Error Resilient Estimation and Adaptive Binary Selection ... - IEEE Xplore
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. IEEE TRANSACTIONS ON MOBILE COMPUTING
1
Error Resilient Estimation and Adaptive Binary Selection for Fast and Reliable Identification of RFID Tags in Error-prone Channel Jongho Park and Tae-Jin Lee, Member, IEEE
Abstract In RFID systems, far field passive tags send information using back scattering. The signal level is typically very small, so channel error during transmission may occur frequently. Due to channel error, performance of RFID tag identification under error-prone channel is degraded compared to that under error-free channel. In this paper, we propose a novel error resilient estimation and adaptive binary selection to overcome the problem of channel errors. Our proposed error resilient estimation algorithm can estimate the number of tags and the channel state accurately regardless of frame errors. And our proposed adaptive binary selection reduces the idle slots caused by frame errors. Performance analysis and simulation results show that the proposed algorithm consumes up to 20% less time slots than the binary tree protocol and dynamic framed slotted ALOHA (DFSA) in various packet error rate (PER) conditions.
Index Terms anti-collision, channel error, collision resolution, RFID, tag estimation.
I. I NTRODUCTION In Radio Frequency IDentification (RFID) [1] systems, passive tags transmit their IDs using back scattering in response to the reader’s request [13]. The back scattering signals of tags are significantly attenuated by distance [19], [20]. So the signal strength of tags is typically very
The authors are with the School of Information and Communication Engineering, Sungkyunkwan University, Suwon 440-746, Korea (e-mail:[email protected]) March 30, 2011
Digital Object Indentifier 10.1109/TMC.2011.112
DRAFT
1536-1233/11/$26.00 © 2011 IEEE
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. IEEE TRANSACTIONS ON MOBILE COMPUTING
2
small, and frame errors may occur due to noise or interference. In addition, the signal power of RFID readers is much bigger than the response signal power of tags, so the signals of readers may interfere with one another and with those of tags [11]. In fact, the interference from other readers may be significant. In [21], it is stated that the interference range of a reader may be larger than the identification range of a reader. So a reader may interfere with other readers although the reader is not within the identification range of other readers. And readers’ interference may cause high frame errors. The performance of frame errors is reflected in the form of packet error rate (PER) which is determined by the signal to noise ratio (SNR) [12]. RFID systems require relatively high SNR (low PER) for reliable operation [17] or a robust mechanism to operate under low SNR. If RFID tags transmit their IDs simultaneously, an RFID reader can not identify the IDs due to collision. So the anti-collision protocol is significantly important for fast tag identification. There are two types of anti-collision protocols, ALOHA-based [2] and tree-based [3]. ALOHA-based protocols can be classified into ALOHA, slotted ALOHA and framed-slotted ALOHA (FSA). Among ALOHA-based protocols, the performance of FSA is superior to those of others because it can control the probability of transmission of tags in a frame. However, RFID readers need to estimate the number of tags to decide an appropriate frame size since the performance of FSA is known to be optimal when the frame size equals the number of tags [7]. Although enhanced tag estimation-based dynamic framed-slotted ALOHA (DFSA) is proposed in [7], [8], [6], and [22] if the number of tags is very large compared to the initial frame size, a reader can not estimate the number of tags (frame size) with accuracy due to many collisions. Thus careful decision on an initial frame size is very important. Most existing DFSA algorithms, however, do not provide an appropriate basis for decision on an initial frame size. In [4], a robust estimation mechanism is proposed, in which the number of responding tags in a frame is reduced by a certain ratio until an appropriate probability of collision is achieved. Then the number of tags (frame size) is estimated from the collision probability. In [23], the estimation algorithm using Tree Slotted ALOHA (TSA), i.e., Dynamic TSA (DyTSA) is proposed. DyTSA performs tag estimation at each child frame in a collision tree, which is different from the original TSA. Although the initial frame size is small compared to the number of tags, DyTSA can estimate the number of tags accurately at a child frame since the number of tags is decreased rapidly at each branch of the tree. In [24], Binary Splitting Tree Slotted Aloha (BSTSA) is March 30, 2011
DRAFT
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. IEEE TRANSACTIONS ON MOBILE COMPUTING
3
proposed. After the identification of a tag in a branch of a binary splitting tree, BSTSA estimates the number of tags in a branch accurately without concern on the initial frame size. Tag (frame) estimation and identification may have problems under unreliable channel conditions. Most tag estimation methods are based on the probability of collision which is very sensitive to error-prone channel. An RFID reader can not distinguish a collision with a frame error. Thus the reader may consider a channel error as a collision. Hence when there are frame errors, the measured probability of collision becomes high. As a result, the estimated number of tags based on the probability of collision is larger than the actual number of tags under error-free channel. Then the estimated frame size also becomes larger than the actually required frame size, and the performance of tag identification is degraded because there are many unnecessary idle slots in the over-estimated frame. Tree-based protocols solve the tag collision problem by binary trees, in which collided tags are separated into two groups randomly. Binary tree protocols may require a large number of time slots because tags experience many collisions at the early level of a binary collision tree. In [9], [14], an algorithm to decide the transmission order of tags by using the binary tree protocol and to use the order at the next identification process is proposed. M-ary tree protocols based on breadth-first tree search and tag estimation mechanism is proposed in [15], [16]. Framed-Slotted ALOHA with robust Estimation and Binary selection (EB-FSA) [4] operates FSA with only one identification frame based on robust estimation of the frame size from small pilot frames, and the collided tags resolve collisions by the binary tree algorithm. Framed Slotted ALOHA with Pilot Frame and Binary Selection (FSAFB) [5] partitions tags into several groups and operates a single frame FSA with binary tree collision resolution. Since there are 2 or 3 tags in most collided slots, applying the binary tree algorithm is efficient under error-free channel. When a frame error occurs, in the binary tree protocol, it is regarded as a collision and the corresponding tag, i.e., the tag that fails by the frame error, randomly selects 0 or 1 according to the collision resolution of the binary tree algorithm. So the failure of identification caused by a frame error induces an additional idle slot resulting in performance deterioration. Thus existing RFID anti-collision protocols require special attention for reliable operation under practical channel error conditions. In this paper, we propose a novel error-resilient anti-collision algorithm for RFID tag identification under unreliable channels. The proposed Error Resilient Estimation (ERE) and Adaptive March 30, 2011
DRAFT
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. IEEE TRANSACTIONS ON MOBILE COMPUTING
4
Binary Selection (ABS) algorithm reduces the identification time by reducing unnecessary idle and collision slots under error-prone channel. The ERE mechanism estimates the number of tags and the channel condition (PER) accurately under error-prone channel regardless of an initial frame size with relatively low computational load. And the ABS algorithm reduces the unnecessary idle slots caused by frame errors using the estimated number of tags and PER in the ERE process. This paper is organized as follows. In Section II, we explain the proposed error resilient estimation and adaptive binary selection algorithm. In Section III, mathematical analysis of DFSA, binary tree, EB-FSA, and ERE-ABS algorithms under error-prone channels is presented. Performance evaluation and comparison via both analysis and simulations is conducted in Section IV. We conclude in Section V. II. ERE-ABS A LGORITHM We propose an algorithm for accurate estimation of channel status and fast identification to enhance performance under unreliable channel condition. A. Error Resilient Estimation (ERE) When the frame size is equal to the number of tags, identification performance of FSA is known to be optimal. Thus an accurate estimation of the number of tags is very important in tag estimation-based DFSA. If the estimated number of tags in a frame is very small compared with the actual number of tags, the next frame size becomes inaccurate. So an initial frame size for initial estimation of the number of tags has to be decided with care. DFSA in [8] estimates the number of tags by using the measured probability of collision. Tag estimation method in [22] uses the probability of collision and idle probability. In [6], the number of tags is estimated by the numbers of idle, success and collision slots. The algorithm in [6] computes the estimated number of tags by minimizing the vector distance between the measured number of idle/success/collision slots and the estimated number of idle/success/collision slots. In [23], DyTSA estimates the number of tags at each child frame of a collision tree noting that the number of tags in a child frame in smaller than that in a parent frame. However, the (measured) probability of collision or success becomes inaccurate when there are channel errors since a reader regards the failure of identification due to a channel error as March 30, 2011
DRAFT
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. IEEE TRANSACTIONS ON MOBILE COMPUTING
5
a collision and the reader can not distinguish a collision with a channel error. So, under errorprone channel, the measured probability of collision is bigger and the measured probability of success is smaller than the actual collision/success probabilities. In this sense, if a reader estimates the number of tags from the measured inaccurate probabilities, it may be over-estimated compared with the actual number of tags. Thus existing estimation methods under channel error causes performance degradation of DFSA. So we propose a new error resilient estimation (ERE) mechanism for the number of tags. ERE uses the probability of idle slots instead of that of collision or success slots, e.g., Zero Estimator (ZE) [22], since the probability of idle slots is not affected by channel errors. Thus the estimation based on the probability of idle slots is more accurate in spite of channel errors. Let’s assume that we have n tags and the size of a frame is L. The probability of collision Pcoll is then Pcoll = 1 − Psucc − Pidle , Psucc
1 =n· L
Pidle =
n−1 1 1− , L 1 1− L
(1) (2)
n ,
(3)
where Pidle and Psucc are the probabilities of idle and success slots, respectively. If the number of tags becomes larger, the measured probability of idle slots Pˆidle becomes quickly closer to 0. So if Pˆidle is used for estimation, the number of tags that can be estimated might be inaccurate if Pˆidle becomes small. And when the measured probability of idle slots Pˆidle becomes small, the estimated number of tags n may be inaccurate, since small measurement error of Pˆidle may lead to large estimation error for the estimated number of tags (see Fig. 1), i.e., when the probability of idle is small, variation of the number of estimated tags may be very large for small variation of measurement error of the probability of idle slots. In order to prevent this problem, in our ERE, we use the threshold Pidle th to decide if the measured probability of idle slots Pˆidle is appropriate. The ERE uses a fixed frame size Lest , and if the probability of idle slots Pˆidle is smaller than the threshold Pidle th (i.e., if the number of responding tags is large), a reader attempts to decrease the number of tags to respond to the reader by a factor of fd using a bit mask in a query frame to tags. Lest , Pidle th and fd are March 30, 2011
DRAFT
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. IEEE TRANSACTIONS ON MOBILE COMPUTING
6 1 P
coll
P
0.9
idle
P
succ
0.8 0.7
Probability
0.6 0.5 0.4 0.3 0.2 0.1 0
Fig. 1.
0
1
2
3
4
5 n/L
6
7
8
9
10
Characteristics of Psucc , Pidle and Pcoll with the number of tags n and the frame size L.
system parameters. We decide appropriate values by extensive simulation. The reader repeats this process until the probability of idle slots exceeds the threshold and then estimates the number of tags n. The tags identified in ERE are not muted during the ERE phase for accuracy of tag estimation. So ERE can estimate the number of tags accurately under unreliable channel regardless of the initial frame size and the number of tags. Furthermore estimation of n using Pˆidle is very simple to compute. Eq. (3) can be expressed by the following equation. n=
log(Pidle ) . log((L − 1)/L)
(4)
So in practice the estimated number of tags nest for n is found as ∗
nest = nest,j ∗ × (fd )j ,
(5)
where j ∗ is the number of repeated estimation frames until the probability of idle slots exceeds the threshold Pidle th , and nest,j ∗ is the estimated number of tags in the j ∗ th estimation frame. nest,j ∗ =
log(Pˆidle,j ∗ ) . log((Lest − 1)/Lest )
(6)
Then the estimated frame size L = nest for the identification process. Note that unlike ours, complicated numerical methods are required to estimate n from Pˆcoll or Pˆsucc . And, for a given Pˆsucc , two estimated numbers for n are mapped, so Pˆsucc can not be used alone to uniquely estimate the number of tags (see Fig. 1). Fig. 2 shows the structure of the estimation process. March 30, 2011
DRAFT
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. IEEE TRANSACTIONS ON MOBILE COMPUTING
7 Error Resilient Estimation (ERE) 0
1
……
Pˆ idle>Pidle_th Adaptive Binary Selection (ABS)
j*
L = nest
Lest
S Success F I S I F
……
F
Measure Pˆ idle,j*, Pˆ succ,j*, Pˆ coll,j* Compute nest and PER
Fig. 2.
I Idle
S I Failure (collision or channel error )
The structure of estimation process of our proposed algorithm
A reader can now take advantage of the estimated nest,j ∗ from Pˆidle,j ∗ to estimate the channel characteristics, PER. First, the probability of success under error-free channel P˜succ,j ∗ can be computed from the estimated nest,j ∗ in Eq. (2). 1 P˜succ,j ∗ = nest,j ∗ Lest
1 1− Lest
nest,j∗ −1
.
(7)
The estimation probability of no collision but channel error which should have been success if there had not been channel error, is P˜j ∗ ((no collision) and (channel error)) = P˜succ,j ∗ − Pˆsucc,j ∗ ,
(8)
where Pˆsucc,j ∗ is the measured (actual) probability of success and P˜succ,j ∗ is the estimated probability that a transmission is successful if there is no channel error. Then PER can be defined as the probability of failure of identification due to a channel error when a tag transmits an ID in a slot without collision. Thus PER can be estimated by ˜ P ER= P˜j ∗ (channel error|success if there is no channel error) P˜ ((no collision) and (channel error)) = P˜succ,j ∗ P˜succ,j ∗ − Pˆsucc,j ∗ . (9) P˜succ,j ∗ Fig. 3 shows the relationship among Pˆsucc,j ∗ , P˜succ,j ∗ and Pˆcoll,j ∗ . The pseudo code to compute =
the frame error rate PER is illustrated in Algorithm 1. B. Adaptive Binary Selection (ABS) In the binary tree collision resolution protocol, when a channel error occurs, a reader and a tag regard the error as a collision and the corresponding tag randomly selects 0 or 1 to resolve March 30, 2011
DRAFT
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. IEEE TRANSACTIONS ON MOBILE COMPUTING
8
the collision resulting in two time slots. So the failure of identification caused by a channel error incurs an additional idle slot. This is the main reason of performance degradation of the binary tree protocol under error-prone channel condition. In this paper, we propose adaptive binary selection mechanism which determines whether a collision is likely to be real or virtual (by channel error) based on the estimated PER. In the identification process, a reader and tags operate FSA with the estimated frame size L (L = nest ) in ERE. When a collision occurs our proposed ABS algorithm statistically determines if a collision is actual collision or it is virtual collision by channel error. ABS determines direct retransmission for a channel error or binary selection for a collision based on the estimated PER in the ERE stage or the probability of collision. Each tag selects a random counter value as in framed-slotted ALOHA after a reader decides the optimal frame size L determined from the estimated nest in the ERE stage. Tags decrease their counter values at each time slot. When a counter becomes 0, the tag transmits its ID. If a collision is observed, the reader operates the ABS algorithm. Note that a collision may be either real collision or channel error. Then a reader transmits a control message which commands to do retransmission or binary selection for the tag(s) at the start of the next slot. The reader maintains two counters, Nc and Ne (initially both are 0). If the reader determines that a collision is due to channel error, it decides to do retransmission and increases Ne by 1. The counter Ne represents the number of consecutive retransmissions. If the reader determines that a collision is real, it decides to do binary selection, and it increases Nc by 1, and resets Ne to 0. If a tag transmits its ID successfully, the reader decreases Nc by 1. The counter Nc denotes the current level of a binary collision tree.
Collision
Success if there is no channel error No collision Success and Chanel error
Pˆ coll,j*
Idle
ˆ succ,j* P ~
Psucc,j* Fig. 3.
Relationship among Pˆsucc,j ∗ , P˜succ,j ∗ and Pˆcoll,j ∗ .
March 30, 2011
DRAFT
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. IEEE TRANSACTIONS ON MOBILE COMPUTING
9
Algorithm 1 Error Resilient Estimation (ERE) ˜ 1: procedure E STIMATION OF FRAME SIZE L(=nest ) AND P ER FSA operation frame with a frame of size Lest Measure(Pˆcoll , Pˆsucc , Pˆidle )
2: 3:
count ← 0 while (Pˆidle ≥ Pidle th ) do
4: 5:
counter for iteration
Transmit a query frame with a bit mask to reduce the number of tags by a factor of
6:
fd 8:
FSA operation frame with a frame of size Lest Measure(Pˆcoll , Pˆsucc, Pˆidle )
9:
count + +
7:
10:
end while
11:
j ∗ ← count, Pˆsucc,j ∗ ← Pˆsucc
12:
Compute nest,j ∗
by Eq. (6)
13:
Compute nest
by Eq. (5)
14:
L ← nest
15: 16: 17:
Compute P˜succ,j ∗ ˜ Compute P ER
by Eq. (7) by Eq. (9)
end procedure
The reader determines retransmission or binary selection in an attempt to increase the efficiency at each decision point (i.e., for each observed collision). The efficiency is defined as the ratio between the number of successfully transmitted tags and the total number of time slots. The efficiency of retransmission ηr (Nc , Ne ) if a collision is estimated as channel error, and that of binary selection ηb (Nc ) if a collision is estimated as actual collision are defined as follows. ηb (Nc ) = Eb (Nc )
(10)
˜ Ne −1 , ηr (Nc , Ne ) = Er (Nc ) · P ER
(11)
where ηb (Nc ) is the efficiency at the (Nc + 1)th level and ηr (Nc , Ne ) is the efficiency at the (Ne + 1)th retransmission and at the Nc th level. The efficiency ηr (Nc , Ne ) is decreased by the March 30, 2011
DRAFT
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. IEEE TRANSACTIONS ON MOBILE COMPUTING
10
number of retransmissions because the number of tags which occupy a slot is decreased by the ˜ rate of P ER. And Eb (i) and Er (i) are defined as follows. Eb (i) =
∞
Pf ail (k) · Eb (k, i),
(12)
k=1
Eb (k, i) =
e ˜ (1 − P ER) · SBIN (k, i + 1) , e ˜ · S e (k, i)) 2 · (CBIN (k, i) + P ER BIN
Er (i) =
∞
Pf ail (k) · Er (k, i),
(13)
(14)
k=1
˜ ˜ · S e (k, i) (1 − P ER) · P ER BIN , e ˜ · S e (k, i)) (CBIN (k, i) + P ER BIN ⎧ ˜ P˜succ P ER· ⎨ k=1 ˜ P˜succ , P˜coll +P ER· Pf ail (k) = nest 1 k 1 nest −k ( ) (1− ) L ⎩ ( k )L , k > 1, ˜ P˜ P˜ +P ER·
Er (k, i) =
coll
(15)
(16)
succ
where Eb (k, i) and Er (k, i) denote the efficiency of binary selection and retransmission, respectively, at the ith binary collision tree level when there are k tags in the binary tree. P˜succ and P˜coll are the computed probabilities of success and collision, respectively, using the estimated number of tags nest . The numerators of Eq. (13) and Eq. (15) are the number of successful e slots because SBIN (k, i) is the number of successful slots in level i when there are k tags in
the binary tree under error-prone channel. If the reader chooses binary selection, the level of the e ˜ tree is increased. So, the number of successful slots is (1 − P ER)S BIN (k, i + 1). If the reader chooses retransmission, the level of the tree does not change. Thus the number of successful ˜ P ER ˜ · S e (k, i). And the denominators of Eq. (13) and Eq. (15) represent slots is (1 − P ER) BIN the total number of used slots. Fig. 4 shows an example for the equations. In the binary selection case, the number of failed (collision or channel error) slots at the ith level is increased by a factor of two at the (i+1)th level, and the number of success slots is the number of real success slots without errors in the binary tree. In the retransmission case, the number of failed (collision or channel error) slots at the ith level is equal to the number of slots used for retransmission, and the number of success slots is the number of no error slots among the failed slots by channel errors. Then Eb (i) and Er (i) denote the efficiency of binary e (k, i) is selection and retransmission, respectively, at the ith binary collision tree level. CBIN
March 30, 2011
DRAFT
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. IEEE TRANSACTIONS ON MOBILE COMPUTING
11 Binary Selection
~ PER· SeBIN (k,i)
C
Se BIN (k,i) Level i
Level i+1
E
E
E
I
S
e
S
BIN
(k,i)
C
I
C
~ 2·(C BIN (k,i)+PER· SeBIN (k,i)) e
I
S
~ (1-PER)·Se BIN (k,i+1)
Retransmission ~ PER· SeBIN (k,i) S E
E
E
S
e BIN
Ce BIN(k,i)
(k,i) S
C
I
Level i C
~ Ce BIN (k,i)+PER·Se BIN (k,i) S
Fig. 4.
Success
I
idle
~ ~ (1-PER)·PER· SeBIN (k,i) C
Real Collision
E
Virtual collision (channel error )
An example of efficiency computation Eqs. (13) and (15).
the number of collided and channel error time slots at depth i when the number of tags is k e at depth i of a binary tree, and SBIN (k, i) is the number of actual successful slots when the e e number of tags is k at depth i of a binary tree. CBIN (k, i) and SBIN (k, i) are defined in Eqs. ˜ in Section III. And Pf ail (k) is the probability of (26)-(27) with n = k and P ER = P ER
identification failure when k tags are in a slot. When k = 1, the identification failure occurs by channel error only. When k > 1, the identification failure occurs due to collision of k tags. So the probability of identification failure can be computed by binomial probability. When k = 1, the identification failure occurs due to channel error. The denominator of Eq. (16) is the total probability of identification failure. If the estimated number of tags nest is accurate and L=nest , Pf ail (k) does not change because the average number of tags in a slot does not change by k. Thus we can obtain ηb (Nc ) and ηr (Nc , Ne ) if an identification fails (collision is observed), and compare them to decide which one yields better efficiency. If ηb (Nc ) is larger than ηr (Nc , Nb ) the reader determines binary selection, and if ηb (Nc ) is smaller than ηr (Nc , Nb ) the reader determines retransmission. If the tags receive the binary selection command, the collided tags randomly select their counter values from 0 or 1, and others increase their counter values by 1. And March 30, 2011
DRAFT
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. IEEE TRANSACTIONS ON MOBILE COMPUTING
12
if the tags receive the retransmission command, the corresponding tags retransmit their IDs immediately without any change of counter values. The pseudo code of the ABS algorithm is given in Algorithm 2. Fig. 5 shows an example and time diagram of our proposed ABS algorithm. In Fig. 5, three tags 1, 2, and 3 transmit simultaneously at a slot in the identification process. The reader compares efficiency of binary selection and that of retransmission and selects binary selection (in this example, the efficiency of binary selection (ηb ) is larger than the efficiency of retransmission (ηr )). And all three tags select 0 and collide again and the reader chooses binary selection again. One of the tags (tag 2) now selects 0 and the others (tag 1 and tag 3) select 1. Tag 2 is now successful and the others (tag 1 and tag 3) collide again. The reader decides binary selection, and one of the tags (tag 1) selects 0 and the other (tag 3) selects 1. Tag 1 which selects 0 occupies a slot alone, but its transmission fails twice due to channel error. The reader decides retransmission for both of failures based on the efficiency (in this example, the efficiency of binary selection (ηb ) is smaller than the efficiency of retransmission (ηr )). Then ηb > ηr and the reader decides binary selection and tag 1 now succeeds. Now tag 3 occupies a slot alone, but its transmission fails due to channel error. The reader then decides retransmission from efficiency and tag 3 finally succeeds. III. P ERFORMANCE A NALYSIS We analyze the average total number of slots used in the identification of DFSA, binary tree, EB-FSA and ERE-ABS under channel error condition. We define P ER as the probability that a reader can not identify a tag’s ID due to channel error. P ER = 1 − (1 − BER)Nbit ,
(17)
where Nbit is the number of bits in a frame with RFID tag ID, and BER is the bit error rate. A. DFSA We assume that DFSA estimates the number of tags n using the probability of collision. The estimation function is defined as fest (Pf ail , L), which is a function of the probability of failure by collision or channel error and the size of a frame. Let ni , Li , Psucc,i and Pcoll,i be the number
March 30, 2011
DRAFT
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. IEEE TRANSACTIONS ON MOBILE COMPUTING
13
Algorithm 2 Adaptive Binary Selection (ABS) 1: procedure TAG I DENTIFICATION ˜ in ERE algorithm initialize the efficiency with L and P ER
2:
INIT
3:
Nc ← 0, Ne ← 0
4:
while there are unidentified tags do
5:
Tag identification process
6:
if an identification fails then
collision observed
if ηr (Nc , Ne ) > ηb (Nc ) then
7:
Reader transmits the retransmission command, by which the tags retransmit
8:
IDs directly Ne + +
9:
else
10:
Reader transmits the binary selection command, by which the tags select 0
11:
or 1 randomly Nc + +, Ne = 0
12:
end if
13:
else
14: 15:
Success of tag identification
16:
Nc − −
17:
end if
18:
end while
19:
end procedure
20:
procedure I NIT
success observed
21:
Compute Pf ail
by Eq. (16)
22:
Compute Eb
by Eq. (12)
23:
Compute Er
by Eq. (14)
24:
Compute ηb
by Eq. (10)
25:
Compute ηr
by Eq. (11)
26:
end procedure
March 30, 2011
DRAFT
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. IEEE TRANSACTIONS ON MOBILE COMPUTING
14 Tag 1, 2, 3
F Tag 1, 2, 3
F
No tags
r(Nc,Ne) 1, i = 0.
(43)
where the number of retransmissions at level i is k ∗ (n, i) = arg max{(Eb (n, i) − Er (n, i)P ERk−1) < 0}. k∈N
(44)
Because the reader chooses retransmission until the efficiency of binary selection becomes larger than that of retransmission. The number of actual success time slots at depth i excluding the success time slots at retransmission is found as ⎧ ⎪ (1 − P ER) · (S(n, 2i ) ⎪ ⎪ ⎪ ⎪ ⎨ −S n > 1, i > 0 ABS c (n, i − 1)), SABS (n, i) = ⎪ ⎪ 0, n > 1, i = 0 ⎪ ⎪ ⎪ ⎩ (1 − P ER), n = 1, i = 0.
(45)
Then the number of actual collided time slots and the channel error time slots without retransmission slots at depth i is CABS (n, i) = C(n, 2i ) + P ERk
March 30, 2011
∗ (n,i)
·
SABS (n, i) . (1 − P ER)
(46)
DRAFT
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. IEEE TRANSACTIONS ON MOBILE COMPUTING
19
So the total number of collided time slots in the binary tree with ABS under channel error becomes CABS (n) =
∞
CABS (n, i).
(47)
i=0
The number of retransmission time slots in the ABS with n tags is R(n) =
∞
R(n, i),
(48)
i=0
where the number of retransmission time slots in the binary tree with ABS at level i for n tags ⎧ P ER i ⎪ ⎪ C(n, 2 ) + S (n, i) · ABS ⎪ (1−P ER) ⎪ ⎪
∗ (n,i)−1 ⎪ k ⎪ ⎪ {S(n, 2i ) − SABS c (n, i − 1)} ⎪ ⎨ − k=0 R(n, i) = ·(1 − P ER) · P ERk , k ∗ (n, i) > 1 ⎪ ⎪ ⎪ ⎪ 0, k ∗ (n, i) = 0 ⎪ ⎪ ⎪ ⎪ ⎪ ⎩ C(n, 2i ) + SABS (n, i) · P ER , k ∗ (n, i) = 1. (1−P ER)
is
(49)
The number of retransmission time slots is the same as the number of failed slots at the previous step. So the number of first retransmission slots at level i is the number of failed slots (collision or channel error) at level i of the binary tree and the number of the next retransmission slots is the number of failed slots at the previous retransmission. Therefore total used time slots for identification under channel error in the proposed ERE-ABS algorithm is found from Eqs. (41) and (42). TERE−ABS = TERE (n) + TABS (n).
(50)
Fig. 6 shows an example for Eq. (44), Eq. (45) and Eq. (48). In our proposed ERE-ABS algorithm, the level of a binary tree includes the retransmitted slots. No matter how many retransmissions there are, the level of a binary tree does not change. In this example the number of retransmissions k ∗ (n, i) at the ith level, is 2. SABS (n, i) is the number of success slots at the ith level including retransmitted slots, which is 3 in the example. The number of retransmission slots R(n, i) at the ith level is 5. IV. S IMULATION R ESULTS We evaluate the performance of DFSA, binary tree, EB-FSA and the proposed ERE-ABS algorithm in channel error condition under varying number of tags and PER via simulations March 30, 2011
DRAFT
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. IEEE TRANSACTIONS ON MOBILE COMPUTING
20
C
C
C
C
E
S
E
R(n,i) K*(n,i)
S
C
E
C
S
ith level
SABS(n,i)
S S
Fig. 6.
S
Success
C
Real Collision
E
Channel Error
An example for Eq. (44), Eq. (45) and Eq. (48).
and compare it with the analysis in section III. In DFSA, an initial frame size is set to the minimum frame size that can estimate n using the tag estimation method in [8] (L=0.2×n). In the EB-FSA algorithm, we set Lest =128, Pcoll th =0.7 and fd =4, and the typical value of i∗ (n) is 3 when n=1000 and P ER=0.7. In our ERE-ABS algorithm, we set Lest =128, Pidle th =0.2 and fd =4. We assume that the tag estimation is accurate in the analysis of DFSA, EB-FSA with ERE and ERE-ABS algorithm, and that the PERs of all tags are the same for simplicity although the PERs of tags may be different in practice. Since our ERE-ABS algorithm depends on the average PER and is designed to estimate the average PER of tags by Eq. 9 not to investigate and estimate the individual PERs, the assumption is not entirely unreasonable. So our ERE-ABS can be applied to the tags with different PERs. Fig. 7 and Table I show that the simulation results are closely matched with the analysis in Section III. We have shown 99% confidence interval with 10000 simulation iterations. In Fig. 7 and Table I, ERE-ABS uses the smallest number of time slots for all cases of PER. In low PER condition, performance enhancement of ERE-ABS is rather small. The proposed ABS algorithm decides retransmission or binary selection based on the efficiency, but when the PER is low, the efficiency of retransmission may be lower than binary selection. Because most decision is expected to be binary selection, the performance of ABS is similar to that of EB-FSA. However in high PER condition, the proposed ABS algorithm can reduce the unnecessary idle slots. So the performance of ERE-ABS is better than other existing protocols. March 30, 2011
DRAFT
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. IEEE TRANSACTIONS ON MOBILE COMPUTING
21 4
x 10 2 1.8
Total Number of Timeslots
1.6 1.4 1.2
ERE−ABS, PER=0.5 (analysis) ERE−ABS, PER=0.7 (analysis) ERE−ABS, PER=0.9 (analysis) EB−FSA with ERE, PER=0.5 (analysis) EB−FSA with ERE, PER=0.7 (analysis) EB−FSA with ERE, PER=0.9 (analysis) ERE−ABS, PER=0.5 (simulation) ERE−ABS, PER=0.7 (simulation) ERE−ABS, PER=0.9 (simulation) EB−FSA with ERE, PER=0.5 (simulation) EB−FSA with ERE, PER=0.7 (simulation) EB−FSA with ERE, PER=0.9 (simulation)
1 0.8 0.6 0.4 0.2 0 100
Fig. 7.
200
300
400
500 600 Number of Tags
700
800
900
1000
Total number of time slots used to identify tags with varying number of tags and the PER. TABLE I S LOPES OF THE CURVES IN F IG . 7
Algorithm
Slope
ERE-ABS, PER=0.5
4.3317
ERE-ABS, PER=0.7
6.4220
ERE-ABS, PER=0.9
14.5359
EB-FSA with ERE, PER=0.5
4.4377
EB-FSA with ERE, PER=0.7
7.1043
EB-FSA with ERE, PER=0.9
20.4370
In Fig. 8 and Table II, we compare the proposed ERE-ABS algorithm with EB-FSA, the binary tree protocol, DFSA and DyTSA [23] algorithm under various PERs. Our proposed ERE-ABS algorithm is superior to other existing algorithms under error-prone channel. In channel error conditions, tags tend to spend relatively long time in a binary tree. So performance of EB-FSA without ERE is very close to that of binary tree algorithm under error-prone channel. The total number of slots used by DyTSA is smaller than that of DFSA but larger than those of the others. DyTSA makes small child frames for each collided slot and estimates the number of slots for each child frame. Over-estimation caused by channel error is relatively small for small frames.
March 30, 2011
DRAFT
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. IEEE TRANSACTIONS ON MOBILE COMPUTING
22 4
x 10
Total Number of Timeslots
2
1.5
ERE−ABS, PER=0.5 ERE−ABS, PER=0.7 ERE−ABS, PER=0.9 EB−FSA, PER=0.5 EB−FSA, PER=0.7 EB−FSA, PER=0.9 Bin. tree, PER=0.5 Bin. tree, PER=0.7 Bin. tree, PER=0.9 DFSA, PER=0.5 DFSA, PER=0.7 DFSA, PER=0.9 DyTSA, PER=0.5 DyTSA, PER=0.7 DyTSA, PER=0.9
1
0.5
0 100
Fig. 8.
200
300
400
500 600 Number of Tags
700
800
900
1000
Total number of time slots used to identify tags with varying number of tags and the PER.
However the top level of the tree is operated with DFSA, which causes performance degradation under error prone channel. Fig. 8 and Table II show that the performance of the binary tree protocol is better than that of DFSA in channel error conditions. And ERE-ABS achieves about 21.5% (at PER=0.5) performance gain over DFSA protocol. The higher the PER is, the larger the gap of the performance improvement becomes because the proposed ABS algorithm reduces the number of unnecessary idle slots. In Fig. 9, we compare the estimation performance of the proposed ERE algorithm with that of EB-FSA and DFSA under various PERs. Our proposed ERE algorithm can estimate the number of tags accurately in spite of high PER. But EB-FSA and DFSA tend to over-estimate the number of tags. In addition, DFSA is assumed to use the optimal initial frame size to estimate the number of tags. In practice, however, the estimation accuracy of DFSA will not be guaranteed. The estimation accuracy of EB-FSA becomes worse after n=300. This is related with the deceasing factor fd and the estimation frame size Lest since the number of estimation frames varies according to fd and Lest . Depending on fd and Lest , i∗ (n) is changed. If i∗ (n) becomes larger, the estimation error also becomes larger because the estimated number of tags is determined by multiplying Lest by i∗ (n). In Fig. 10, we compare total number of time slots used for identification of the proposed
March 30, 2011
DRAFT
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. IEEE TRANSACTIONS ON MOBILE COMPUTING
23
TABLE II S LOPES OF THE CURVES IN F IG . 8
Algorithm
Slope
ERE-ABS, PER=0.5
4.3317
ERE-ABS, PER=0.7
6.4220
ERE-ABS, PER=0.9
14.5359
EB-FSA, PER=0.5
4.7732
EB-FSA, PER=0.7
7.4398
EB-FSA, PER=0.9
20.7725
Bin. tree, PER=0.5
4.8854
Bin. tree, PER=0.7
7.5521
Bin. tree, PER=0.9
20.8853
DFSA, PER=0.5
5.5298
DFSA, PER=0.7
9.1492
DFSA, PER=0.9
27.2493
DyTSA, PER=0.5
4.4689
DyTSA, PER=0.7
7.2222
DyTSA, PER=0.9
20.8822
ERE-ABS algorithm with that of EB-FSA, the binary tree protocol and the DFSA algorithm for varying PER. The number of consumed time slots of our proposed ERE-ABS algorithm is the lowest at all PER conditions. The ERE-ABS consumes about 16% less time slots (at PER=0.1) than DFSA protocol. The ERE mechanism contributes to performance enhancement under low PER (below 0.5), and the ABS mechanism contributes to performance enhancement under high PER (above 0.5). In Fig. 10, performance of EB-FSA and the binary tree protocol are shown to be very similar. EB-FSA spends less time slots under low PER. Under high PER, performance of EB-FSA becomes very close to that of the binary tree protocol. Fig. 11 shows the number of idle slots. The number of idle slots of our ERE-ABS is the smallest. The proposed ERE-ABS algorithm allows retransmission of failed IDs of tags appropriately, so the number of idle slots is greatly reduced. The reduced number of idle slots is the main portion of the gain of the proposed ABS algorithm. The number of idle slots of other algorithms is increasing as PER increases, resulting in increased identification time. In Fig. 12, we compare the average difference between the number of tags and the estimated
March 30, 2011
DRAFT
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. IEEE TRANSACTIONS ON MOBILE COMPUTING
24
Estimated number of tags
2000
ERE, PER=0.5 ERE, PER=0.7 ERE, PER=0.9 EB−FSA, PER=0.5 EB−FSA, PER=0.7 EB−FSA, PER=0.9 DFSA, PER=0.5 DFSA, PER=0.7 DFSA, PER=0.9
1500
1000
500
0 100
Fig. 9.
200
300
400
500 600 Number of Tags
700
800
900
1000
Estimated number of tags with varying number of tags and the PER.
14000
12000
ERE−ABS EB−FSA Bin. tree DFSA
Total Number of Timeslots
10000
8000
6000
4000
2000
0 0.1
Fig. 10.
0.2
0.3
0.4
0.5 0.6 Packet Error Rate
0.7
0.8
0.9
Total number of time slots used to identify tags for varying PER when n = 500.
number of tags of the proposed ERE-ABS with that of EB-FSA, DFSA and ZE [22]. DFSA and ZE use ideal frame size (L=500). The ZE and ERE algorithm have very small difference regardless of PER. ZE shows the smallest difference since the ZE algorithm uses the idle probability and adopts the ideal frame size. The difference of EB-FSA is similar to that of DFSA in a specific condition (Lest = 128 and fd = 4). And the difference of ERE algorithm is influenced by Lest . When Lest is large, the sample space to measure the probabilities is large March 30, 2011
DRAFT
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. IEEE TRANSACTIONS ON MOBILE COMPUTING
25 5000 4500
ERE−ABS EB−FSA Bin. tree DFSA
4000
Total Number of Idle Slots
3500 3000 2500 2000 1500 1000 500 0 0.1
Fig. 11.
0.2
0.3
0.4
0.5 0.6 Packet Error Rate
0.7
0.8
0.9
Total number of idle slots with the varying PER when n = 500.
because the granularity of the probabilities is large. The number of iterations, i.e., number of estimation frames in the estimation phase of EB-FSA may vary according to n, Lest , fd , and PER. V. C ONCLUSION We have proposed an error-resilient mechanism to estimate tag and frame size as well as channel state (PER), and an adaptive binary selection algorithm to overcome channel error problems. The proposed error resilient estimation method can estimate the number of tags accurately regardless of channel error and has low computational complexity. And the proposed adaptive binary selection algorithm can save time slots by reducing unnecessary waste of idle slots under channel error. We have provided analytical models of existing anti-collision algorithms under error-prone channels. Performance analysis and simulation results indicate that the proposed ERE-ABS algorithm consumes less time slots than other DFSA and binary tree protocols under error-prone channel conditions. R EFERENCES [1] R. Want, “An Introduction to RFID Technology,” IEEE Pervasive Computing, vol. 5, no. 1, pp. 25-33, Jan. 2006. [2] “EP C T M Radio-frequency Identification Protocols Class-1 Generation-2 UHF RFID Protocol For Communications at 860MHz-960MHz Version 1.0.9,” EPCglobal, Jan. 2005. March 30, 2011
DRAFT
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. IEEE TRANSACTIONS ON MOBILE COMPUTING
26 800 ERE L
=64, f =2
est
700
d
ERE Lest=64, fd=4 ERE Lest=128, fd=2
Difference between n and estimated n
ERE Lest=128, fd=4 600
EB−FSA L
=64, f =2
est
d
EB−FSA Lest=64, fd=4 EB−FSA L
=128, f =2
EB−FSA L
=128, f =4
est
500
d
est
d
DFSA ZE 400
300
200
100
0 0.1
0.2
0.3
0.4
0.5 Packet error rate
0.6
0.7
0.8
0.9
Fig. 12. Difference between the actual number of tags and the estimated number of tags with the varying PER when n = 500.
[3] “Information Technology Automatic Identification and Data Capture Techniques - Radio Frequency Identification For Item Management Air Interface - Part 6: Parameters For Air Interface Communications at 860-960 MHz,” ISO/IEC FDIS 18000-6, Nov. 2003. [4] J. Park, M. Y. Chung, and T.-J. Lee, “Identification of RFID Tags in Framed-Slotted ALOHA with Robust Estimation and Binary Selection,” IEEE Comm. Letters, vol. 11, no. 5, pp. 452-454, May 2007. [5] J. Eom, T.-J. Lee, R. Rietman, and A. Yener, “An Efficient Framed Slotted ALOHA Algorithm with Pilot Frame and Binary Selection for Anti-Collision of RFID Tag,” IEEE Comm. Letters, vol. 12, no. 11, pp. 861-863, Nov. 2008. [6] H. Vogt, “Efficient Object Identification with Passive RFID Tags,” in Proc. of Pervasive Computing, pp. 98-113, Aug. 2002. [7] S. Lee, S. Joo and C. Lee, “An Enhanced Dynamic Framed Slotted ALOHA Algorithm for RFID Tag Identification,” in Proc. of MobiQuitous, pp. 166-172, Jul. 2005. [8] J. Cha and J. Kim, “Novel Anti-collision Algorithms for Fast Object Identification in RFID System,” in Proc. of Parallel and Distributed System, vol. 2, pp. 63-67, Jul. 2005. [9] J. Myung, W. Lee and J. Srivastava, “Adaptive Binary Splitting for Efficient RFID Tag Anti-Collision,” IEEE Comm. Letters, vol. 10, no. 3, pp. 144-146, Mar. 2006. [10] A. J. E. M. Janssen and M. J. M. de Jong, “Analysis of Contention Tree Algorithms,” IEEE Trans. on Information Theory, vol. 46, no. 6, pp. 2163-2172, Sep. 2000. [11] D.-Y. Kim, B.-J. Jang, H.-G. Yoon, J.-S. Park, and J.-G. Yook, “Effects of Reader Interference on the RFID Interrogation Range,” in Proc. of European Conference on Wireless Technology(EcWT07), pp.728-732, Oct. 2007. [12] S. M. Yeo, B. W. Jeon, J. H. Bae, Y, J. Moon, Y. J. Kim, H. H. Ron, J. S. Park, Y. R. Seong, H. R. Oh, J. S. Kim, C. W. Park, and G. Y. Choi, “A Channel Allocation Scheme Considering with Collisions and Interferences in Practical UHF RFID applied Communication Fields,” in Proc. of IEEE International Conference on RFID, pp.258-268, Apr. 2008. [13] M. Simon. and D. Divsalar, “Some Interesting Observations for Certain Line Codes With Application to RFID,” IEEE Trans. on Comm., vol. 54, no. 4, pp. 583-586, Apr. 2006.
March 30, 2011
DRAFT
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. IEEE TRANSACTIONS ON MOBILE COMPUTING
27
[14] Y-C. Lai, and C-C. Lin, “A Pair-Resolution Blocking Algorithm on Adaptive Binary Splitting for RFID Tag Identification,” IEEE Comm. Letters, vol. 12, no. 6, pp. 432-434, June 2008. [15] Y.-C. Ko, S. Roy, J. R. Smith, H.-W. Lee, and C.-H. Cho, “An Enhanced RFID Multiple Access Protocol for Fast Inventory,” in Proc. IEEE Globecom, 2007. [16] Y.-C. Ko, S. Roy, J. R. Smith, H.-W. Lee, and C.-H. Cho, “RFID MAC Performance Evaluation Based on ISOIEC 18000-6 Type C,” IEEE Comm. Letters, vol. 12, no. 6, pp. 426-428, June 2008. [17] M. Mohaisen, H. Yoon, and K. Chang, “Radio Transmission Performance of EPCglobal Gen-2 RFID System,” in Proc. of ICACT 2008, pp. 1423-1428, Feb. 2008 [18] C. Jin, S. H. Cho, and K. Y. Jeon, “Performance Evaluation of RFID EPC Gen2 Anti-collision Algorithm in AWGN Environment,” in Proc. of ICMA 2007, pp. 2066-2070 Aug. 2007. [19] P. V. Nikitin and K. Rao, “Theory and Measurement of Backscattering from RFID Tags,” IEEE Antennas Propag. Mag., vol. 48, no. 6, pp. 212-218, Dec. 2006. [20] A. Safarian, A. Shameli, A. Rofougaran, M. Rofougaran, and F. De Flaviis, “RF Identification (RFID) Reader Front Ends With Active Blocker Rejection,” IEEE Trans. on Microwave Theory and Techniques, vol. 57, no. 4, pp. 1-10, 2009. [21] S. M. Yeo, B. W. Jeon, J. H. Bae, Y, J. Moon, Y. J. Kim, H. H. Ron, J. S. Park, Y. R. Seong, H. R. Oh, J. S. Kim, C. W. Park, and G. Y. Choi, “A Channel Allocation Scheme Considering with Collisions and Interferences in Practical UHF RFID applied Communication Fields,” in Proc. of IEEE International Conference on RFID, pp. 258-268, Apr. 2008. [22] K. Kodialam, and T. Nandagopal, “Fast and Reliable Estimation Schemes in RFID Systems,” in Proc. of MobiCom, pp. 322-333, Sep. 2006. [23] G. Maselli, C. Petrioli, and C. Vicari, “Dynamic Tag Estimation for Optimizing Tree Slotted Aloha in RFID Networks,” in Proc. of ACM MSWiM, pp. 315-322, Oct. 2008. [24] T. F. La Porta, G. Maselli, and C. Petrioli, “Anticollision Protocols for Single-Reader RFID Systems: Temporal Analysis and Optimization,” IEEE Transactions on Mobile Computing, vol. 10, no. 2, pp. 267-279, Feb. 2011.
Jongho Park received the B.S. and M.S. degrees in electrical and computer engineering from Sungkyunkwan University, Korea, in 2004 and 2006, respectively. He is currently pursuing his Ph.D. degree in department of electrical and computer engineering at Sungkyunkwan University since March 2006. His research interests include Medium Access Control (MAC) of RFID, wireless communication networks, Wireless LAN, Wireless PAN, and ad hoc networks.
March 30, 2011
DRAFT
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. IEEE TRANSACTIONS ON MOBILE COMPUTING
28
Tae-Jin Lee received his B.S. and M.S. in electronics engineering from Yonsei University, Korea in 1989 and 1991, repectively, and the M.S.E. degree in electrical engineering and computer science from University of Michigan, Ann Arbor, in 1995. He received the Ph.D. degree in electrical and computer engineering from the University of Texas, Austin, in May 1999. In 1999, he joined Corporate R&D Center, Samsung Electronics where he was a senior engineer. Since 2001, he has been an Associate Professor in the School of Information and Communication Engineering at Sungkyunkwan University, Korea. He was a visiting professor in Pennsylvania State University from 2007 to 2008. His research interests include RFID, performance evaluation, resource allocation, Medium Access Control (MAC), and design of communication networks and systems, wireless MAN/LAN/PAN, home/ad-hoc/sensor networks, next generation wireless communication systems, and optical networks. Since 2004, he has been a voting member of IEEE 802.11 WLAN Working Group, and is a memeber of IEEE.
March 30, 2011
DRAFT
1
Error Resilient Estimation and Adaptive Binary Selection for Fast and Reliable Identification of RFID Tags in Error-prone Channel Jongho Park and Tae-Jin Lee, Member, IEEE
Abstract In RFID systems, far field passive tags send information using back scattering. The signal level is typically very small, so channel error during transmission may occur frequently. Due to channel error, performance of RFID tag identification under error-prone channel is degraded compared to that under error-free channel. In this paper, we propose a novel error resilient estimation and adaptive binary selection to overcome the problem of channel errors. Our proposed error resilient estimation algorithm can estimate the number of tags and the channel state accurately regardless of frame errors. And our proposed adaptive binary selection reduces the idle slots caused by frame errors. Performance analysis and simulation results show that the proposed algorithm consumes up to 20% less time slots than the binary tree protocol and dynamic framed slotted ALOHA (DFSA) in various packet error rate (PER) conditions.
Index Terms anti-collision, channel error, collision resolution, RFID, tag estimation.
I. I NTRODUCTION In Radio Frequency IDentification (RFID) [1] systems, passive tags transmit their IDs using back scattering in response to the reader’s request [13]. The back scattering signals of tags are significantly attenuated by distance [19], [20]. So the signal strength of tags is typically very
The authors are with the School of Information and Communication Engineering, Sungkyunkwan University, Suwon 440-746, Korea (e-mail:[email protected]) March 30, 2011
Digital Object Indentifier 10.1109/TMC.2011.112
DRAFT
1536-1233/11/$26.00 © 2011 IEEE
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. IEEE TRANSACTIONS ON MOBILE COMPUTING
2
small, and frame errors may occur due to noise or interference. In addition, the signal power of RFID readers is much bigger than the response signal power of tags, so the signals of readers may interfere with one another and with those of tags [11]. In fact, the interference from other readers may be significant. In [21], it is stated that the interference range of a reader may be larger than the identification range of a reader. So a reader may interfere with other readers although the reader is not within the identification range of other readers. And readers’ interference may cause high frame errors. The performance of frame errors is reflected in the form of packet error rate (PER) which is determined by the signal to noise ratio (SNR) [12]. RFID systems require relatively high SNR (low PER) for reliable operation [17] or a robust mechanism to operate under low SNR. If RFID tags transmit their IDs simultaneously, an RFID reader can not identify the IDs due to collision. So the anti-collision protocol is significantly important for fast tag identification. There are two types of anti-collision protocols, ALOHA-based [2] and tree-based [3]. ALOHA-based protocols can be classified into ALOHA, slotted ALOHA and framed-slotted ALOHA (FSA). Among ALOHA-based protocols, the performance of FSA is superior to those of others because it can control the probability of transmission of tags in a frame. However, RFID readers need to estimate the number of tags to decide an appropriate frame size since the performance of FSA is known to be optimal when the frame size equals the number of tags [7]. Although enhanced tag estimation-based dynamic framed-slotted ALOHA (DFSA) is proposed in [7], [8], [6], and [22] if the number of tags is very large compared to the initial frame size, a reader can not estimate the number of tags (frame size) with accuracy due to many collisions. Thus careful decision on an initial frame size is very important. Most existing DFSA algorithms, however, do not provide an appropriate basis for decision on an initial frame size. In [4], a robust estimation mechanism is proposed, in which the number of responding tags in a frame is reduced by a certain ratio until an appropriate probability of collision is achieved. Then the number of tags (frame size) is estimated from the collision probability. In [23], the estimation algorithm using Tree Slotted ALOHA (TSA), i.e., Dynamic TSA (DyTSA) is proposed. DyTSA performs tag estimation at each child frame in a collision tree, which is different from the original TSA. Although the initial frame size is small compared to the number of tags, DyTSA can estimate the number of tags accurately at a child frame since the number of tags is decreased rapidly at each branch of the tree. In [24], Binary Splitting Tree Slotted Aloha (BSTSA) is March 30, 2011
DRAFT
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. IEEE TRANSACTIONS ON MOBILE COMPUTING
3
proposed. After the identification of a tag in a branch of a binary splitting tree, BSTSA estimates the number of tags in a branch accurately without concern on the initial frame size. Tag (frame) estimation and identification may have problems under unreliable channel conditions. Most tag estimation methods are based on the probability of collision which is very sensitive to error-prone channel. An RFID reader can not distinguish a collision with a frame error. Thus the reader may consider a channel error as a collision. Hence when there are frame errors, the measured probability of collision becomes high. As a result, the estimated number of tags based on the probability of collision is larger than the actual number of tags under error-free channel. Then the estimated frame size also becomes larger than the actually required frame size, and the performance of tag identification is degraded because there are many unnecessary idle slots in the over-estimated frame. Tree-based protocols solve the tag collision problem by binary trees, in which collided tags are separated into two groups randomly. Binary tree protocols may require a large number of time slots because tags experience many collisions at the early level of a binary collision tree. In [9], [14], an algorithm to decide the transmission order of tags by using the binary tree protocol and to use the order at the next identification process is proposed. M-ary tree protocols based on breadth-first tree search and tag estimation mechanism is proposed in [15], [16]. Framed-Slotted ALOHA with robust Estimation and Binary selection (EB-FSA) [4] operates FSA with only one identification frame based on robust estimation of the frame size from small pilot frames, and the collided tags resolve collisions by the binary tree algorithm. Framed Slotted ALOHA with Pilot Frame and Binary Selection (FSAFB) [5] partitions tags into several groups and operates a single frame FSA with binary tree collision resolution. Since there are 2 or 3 tags in most collided slots, applying the binary tree algorithm is efficient under error-free channel. When a frame error occurs, in the binary tree protocol, it is regarded as a collision and the corresponding tag, i.e., the tag that fails by the frame error, randomly selects 0 or 1 according to the collision resolution of the binary tree algorithm. So the failure of identification caused by a frame error induces an additional idle slot resulting in performance deterioration. Thus existing RFID anti-collision protocols require special attention for reliable operation under practical channel error conditions. In this paper, we propose a novel error-resilient anti-collision algorithm for RFID tag identification under unreliable channels. The proposed Error Resilient Estimation (ERE) and Adaptive March 30, 2011
DRAFT
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. IEEE TRANSACTIONS ON MOBILE COMPUTING
4
Binary Selection (ABS) algorithm reduces the identification time by reducing unnecessary idle and collision slots under error-prone channel. The ERE mechanism estimates the number of tags and the channel condition (PER) accurately under error-prone channel regardless of an initial frame size with relatively low computational load. And the ABS algorithm reduces the unnecessary idle slots caused by frame errors using the estimated number of tags and PER in the ERE process. This paper is organized as follows. In Section II, we explain the proposed error resilient estimation and adaptive binary selection algorithm. In Section III, mathematical analysis of DFSA, binary tree, EB-FSA, and ERE-ABS algorithms under error-prone channels is presented. Performance evaluation and comparison via both analysis and simulations is conducted in Section IV. We conclude in Section V. II. ERE-ABS A LGORITHM We propose an algorithm for accurate estimation of channel status and fast identification to enhance performance under unreliable channel condition. A. Error Resilient Estimation (ERE) When the frame size is equal to the number of tags, identification performance of FSA is known to be optimal. Thus an accurate estimation of the number of tags is very important in tag estimation-based DFSA. If the estimated number of tags in a frame is very small compared with the actual number of tags, the next frame size becomes inaccurate. So an initial frame size for initial estimation of the number of tags has to be decided with care. DFSA in [8] estimates the number of tags by using the measured probability of collision. Tag estimation method in [22] uses the probability of collision and idle probability. In [6], the number of tags is estimated by the numbers of idle, success and collision slots. The algorithm in [6] computes the estimated number of tags by minimizing the vector distance between the measured number of idle/success/collision slots and the estimated number of idle/success/collision slots. In [23], DyTSA estimates the number of tags at each child frame of a collision tree noting that the number of tags in a child frame in smaller than that in a parent frame. However, the (measured) probability of collision or success becomes inaccurate when there are channel errors since a reader regards the failure of identification due to a channel error as March 30, 2011
DRAFT
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. IEEE TRANSACTIONS ON MOBILE COMPUTING
5
a collision and the reader can not distinguish a collision with a channel error. So, under errorprone channel, the measured probability of collision is bigger and the measured probability of success is smaller than the actual collision/success probabilities. In this sense, if a reader estimates the number of tags from the measured inaccurate probabilities, it may be over-estimated compared with the actual number of tags. Thus existing estimation methods under channel error causes performance degradation of DFSA. So we propose a new error resilient estimation (ERE) mechanism for the number of tags. ERE uses the probability of idle slots instead of that of collision or success slots, e.g., Zero Estimator (ZE) [22], since the probability of idle slots is not affected by channel errors. Thus the estimation based on the probability of idle slots is more accurate in spite of channel errors. Let’s assume that we have n tags and the size of a frame is L. The probability of collision Pcoll is then Pcoll = 1 − Psucc − Pidle , Psucc
1 =n· L
Pidle =
n−1 1 1− , L 1 1− L
(1) (2)
n ,
(3)
where Pidle and Psucc are the probabilities of idle and success slots, respectively. If the number of tags becomes larger, the measured probability of idle slots Pˆidle becomes quickly closer to 0. So if Pˆidle is used for estimation, the number of tags that can be estimated might be inaccurate if Pˆidle becomes small. And when the measured probability of idle slots Pˆidle becomes small, the estimated number of tags n may be inaccurate, since small measurement error of Pˆidle may lead to large estimation error for the estimated number of tags (see Fig. 1), i.e., when the probability of idle is small, variation of the number of estimated tags may be very large for small variation of measurement error of the probability of idle slots. In order to prevent this problem, in our ERE, we use the threshold Pidle th to decide if the measured probability of idle slots Pˆidle is appropriate. The ERE uses a fixed frame size Lest , and if the probability of idle slots Pˆidle is smaller than the threshold Pidle th (i.e., if the number of responding tags is large), a reader attempts to decrease the number of tags to respond to the reader by a factor of fd using a bit mask in a query frame to tags. Lest , Pidle th and fd are March 30, 2011
DRAFT
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. IEEE TRANSACTIONS ON MOBILE COMPUTING
6 1 P
coll
P
0.9
idle
P
succ
0.8 0.7
Probability
0.6 0.5 0.4 0.3 0.2 0.1 0
Fig. 1.
0
1
2
3
4
5 n/L
6
7
8
9
10
Characteristics of Psucc , Pidle and Pcoll with the number of tags n and the frame size L.
system parameters. We decide appropriate values by extensive simulation. The reader repeats this process until the probability of idle slots exceeds the threshold and then estimates the number of tags n. The tags identified in ERE are not muted during the ERE phase for accuracy of tag estimation. So ERE can estimate the number of tags accurately under unreliable channel regardless of the initial frame size and the number of tags. Furthermore estimation of n using Pˆidle is very simple to compute. Eq. (3) can be expressed by the following equation. n=
log(Pidle ) . log((L − 1)/L)
(4)
So in practice the estimated number of tags nest for n is found as ∗
nest = nest,j ∗ × (fd )j ,
(5)
where j ∗ is the number of repeated estimation frames until the probability of idle slots exceeds the threshold Pidle th , and nest,j ∗ is the estimated number of tags in the j ∗ th estimation frame. nest,j ∗ =
log(Pˆidle,j ∗ ) . log((Lest − 1)/Lest )
(6)
Then the estimated frame size L = nest for the identification process. Note that unlike ours, complicated numerical methods are required to estimate n from Pˆcoll or Pˆsucc . And, for a given Pˆsucc , two estimated numbers for n are mapped, so Pˆsucc can not be used alone to uniquely estimate the number of tags (see Fig. 1). Fig. 2 shows the structure of the estimation process. March 30, 2011
DRAFT
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. IEEE TRANSACTIONS ON MOBILE COMPUTING
7 Error Resilient Estimation (ERE) 0
1
……
Pˆ idle>Pidle_th Adaptive Binary Selection (ABS)
j*
L = nest
Lest
S Success F I S I F
……
F
Measure Pˆ idle,j*, Pˆ succ,j*, Pˆ coll,j* Compute nest and PER
Fig. 2.
I Idle
S I Failure (collision or channel error )
The structure of estimation process of our proposed algorithm
A reader can now take advantage of the estimated nest,j ∗ from Pˆidle,j ∗ to estimate the channel characteristics, PER. First, the probability of success under error-free channel P˜succ,j ∗ can be computed from the estimated nest,j ∗ in Eq. (2). 1 P˜succ,j ∗ = nest,j ∗ Lest
1 1− Lest
nest,j∗ −1
.
(7)
The estimation probability of no collision but channel error which should have been success if there had not been channel error, is P˜j ∗ ((no collision) and (channel error)) = P˜succ,j ∗ − Pˆsucc,j ∗ ,
(8)
where Pˆsucc,j ∗ is the measured (actual) probability of success and P˜succ,j ∗ is the estimated probability that a transmission is successful if there is no channel error. Then PER can be defined as the probability of failure of identification due to a channel error when a tag transmits an ID in a slot without collision. Thus PER can be estimated by ˜ P ER= P˜j ∗ (channel error|success if there is no channel error) P˜ ((no collision) and (channel error)) = P˜succ,j ∗ P˜succ,j ∗ − Pˆsucc,j ∗ . (9) P˜succ,j ∗ Fig. 3 shows the relationship among Pˆsucc,j ∗ , P˜succ,j ∗ and Pˆcoll,j ∗ . The pseudo code to compute =
the frame error rate PER is illustrated in Algorithm 1. B. Adaptive Binary Selection (ABS) In the binary tree collision resolution protocol, when a channel error occurs, a reader and a tag regard the error as a collision and the corresponding tag randomly selects 0 or 1 to resolve March 30, 2011
DRAFT
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. IEEE TRANSACTIONS ON MOBILE COMPUTING
8
the collision resulting in two time slots. So the failure of identification caused by a channel error incurs an additional idle slot. This is the main reason of performance degradation of the binary tree protocol under error-prone channel condition. In this paper, we propose adaptive binary selection mechanism which determines whether a collision is likely to be real or virtual (by channel error) based on the estimated PER. In the identification process, a reader and tags operate FSA with the estimated frame size L (L = nest ) in ERE. When a collision occurs our proposed ABS algorithm statistically determines if a collision is actual collision or it is virtual collision by channel error. ABS determines direct retransmission for a channel error or binary selection for a collision based on the estimated PER in the ERE stage or the probability of collision. Each tag selects a random counter value as in framed-slotted ALOHA after a reader decides the optimal frame size L determined from the estimated nest in the ERE stage. Tags decrease their counter values at each time slot. When a counter becomes 0, the tag transmits its ID. If a collision is observed, the reader operates the ABS algorithm. Note that a collision may be either real collision or channel error. Then a reader transmits a control message which commands to do retransmission or binary selection for the tag(s) at the start of the next slot. The reader maintains two counters, Nc and Ne (initially both are 0). If the reader determines that a collision is due to channel error, it decides to do retransmission and increases Ne by 1. The counter Ne represents the number of consecutive retransmissions. If the reader determines that a collision is real, it decides to do binary selection, and it increases Nc by 1, and resets Ne to 0. If a tag transmits its ID successfully, the reader decreases Nc by 1. The counter Nc denotes the current level of a binary collision tree.
Collision
Success if there is no channel error No collision Success and Chanel error
Pˆ coll,j*
Idle
ˆ succ,j* P ~
Psucc,j* Fig. 3.
Relationship among Pˆsucc,j ∗ , P˜succ,j ∗ and Pˆcoll,j ∗ .
March 30, 2011
DRAFT
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. IEEE TRANSACTIONS ON MOBILE COMPUTING
9
Algorithm 1 Error Resilient Estimation (ERE) ˜ 1: procedure E STIMATION OF FRAME SIZE L(=nest ) AND P ER FSA operation frame with a frame of size Lest Measure(Pˆcoll , Pˆsucc , Pˆidle )
2: 3:
count ← 0 while (Pˆidle ≥ Pidle th ) do
4: 5:
counter for iteration
Transmit a query frame with a bit mask to reduce the number of tags by a factor of
6:
fd 8:
FSA operation frame with a frame of size Lest Measure(Pˆcoll , Pˆsucc, Pˆidle )
9:
count + +
7:
10:
end while
11:
j ∗ ← count, Pˆsucc,j ∗ ← Pˆsucc
12:
Compute nest,j ∗
by Eq. (6)
13:
Compute nest
by Eq. (5)
14:
L ← nest
15: 16: 17:
Compute P˜succ,j ∗ ˜ Compute P ER
by Eq. (7) by Eq. (9)
end procedure
The reader determines retransmission or binary selection in an attempt to increase the efficiency at each decision point (i.e., for each observed collision). The efficiency is defined as the ratio between the number of successfully transmitted tags and the total number of time slots. The efficiency of retransmission ηr (Nc , Ne ) if a collision is estimated as channel error, and that of binary selection ηb (Nc ) if a collision is estimated as actual collision are defined as follows. ηb (Nc ) = Eb (Nc )
(10)
˜ Ne −1 , ηr (Nc , Ne ) = Er (Nc ) · P ER
(11)
where ηb (Nc ) is the efficiency at the (Nc + 1)th level and ηr (Nc , Ne ) is the efficiency at the (Ne + 1)th retransmission and at the Nc th level. The efficiency ηr (Nc , Ne ) is decreased by the March 30, 2011
DRAFT
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. IEEE TRANSACTIONS ON MOBILE COMPUTING
10
number of retransmissions because the number of tags which occupy a slot is decreased by the ˜ rate of P ER. And Eb (i) and Er (i) are defined as follows. Eb (i) =
∞
Pf ail (k) · Eb (k, i),
(12)
k=1
Eb (k, i) =
e ˜ (1 − P ER) · SBIN (k, i + 1) , e ˜ · S e (k, i)) 2 · (CBIN (k, i) + P ER BIN
Er (i) =
∞
Pf ail (k) · Er (k, i),
(13)
(14)
k=1
˜ ˜ · S e (k, i) (1 − P ER) · P ER BIN , e ˜ · S e (k, i)) (CBIN (k, i) + P ER BIN ⎧ ˜ P˜succ P ER· ⎨ k=1 ˜ P˜succ , P˜coll +P ER· Pf ail (k) = nest 1 k 1 nest −k ( ) (1− ) L ⎩ ( k )L , k > 1, ˜ P˜ P˜ +P ER·
Er (k, i) =
coll
(15)
(16)
succ
where Eb (k, i) and Er (k, i) denote the efficiency of binary selection and retransmission, respectively, at the ith binary collision tree level when there are k tags in the binary tree. P˜succ and P˜coll are the computed probabilities of success and collision, respectively, using the estimated number of tags nest . The numerators of Eq. (13) and Eq. (15) are the number of successful e slots because SBIN (k, i) is the number of successful slots in level i when there are k tags in
the binary tree under error-prone channel. If the reader chooses binary selection, the level of the e ˜ tree is increased. So, the number of successful slots is (1 − P ER)S BIN (k, i + 1). If the reader chooses retransmission, the level of the tree does not change. Thus the number of successful ˜ P ER ˜ · S e (k, i). And the denominators of Eq. (13) and Eq. (15) represent slots is (1 − P ER) BIN the total number of used slots. Fig. 4 shows an example for the equations. In the binary selection case, the number of failed (collision or channel error) slots at the ith level is increased by a factor of two at the (i+1)th level, and the number of success slots is the number of real success slots without errors in the binary tree. In the retransmission case, the number of failed (collision or channel error) slots at the ith level is equal to the number of slots used for retransmission, and the number of success slots is the number of no error slots among the failed slots by channel errors. Then Eb (i) and Er (i) denote the efficiency of binary e (k, i) is selection and retransmission, respectively, at the ith binary collision tree level. CBIN
March 30, 2011
DRAFT
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. IEEE TRANSACTIONS ON MOBILE COMPUTING
11 Binary Selection
~ PER· SeBIN (k,i)
C
Se BIN (k,i) Level i
Level i+1
E
E
E
I
S
e
S
BIN
(k,i)
C
I
C
~ 2·(C BIN (k,i)+PER· SeBIN (k,i)) e
I
S
~ (1-PER)·Se BIN (k,i+1)
Retransmission ~ PER· SeBIN (k,i) S E
E
E
S
e BIN
Ce BIN(k,i)
(k,i) S
C
I
Level i C
~ Ce BIN (k,i)+PER·Se BIN (k,i) S
Fig. 4.
Success
I
idle
~ ~ (1-PER)·PER· SeBIN (k,i) C
Real Collision
E
Virtual collision (channel error )
An example of efficiency computation Eqs. (13) and (15).
the number of collided and channel error time slots at depth i when the number of tags is k e at depth i of a binary tree, and SBIN (k, i) is the number of actual successful slots when the e e number of tags is k at depth i of a binary tree. CBIN (k, i) and SBIN (k, i) are defined in Eqs. ˜ in Section III. And Pf ail (k) is the probability of (26)-(27) with n = k and P ER = P ER
identification failure when k tags are in a slot. When k = 1, the identification failure occurs by channel error only. When k > 1, the identification failure occurs due to collision of k tags. So the probability of identification failure can be computed by binomial probability. When k = 1, the identification failure occurs due to channel error. The denominator of Eq. (16) is the total probability of identification failure. If the estimated number of tags nest is accurate and L=nest , Pf ail (k) does not change because the average number of tags in a slot does not change by k. Thus we can obtain ηb (Nc ) and ηr (Nc , Ne ) if an identification fails (collision is observed), and compare them to decide which one yields better efficiency. If ηb (Nc ) is larger than ηr (Nc , Nb ) the reader determines binary selection, and if ηb (Nc ) is smaller than ηr (Nc , Nb ) the reader determines retransmission. If the tags receive the binary selection command, the collided tags randomly select their counter values from 0 or 1, and others increase their counter values by 1. And March 30, 2011
DRAFT
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. IEEE TRANSACTIONS ON MOBILE COMPUTING
12
if the tags receive the retransmission command, the corresponding tags retransmit their IDs immediately without any change of counter values. The pseudo code of the ABS algorithm is given in Algorithm 2. Fig. 5 shows an example and time diagram of our proposed ABS algorithm. In Fig. 5, three tags 1, 2, and 3 transmit simultaneously at a slot in the identification process. The reader compares efficiency of binary selection and that of retransmission and selects binary selection (in this example, the efficiency of binary selection (ηb ) is larger than the efficiency of retransmission (ηr )). And all three tags select 0 and collide again and the reader chooses binary selection again. One of the tags (tag 2) now selects 0 and the others (tag 1 and tag 3) select 1. Tag 2 is now successful and the others (tag 1 and tag 3) collide again. The reader decides binary selection, and one of the tags (tag 1) selects 0 and the other (tag 3) selects 1. Tag 1 which selects 0 occupies a slot alone, but its transmission fails twice due to channel error. The reader decides retransmission for both of failures based on the efficiency (in this example, the efficiency of binary selection (ηb ) is smaller than the efficiency of retransmission (ηr )). Then ηb > ηr and the reader decides binary selection and tag 1 now succeeds. Now tag 3 occupies a slot alone, but its transmission fails due to channel error. The reader then decides retransmission from efficiency and tag 3 finally succeeds. III. P ERFORMANCE A NALYSIS We analyze the average total number of slots used in the identification of DFSA, binary tree, EB-FSA and ERE-ABS under channel error condition. We define P ER as the probability that a reader can not identify a tag’s ID due to channel error. P ER = 1 − (1 − BER)Nbit ,
(17)
where Nbit is the number of bits in a frame with RFID tag ID, and BER is the bit error rate. A. DFSA We assume that DFSA estimates the number of tags n using the probability of collision. The estimation function is defined as fest (Pf ail , L), which is a function of the probability of failure by collision or channel error and the size of a frame. Let ni , Li , Psucc,i and Pcoll,i be the number
March 30, 2011
DRAFT
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. IEEE TRANSACTIONS ON MOBILE COMPUTING
13
Algorithm 2 Adaptive Binary Selection (ABS) 1: procedure TAG I DENTIFICATION ˜ in ERE algorithm initialize the efficiency with L and P ER
2:
INIT
3:
Nc ← 0, Ne ← 0
4:
while there are unidentified tags do
5:
Tag identification process
6:
if an identification fails then
collision observed
if ηr (Nc , Ne ) > ηb (Nc ) then
7:
Reader transmits the retransmission command, by which the tags retransmit
8:
IDs directly Ne + +
9:
else
10:
Reader transmits the binary selection command, by which the tags select 0
11:
or 1 randomly Nc + +, Ne = 0
12:
end if
13:
else
14: 15:
Success of tag identification
16:
Nc − −
17:
end if
18:
end while
19:
end procedure
20:
procedure I NIT
success observed
21:
Compute Pf ail
by Eq. (16)
22:
Compute Eb
by Eq. (12)
23:
Compute Er
by Eq. (14)
24:
Compute ηb
by Eq. (10)
25:
Compute ηr
by Eq. (11)
26:
end procedure
March 30, 2011
DRAFT
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. IEEE TRANSACTIONS ON MOBILE COMPUTING
14 Tag 1, 2, 3
F Tag 1, 2, 3
F
No tags
r(Nc,Ne) 1, i = 0.
(43)
where the number of retransmissions at level i is k ∗ (n, i) = arg max{(Eb (n, i) − Er (n, i)P ERk−1) < 0}. k∈N
(44)
Because the reader chooses retransmission until the efficiency of binary selection becomes larger than that of retransmission. The number of actual success time slots at depth i excluding the success time slots at retransmission is found as ⎧ ⎪ (1 − P ER) · (S(n, 2i ) ⎪ ⎪ ⎪ ⎪ ⎨ −S n > 1, i > 0 ABS c (n, i − 1)), SABS (n, i) = ⎪ ⎪ 0, n > 1, i = 0 ⎪ ⎪ ⎪ ⎩ (1 − P ER), n = 1, i = 0.
(45)
Then the number of actual collided time slots and the channel error time slots without retransmission slots at depth i is CABS (n, i) = C(n, 2i ) + P ERk
March 30, 2011
∗ (n,i)
·
SABS (n, i) . (1 − P ER)
(46)
DRAFT
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. IEEE TRANSACTIONS ON MOBILE COMPUTING
19
So the total number of collided time slots in the binary tree with ABS under channel error becomes CABS (n) =
∞
CABS (n, i).
(47)
i=0
The number of retransmission time slots in the ABS with n tags is R(n) =
∞
R(n, i),
(48)
i=0
where the number of retransmission time slots in the binary tree with ABS at level i for n tags ⎧ P ER i ⎪ ⎪ C(n, 2 ) + S (n, i) · ABS ⎪ (1−P ER) ⎪ ⎪
∗ (n,i)−1 ⎪ k ⎪ ⎪ {S(n, 2i ) − SABS c (n, i − 1)} ⎪ ⎨ − k=0 R(n, i) = ·(1 − P ER) · P ERk , k ∗ (n, i) > 1 ⎪ ⎪ ⎪ ⎪ 0, k ∗ (n, i) = 0 ⎪ ⎪ ⎪ ⎪ ⎪ ⎩ C(n, 2i ) + SABS (n, i) · P ER , k ∗ (n, i) = 1. (1−P ER)
is
(49)
The number of retransmission time slots is the same as the number of failed slots at the previous step. So the number of first retransmission slots at level i is the number of failed slots (collision or channel error) at level i of the binary tree and the number of the next retransmission slots is the number of failed slots at the previous retransmission. Therefore total used time slots for identification under channel error in the proposed ERE-ABS algorithm is found from Eqs. (41) and (42). TERE−ABS = TERE (n) + TABS (n).
(50)
Fig. 6 shows an example for Eq. (44), Eq. (45) and Eq. (48). In our proposed ERE-ABS algorithm, the level of a binary tree includes the retransmitted slots. No matter how many retransmissions there are, the level of a binary tree does not change. In this example the number of retransmissions k ∗ (n, i) at the ith level, is 2. SABS (n, i) is the number of success slots at the ith level including retransmitted slots, which is 3 in the example. The number of retransmission slots R(n, i) at the ith level is 5. IV. S IMULATION R ESULTS We evaluate the performance of DFSA, binary tree, EB-FSA and the proposed ERE-ABS algorithm in channel error condition under varying number of tags and PER via simulations March 30, 2011
DRAFT
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. IEEE TRANSACTIONS ON MOBILE COMPUTING
20
C
C
C
C
E
S
E
R(n,i) K*(n,i)
S
C
E
C
S
ith level
SABS(n,i)
S S
Fig. 6.
S
Success
C
Real Collision
E
Channel Error
An example for Eq. (44), Eq. (45) and Eq. (48).
and compare it with the analysis in section III. In DFSA, an initial frame size is set to the minimum frame size that can estimate n using the tag estimation method in [8] (L=0.2×n). In the EB-FSA algorithm, we set Lest =128, Pcoll th =0.7 and fd =4, and the typical value of i∗ (n) is 3 when n=1000 and P ER=0.7. In our ERE-ABS algorithm, we set Lest =128, Pidle th =0.2 and fd =4. We assume that the tag estimation is accurate in the analysis of DFSA, EB-FSA with ERE and ERE-ABS algorithm, and that the PERs of all tags are the same for simplicity although the PERs of tags may be different in practice. Since our ERE-ABS algorithm depends on the average PER and is designed to estimate the average PER of tags by Eq. 9 not to investigate and estimate the individual PERs, the assumption is not entirely unreasonable. So our ERE-ABS can be applied to the tags with different PERs. Fig. 7 and Table I show that the simulation results are closely matched with the analysis in Section III. We have shown 99% confidence interval with 10000 simulation iterations. In Fig. 7 and Table I, ERE-ABS uses the smallest number of time slots for all cases of PER. In low PER condition, performance enhancement of ERE-ABS is rather small. The proposed ABS algorithm decides retransmission or binary selection based on the efficiency, but when the PER is low, the efficiency of retransmission may be lower than binary selection. Because most decision is expected to be binary selection, the performance of ABS is similar to that of EB-FSA. However in high PER condition, the proposed ABS algorithm can reduce the unnecessary idle slots. So the performance of ERE-ABS is better than other existing protocols. March 30, 2011
DRAFT
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. IEEE TRANSACTIONS ON MOBILE COMPUTING
21 4
x 10 2 1.8
Total Number of Timeslots
1.6 1.4 1.2
ERE−ABS, PER=0.5 (analysis) ERE−ABS, PER=0.7 (analysis) ERE−ABS, PER=0.9 (analysis) EB−FSA with ERE, PER=0.5 (analysis) EB−FSA with ERE, PER=0.7 (analysis) EB−FSA with ERE, PER=0.9 (analysis) ERE−ABS, PER=0.5 (simulation) ERE−ABS, PER=0.7 (simulation) ERE−ABS, PER=0.9 (simulation) EB−FSA with ERE, PER=0.5 (simulation) EB−FSA with ERE, PER=0.7 (simulation) EB−FSA with ERE, PER=0.9 (simulation)
1 0.8 0.6 0.4 0.2 0 100
Fig. 7.
200
300
400
500 600 Number of Tags
700
800
900
1000
Total number of time slots used to identify tags with varying number of tags and the PER. TABLE I S LOPES OF THE CURVES IN F IG . 7
Algorithm
Slope
ERE-ABS, PER=0.5
4.3317
ERE-ABS, PER=0.7
6.4220
ERE-ABS, PER=0.9
14.5359
EB-FSA with ERE, PER=0.5
4.4377
EB-FSA with ERE, PER=0.7
7.1043
EB-FSA with ERE, PER=0.9
20.4370
In Fig. 8 and Table II, we compare the proposed ERE-ABS algorithm with EB-FSA, the binary tree protocol, DFSA and DyTSA [23] algorithm under various PERs. Our proposed ERE-ABS algorithm is superior to other existing algorithms under error-prone channel. In channel error conditions, tags tend to spend relatively long time in a binary tree. So performance of EB-FSA without ERE is very close to that of binary tree algorithm under error-prone channel. The total number of slots used by DyTSA is smaller than that of DFSA but larger than those of the others. DyTSA makes small child frames for each collided slot and estimates the number of slots for each child frame. Over-estimation caused by channel error is relatively small for small frames.
March 30, 2011
DRAFT
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. IEEE TRANSACTIONS ON MOBILE COMPUTING
22 4
x 10
Total Number of Timeslots
2
1.5
ERE−ABS, PER=0.5 ERE−ABS, PER=0.7 ERE−ABS, PER=0.9 EB−FSA, PER=0.5 EB−FSA, PER=0.7 EB−FSA, PER=0.9 Bin. tree, PER=0.5 Bin. tree, PER=0.7 Bin. tree, PER=0.9 DFSA, PER=0.5 DFSA, PER=0.7 DFSA, PER=0.9 DyTSA, PER=0.5 DyTSA, PER=0.7 DyTSA, PER=0.9
1
0.5
0 100
Fig. 8.
200
300
400
500 600 Number of Tags
700
800
900
1000
Total number of time slots used to identify tags with varying number of tags and the PER.
However the top level of the tree is operated with DFSA, which causes performance degradation under error prone channel. Fig. 8 and Table II show that the performance of the binary tree protocol is better than that of DFSA in channel error conditions. And ERE-ABS achieves about 21.5% (at PER=0.5) performance gain over DFSA protocol. The higher the PER is, the larger the gap of the performance improvement becomes because the proposed ABS algorithm reduces the number of unnecessary idle slots. In Fig. 9, we compare the estimation performance of the proposed ERE algorithm with that of EB-FSA and DFSA under various PERs. Our proposed ERE algorithm can estimate the number of tags accurately in spite of high PER. But EB-FSA and DFSA tend to over-estimate the number of tags. In addition, DFSA is assumed to use the optimal initial frame size to estimate the number of tags. In practice, however, the estimation accuracy of DFSA will not be guaranteed. The estimation accuracy of EB-FSA becomes worse after n=300. This is related with the deceasing factor fd and the estimation frame size Lest since the number of estimation frames varies according to fd and Lest . Depending on fd and Lest , i∗ (n) is changed. If i∗ (n) becomes larger, the estimation error also becomes larger because the estimated number of tags is determined by multiplying Lest by i∗ (n). In Fig. 10, we compare total number of time slots used for identification of the proposed
March 30, 2011
DRAFT
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. IEEE TRANSACTIONS ON MOBILE COMPUTING
23
TABLE II S LOPES OF THE CURVES IN F IG . 8
Algorithm
Slope
ERE-ABS, PER=0.5
4.3317
ERE-ABS, PER=0.7
6.4220
ERE-ABS, PER=0.9
14.5359
EB-FSA, PER=0.5
4.7732
EB-FSA, PER=0.7
7.4398
EB-FSA, PER=0.9
20.7725
Bin. tree, PER=0.5
4.8854
Bin. tree, PER=0.7
7.5521
Bin. tree, PER=0.9
20.8853
DFSA, PER=0.5
5.5298
DFSA, PER=0.7
9.1492
DFSA, PER=0.9
27.2493
DyTSA, PER=0.5
4.4689
DyTSA, PER=0.7
7.2222
DyTSA, PER=0.9
20.8822
ERE-ABS algorithm with that of EB-FSA, the binary tree protocol and the DFSA algorithm for varying PER. The number of consumed time slots of our proposed ERE-ABS algorithm is the lowest at all PER conditions. The ERE-ABS consumes about 16% less time slots (at PER=0.1) than DFSA protocol. The ERE mechanism contributes to performance enhancement under low PER (below 0.5), and the ABS mechanism contributes to performance enhancement under high PER (above 0.5). In Fig. 10, performance of EB-FSA and the binary tree protocol are shown to be very similar. EB-FSA spends less time slots under low PER. Under high PER, performance of EB-FSA becomes very close to that of the binary tree protocol. Fig. 11 shows the number of idle slots. The number of idle slots of our ERE-ABS is the smallest. The proposed ERE-ABS algorithm allows retransmission of failed IDs of tags appropriately, so the number of idle slots is greatly reduced. The reduced number of idle slots is the main portion of the gain of the proposed ABS algorithm. The number of idle slots of other algorithms is increasing as PER increases, resulting in increased identification time. In Fig. 12, we compare the average difference between the number of tags and the estimated
March 30, 2011
DRAFT
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. IEEE TRANSACTIONS ON MOBILE COMPUTING
24
Estimated number of tags
2000
ERE, PER=0.5 ERE, PER=0.7 ERE, PER=0.9 EB−FSA, PER=0.5 EB−FSA, PER=0.7 EB−FSA, PER=0.9 DFSA, PER=0.5 DFSA, PER=0.7 DFSA, PER=0.9
1500
1000
500
0 100
Fig. 9.
200
300
400
500 600 Number of Tags
700
800
900
1000
Estimated number of tags with varying number of tags and the PER.
14000
12000
ERE−ABS EB−FSA Bin. tree DFSA
Total Number of Timeslots
10000
8000
6000
4000
2000
0 0.1
Fig. 10.
0.2
0.3
0.4
0.5 0.6 Packet Error Rate
0.7
0.8
0.9
Total number of time slots used to identify tags for varying PER when n = 500.
number of tags of the proposed ERE-ABS with that of EB-FSA, DFSA and ZE [22]. DFSA and ZE use ideal frame size (L=500). The ZE and ERE algorithm have very small difference regardless of PER. ZE shows the smallest difference since the ZE algorithm uses the idle probability and adopts the ideal frame size. The difference of EB-FSA is similar to that of DFSA in a specific condition (Lest = 128 and fd = 4). And the difference of ERE algorithm is influenced by Lest . When Lest is large, the sample space to measure the probabilities is large March 30, 2011
DRAFT
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. IEEE TRANSACTIONS ON MOBILE COMPUTING
25 5000 4500
ERE−ABS EB−FSA Bin. tree DFSA
4000
Total Number of Idle Slots
3500 3000 2500 2000 1500 1000 500 0 0.1
Fig. 11.
0.2
0.3
0.4
0.5 0.6 Packet Error Rate
0.7
0.8
0.9
Total number of idle slots with the varying PER when n = 500.
because the granularity of the probabilities is large. The number of iterations, i.e., number of estimation frames in the estimation phase of EB-FSA may vary according to n, Lest , fd , and PER. V. C ONCLUSION We have proposed an error-resilient mechanism to estimate tag and frame size as well as channel state (PER), and an adaptive binary selection algorithm to overcome channel error problems. The proposed error resilient estimation method can estimate the number of tags accurately regardless of channel error and has low computational complexity. And the proposed adaptive binary selection algorithm can save time slots by reducing unnecessary waste of idle slots under channel error. We have provided analytical models of existing anti-collision algorithms under error-prone channels. Performance analysis and simulation results indicate that the proposed ERE-ABS algorithm consumes less time slots than other DFSA and binary tree protocols under error-prone channel conditions. R EFERENCES [1] R. Want, “An Introduction to RFID Technology,” IEEE Pervasive Computing, vol. 5, no. 1, pp. 25-33, Jan. 2006. [2] “EP C T M Radio-frequency Identification Protocols Class-1 Generation-2 UHF RFID Protocol For Communications at 860MHz-960MHz Version 1.0.9,” EPCglobal, Jan. 2005. March 30, 2011
DRAFT
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. IEEE TRANSACTIONS ON MOBILE COMPUTING
26 800 ERE L
=64, f =2
est
700
d
ERE Lest=64, fd=4 ERE Lest=128, fd=2
Difference between n and estimated n
ERE Lest=128, fd=4 600
EB−FSA L
=64, f =2
est
d
EB−FSA Lest=64, fd=4 EB−FSA L
=128, f =2
EB−FSA L
=128, f =4
est
500
d
est
d
DFSA ZE 400
300
200
100
0 0.1
0.2
0.3
0.4
0.5 Packet error rate
0.6
0.7
0.8
0.9
Fig. 12. Difference between the actual number of tags and the estimated number of tags with the varying PER when n = 500.
[3] “Information Technology Automatic Identification and Data Capture Techniques - Radio Frequency Identification For Item Management Air Interface - Part 6: Parameters For Air Interface Communications at 860-960 MHz,” ISO/IEC FDIS 18000-6, Nov. 2003. [4] J. Park, M. Y. Chung, and T.-J. Lee, “Identification of RFID Tags in Framed-Slotted ALOHA with Robust Estimation and Binary Selection,” IEEE Comm. Letters, vol. 11, no. 5, pp. 452-454, May 2007. [5] J. Eom, T.-J. Lee, R. Rietman, and A. Yener, “An Efficient Framed Slotted ALOHA Algorithm with Pilot Frame and Binary Selection for Anti-Collision of RFID Tag,” IEEE Comm. Letters, vol. 12, no. 11, pp. 861-863, Nov. 2008. [6] H. Vogt, “Efficient Object Identification with Passive RFID Tags,” in Proc. of Pervasive Computing, pp. 98-113, Aug. 2002. [7] S. Lee, S. Joo and C. Lee, “An Enhanced Dynamic Framed Slotted ALOHA Algorithm for RFID Tag Identification,” in Proc. of MobiQuitous, pp. 166-172, Jul. 2005. [8] J. Cha and J. Kim, “Novel Anti-collision Algorithms for Fast Object Identification in RFID System,” in Proc. of Parallel and Distributed System, vol. 2, pp. 63-67, Jul. 2005. [9] J. Myung, W. Lee and J. Srivastava, “Adaptive Binary Splitting for Efficient RFID Tag Anti-Collision,” IEEE Comm. Letters, vol. 10, no. 3, pp. 144-146, Mar. 2006. [10] A. J. E. M. Janssen and M. J. M. de Jong, “Analysis of Contention Tree Algorithms,” IEEE Trans. on Information Theory, vol. 46, no. 6, pp. 2163-2172, Sep. 2000. [11] D.-Y. Kim, B.-J. Jang, H.-G. Yoon, J.-S. Park, and J.-G. Yook, “Effects of Reader Interference on the RFID Interrogation Range,” in Proc. of European Conference on Wireless Technology(EcWT07), pp.728-732, Oct. 2007. [12] S. M. Yeo, B. W. Jeon, J. H. Bae, Y, J. Moon, Y. J. Kim, H. H. Ron, J. S. Park, Y. R. Seong, H. R. Oh, J. S. Kim, C. W. Park, and G. Y. Choi, “A Channel Allocation Scheme Considering with Collisions and Interferences in Practical UHF RFID applied Communication Fields,” in Proc. of IEEE International Conference on RFID, pp.258-268, Apr. 2008. [13] M. Simon. and D. Divsalar, “Some Interesting Observations for Certain Line Codes With Application to RFID,” IEEE Trans. on Comm., vol. 54, no. 4, pp. 583-586, Apr. 2006.
March 30, 2011
DRAFT
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. IEEE TRANSACTIONS ON MOBILE COMPUTING
27
[14] Y-C. Lai, and C-C. Lin, “A Pair-Resolution Blocking Algorithm on Adaptive Binary Splitting for RFID Tag Identification,” IEEE Comm. Letters, vol. 12, no. 6, pp. 432-434, June 2008. [15] Y.-C. Ko, S. Roy, J. R. Smith, H.-W. Lee, and C.-H. Cho, “An Enhanced RFID Multiple Access Protocol for Fast Inventory,” in Proc. IEEE Globecom, 2007. [16] Y.-C. Ko, S. Roy, J. R. Smith, H.-W. Lee, and C.-H. Cho, “RFID MAC Performance Evaluation Based on ISOIEC 18000-6 Type C,” IEEE Comm. Letters, vol. 12, no. 6, pp. 426-428, June 2008. [17] M. Mohaisen, H. Yoon, and K. Chang, “Radio Transmission Performance of EPCglobal Gen-2 RFID System,” in Proc. of ICACT 2008, pp. 1423-1428, Feb. 2008 [18] C. Jin, S. H. Cho, and K. Y. Jeon, “Performance Evaluation of RFID EPC Gen2 Anti-collision Algorithm in AWGN Environment,” in Proc. of ICMA 2007, pp. 2066-2070 Aug. 2007. [19] P. V. Nikitin and K. Rao, “Theory and Measurement of Backscattering from RFID Tags,” IEEE Antennas Propag. Mag., vol. 48, no. 6, pp. 212-218, Dec. 2006. [20] A. Safarian, A. Shameli, A. Rofougaran, M. Rofougaran, and F. De Flaviis, “RF Identification (RFID) Reader Front Ends With Active Blocker Rejection,” IEEE Trans. on Microwave Theory and Techniques, vol. 57, no. 4, pp. 1-10, 2009. [21] S. M. Yeo, B. W. Jeon, J. H. Bae, Y, J. Moon, Y. J. Kim, H. H. Ron, J. S. Park, Y. R. Seong, H. R. Oh, J. S. Kim, C. W. Park, and G. Y. Choi, “A Channel Allocation Scheme Considering with Collisions and Interferences in Practical UHF RFID applied Communication Fields,” in Proc. of IEEE International Conference on RFID, pp. 258-268, Apr. 2008. [22] K. Kodialam, and T. Nandagopal, “Fast and Reliable Estimation Schemes in RFID Systems,” in Proc. of MobiCom, pp. 322-333, Sep. 2006. [23] G. Maselli, C. Petrioli, and C. Vicari, “Dynamic Tag Estimation for Optimizing Tree Slotted Aloha in RFID Networks,” in Proc. of ACM MSWiM, pp. 315-322, Oct. 2008. [24] T. F. La Porta, G. Maselli, and C. Petrioli, “Anticollision Protocols for Single-Reader RFID Systems: Temporal Analysis and Optimization,” IEEE Transactions on Mobile Computing, vol. 10, no. 2, pp. 267-279, Feb. 2011.
Jongho Park received the B.S. and M.S. degrees in electrical and computer engineering from Sungkyunkwan University, Korea, in 2004 and 2006, respectively. He is currently pursuing his Ph.D. degree in department of electrical and computer engineering at Sungkyunkwan University since March 2006. His research interests include Medium Access Control (MAC) of RFID, wireless communication networks, Wireless LAN, Wireless PAN, and ad hoc networks.
March 30, 2011
DRAFT
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. IEEE TRANSACTIONS ON MOBILE COMPUTING
28
Tae-Jin Lee received his B.S. and M.S. in electronics engineering from Yonsei University, Korea in 1989 and 1991, repectively, and the M.S.E. degree in electrical engineering and computer science from University of Michigan, Ann Arbor, in 1995. He received the Ph.D. degree in electrical and computer engineering from the University of Texas, Austin, in May 1999. In 1999, he joined Corporate R&D Center, Samsung Electronics where he was a senior engineer. Since 2001, he has been an Associate Professor in the School of Information and Communication Engineering at Sungkyunkwan University, Korea. He was a visiting professor in Pennsylvania State University from 2007 to 2008. His research interests include RFID, performance evaluation, resource allocation, Medium Access Control (MAC), and design of communication networks and systems, wireless MAN/LAN/PAN, home/ad-hoc/sensor networks, next generation wireless communication systems, and optical networks. Since 2004, he has been a voting member of IEEE 802.11 WLAN Working Group, and is a memeber of IEEE.
March 30, 2011
DRAFT
