6
0.2 TID, M8, Tari=12.5us
TID-M4 TID-M4
EPC, M4, Tari=25.0us
EPC-M8 EPC-M8 TID, M4, Tari=25.0us
TID-M8 TID-M8 0.5
0.0 50 60 70 80 90 100 110 120 130
60 80 100 120 60 80 100 120 Vehicle speed, kmph
Speed, kmph Speed, kmph

Fig. 17. Vehicle identification probability, each vehicle is identified if any

Fig. 15. License plate tag identification probability, Tari = 12.5 µs license plate tag was read

Front Plates Back Plates

1.0 VII. C ONCLUSION
0.9 The UHF RFID is a promising technology for automatic
0.8
vehicle identification. However, to maximize its performance
propagation media conditions, equipment placement and vari-
Tag Read Probability

0.7
ous protocol parameters must be taken into account. Doppler
0.6 shift and multipath propagation affect the system significantly
0.5
as well as the Q, Tari, tag encoding and session values
selection. It was shown that the Target flag switching allows to
0.4 EPC, with Doppler effect EPC with Doppler effect
TID with Doppler effect TID with Doppler effect
increase the tag identification probability. Another observation
0.3 EPC, without Doppler effect EPC without Doppler effect is a degrading system performance with large M and Tari val-
TID without Doppler effect TID without Doppler effect
0.2 ues. According to the results provided, average values M = 4
60 80 100 120 60 80 100 120 and Tari equal to 12.5 µs or 18.25 µs should be preferred
Speed, kmph Speed, kmph
in most cases. While the demonstrated results were obtained
Fig. 16. Doppler effect influence on tag identification probability for M=4,
with typical protocol and RFID equipment parameters, more
Tari = 12.5 µs accurate data should be used to get higher precision results.
It should be also noted, that the system doesn’t suffer from
collisions since most of the time there is one tag operating in
could be described by decreasing the rounds number the tag the interrogator field at most.
participates in due to long commands and replies durations. It was shown that the system performance is satisfactory for
Doppler shift has significant impact on tags read rate (see the vehicles moving with average speed up to 100 kilometers
Fig. 16). The difference becomes especially large for tags per hour, but becomes too low for fast moving vehicles.
moving on high speed, when Doppler effect decreases the Different solutions exist for this problem mitigation. One way
license plate identification probability by 10–20%. It should is to use better antennas with higher gains. Another way is
also be mentioned that Doppler to optimize the tags querying procedures, e.g. repeating Read
The overall vehicle identification probabilities are shown commands several times. It is a focus of the future work to
on figure 17. Each vehicle was treated identified if any one study various tags querying scenarios and their influence on
license plate TID was read. The estimations coincide with the the system performance.
experimental data for normal vehicle speed. The best results
were obtained with M = 4 value and Tari = 12.5 µs. While the ACKNOWLEDGMENT
vehicle identification probabilities for Tari = 25 µs and M = 8 This work has been financially supported by the Russian
were high in the beginning, they degraded along with the Science Foundation and the Department of Science and Tech-
vehicle speed growth. While the results provided demonstrate nology (India) via grant 16-49-02021 for the joint research
that the vehicle identification rate isn’t very high for fast project by the V.A.Trapeznikov Institute of control Sciences
moving vehicles, it can be improved by using the reader and the CMS College Kottayam.
antennas with higher gain and tags with better sensitivity.
Another reason for low identification probability is the choice
R EFERENCES
of Rayleigh distribution in BER evaluation since it gives a
pessimistic BER estimation in the environment considered, [1] EPCGlobal, “EPC radio-frequency identity protocols generation-2 UHF
RFID. specification for RFID air interface. protocol for communications
and the values obtained within the simulation may slightly at 860 mhz 960 mhz. version 2.0.1 ratified,” GS1 EPCglobal Inc.,
differ if a more realistic model is chosen. EPCGlobal Gen2 Specification, 2015.

2469-7281 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JRFID.2017.2751592, IEEE Journal
of Radio Frequency Identification
10

[2] D. M. Dobkin, The RF in RFID: Passive UHF RFID in Practice. [27] L. Azpilicueta, C. Vargas-Rosales, and F. Falcone, “Intelligent Vehicle
Elsevier Inc, 2008. Communication: Deterministic Propagation Prediction in Transportation
[3] K. Finkenzeller, RFID Handbook: Fundamentals and Applications in Systems,” IEEE Vehicular Technology Magazine, vol. 11, pp. 29–37,
Contactless Smart Cards and Identification, 2nd ed. Wiley Publishing, Sept. 2016.
2003. [28] J. D. Griffin and G. D. Durgin, “Complete Link Budgets for Backscatter-
[4] V. M. Vishnevsky, A. A. Larionov, and R. E. Ivanov, “Architecture of ap- Radio and RFID Systems,” IEEE Antennas and Propagation Magazine,
plication platform for RFID-enabled traffic law enforcement system,” in vol. 51, no. 2, pp. 11–25, April 2009.
Proceedings of the 2014 7th International Workshop on Communication [29] G. Matz and F. Hlawatsch, Chapter 1. Fundamentals of Time-Varying
Technologies for Vehicles (Nets4Cars-Fall). IEEE, 2014. Communication Channels. Elsevier, 2011, ch. 1, pp. 1–63.
[5] P. V. Nikitin and K. V. S. Rao, “Antennas and propagation in UHF [30] C. A. Balanis, Antenna Theory: Analysis and Design, 4th ed., 2016.
RFID systems,” in Proceedings of the IEEE International Conference [31] A. Larionov and R. Ivanov, “RFID Simulator,” 2017. [Online].
on RFID. IEEE, 2008, pp. 277–288. Available: https://github.com/larioandr/rfidsim
[6] K. V. S. Rao, P. V. Nikitin, and S. F. Lam, “Antenna design for UHF
RFID tags: a review and a practical application,” IEEE Transactions on
Antennas and Propagation, vol. 53, no. 12, pp. 3870–3876, 2005.
[7] P. V. Nikitin and K. V. S. Rao, “Performance limitations of passive
UHF RFID systems,” in IEEE Antennas and Propagation Society Andrey A. Larionov received an Engineering de-
International Symposium. IEEE, 2006, pp. 1011–1014. gree in applied mathematics and computer science
[8] P. Nikitin, K. Rao, E. Lam, and P. Pinc, “UHF RFID tag characterization: in 2007 from Faculty of Computational Mathemat-
Overview and state-of-the-art,” in AMTA, Seattle, WA, 2012. ics and Cybernetics of Lomonosov Moscow State
[9] P. V. Nikitin and K. V. S. Rao, “Measurement of backscattering from University (CMC MSU), Russia. Currently he is
RFID tags,” in Proceedings of Antennas Measurement Techniques As- a research fellow of V.A. Trapeznikov Institute of
sociation Symposium, 2005. Control Sciences of Russian Academy of Sciences
(ICS RAS). His research interests cover wireless net-
[10] R. V. A. Lazaro, D. Girbau, “Effects of interferences in UHF RFID
systems,” Progress In Electromagnetics Research, vol. 98, pp. 425–443, works and UHF RFID systems, simulation modeling
2009. and queueing theory.
[11] S. R. M. Gonzalez and R. L. Miranda, Chapter 9. Passive UHF
RFID Technology Applied to Automatic Vehicle Identification: Antennas,
Propagation Models and Some Problems Relative to Electromagnetic
Compatibility. IGI Global, 2013, ch. 9, pp. 188–220.
[12] M. Mohaisen, H. Yoon, and K. Chang, “Radio transmission performance Roman E. Ivanov received B.E. degree in applied
of EPCglobal Gen-2 RFID system,” CoRR, vol. abs/0911.0542, 2009. physics and mathematics in 2013 and M.E. degree
[13] P. V. Nikitin and K. V. S. Rao, “Effect of Gen2 protocol parameters in telecommunication systems and networks in 2015
on RFID tag performance,” in Proceedings of the IEEE International from Moscow Institute of Physics and Technology,
Conference on RFID. IEEE, 2009. Russia. Currently he is a research fellow of V.A.
[14] “OMNeT++ Discrete Event Simulator,” 2017. [Online]. Available: Trapeznikov Institute of Control Sciences of Russian
https://omnetpp.org/ Academy of Sciences. He was heavily involved in a
[15] A. Larionov and R. Ivanov, “PyONS: Python Open Network Simulator,” radio propagation analysis of UHF RFID channels.
2017. [Online]. Available: https://github.com/larioandr/pyons His interests include radio propagation, resource
[16] J. Landt, “The history of RFID,” IEEE Potentials, 24, vol. 24, pp. 8–11, allocation, performance analysis of IEEE 802.11 and
2005. mesh networks.
[17] P. Blythe, “RFID for road tolling, road-use pricing and vehicle access
control,” IEEE Colloquium on RFID Technology (Ref. No, pp. 8–16,
1999.
[18] A. A. Khan, A. I. E. Yakzan, and M. Ali, “Radio frequency identification
(RFID) based toll collection system,” in Proceedings of the 2011 Third Vladimir M. Vishnevsky received Engineering de-
International Conference on Computational Intelligence, Communica- gree in applied mathematics in 1971 from Moscow
tion Systems and Networks, ser. CICSYN ’11. Washington, DC, USA: Institute of Electronics and Mathematics, Russia,
IEEE, 2011, pp. 103–107. Ph.D. degree in queuing theory and telecommuni-
[19] W.-J. Yoon, S.-H. Chung, and S.-J. Lee, “Implementation and perfor- cation networks in 1974 and Sc.D. in telecommu-
mance evaluation of an active RFID system for fast tag collection,” nication networks in 1988 from both from V.A.
Comput. Commun., vol. 31, no. 17, pp. 4107–4116, Nov. 2008. Trapeznikov Institute of Control Sciences of Russian
[20] H. A.-A. Fawzi M. Al-Naima, “Vehicle location system based on RFID,” Academy of Sciences (ICS RAS). He became a
in Proceedings of the Developments in E-systems Engineering (DeSE), Professor at ICS RAS in 1989 and at Moscow
2011. IEEE, 2011. Institute of Physics and Technology in 1990.
[21] R.-J. K. Jan-Dong Tseng, Wen-De Wang, “An UHF band RFID vehicle Currently he is a head of telecommunication net-
management system,” in Proceedings of the 2007 IEEE International works laboratory at ICS RAS. He was an Assistant Head of the Institute of
Workshop on Anti-counterfeiting, Security, Identification. IEEE, 2007. Information Transmission Problems of RAS from 1990 to 2010 and Assistant
[22] V. M. Vishnevsky and A. A. Larionov, “Design concepts of an ap- Head of laboratory in ICS RAS from 1971 to 1990. He is the author of over
plication platform for traffic law enforcement and vehicles registration 300 papers in queuing theory and telecommunications. He is a co-Chair of
comprising RFID technology,” in 2012 IEEE International Conference ICUMT, IEEE conference, RTUWO, IEEE conference and a General Chair
on RFID-Technologies and Applications (RFID-TA). IEEE, 2012. of DCCN. His main interests include queuing systems, telecommunications
and mathematical modeling.
[23] C. Floerkemeier and S. Sarma, “RFIDSim - a physical and logical layer
simulation engine for passive RFID,” IEEE Transactions on Automation
Science and Engineering, vol. 6, no. 1, pp. 33–43, January 2009.
[24] D. Arnitz, U. Muehlmann, T. Gigl, and K. Witrisal, “Wideband system-
level simulator for passive UHF RFID,” in IEEE International Confer-
ence on RFID. Orlando, FL, USA: IEEE, 2009, pp. 28–33.
[25] Z. Zhang, Z. Lu, Q. Chen, X. Yan, and L.-R. Zheng, “COSMO:
CO-Simulation with MATLAB and OMNeT++ for Indoor Wireless
Networks,” in GLOBECOM. IEEE, 2010, pp. 1–6.
[26] A. G. Dimitrio, S. Siachalou, A. Bletsas, and J. N. Sahalos, “A site-
specific stochastic propagation model for passive UHF RFID,” IEEE
Antennas and Wireless Propagation Letters, vol. 13, pp. 623–626, March
2014.

2469-7281 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.