78
Int. J. Ad Hoc and Ubiquitous Computing, Vol. 8, Nos. 1/2, 2011
EBQP: Enhanced Binary Query Protocol for RFID tag collision resolution with progressive population estimation Xinqing Yan* and Guiliang Zhu School of Information Engineering, North China University of Water Resources and Electric Power, Zhengzhou, 450011, China Fax: 86-371-65790034 Fax: 86-371-69127635 E-mail:
[email protected] E-mail:
[email protected] *Corresponding author Abstract: Rapid tag collision resolution is a key for the universal adoption of the RFID systems. Based on the function devised to map the tag identifiers and query strings to real values in [0, 1), an Enhanced Binary Query Protocol (EBQP) is proposed. During collision resolution, according to the number of tags identified and the mapped value of the binary query string, the overall tag population is estimated, and the optimal query string for the reader to broadcast is calculated. This process is performed repeatedly until all tags are identified. Numeric simulations verify that this protocol can achieve better throughput and latency. Keywords: collision resolution; progressive estimation; query tree protocol; RFID. Reference to this paper should be made as follows: Yan, X.Q. and Zhu, G.L. (2011) ‘EBQP: Enhanced Binary Query Protocol for RFID tag collision resolution with progressive population estimation’, Int. J. Ad Hoc and Ubiquitous Computing, Vol. 8, Nos. 1/2, pp.78–84. Biographical notes: X.Q. Yan received his PhD in 2006 from the School of Logistics Engineering in Wuhan University of Science and Technology (WHUT), Wuhan, China. He is presently working as an Associate Professor within the School of Information Engineering, North China University of Water Resource and Electric Power (NCWU), Zhenzhou. Prior to NCWU, he worked as a post-doctor researcher in the State Key Laboratory of Digital Manufacturing Equipment and Technology, Huazhong University of Science and Technology (HUST). His research interest includes ubiquitous computing, wireless sensor networks and RFID system. G.L. Zhu graduated from the Department of Computer Science and Technology in Peking University in 1977. He is presently a Professor within the School of Information Engineering, NCWU, Zhengzhou. His research interest includes image and signal processing, communication system and networking.
1 Introduction Radio Frequency Identification (RFID) enables the rapid tracking and tracing of multiple physical objects with electronic tags attached through contactless and automatic gathering of the unique digital identifier stored in each electronic tag, and bridges the huge gap between the physical object world and the virtual digital space. RFID technology is regarded as the main enabler of the upcoming pervasive computing (Stanford, 2003) and backstone of the “Internet of Things” (Engels et al., 2001). Nowadays, RFID systems have been widely deployed in a lot of application fields. One of key issues that affect the universal adoption of the RFID system is the collision caused by multiple
Copyright © 2011 Inderscience Enterprises Ltd.
tags trying to transmit their identifiers to the reader simultaneously through the air interface, which is a shared wireless communication channel between the reader and tags (Wua et al., 2006). The modulated waveforms from these tags will interfere in the air interface, and result in that what the reader can get is only a collision signal but no useful information. Because of collisions, RFID tag identification process suffers from low accuracy with typical read rate 60–70% (Jeffery et al., 2008) and long delay. Proposed RFID tag collision resolution protocols can be categorised as the probabilistic frame slotted ALOHA based protocols, the deterministic splitting tree based protocols (Shih et al., 2006) and some hybrid protocols (Bonuccelli et al., 2006). Due to the
EBQP: Enhanced Binary Query Protocol for RFID tag collision resolution randomness introduced in them, the protocols in the frame slotted ALOHA based and hybrid categories suffer from the ‘tag starvation’ problem, in which case some tags may continuously collide with other tags and cannot be identified in quite a long time, so protocols in both categories might be unable to resolve all tag collisions occurred in the RFID system and will not be discussed further in this paper. The splitting tree based protocols recursively divide the set of tags which cause collisions into subsets until in each subset, there is at most one tag, so no confliction occurs in the final subsets and all tags can be identified successfully (Kim and Park, 2007; Myung et al., 2006). As a category of the splitting tree based protocols, the query tree based protocols perform the subset division through asking the RFID reader to broadcast some deliberately designed binary query strings, and performs perfectly well in resolving the collisions caused by memoryless passive RFID tags. Due to the simplicity, some query tree based RFID tags collision resolution protocols have been accepted as international or industrial standards. However, the query tree based tag collision resolution protocols suffer from scalability and might be unacceptable in the case of a large number of tags (Bin et al., 2005). In some applications, it is required to gather the identifiers of a large amount of RFID tags accurately and rapidly. For example, when a truck loading hundreds tagged items passes the gate of a warehouse, the readers deployed there will be required to gather all tag identifiers within an instant. Besides, rapid and reliable RFID tag collision resolution protocol is always the pursuit of researchers. In this paper, EBQP is suggested to facilitate the tag collision resolution process, in which a function is devised to map the binary identifiers stored in tags and binary query strings broadcasted by the reader to real values in the interval [0, 1). In a collision resolution cycle of this protocol, based on the amount of tag identifiers already gathered and the value which the binary query string broadcasted by the reader maps to, the overall population of tags within the vicinity of the reader is estimated, and the optimal binary query string for the reader to broadcast in the next frame is calculated. This process is performed repeatedly and progressively until all tag collisions are resolved and all tag identifiers are gathered successfully. Theoretical analysis and numerical simulation verify that this enhanced protocol outperforms other query tree base protocols regarding to the metrics of throughput and latency. The rest of this paper is organised as follows: Section 2 introduces the basic assumptions which we hold in the research and discusses briefly the query tree based RFID tag collision resolution protocols. Section 3 presents the progressive estimation of tag population and the propose the EBQP protocol for RFID tag collision resolution. Section 4 evaluates and compares the performance of this protocol with
79
other query tree based protocols regarding to the metrics of throughput and time delay through numeric simulations and theoretical analysis. Finally in Section 5 we conclude.
2 Background of the research 2.1 Basic assumptions In this paper, as presented in Kaplan and Gulko (1985) on the research of multiple access channel, the following basic assumptions are hold: •
Time is slotted. A collision resolution frame is consisted of a series discrete time slots, and each slot can be either a command slot for the reader to broadcast message or a data slot for the tags to transmit their identifiers.
•
One tag group is interrogated in a data slot.
•
The air interface is ternary, and the interrogation of a tag group reveals the presence of zero, one or two or more tags.
•
The air interface is perfect, and no signal transmission error or lose occurs in it.
But unlike (Kaplan and Gulko, 1985), the air interface is assumed to be instantaneous, rather than delayed, due to the limited distance between tags and reader. However, short time intervals are still needed in the RFID system for the reader or tag to modulate and transmit a bit. Besides, as in other research works on tag collision resolution protocols, it is assumed that the binary identifiers stored in all tags are uniformly distributed and are of the same length. Due to the short interval, in a collision resolution cycle, the tags remain stable, neither newly arrived tags enter nor do existing tags leave the interrogation zone of the reader.
2.2 The query tree based protocols In the query tree based protocols, the collisions caused multiple RFID tags are resolved in a cycle through a series of frames, and in each frame, there are a command slot and a data slot. In frame i, the RFID reader broadcasts a message containing a binary query string qi (with initial value q1 = ε) in the command slot to all tags in its vicinity. Each tag, upon receiving this message and decoding the binary query string qi , compares qi with the binary identifier it stores. If its binary identifier starts with qi , the tag modulates its binary identifier and transmits the waveform back to the reader in the data slot through the air interface, otherwise, it keeps silent and does not respond in the frame. If no RFID tag answer in the frame, the RFID reader can detect the idle state of the data slot and the protocol may continue query with other binary query string in the next frame. Otherwise if only one tag answers in
80
X.Q. Yan and G.L. Zhu
the frame, no collision occurs. In this case the reader can receive the waveform and decode the tag identifier successfully. If two or more tags answer in the frame, their waveforms will collide in the air interface, the reader can detect the collision occurred in the frame and these tags will be divided into two subsets. When the collision caused by two or more tags is detected, the query tree protocol appends 0 to the binary query string qi adopted in the frame, asks the reader to broadcast qi 0 in the next frame and begins to resolve the collisions caused by tags whose binary identifiers start with qi 0 in the following frames. After all collisions caused by tags whose identifiers start with qi 0 are resolved, the query tree protocol appends 1 to the binary query string qi , and begins to identify these tags whose binary identifiers start with qi 1. In such a way, the set of tags which cause collision in a frame is divided into two subsets by the protocol through appending ‘0’ or ‘1’ to the binary query strings qi broadcasted. This ‘divide-and-conquer’ process is performed repeatedly and recursively until in each of the final subsets, there is at most one tag and no collision occurs. The working process of this protocol looks like a binary search tree, hence the query tree based protocol is named, and this protocol is called the standard query tree protocol. Figure 1 shows the collision resolution process of RFID tags with binary identifiers 010, 011, 100 and 111 using the standard query tree protocol. Figure 1 Frames used in resolving the collisions caused by tags with binary identifier 010, 011, 100 and 111
To divide t RFID tags into 2 subsets, the probability that a subset contains k (k ≥ 0 and k ≤ t) tags can be calculated according to the Binomial distribution as 1 t B(k) = t . (1) 2 k For the standard binary query tree protocol to resolve the collision caused by t RFID tags, according to our previous research (Yan et al., 2008), the number of frame Lt needed in a collision resolution cycle can be calculated recursively as t≤1 1 t . (2) Lt = B(i)Li t ≥ 2 1 + 2 i=0
The throughput of the tag collision resolution protocol is calculated as Ltt × 100%. It is reported that the stable throughput of this standard binary query tree protocol is 34.6% (Hush and Wood, 1998). It can be observed that if the frame with binary query string qi results in collision, and the following frame with the binary query string qi 0 results in idle, the broadcast of the binary query string qi 1 can be predetermined to result in collision and thus can be skipped to facilitate the collision resolution without affecting the result, such as the fourth slot in Figure 1. In such case, binary query strings qi 10 can be broadcasted directly following the frame with binary query string qi 0. This protocol is called the modified query tree protocol, and the number of frames needed by this protocol in resolving the collision caused by t tags can be calculated also according to Yan et al. (2008) as t≤1 1 t Lt = . (3) B(i)Li − B(0) t ≥ 2 1 + 2 i=0
The stable throughput achieved in this modified query tree protocol is 37.5% (Hush and Wood, 1998), near 3% improvement over the standard query tree protocol.
3 The enhanced binary query protocol If the population of RFID tags within the vicinity of the reader is known or can be estimated in advance, these tags can be divided into multiple subsets initially using some deliberately calculated binary query strings, some intermediate frames which probably result in collision can be skipped directly, and the overall tag collision resolution cycle can be facilitated. However, in typical RFID systems and applications, the population of tags near the reader is usually unknown in advance. In this section, we suggest a novel method to estimate progressively the population of RFID tags within the vicinity of the reader during a collision resolution cycle, and based on this estimation, an EBQP is proposed to facilitate the gathering of tag identifiers.
3.1 Estimation of the RFID tag population In order to estimate the tag population during a collision resolution cycle, we define a function which maps the binary identifiers stored in the tags and the binary query strings broadcasted by the reader to real values in the interval [0, 1). For any binary string a1 . . . an , the map function f is defined as f (a1 . . . an ) =
n ai i=1
2i
(4)
81
EBQP: Enhanced Binary Query Protocol for RFID tag collision resolution which maps the unique binary identifier stored in each RFID tag to a unique real value in the interval [0, 1). However, multiple binary query strings broadcasted by the reader may be mapped to a common real value, for example, all query strings 01, 010, 0100 are mapped to the same value 0.25. To finish the definition of this function, especially for the empty binary string ε, we define f (ε) = 0. A RFID tag is said to be contained in an interval [p, p + x) if its binary identifier a satisfies p ≤ f (a) < p + x. During a collision resolution cycle, if for a frame, the binary query string to be broadcasted is qi , and f (qi ) = p, then the collisions caused by tags contained in the interval [0, p) should have been resolved, and all binary identifiers stored in these tags should have already been gathered successfully. Suppose that there are t tags within the vicinity of the reader, according to the Bernoulli and Binomial distributions, the probability that k tags are contained in the interval [0, p) can be calculated as t k t−k P (k, p) = (p) (1 − p) . (5) k
exactly one tag) and collision (the subinterval contains two or more tags) can be calculated with Pidle (x) = e−
(10)
Psucc (x) = e−
(11)
Pcoll (x) = 1 − e− − e−
(12)
where = xtˆ, specifies the average number of tags contained in the subinterval with length x. Figure 2 depicts the values of Pidle (x), Psucc (x) and Pcoll (x) as changes from 0 to 2. Figure 2 The values of Pidle (x), Psucc (x), Pcoll (x) as varies from 0 to 2 (see onlie version for colours)
If t is unknown and k > 0, according to the Maximum Likelihood principle, the tag population tˆ can be estimated as k tˆ = M axLikeli p(k) = . p
(6)
Such RFID tag population estimation owns the following important properties: E|tˆ| = t Vartˆ =
(7) t(1 − p) p
(8)
and for any δ > 0, according to the Chebyshev inequality we have tˆ
1−p P − 1 ≥ δ ≤ 2 . (9) t δ tp From equation (9), we can drawn the conclusion that as p increases and approaches to 1, the estimated tag population tˆ approaches the actual tag population t.
3.2 Calculation of the next subinterval to resolve Once the resolution of interval [0, p) results in k (k > 0) tags identified, and the overall tag population is estimated as tˆ = kp , the the next thing for the protocol to do is to calculate and find an appropriate subinterval length x (x > 0 and p + x < 1) so that the subinterval [p, p + x) can be resolved with maximum throughput. For one query over the subinterval [p, p + x), the probabilities that the query results in idle (the subinterval contains no tag), successful (the subinterval contains
To find the appropriate x, a natural thought is to maximise Psucc , so that the probability that the subinterval [p, p + x) contains exactly one tag and can be resolved with just one query is also maximised. This leads to = 1, Psucc (x) = 0.368, and x = = kp . Adopting tˆ this x, the throughput achieved in the resolution of this subinterval [p, p + x) is 36.8%. However, it can also be observed in Figure 2 that with the same probability 36.8%, this subinterval may contain no tag and the corresponding query will result in idle. Idle query will reduce the overall throughput of the protocol significantly. So for the protocol to achieve maximum throughput, we wish to find an optimal value for , so that the function g() used to calculate the frames needed to resolve a tag in the subinterval is minimised. The function g() can be calculated according to the Poisson distribution with ∞ ∞ P (k, )Lk k−1 − g() = k=0 ≈ (13) e Lk k! k=0
where Lk specifies the amount of frames needed to resolve the collisions caused by k tags. Suppose that the modified binary query tree protocol is used to resolve the collision occurred in the subinterval [p, p + x), and Lk can be calculated using equation (3), numerical evaluation shows that when = 1.26, g() is minimised to be about 2.16.
82
X.Q. Yan and G.L. Zhu
It is worthy noticed that this also maximises Psucc (x) − Pidle (x) while keeps Psucc (x) ≥ Pcoll (x), as depicted in Figure 2. For such , the corresponding x for next optimal subinterval [p, p + x) which the protocol should resolve in the next frame can be calculated as x=
1.26p 1.26 = . k t
(14)
3.3 Calculation of the next binary query string From the above discussion, we can draw the conclusion that if after interval [0, p) has been resolved, and k (k > 0) RFID tags have been identified, the optimal subinterval for the protocol to resolve in the next frame should be [p, p + 1.26p k ). However, due to the extreme constraint on computation put on the them, most RFID tags are unable to map their identifiers to real values in the interval [0, 1) nor make real values comparison. This subinterval [p, p + 1.26p k ) must be converted to a binary query string in the protocol so that the reader can broadcast in the command slot of next frame. But quite often for all binary strings mapped in the subinterval [p, p + 1.26p k ), there may not exist a non-empty common binary prefix string. For example, for the subinterval [0.250, 0.725), there does no exist a non-empty common binary prefix string. To overcome this issue and find out the optimal binary query string for the reader to broadcast in the next frame, we suggest to divide the whole interval [0, 1) into 2m subintervals, where m is calculated with ˆ t
k
m = round(log21.26 ) = round(log21.26p ).
(15)
With such division, we get a series of subintervals 1 [0, 21m ), . . . , [ 2jm , j+1 2m ), . . . , [1 − 2m , 1). For all binary strings contained in a subinterval [ 2im , i+1 2m ), there is a non-empty common binary prefix string. If the interval [0, p) has been resolved with one or more tags identified, and the interval [0, 1) is divided into a series of intervals, then the subinterval [ 2im , i+1 2m ) can . The maximum-lengthed be found, where 2im ≤ p < i+1 2m common binary prefix string for the subinterval [p, i+1 2m ) should be adopted by the protocol as the optimal binary query string for the reader to broadcast in the next frame.
3.4 EBQP: the Enhanced Binary Query Protocol According to the discussion presented above, the Enhanced Binary Query Protocol (EBQP) based on the progressive estimation of RFID tag population is proposed, in which after one or more tag identifiers have been collected using the some other query tree protocol, the EBQP protocols maps the binary query string broadcasted in the frame to a real value using the function defined in equation (4). And afterwards, when an interval [0, p) has been resolved successfully, the
overall RFID tag population tˆ is estimated, the whole interval [0, 1) is divided into a series of subintervals, the optimal subinterval [p, 2im ) for the protocol to resolve next is located, and the maximum-lengthed common binary prefix string of this subinterval is adopted in the protocol for the RFID reader to broadcast in the next frame. If the broadcast of this binary query string results in idle or successful, the process is repeated. Otherwise, collision occurs in the subinterval [p, 2im ), the EBQP protocol divides this subinterval into two halves, starts to i copy with the first half [p, 2m+1 ), and asks the reader to broadcast the maximum-lengthed common binary prefix string of this half in the next frame. This process is performed repeatedly and progressively until the frame whose binary query string consists of binary digits ‘1’ only results in idle or successful. In such case, all collisions caused by tags within the vicinity of the reader should have been resolved and all tag identifiers should have been collected successfully.
4 Performance evaluation The performance of RFID tag collision resolution protocol is often evaluated regarding to the metrics of throughput and latency. Throughput reflects the efficient use of the wireless communication channel as well as the efficiency of the protocol, while latency reflects the time needed in a collision resolution cycle. To evaluate the performance of the EBQP protocol, we compare the throughput and time needed in a collision resolution cycle of this protocol with the standard and modified binary query tree protocol. The throughput and latency of a collision resolution cycle for the standard and modified binary query tree protocols can be calculated directly, but for the EBQP protocol proposed in this paper, no direct calculation method has been found at present, so instead, we perform numerical simulations to evaluate its performance. In the numerical simulations, 100 different data sets are adopted, and in each data set there are 1000 96bit randomly generated binary strings representing 1000 tag identifers. The EBQP protocol is implemented and the process is simulated for this protocol to resolve the collisions caused by 1 to 1000 tags in each data set. The amount of frames as well as bits broadcasted by the reader and transmitted by the tags in the collision resolution process for each data set are recorded and averaged at last.
4.1 Throughput of protocols Figure 3 shows the throughput of these protocols achieved through theoretical analysis and numerical simulations. It can be seen that the throughput of the EBQP protocol we proposed is the highest, and
83
EBQP: Enhanced Binary Query Protocol for RFID tag collision resolution as the tag population exceeds 50, the throughput of the standard binary query tree protocol is stabilised to 34.6%, and that of the modified binary query tree protocol is stabilised to 37.5%, as we have discussed in Subsection 2.2. But the throughput of the EBQP protocol varies around 44.9%, with minimum 43.1% and maximum 46.5%. This variance may be due to that the amount of subintervals for the whole interval [0, 1) to divide must be 2m (where m is a non-negative integer) in order to find the optimal binary query string for reader to broadcast in the next frame.
For the modified binary query tree protocol, the corresponding DCt and DDt can be calculated using t≤1 0 t DCt = (18) B(i)DCi − B(0) t ≥ 2 2 + 2
Figure 3 Throughput achieved in the EBQP and both query tree protocols (see online version for colours)
For the EBQP protocol proposed in this paper, the amounts of bits broadcasted by the reader, DCt , and transmitted by all tags, DDt , in a collision resolution cycle are achieved through the numerical simulations. Suppose that in the RFID system, the time needed for the reader to broadcast a bit is τc , and that for a tag to transmit a bit is τd , the time needed for the protocols to resolve all collision caused by t tags can be calculated with
i=0
and
0 td DDt = t B(i)DDi − B(0)td t td t + 2
The data transferred in an identification cycle can be divided into two categories: the bits contained in the binary query string broadcasted by the reader and the bits contained in the answers transmitted by the tags. For the standard binary query tree protocol, to resolve the collision caused by t tags, the amount of bits contained in the query strings broadcasted by the reader in a collision resolution cycle can be calculated recursively as t≤1 0 t DCt = (16) B(i)DCi t ≥ 2 2 + 2
. (19)
t≥2
i=0
T = DDt τc + Lt td τd .
4.2 Identification latency
t=0 t=1
(20)
Without loss of generality, the 915MHz EPCglobal class 1 RFID system is taken as an example in this paper to compare the performance of these protocols, in which td = 120 bits, including 96bit binary identifier stored in a tag and 24bit CRC code, τc = 12.5 µs, and τd = 4 µs. Similar comparison can also be performed to other RFID systems with corresponding values. Figure 4 shows the time needed in a collision resolution cycle for these protocols, from which the conclusion can be drawn that the time needed in a collision resolution cycle by all these protocols is almost linearly related with the tag population, and the EBQP protocol also outperforms the other two protocols. Figure 4 Collision resolution latency in the EBQP and both query tree based protocols (see online version for colours)
i=0
and the amount of bits contained in the answers transmitted by all tags in a collision resolution cycle can be calculated as 0 t=0 t=1 td (17) DDt = t B(i)DDi t ≥ 2 td t + 2 i=0
where td denotes the amount of bits transmitted in an answering by a tag, these bits include its binary identifier and CRC bits.
From Figures 3 and 4, it can be observed that the EBQP protocol outperforms the standard and modified query tree protocols regarding to both metrics of throughput and latency.
84
X.Q. Yan and G.L. Zhu
5 Conclusion and future research
References
RFID technology holds the promise to enable human beings to monitor the world with fine granularity, however efficient and reliable RFID tag collision resolution protocol is a necessity for this promise to make a reality. In this paper, a function is defined to map the binary identifiers of tags and binary query strings broadcasted by the reader to real values in the interval [0, 1), and an EBQP for reliable and efficient tag collision resolution is presented. In a collision resolution cycle of this protocol, based on the amount of tag identifiers already gathered and the binary query string broadcasted by the reader, the overall population of tags within the vicinity of the reader is estimated, and the optimal binary query string for the reader to broadcast in the next frame is calculated. This process is performed repeatedly and progressively until all tag collisions are resolved and all tag identifiers are collected. Theoretical analysis and numeric evaluations verify that this protocol can achieve better performance than the standard and modified binary query tree protocols regarding to both metrics of throughput and latency. Besides, this protocol does not require the tags to remember anything, and can be adopted to resolve the collisions caused by the memoryless passive RFID tags. In the future research, by combining the method introduced in this paper with other recent works, such as that proposed in Sheng et al. (2008), we think that some new RFID tag collision resolution protocols can be proposed and better result may be achieved.
Bin, Z., Mamoru, K. and Masashi, S. (2005) ‘Framed ALOHA for Multiple RFID objects identification’, IEICE Transactions on Communications, Vol. E88CB, No. 3, pp.991–999. Bonuccelli, M.A., Lonetti, F. and Martelli. P.F. (2006) ‘Tree slotted ALOHA: a new protocol for tag identification in RFID networks’, Proceedings of the 2006 International Symposium on World of Wireless, Mobile and Multimedia Networks, IEEE, Buffalo-Niagara Falls, NY, USA, pp.603–608. Engels D.W., Foley, J., Waldrop, J., Sarma, S.E. and Brock, D. (2001) ‘The networked physical world: an automated identication architecture’, Proceedings of the Second IEEE Workshop on Internet Applications, IEEE, San Jose, CA, USA, pp.76–77. Hush, D.R. and Wood. C. (1998) ‘Analysis of tree algorithms for RFID arbitration’, Proceedings of the IEEE International Symposium on Information Theory, Cambridge, MA, USA, p.107. Jeffery, S.R., Franklin, M.J. and Gaorfalakis, M. (2008) ‘An adaptive RFID middleware for supporting metaphysical data independence’, The VLDB Journal, Vol. 17, No. 2, pp.265–289. Kaplan, M.A. and Gulko, E. (1985) ‘Analytic properties of multiple-access trees’, IEEE Transactions on Information Theory, Vol. IT-31, No. 2, pp.255–263. Kim, S.H. and Park, P. (2007) ‘An efficient tree-Based tag anti-Collision protocol for RFID systems’, IEEE Communications Letters, Vol. 11, No. 5, pp.449–451. Myung, J., Lee, W. and Srivastava, J. (2006) ‘Adaptive binary splitting for efficient RFID tag anti-collision’, IEEE Communications Letters, Vol. 10, No. 3, pp.144–146. Sheng, B., Tan, C.C., Li, Q. and Mao, W. (2008) ‘Finding popular categories for RFID tags’, Proceedings of the 9th ACM International Symposium on Mobile Ad Hoc Networking and Computing, ACM, NY, USA, pp.159–168. Shih, D.H., Sun, P.L., Yen, D.C. and Huang, S.M. (2006) ‘Taxonomy and survey of RFID anti-collision protocols’, Computer Communications, Vol. 29, No. 1, pp.2150–2166. Stanford, V. (2003) ‘Pervasive computing goes the last hundred feet with RFID systems’, IEEE Pervasive Computing, Vol. 2, No. 2, pp.1–14. Wua, N.C., Nystroma, M.A., Lina, T.R. and Yu, H.C. (2006) ‘Challenges to global RFID adoption’, Technovation, Vol. 26, No. 12, pp.1317–1323. Yan, X.Q., Yin, Z.P. and Xiong, Y.L. (2008) ‘A comparative study on the splitting tree based protocols for RFID tag identification’, Proceedings of the IEEE International Conference on Communication System, IEEE, Guangzhou, China, pp.332–336.
Acknowledgements The research of this paper is partly sponsored by the Science Foundation for High Level Talents in North China University of Water Resources and Electric Power under Grant No. 200923, the Science Technology Research Program of Henan Province under Grant No. 102102210224 and the Program for Science & Technology Innovation Talents in Universities of Henan Province under Grant No. 2011HASTIT020. The authors of this paper would like to express their thanks to the editors and reviewers for their insightful comments and suggestions. This work was presented in part at the Second IEEE International Workshop on Wireless Network Algorithm and Theory, October 12, 2009, Macau SAR, China.
