Energy Efficient Tag Identification Algorithms For RFID: Survey, Motivation And New Design

Abstract

RFID is widely applied in massive tag based applications, thus effective anti-collision algorithms to reduce communication overhead are of great importance to RFID in achieving energy and time efficiency. Existing MAC algorithms are primarily focusing on improving system throughput or reducing total identification time. However, with the advancement of embedded systems and mobile applications, the energy consumption aspect is increasingly important and should be considered in the new design. In this article, we start with a comprehensive review and analysis of the state-of-the-art anti-collision algorithms. Based on our existing works, we further discuss a novel design of anti-collision algorithm and show its effectiveness in achieving energy efficiency for the RFID system using EPCglobal C1 Gen2 UHF standard.

Full text

1536-1284/19/$25.00 © 2019 IEEE IEEE Wireless Communications • Accepted for Publication
1
AbstrAct
RFID is widely applied in massive tag based
applications, thus effective anti-collision algorithms
to reduce communication overhead are of great
importance to RFID in achieving energy and time
efficiency. Existing MAC algorithms are primarily
focusing on improving system throughput or reduc-
ing total identification time. However, with the
advancement of embedded systems and mobile
applications, the energy consumption aspect is
increasingly important and should be considered
in the new design. In this article, we start with a
comprehensive review and analysis of the state-
of-the-art anti-collision algorithms. Based on our
existing works, we further discuss a novel design of
anti-collision algorithm and show its effectiveness
in achieving energy efficiency for the RFID system
using EPCglobal C1 Gen2 UHF standard.
IntroductIon
The Internet of Things (IoT) is an emerging appli-
cation of Internet and traditional telecommunica-
tion networks, which allows embedded objects
to implement interconnection and interopera-
bility. Radio frequency identification (RFID),
which is a key technology to enable Internet of
Things (IoT), can identify objects automatically
by employing wireless communication. Typical-
ly, an RFID system includes tags, a reader and
back-end systems. A tag is made up of antenna,
coupling component, and microchip. Enclosed in
an adhesive sticker, every tag is attached to an
item with a unique identifier (UID). The reader
initializes an identification process by broadcast-
ing a query command. After receiving the query
command, the tags in the vicinity respond to
the reader with their IDs. Accordingly, RFID can
identify multiple items without line of sight and
easily map the physical world to the cyber world.
According to the power supply mode, an RFID
tag can be categorized as either a passive or
active tag. A passive tag with small size and low
cost has no onboard power supply, its operating
energy is from the continuous wave transmitted
by the reader. Thus, the transmission distance is
quite limited. In contrast, an active tag has an
internal battery to provide energy for the micro-
chip and ensure communication between tag
and reader. The potential transmission range can
thus reach several hundred meters. However,
the production cost is high, and the service life-
time is short because the battery needs to be
periodically replaced.
RFID drives many IoT applications. For exam-
ple, by accurately tracking a good expiry date or
item leakage, RFID can help reduce waste and
energy consumption in operations ranging from
monitoring to packaging and refrigeration, which in
turns enables more extensive deployment of RFID
systems [1]. With a future trend to integrate RFID
into IoT system, the format of the reader may not
necessarily be a fixed device. A mobile reader or
even battery powered wireless sensor nodes can
be enabled as reader devices. Thus, energy effi-
ciency is an important metric to evaluate the over-
all performance of RFID systems [2-4]. An energy
efficient RFID protocol can prolong the operating
lifetimes of readers and tags (if they are active) and
promote the growth of green RFID and its various
applications that have been envisioned. In order to
achieve that, the reader needs to adopt an ener-
gy-efficient anti-collision algorithm to optimize tag
cardinality (the number of unread tags) estimation,
adaptively modulate transmission power level, and
reduce tag collision and eavesdropping, etc. [4]. In
this article, we survey the state-of-the-art anti-col-
lision algorithms and demonstrate our effort in
developing an energy efficient RFID anti-collision
algorithm.
Existing RFID anti-collision solutions can be
mainly divided into Aloha-based [5-6] and tree-
based. A tree-based [7] algorithm is especially
operated by recursively dividing the contending
tags into smaller groups until each group contains
up to one tag. An Aloha-based algorithm employs
a frame structure that contains a certain number
of time intervals (called time slots) per frame, and
tags randomly pick up a time slot to respond to the
reader using their IDs. These previous works pay
more attention to improving system throughput
or reducing identification time rather than ener-
gy consumption. For a static or fixed reader with
power supply, identification time is more important
when a number of tagged items need to be identi-
fied in a continuous scanning manner. However, in
many scenarios like inventory control where a por-
table or mobile reader with limited power is widely
used, the energy consumption is critical, especially
when periodic scanning is needed. Lower energy
consumption can maintain a longer service time
and avoid frequent recharging or replacement.
In the following sections, we start with a review
of various types of anti-collision mechanisms. Fur-
Jian Su, Zhengguo Sheng, Victor C. M. Leung, and Yongrui Chen
E E T I A
F RFID: S, M A N D
ACCEPTED FROM OPEN CALL
Jian Su is with Nanjing University of Information Science and Technology; Zhengguo Sheng is with the University of Sussex;
Victor C. M. Leung is with the University of British Columbia; Yongrui Chen ias with University of Chinese Academy of Sciences.
Digital Object Identifier:
10.1109/MWC.2019.1800249

IEEE Wireless Communications • Accepted for Publication 2
thermore, in order to achieve energy efficiency and
reduce the computational complexity, we present
an anti-collision solution called an Energy-Aware
Frame Adjustment Strategy (EAFAS) based algo-
rithm. The proposed algorithm integrates the
low-cost estimation method, adaptive frame
size calculation strategy and efficient frame size
adjustment policy. To be specific, the presented
algorithm ascertains the optimal frame size based
on both estimated tag cardinality and energy effi-
ciency of RFID system. Moreover, the proposed
in-frame mechanism can also end the improper
frame in advance. The remainder of the article is
organized as follows. The next section reviews and
analyzes the mainstream anti-collision strategies for
RFID systems. Then we discuss the existing ener-
gy-efficient RFID algorithms. A newly energy-effi-
cient anti-collision algorithm is then described.We
then illustrate the performance results. Finally, we
conclude the article.
summAry of AntI-collIsIon AlgorIthms In rfId
With consideration of cost and implementation
complexity, the time-division multiple access
(TDMA) solutions have been mainly used in RFID
systems. That is, each tag occupies the channel
in a separated time interval and communicates
to the reader. The TDMA based solutions can be
divided into Aloha-based and tree-based algo-
rithms which can be further divided into binary
splitting and query tree. A comparison of various
solutions is shown in Table 1.
AlohA-bAsed AlgorIthms
Aloha-based algorithms can be divided into three
types, namely Pure Aloha (PA), Slotted Aloha (SA)
and Framed Slotted Aloha (FSA). Among them,
Dynamic FSA (DFSA) has been widely used in
UHF RFID. The DFSA algorithm is characterized
by the strategy to adapt the frame size along
the identification process [6]. The principle of
the DFSA algorithm is to divide time into several
frame segments, each of which consists of a num-
ber of time slots. A tag responds to the reader
with its ID when it receives a query command
specifying the parameter F (F corresponds to the
number of slots per frame). There are three pos-
sible states in a time slot: single response (single-
ton slot), no response (empty slot), and multiple
responses (collision slot). After reading a frame,
the reader needs to make full use of probabilistic
or statistical methods to estimate the cardinality.
Therefore, the Aloha-based algorithms can also be
called probabilistic algorithms.
The performance of DFSA depends on both
the cardinality (the number of remaining tags) esti-
mation and the setting of the frame size. For a par-
ticular frame, the system throughput is calculated
as the number of identified tags over the frame
size. Specifically, the maximum system throughput
0.368 is attained asymptotically when the frame
size is equal to the number of tags to be identi-
fied [5]. In order to improve the performance of
DFSA, most previous solutions [5-6] require vast
computational costs so that the accuracy of esti-
mation can be ensured. However, most handheld
RFID readers in practice are computation con-
strained due to their low-cost hardware structure
such as single-core microprocessor. Consequently,
anti-collision algorithms with complex estimation
are inefficient in terms of time or energy efficien-
cy. Recently, a number of energy-efficient DFSA
algorithms have been proposed for the purpose
of reducing computational overhead. The litera-
ture [9] presented an anti-collision protocol that
depends on one examination of frame size at a
specific time slot during each identification round.
The authors in [18] introduced an Improved Linear-
ized Combinatorial Model (ILCM) to estimate the
cardinality with modest calculation cost. However,
its performance fluctuates sharply with the number
of tags. In [10], the authors presented a FuzzyQ
method that integrates fuzzy logic with a DFSA
algorithm. A fuzzy rule based system is defined to
model the frame size and the collision rate with
fuzzy sets to adaptively calculate frame size. How-
ever, the performance of FuzzyQ needs to be fur-
ther improved.
ere are three pos-
sible states in a time
slot: single response
(singleton slot), no
response (empty slot),
and multiple responses
(collision slot). Aer
reading a frame, the
reader needs to make
full use of probabilistic
or statistical methods
to estimate the cardi-
nality. erefore, the
Aloha-based algorithms
can also be called
probabilistic algorithms.
TABLE 1. The Characteristics of various anti-collision algorithms.
Categories Methods No Tag
starvation
Good
stability
High
throughput
Low
complexity
Energy
efficient
Compatible
to UHF
standard
DFSA
MAP [5] ü ü
ILCM [10] ü ü
EACAEA [9] ü ü
FuzzyQ [11] ü ü ü
SUBF-DFSA [12] ü ü ü ü
ds-DFSA [14] ü ü ü ü
EAFSA ü ü ü ü ü
QT
CT [7] ü ü ü ü
MQT [13] ü ü ü
BS
BS [15] ü ü ü ü
ISE-BS [15] ü ü ü ü

IEEE Wireless Communications • Accepted for Publication
3
The overall feature of the Aloha-based algo-
rithms is easy to implement. The reader can sta-
tistically analyze the distribution of tags within the
frame thereby estimating the number of unread
tags. The disadvantages are the tag starvation prob-
lem and high complexity in tag cardinality estima-
tion. The tag cardinality estimation methods with
high complexity cause higher energy consumption
at the reader side.
Query tree AlgorIthms
In query tree (QT) algorithms, every tag is
assigned with a unique ID. The QT algorithm is
working as a virtual traversal tree. The depth is
defined as the number of branches from the root
node to the leaf node. Each branch is marked
with the method of “left 0 right 1.” The read-
er first sends a probe command with a prefix
0; tags with prefix ID 0 will transmit their IDs to
the reader. When a collision occurs, the read-
er iteratively divides the collided tags into two
subsets according the position of collided bits
of tags. The subsets become smaller until each
subset contains only one tag. Such algorithms
require a stack inside the reader to store the
query prefix information. The reader constantly
updates the query prefix based on the collision
bits and pushes the query prefix onto the stack
until the stack is empty. The current research
on the QT algorithm focuses on how to use the
collision information to update the query prefix.
The authors in [7] presented a collision tree (CT)
algorithm that generates prefix and splits collid-
ed tags according to the first collided bit. M-ary
Query Tree (MQT) has been proposed in [13],
by forming an M-ary traversal tree instead of a
binary traversal tree for collided tags. Although
the number of probed slots is reduced, the com-
munication overhead is increased by containing
both a mapped M-bits string and ID sequence.
Compared to Aloha-based algorithms, QT algo-
rithms use the tag ID to separate the collided tags.
Hence, they are deterministic in nature and not
suffering from the tag starvation problem. How-
ever, QT algorithms require a large number of
reader queries and tag responses, and rely on col-
lided responses to determine subsequent queries,
which causes higher energy consumption at both
the reader and tags (if the active tags are used).
Also, QT algorithms require strict synchronization
of the date responded by multiple tags. Therefore,
its application is limited and it is difficult to apply to
UHF RFID systems.
bInAry splIttIng AlgorIthms
Binary splitting (BS) algorithms were original-
ly developed for random access networks. The
BS algorithm continuously divides a collided
tag set into smaller subsets by using a random
binary number. Essentially, a BS algorithm also
belongs to a probabilistic algorithm because it has
a strong randomness. It has been widely used in
the ISO/IEC 18000-6B standard. Different from
the Aloha-based algorithm, contending tags in
the BS will be repeatedly divided into groups until
each group owns only one tag. Compared to Alo-
ha-based algorithms, BS algorithms are insensitive
to tag cardinality because when the tag number
is increased, the system throughput is almost
converged to a constant value. Although the BS
approach can tackle tag starvation, it has relative-
ly long identification latency due to the splitting
procedure starting from a single set with all tags.
Moreover, such an algorithm always uses a tag ID
to perform collision arbitration, and hence reduc-
es the time efficiency.
exIstIng energy-effIcIent
AntI-collIsIon AlgorIthms
dfsA AlgorIthms wIth
low cost estImAtIon And sub-frAme observAtIon
In most RFID application scenarios, the reader
needs to estimate the cardinality accurately to
maximize the system throughput. Previous works
focus on how to improve the estimation accu-
racy at the end of the frame, then update the
frame size accordingly [5–6, 8]. If the foregoing
frame size is improper, the accumulated estima-
tion error will degrade the whole performance.
Moreover, the estimation with high complexity
will consume more energy. In [11], we proposed
a sub-frame based algorithm (SUBF-DFSA) to
overcome the accumulated estimation error.
Specifically, the tag cardinality is estimated
based on the linear relation between the empty
and collision slot statistically counted in a sub-
frame. Since the computational complexity of
the estimation is reduced, the energy efficiency
of SUBF-DFSA can be improved compared to
the estimation methods with high complexity.
However, since the usage of empirical correla-
tion is not based on theoretical calculation, the
accuracy of estimation is not sufficient. In [12],
we further proposed a two-phase anti-collision
algorithm called detected sector based DFSA
(ds-DFSA) to enhance the identification perfor-
mance. The ds-DFSA algorithm effectively uses
empty, singleton and collision statistics in an
early observation phase to recursively determine
an optimal frame size. After that, the simple
calculation is used to estimate the number of
concurrent tags contained in each collision slot
nave. Then, the frame size for each collision slot
is obtained as the closest power-of-two value to
nave. Benefiting from such divide-and-conquer
frame size assignment and low-cost estimation
strategy, both time and energy efficiency have
been improved. Moreover, due to the require-
ment of a new command, modification to the
existing UHF RFID standard is needed for the
proposed algorithm, thus making it difficult to be
implemented in off-the-shelf RFID systems.
bs AlgorIthm wIth Idle slots elImInAtIon
In BS-based algorithms, a single set is usually
formed in concurrent tags. If a collision is detect-
ed, the reader will repeatedly divide the collided
tag set into multiple subsets and resolve them one
by one. Although BS-based solutions are robust to
tag starvation, its latency is high due to the large
number of concurrent tags involved in each col-
lision response. Moreover, the tag ID is used for
collision arbitration in BS-based solutions, which
increases the total collision arbitration time and
wastes the energy consumption. In [14], by elim-
inating the empty slots, we presented a binary
splitting protocol (ISE-BS) to enhance the perfor-
mance of the RFID system. The ISE-BS algorithm
Compared to Alo-
ha-based algorithms,
QT algorithms use the
tag ID to separate the
collided tags. Hence,
they are determinis-
tic in nature and not
suffering from the tag
starvation problem.
However, QT algo-
rithms require a large
number of reader que-
ries and tag responses,
and rely on collided
responses to determine
subsequent queries,
which causes higher
energy consumption
at both the reader and
tags (if the active tags
are used).

IEEE Wireless Communications • Accepted for Publication 4
is a variant of binary splitting by means of intro-
ducing 1-bit Q signal to pre-split contending tags
set. Since the empty slots in the splitting process
and the time duration used for collision arbitra-
tion are eliminated, the performance of ISE-BS
can be improved on the basis of time and energy
efficiency.
Although the above discussed algorithms can
improve the energy efficiency to some extent,
they are unable to optimize the frame size set-
ting according to the energy consumption. In the
following, an improved anti-collision algorithm is
proposed to overcome these drawbacks. The pro-
posed algorithm aims at achieving both robust esti-
mation with low-cost and energy efficient frame
size setting.
A new desIgn of
energy effIcIent IdentIfIcAtIon strAtegy
tAg cArdInAlIty estImAtIon
In order to guarantee the estimation accuracy,
here we also refer to the maximum a posteriori
probability (MAP) [5] method to calculate the
tag cardinality based on feedback from a sub-
frame. Although MAP can achieve an accurate
estimation, its high computational overhead hin-
ders its application in low-cost RFID platforms
such as a handheld reader. In the proposed
estimation method, we design look-up tables
(LUT) to pre-store intermediate variables of
estimation results. Restricted by the sub-frame
size and the item quantity in the tables, the pro-
posed estimation strategy is space-efficient and
implementable. Considering n tags allocated in F
slots, the probability that an empty slot occurs e
times, singleton slot occurs s times, and collision
slot occurs c times in a sub-frame Fsub can be
expressed as P(n|e, s, c) by using multinomial
distribution, that is
P(n|e,s,c)=F
sub
!
e!s!c!
P
i
e
P
s
s
P
c
c
(1)
The tag cardinality involved in a sub-frame is
determined when the value of P(n|e, s, c) is max-
imized. So, the estimation result in a sub-frame is
^
nsub. Then the estimated cardinality involved in the
full frame is calculated as
^
nest = ^
nsub · (F/Fsub) (2)
The recommendation setting of Fsub can be
referred to [12].
AdAptIve frAme sIze cAlculAtIon
Traditionally, most existing DFSA algorithms set
the frame size as the proximal value of the esti-
mated tag cardinality with the aim of maximizing
the system throughput [9, 11, 12, 15]. However,
such frame size setting strategy is only applied
when equal time duration for each slot type is
assumed. The EPCglobal C1 Gen2 UHF RFID
standard specifies different duration for empty,
singleton, and collision slots, defined as Te, Ts, and
Tc, respectively. Therefore, the traditional system
throughput is not an appropriate metric to eval-
uate the performance of the RFID anti-collision
algorithm. Moreover, such metric does not con-
sider the energy consumption. Unlike the pre-
vious DFSA algorithms, the proposed algorithm
calculates the frame size by maximizing the ener-
gy efficiency, which can be defined as
η
effi
=S⋅(P
RT
+P
Rr
)⋅T
EPC
[ ]
T
total
⋅P
Rt
+T
received
⋅P
Rr
+E
est
(3)
wherein S and C denote the number of singleton
slots and collision slots, respectively. TS, TRN16, and
TEPC denote the time duration of a singleton slot,
a 16-bits random number and a EPC (UID). Ttotal
denotes the required time for a whole identifica-
tion process. Treceived denotes the receiving time
duration of the reader for identifying all tags. PRt
and PRr represent the transmitting and receiving
power of the reader when it communicates with
tags, respectively. Eest is the energy consumption
during the tag cardinality estimation process. If
the frame size F is assumed large enough, the
probability distribution of r tags allocated in a slot
can be approximated as a Poisson distribution
with mean l = n/F. Then S, C, Treceived and Ttotal
can be approximated as the functions of the tag
number n and frame size F. By maximizing Eq. 1,
the value of l corresponding to the different time
parameters can be derived. The optimal frame
size can be determined by the estimated tag car-
dinality nest and l.
frAme sIze Adjustment strAtegy
The general frame size adjustment strategy can
be divided into three categories. First is Frame-by-
Frame (FbF) [5, 8] in which the reader calculates
a new frame size in the last slot of the current
frame. The FbF strategy is not efficient when the
frame size (the number of slots within the frame)
is far away from the number of tags. Second is
Slot-by-Slot (SbS) in which the reader calculates
the new frame size at every slot in the current
frame. The SbS strategy suffers from a rather high
complexity. Finally, the sub-frame solution pro-
vides the flexibility of ending the current frame
in advance to maintain the performance stability
with a reduced computational complexity. Third is
Point-by-Point (PbP) in which the reader chooses
some particular slots within the frame, referred to
as the point in the presented works [9, 11, 12].
The reader updates the frame size at the point
which is usually set as a fraction of the current
frame. In our proposed algorithm, we adopt a
hybrid strategy combining sub-frame observation
and SbS. At every slot, the reader keeps track of
the relation between E and C, and then the read-
er will reset the sub-frame size if the difference
value between E and C is above the threshold
value. After the reading of Fsub slots, the reader
estimates the tag cardinality and updates the new
frame size for the next identification round. Then
the reader computes the energy efficiency heffi1
and heffi2 with the current frame size and the new
frame size, respectively. The reader ends the cur-
rent frame and enables the new frame only if
heffi1 < heffi2. Otherwise, the reader will continue
reading the next slot in the current frame. The
identification process continues until no collision
occurs. According to the hybrid frame size adjust-
ment strategy, the algorithm can achieve a better
and stable performance.
By combining tag cardinality estimation and
adaptive frame size calculation, the Energy-Aware
Frame Adjustment Strategy (EAFAS) based algo-
e traditional system
throughput is not an
appropriate metric to
evaluate the perfor-
mance of the RFID
anti-collision algorithm.
Moreover, such metric
does not consider the
energy consumption.
Unlike the previous
DFSA algorithms, the
proposed algorithm
calculates the frame
size by maximizing the
energy efficiency.

IEEE Wireless Communications • Accepted for Publication
5
rithm is proposed. The flowchart of the EAFAS is
described in Fig. 1.
evAluAtIon dIscussIon
This section evaluates the performance of the
EAFAS algorithm in metrics, system throughput
and energy efficiency, and compares it with state-
of-the-art methods including MAP [5], ILCM [8],
EACAEA [9], FuzzyQ [10], and SUBF-DFSA [11].
Simulation scenarios with a reader and a various
number of tags have been evaluated using MAT-
LAB, where the tags are uniformly distributed
in the reader vicinity so that all tags can receive
the reader’s command. In our simulations, the
tag number is chosen between 100 and 1000.
This article mainly focuses on the MAC layer,
whereas the physical layer effects are not consid-
ered [8–12, 14]. To reduce the randomness and
ensure the convergence, the simulation results
are averaged over 1000 iterations. The parame-
ters used in the MATLAB simulation are listed in
Table 2, which are aligned with the EPCglobal C1
Gen2 UHF RFID and commercial Impinj solution
specifications.
Figure 2a compares the system throughput
of various algorithms. The number of tags varies
between 5 and 100. The frame size is initialized
as 16. As can be observed from Fig. 2a, the pro-
posed EAFAS achieves more stable performance,
especially when the number of tags increases. Such
an improvement is due to the reasons that the pro-
posed estimation strategy can obtain the accurate
result and the proposed frame setting mechanism
can ease estimation error, which in turn reduc-
es the total number of slots. The performance of
MAP and ILCM are very close because their frame
size adjustment is based on the same estimation
derived from a full frame. In contrast, the EACAEA,
FuzzyQ and SUBF-DFSA adopt the PbP strategy
to adjust the frame size. Thus, they can provide
more stable performance than the previous two
algorithms. Figure 2b plots the system throughput
when the tag number ranges between 100 and
1000. The frame size is also initialized as 16. By
comparing both Figs. 2a and 2b, most algorithms
show discrepant performance. For example, the
average system throughput of SUBF-DFSA is lower
than MAP, ILCM and EACAEA when the number
of tags is between 5 and 100. As the number of
tags increases, the impact of the initial frame on sys-
tem throughput will be weakened. The SUBF-DFSA
algorithm is capable of interrupting inappropriate
frames through PbP observation for achieving a
considerable performance improvement. Since the
frame size can be adaptively adjusted to a different
tag cardinality, the EAFAS can always hold the best
system throughput compared to other algorithms.
To further compare the performance of the
EAFAS to other methods, a 95 percent confi-
dence interval of system throughput for various
algorithms is summarized in the Table 3. As can
be observed, the lower bound of the EAFAS is
even higher than the upper bound of MAP, ILCM,
EACAEA, and FuzzyQ at any number of simulation
runs. Therefore, we believe our proposed method
shows robust performance under different simula-
tion runs.
For the purpose of evaluating the energy
consumption of anti-collision algorithms, Fig. 3
shows the energy efficiency of various algo-
rithms. We can observe from Fig. 3a, the curves
of all algorithms fluctuate when the number of
tags is small. As the number of tags is above 50,
their performance becomes more stable; this is
especially true when the number is above 100
in Fig. 3b. When the number of tags is close to
the initial frame size, all algorithms can achieve
the highest performance. However, with a con-
tinuing increase of tag numbers, all algorithms
show deteriorating performance because of the
extra slots used to estimate the unread tags. It
is also noted that energy efficiency depends on
both identification time and estimation complexi-
ty; the proposed EAFAS is the only algorithm that
can maintain good performance in both through-
put and energy efficiency. Moreover, since our
FIGURE 1. The flowchart of the proposed algo-
rithm.
Reader broadcasts
Query/QueryAdj (F,Fsub)
Tag response
E=E+1
slot_index++
C=C+1
slot_index++
Identify the tag
S=S+1
slot_index++
E-3.2C/ threshold or
E-0.6C/-threshold
Fsub=slot_index
slot_index F
Slot_index==Fsub
effi1<effi2
Compute effi1and effi2
Is C==0?
Identification ends
Y
N
Y
N
Y
N
Y
N
Y
N
singleton
collision
empty
estimates tag cardinality
TABLE 2. The Parameters used in MATLAB simu-
lation.
Parameters Value Parameters Value
R->T modulation DSB-ASK T->R modulation FM0
R->T preamble
(ms) 312. 5 T->R preamble
(ms) 150
R->T frame-syns
(ms) 237.5 BLF (kHz) 40
Tari (ms) 25 Data rate (kb/s) 40
DR 8 M 1
Iteration times 1000 T1 (ms) 250
TRcal (ms) 200 T2 (ms) 100
RTcal (ms) 75 T3 (ms) 50
RN16 (ms) 550 EPC (ms) 3350
e SUBF-DFSA algo-
rithm is capable of
interrupting inappropri-
ate frames through PbP
observation for achiev-
ing a considerable
performance improve-
ment. Since the frame
size can be adaptively
adjusted to a different
tag cardinality, the
EAFAS can always
hold the best system
throughput compared
to other algorithms.

IEEE Wireless Communications • Accepted for Publication 6
FIGURE 2. Comparison of system throughput under various algorithms.
0 20 40 60 80 100
0.26
0.28
0.3
0.32
0.34
0.36
0.38
0.4
Number of tags
System throughput
MAP
ILCM
EACAEA
FuzzyQ
SUBF-DFSA
EAFAS
100 200 300 400 500 600 700 800 900 1000
0.3
0.31
0.32
0.33
0.34
0.35
0.36
Number of tags
System throughput
MAP
ILCM
EACAEA
FuzzyQ
SUBF-DFSA
EAFAS
TABLE 3. The 95 percent confidence interval of system throughput for various algorithms.
Algorithms
MAP ILCM EACAEA FuzzyQ SUBF-DFSA EAFAS
ST
100 0.336~0.345 0.314~0.330 0.335~0.339 0.309~0.315 0.344~0.352 0.350~0.356
200 0.335~0.342 0.319~0.333 0.334~0.339 0.309~0.317 0.345~0.352 0.349~0.355
300 0.336~0.344 0.319~0.332 0.334~0.339 0.309~0.317 0.344~0.351 0.349~0.355
400 0.336~0.344 0.319~0.332 0.334~0.339 0.309~0.315 0.344~0.351 0.350~0.356
500 0.336~0.344 0.319~0.332 0.334~0.338 0.309~0.315 0.345~0.351 0.349~0.355
600 0.336~0.344 0.319~0.331 0.335~0.339 0.308~0.317 0.345~0.351 0.350~0.355
700 0.336~0.343 0.319~0.331 0.335~0.339 0.310~0.317 0.345~0.351 0.350~0.355
800 0.336~0.344 0.319~0.332 0.334~0.338 0.310~0.317 0.345~0.351 0.349~0.355
900 0.336~0.344 0.319~0.332 0.335~0.338 0.309~0.316 0.344~0.351 0.349~0.355
1000 0.336~0.344 0.319~0.332 0.335~0.339 0.308~0.316 0.345~0.351 0.349~0.355
FIGURE 3. Comparison of energy efficiency under various algorithms.
0 20 40 60 80 100
0.45
0.46
0.47
0.48
0.49
0.5
0.51
Number of tags
Energy efficiency
MAP
ILCM
EACAEA
FuzzyQ
SUBF-DFSA
EAFAS
100 250 400 550 700 850 100
0
0.45
0.46
0.47
0.48
0.49
0.5
0.51
Number of tags
Energy efficiency
MAP
ILCM
EACAEA
FuzzyQ
SUBF-DFSA
EAFAS

IEEE Wireless Communications • Accepted for Publication
7
presented solution is based on the EPCglobal C1
Gen2 standard, it is compatible with most hard-
ware platforms, thus saving cost.
conclusIon
This tutorial has discussed energy efficient
RFID anti-collision algorithms regarding design
concept and analysis, especially on the perfor-
mance of the MAC layer. We have mainly ana-
lyzed and compared the performance of system
throughput and energy consumption of various
anti-collision algorithms. In our view, our pro-
posed EAFAS solution makes RFID capable of
adapting to energy-aware scenarios and meet-
ing future green IoT application requirements of
energy efficiency and high system throughput.
The acquired new insights on MAC performance
could also provide a precise guideline for the
efficient designs of practical and reliable RFID
communications systems. Hence, these results
will potentially have a broad impact across a
range of areas, including supply chain manage-
ment, inventory control, and asset tracking.
Acknowledgment
This work is supported by the National Natural
Science Foundation of China (No. 61802196); the
Natural Science Foundation of Jiangsu Province
(No. BK20180791); the Natural Science Foun-
dation of Jiangsu Higher Education Institutions
of China (No. 17KJB510036); and the Startup
Foundation for Introducing Talent of NUIST (No.
2243141701031). We would like to thank editor
Han-Chieh Chao and the anonymous reviewers
for their insights that improved the article signifi-
cantly.
references
[1] I. Bose and S. Yan, “The Green Potential of RFID Projects: A
Case-Based Analysis,” IT Professional, vol. 13, no. 1, 2011,
pp. 41–47.
[2] F. Shaikh, S. Zeadally, and E. Exposito, “Enabling Technolo-
gies for Green Internet of Things,” IEEE Systems J., vol. 11,
no. 2, 2017, pp. 983–94.
[3] R. Arshad et al., “Green IoT: An Investigation on Energy on
Energy Saving Practices for 2020 and Beyond,” IEEE Access,
vol. 5, 2017, pp. 15667–81.
[4] C. Zhu et al., “Green Internet of Things for Smart World,”
IEEE Access, vol. 3, 2015, pp. 2151–62.
[5] W. T. Chen, “An Accurate Tag Estimate Method for Improv-
ing the Performance of an RFID Anticollision Algorithm
Based on Dynamic Frame Length Aloha,” IEEE Trans. Auto-
mation Sci. and Eng., vol. 6, no. 1, 2009, pp. 9–15.
[6] B. Knerr et al., “Slot-Wise Maximum Likelihood Estimation of
the Tag Population Size in FSA Protocols,” IEEE Trans. Com-
mun., vol. 58, no. 2, 2010, pp. 578–85.
[7] X. Jia, Q. Feng, and L. Yu, “Stability Analysis of an Efficient
Anti-Collision Protocol for RFID Tag Identification,” IEEE
Trans. Commun., vol. 68, no. 6, 2012, pp. 2285–93.
[8] P. Solic, J. Radic, and N. Rozic, “Energy Efficient Tag Estima-
tion Method for Aloha-based RFID Systems,” IEEE Sensors J.,
vol. 14, no. 10, pp. 3637-3647, 2014.
[9] W. T. Chen, “Optimal Frame Length Analysis and an Effi-
cient Anticollision Algorithm with Early Adjustment of Frame
Length for RFID Systems,” IEEE Trans. Veh. Technol., vol. 65,
no. 5, 2016, pp. 3342–48.
[10] L. Arjona et al., “Fast Fuzzy Anticollision Protocol for the
RFID Standard EPC Gen-2,” Electron. Lett., vol. 52, no. 8,
2016, pp. 663–65.
[11] J. Su et al., “An Efficient Sub-Frame Based Tag Identification
Algorithm for UHF RFID Systems,” Proc. Intl. Conf. Com-
mun. (ICC 2016), 2016, pp. 1–6.
[12] J. Su et al., “An Effective Frame Breaking Policy for Dynam-
ic Framed Slotted Aloha in RFID,” IEEE Commun. Lett., vol.
20. no. 4, 2016, pp. 692–95.
[13] J. Shin, B. Jeon, and D. Yang, “Multiple RFID Tags Identifi-
cation with M-ary Query Tree Scheme,” IEEE Commun. Lett.,
vol. 17. no. 3, 2013, pp. 604–07.
[14] J. Su et al., “Idle-Slots Elimination Based Binary Splitting
Anti-Collision Algorithm for RFID,” IEEE Commun. Lett., vol.
20. no. 12, 2016, pp. 2394–97.
[15] C.-S. Lee, D.-H. Kim, and J.-D. Kim, “An Energy Efficient
Active RFID Protocol to Avoid Overhearing Problem,” IEEE
Sensor J., vol. 14, no. 1, 2014, pp. 15–24.
bIogrAphIes
Jian Su (sj890718@gmail.com) has been a lecturer in the School of
Computer and Software at the Nanjing University of Information
Science and Technology since 2017. He received his Ph.D. with
distinction in communication and information systems at the Uni-
versity of Electronic Science and Technology of China in 2016. He
holds a B.S. in electronic and information engineering from Hankou
University and an M.S. in electronic circuits and systems from Cen-
tral China Normal University. His current research interests cover
Internet of Things, RFID, and wireless sensors networking.
Zhengguo Sheng (Z.Sheng@sussex.ac.uk) has been a senior
lecturer in the Department of Engineering and Design at the
University of Sussex since 2015. He received his Ph.D. and M.S.
with distinction at Imperial College London in 2011 and 2007,
respectively, and his B.Sc. from the University of Electronic Sci-
ence and Technology of China (UESTC) in 2006. His current
research interests cover the Internet of Things (IoT), connected
vehicles, and cloud/edge computing.
Victor c. M . L eung (vleung@ece.ubc.ca) is an emeritus pro-
fessor of electrical and computer engineering at the University
of British Columbia. He has co-authored approximately 1200
technical papers in the areas of wireless networks and mobile
systems, in addition to 43 book chapters and 14 book titles.
He is a Fellow of the Royal Society of Canada, the Canadian
Academy of Engineering, Institute of Electrical and Electronics
Engineers (IEEE), and the Engineering Institute of Canada.
Yongrui c hen (chenyr@ucas.ac.cn) has been an associate pro-
fessor in the Department of Electronical, Electrical and Commu-
nication Engineering at the University of Chinese Academy of
Sciences (UCAS) since 2014. He received his Ph.D. at UCAS in
2011, his M.S. at Tsinghua University in 2007, and his B.Sc. from
Yanshan University in 2001. His current research interests cover
the Internet of Things (IoT) and heterogeneous wireless networks.
e acquired new
insights on MAC
performance could
also provide a pre-
cise guideline for the
efficient designs of
practical and reliable
RFID communications
systems. Hence, these
results will potentially
have a broad impact
across a range of areas,
including supply chain
management, invento-
ry control, and asset
tracking.