Discrete Power-Based Distance Clustering forAnti-Collision Schemes in RFID Systems

Waleed AlsalihComputer Science Department,

College of Computer and Information Sciences,King Saud University, Saudi Arabia

Email: wsalih@ksu.edu.sa

Abstract—Radio Frequency IDentification (RFID) has enableda wide range of automated tracking and monitoring applications.However, RFID tags share a wireless communication mediumto deliver their information to the RFID reader which resultsin tag collisions and, hence, a significant energy consumptionand delay in the interrogation process. Handling tag collisionsis a challenging task because of the limited capabilities availableto passive RFID tags. In classical anti-collision schemes, theRFID reader interrogates all tags in its range at once using itsmaximum transmission range, which results in many collisionsand wastes energy. In a previous work, we proposed a novelapproach in which the interrogation zone of an RFID readeris divided into a number of ring-shaped clusters, and tagsof different clusters are read separately. We also proposed amethod that finds the optimal clustering scheme assuming anideal setting in which the transmission range of the RFID readercan be tuned with high precision. In this paper, we considera more practical scenario in which the RFID reader has afinite set of discrete transmission ranges rather than continuousones. This suits currently existing commercial RFID readersthat come with configurable output power. We present a delaymathematical analysis for this optimization problem and devisean algorithm that finds the optimal clustering efficiently. Theproposed approach can be integrated with any existing anti-collision scheme to improve its performance and, hence, meetthe demand of large scale RFID applications. Simulation resultsshow that our approach is able to make significant improvementsin saving energy and time by reducing collisions.

I. INTRODUCTION

Radio Frequency IDentification (RFID) technology has agreat potential for monitoring and tracking applications [6].It does not require line-of-sight for communication, it cansurvive harsh environmental conditions, and it is cost- andpower- efficient. This gives RFID an advantage over otheridentification systems (e.g., optical identification systems andbar-code systems). These features make RFID a key enabler ofa true ubiquitous computing environment; it can turn objectsinto a network of mobile nodes which can be identified,tracked, and monitored to trigger actions or to respond torequests.

An RFID system is typically composed of an applicationhost, a reader, and a set of tags. A tag is designed to storebetween 96 bits and 64 K bytes of information. Tags can beeither passive or active. A passive tag has no physical powersource. It harvests energy from the reader’s generated radiowaves, using backscattering modulation [6], [4]; and consumes

that energy in carrying out processing and communicationtasks. Even though passive tags dominate the RFID market,they have very limited functionalities for processing andcommunication. A passive tag processes simple state machinesand has no medium sensing capabilities. On the other hand, anactive tag has a power source and may possess certain sensingcapabilities for temperature or pressure.

Tags that can be interrogated by a certain reader are saidto be within that reader’s interrogation zone. To identifytags in its interrogation zone, an RFID reader broadcasts aquery asking tags in its interrogation zone to respond bysending the information they store. RFID tags then replybe sending their information back to the reader. However,at any point of time, at most one tag should be accessingthe wireless medium in order for the reader to receive andsuccessfully decode the signal. Simultaneous wireless mediumaccess by tags results in tag collisions and, hence, causesa significant delay. To maintain high performance operation,efficient mechanisms for Medium Access Control (MAC) areneeded [4]. Conventional collision avoidance methods, such asCarrier Sense Multiple Access (CSMA), can not be adoptedfor RFID systems especially when passive tags are used.Avoidance mechanisms also require more complicated tagswith sensing and/or synchronization capabilities. Proposals forRFID systems have, therefore, favoured reactive anti-collisionapproaches to deal with collisions.

The focus of our work in this paper is to overcome tags-to-reader collisions. While several schemes have been proposedto deal with tags’ collisions in RFID systems, the interrogationdelay is still a problem for some applications that involvedense and/or fast moving passive tags. This causes immensedata collisions at the reader. Therefore, sophisticated anti-collision algorithms need to be sought after to both meetexisting applications requirements and attract new ones.

A promising direction to avoid a significant amount of colli-sions is to partition the interrogation zone spatially into smallerclusters and to have tags in each cluster being read separately(i.e., one cluster at a time). Any existing anti-collision schemecan be used to resolve collisions in a single cluster. That shouldreduce the number of collisions as it reduces the numberof tags that may respond at the same time. Towards thatobjective, we introduced the Power-based Distance Clustering(PDC) scheme [1]. In PDC, the reader tunes the transmission

13th Annual IEEE Workshop on Wireless Local Networks 2013

978-1-4799-0540-9/13/$31.00 ©2013 IEEE 868

power so that tags within the interrogation zone are clusteredbased on their distance from the reader. Tags which are beinginterrogated within the current cluster will not respond to thereader’s queries for the subsequent clusters. Once a tag is read,it is forced into a sleep mode. The main advantage of thePDC approach is its ability to be seamlessly integrated withany existing scheme to enable higher reading rates that meetthe demand of large scale RFID applications. We studied theviability and efficiency of the PDC scheme in [1]. However,it was not clear how to find the best partitioning scheme.Indeed, having too many clusters may result in many emptyclusters, which is an extra overhead, and few clusters mayresult in having crowded clusters; both situations affect theperformance of that approach significantly. In this paper, weuse mathematical optimization to find the optimal number ofclusters for the PDC scheme.

The contributions of this paper are as follows. We for-mulate an optimization problem for finding clustering planthat minimizes the number of collisions and, hence, reducesthe interrogation delay. We assume that the RFID reader hasfinite, discrete transmission ranges; this applies to today’scommercially available RFID readers such as SkyeTek/M10RFID reader [20]. We present a mathematical delay analysisand devise an efficient algorithm to find the optimal clusteringplan for this problem. Our proposed method has been designedto adapt to different system environments such as the numberof tags and their distribution. We also show the results ofseveral experiments, using the ns-2 simulator [19], whichverify the effectiveness and performance improvements of theoptimized PDC scheme.

The rest of the paper is organized as follows. Section IIsurveys existing literature related to RFID anti-collision proto-cols. In section III, we present a delay analysis and an optimalalgorithm for discrete PDC. Section IV presents the results ofour experiments. Finally, Section V concludes our work.

II. LITERATURE SURVEY

Several anti-collision schemes exist in the literature. Theseschemes are generally divided into two categories: probabilis-tic and deterministic.

A. Deterministic anti-collision schemes

The Binary Search Tree (BST) is a prefix-based schemethat relies on the ability of the RFID reader to determine theposition of the colliding bits [6]. An RFID reader initiallybroadcasts a request with all bits being set to 1. In case ofa collision, the reader makes another request in which themost significant colliding bit is set to 0. Only tags with an IDvalue less than or equal to the broadcasted ID respond. Thisprocess continues until exactly one tag is identified. When atag is read successfully, it goes into a sleep mode. This processcontinues until all tags are read successfully. In the BSTscheme, inquiring requests, sent by the reader, and responses,sent by tags, carry full IDs, which is an extra overhead. Toovercome this, in the dynamic binary search algorithm, whichis proposed in [6], tags send only the least significant bits

starting from the last colliding bit. This reduces the numberof bits transferred between the tags and the reader. However,that does not reduce the number of identification cycles. AnEnhanced BST (EBST) with backtracking is proposed in [14].In the EBST scheme, the reader sends the location of the mostsignificant colliding bit rather than sending a complete ID tocompare with. When a tag receives that location, it respondsonly if it has the bit in that location equal to 0. When a tagis identified successfully, the reader backtracks to previousunsuccessful requests.

The Query Tree (QT) is a memoryless scheme that does notrequire the tag to maintain any inquiring history (e.g., a bitpointer) [9]. During each interrogation cycle the reader broad-casts a query (which is a sequence of bits defining a prefix),and only those tags whose IDs matches the broadcasted prefixsend the remaining of their ID bits back to the reader. If thereader detects a collision, it generates two queries: one with 0and one with 1 appended to the prefix of the last query, andpushes them into a stack to be pulled one by one. The readercontinues to pull queries from the stack and broadcast themuntil all tags have been identified. Several variations of theQT algorithm exist [18], [17], and [15]. The scanning-basedpre-processing scheme starts by scanning the IDs of tags tofind the position of the colliding bits [10]. A bit position map,showing locations of the colliding bits, is broadcasted to alltags. This mechanism condenses a tag ID length and its valueto the length and value of the bit position map. The BBT or theQT protocol is used for arbitrating collisions of the condensedIDs.

B. Probabilistic anti-collision schemes

Probabilistic schemes are variations of the framed ALOHAscheme in which the reader broadcasts the frame length, andeach tag picks a time slot and uses it to transmit its ID. Oneof these schemes is the framed slotted ALOHA in which theframe is divided into a number of slots and each tag randomlypicks a slot and uses it to respond to the reader. The probabilityof a collision is then proportional to the number of tagsusing the same frame. The enhanced dynamic framed slottedALOHA algorithm, which was proposed in [11], adjusts theframe size dynamically according to the number of tags.

Probabilistic schemes can be classified into two sub-categories: static and dynamic. In static schemes, the framehas a fixed number of slots, which is suitable only for low tagdensities. While static schemes are easy to implement, theydo not adapt to different system variables such as the totalnumber of tags, and this affects their performance. Dynamicschemes, on the other hand, tune the frame size to be in linewith values of different system variables (e.g., number of tags).For example, the Dynamic Framed Slotted ALOHA (DFSA)scheme [6] sets the size of the current frame based on statisticsfrom previous frames such as the number of successful slotsand that of the collision ones. The EPC Class-1 Gen-2 Q-Algorithm [5] is another example of a probabilistic, dynamicscheme. The Q-Algorithm maintains a variable Q whose valueis between 1 and 15. The frame size is 2Q. At the beginning

13th Annual IEEE Workshop on Wireless Local Networks 2013

869

of a frame, a value for Q is broadcasted to all tags within theinterrogation zone of a reader, and each tag randomly choosesa slot number from 0 to 2Q − 1 to use it for its transmission.Based on numbers of collision slots, idle slots, and successfulslots; a new value for Q is selected for the next frame. Severaldynamic schemes adjust the frame size based on estimates tothe number of tags (tag count) [2], [12], [13], [16], [11].

Statistical algorithms [7], [3], [8] exploit statistical infor-mation to speed up the interrogation process. The AdaptiveSlotted ALOHA Protocol (ASAP) [8] is an example of suchalgorithms. These statistical algorithms resemble the determin-istic anti-collision category and share the same pros and cons.

C. Power-based Distance Clustering (PDC)

The PDC is a divide-and-conquer anti-collision approach inwhich tags are divided into clusters based on their distance tothe reader. Tags in different clusters are then read separately(one cluster at a time). This has the potential to reduce thenumber of tags which can be concurrently active, lower thecollision probability, and, hence, expedite the interrogationprocess. Partitioning the interrogation zone can be achievedby controlling the reader’s antenna power level. The reflectedpower density and the reader range can be computed using thefollowing formulas [6][17]:

S =λ2.Preader.GReader.GTag

(4π)2R4(1)

and

R =λ

4π.

4

√k.Preader.G2

Reader.G2Tag

Pback(2)

where S is the reflected power density, λ is the wavelengthof the emitted electromagnetic wave, Preader is the powersupplied to the reader’s antenna, R is the distance betweenthe reader and the tag, GReader and GTag are respectively theantenna gain for the reader and the tag, and Pback is the powerreceived by the reader from the tag. It follows from (1) thatthe power density reflected back by the antenna is proportionalto the fourth root of the power transmitted by the reader. Alsoit follows from (2) that the reading distance between the tagand the reader can be changed by changing the power suppliedto the reader’s antenna while maintaining Pback. Hence, theinterrogation range of the reader can be reduced by loweringPreader.

An example of such a partitioning is shown in Fig. 1, wherethe interrogation zone is divided into three clusters: D1, D2,and D3. When the reader sends a request to a particular cluster,only those tags in that particular cluster may respond. Anyanti-collision scheme can be used to resolve collisions withina single cluster.

The scheme is sensitive to the size and the number of clus-ters. A large number of clusters in a sparse tag environmentsyields longer delays as a result of many empty clusters andidle cycles. On the other hand, a small number of clusters mayresult in having too many tags in one cluster which rendersthe scheme ineffective, especially in dense tag environments.

Fig. 1. An example of an interrogation zone of 3 clusters.

The Static PDC, which we proposed in [1], divides theinterrogation zone into clusters based on a fixed stepping valued. This means that the transmission range of a cluster is morethan that of the previous cluster by a fixed value d. Whilevery simple, this scheme does not divide the interrogationzone into equal-size clusters and does not adapt to differenttags densities. However, it has the ability to be integratedwith any anti-collision scheme and it has shown acceptableimprovements as shown in Section IV.

We presented the Optimal PDC (O-PDC) scheme in [21] tofind the optimal number of clusters assuming an ideal RFIDreader whose transmission power can be tuned with a highprecision so that the transmission range can be controlled withhigh accuracy. The O-PDC finds the optimal clustering plan byderiving a formula representing the expected number of cycles(i.e., queries), which is an indicator of delay, as a functionof the number of clusters. Mathematical optimization is thenused to find the optimal value of the number of clusters. Whileoptimal and efficient, the O-PDC is suitable for an ideal settingin which RFID tags are uniformly distributed and the RFIDreader has a precisely tunable transmission range. In this paper,however, we present a near-optimal scheme that suits RFIDreaders with discrete transmission ranges, which applies totoday’s commercial RFID readers (e.g., SkyeTek/M10 RFIDreader [20]).

III. OPTIMAL DISCRETE POWER-BASED DISTANCECLUSTERING (OD-PDC)

In this section we present a dynamic-programming algo-rithm that finds a near-optimal clustering plan for an RFIDsystem in which the reader has discrete transmission ranges,and without any assumptions on the distribution of tags.

A. System model and problem definition

We consider an RFID system consisting of an RFID readerand n passive tags. The interrogation zone is modeled as

13th Annual IEEE Workshop on Wireless Local Networks 2013

870

Inactive transmission range

Active transmission range A cluster

RFID reader RFID tag

Fig. 2. An example of 3 transmission ranges and 2 clusters.

a circle centred at the RFID reader with a radius of Runits. The RFID reader has k discrete transmission rangesTr1, T r2, . . . , T rk; where Tr1 < Tr2 < · · · < Trk = R. Wealso use Tr0 = 0 to denote a virtual transmission range. Eachtransmission range may and may not be used in interrogatingRFID tags; this makes it possible to divide the interrogationzone into up to k clusters. A cluster is the area between twoconsecutive active (i.e., used) transmission ranges as shown inFig. 2. For example, when k = 3, we can have 3 clusters if alltransmission ranges are active. We may also have two clustersif only Tr1 and Tr3 are active, and we may have only onecluster if only Tr3 is active. Note that Trk must be activein order to cover the whole interrogation zone. A clusteringscheme is defined by a subset of transmission ranges to beactive. A clustering scheme defines the set of clusters used inreading all tags in the interrogation zone. Now, the problemcan be defined as follows:

Find a clustering scheme that covers the wholeinterrogation zone, such that the total number ofcycles required to read all tags is minimized.

B. Assumptions

To resolve collisions within a single cluster, any anti-collision scheme can be used. We assume that the distribu-tion of tags is known; yet we do not assume a particulardistribution. In fact all what is required by this scheme is theprobability that a particular cluster is empty of tags; a detaileddistribution does not have to be available. We also assume thatthe total number of tags is known to the reader.

C. Delay analysis

This subsection explains how each cluster decreases orincreases the total number of cycles. A cluster is definedby a pair of transmission ranges that bound it (e.g, thecluster (i, j) is the area between Tri and Trj). It is obviousthat more clusters results in less collisions; we can forexample add more clusters until each cluster has at mostone tag and, hence, there will be no collisions. However,increasing the number of clusters arbitrarily results inhaving many empty clusters and, hence, additional idlecycles. Based on this observation, we should maximize thenumber of non-empty clusters and minimize the numberof empty clusters. While the negative effect of an emptycluster is straight forward (which is one additional idlecycle), quantifying the positive effect of a non-empty clusterdepends heavily on the anti-collision scheme used withinsingle clusters. To make a general optimization scheme, weassume that an empty cluster adds one extra idle cycle anda non-empty cluster saves d cycles. Therefore, the objective is:

MAX αd− β, (3)

where α is the number of non-empty clusters and β is thenumber of empty clusters.

For some anti-collision schemes, finding the exact value ofd is trivial. For example, the binary anti-collision algorithmwith backtracking has d = 1 [14]. On the other hand, it is noteasy to find the exact value of d for some probabilistic anti-collision schemes. Nevertheless, we can use d = 1 to reflectthe objective of maximizing the number of non-empty clustersand minimizing the number of empty clusters regardless of theanti-collision scheme being used to resolve collisions withinsingle clusters.

Each cluster will contribute to the objective function in(3). Let’s give each cluster a rank based on its expectedcontribution to the objective function, and let h(i, j) denotethe rank of a cluster (i, j). h(i, j) can be computed as follows.

h(i, j) = P (n(i,j) > 0) ∗ d− P (n(i,j) = 0), (4)

where n(i,j) is the number of tags located in the cluster (i, j)and P (e) is the probability of an event e, which can becomputed based on the distribution of tags. This is based onthe fact that a non-empty cluster saves d cycles and an emptyone adds an extra idle cycle. When the rank of a clusteris negative, it means that the cluster is expected to add anextra cycle rather than to save cycles. The rank of a clusteringscheme cl, which is denoted by H(cl), is the sum of the ranksof clusters composing cl. The optimal clustering scheme is onewith the maximum rank.

D. Optimal clustering algorithm

In this sub-section we present an algorithm that finds theoptimal clustering scheme (a clustering scheme is defined bya subset of transmission ranges to be active). In general whenthere are k transmission ranges, there will be 2k−1 possible

13th Annual IEEE Workshop on Wireless Local Networks 2013

871

clustering schemes that cover the whole interrogation zone;it is actually equivalent to the number of all subsets of aset of k − 1 elements. We present a dynamic programmingalgorithm to solve this problem. This algorithm finds theclustering scheme whose rank is maximum.

We start with some definitions. Let CL[i] denote the setof all clustering schemes in which Tri is active and Trj isinactive for all j > i. Let M [i] denote the maximum rank of aclustering scheme in CL[i] (i.e., M [i] = MAXcl∈CL[i] H(cl)).It is obvious that when we have k transmission ranges, CL[k]will belong to the optimal clustering scheme. The algorithmis described in Algorithm 1.

Algorithm 1: Optimal Discrete Clustering.Function Find Optimal (n, k)Input: n: the number of tags.

k: the number of transmission ranges.Output: The optimal clustering.M [0] = 0;for i = 1 to k do

Next[i] = 0;M [i] = h(i, 0);for j = 1 to i− 1 do

if M [j] + h(i, j) > M [i] thenM [i] =M [j] + h(i, j);Next[i] = j;

endend

endreturn (M [k], Next[ ]);

IV. EXPERIMENTAL RESULTS

In this section, we present the results of a simulation basedstudy we conducted to evaluate the performance of the PDCschemes. We investigate the performance improvements thatcan be made by the three PDC schemes: the static PDC whichwas presented in [1], the Optimal PDC (O-PDC) which waspresented in [21], and the Optimal Discrete PDC (OD-PDC)which is presented in this paper. We study the performance ofthese schemes by generating random RFID networks (topolo-gies) and comparing the performance of a particular pureclassical anti-collision scheme (i.e., without PDC) with that ofthe PDC scheme integrated with the same classical scheme.We may, for example, compare the performance of the pureQT scheme with that of the O-PDC integrated with the QTscheme (i.e., the O-PDC scheme with the QT scheme beingused to resolve collisions within single clusters).

We extended the ns-2 simulator [19] to implement RFID.We generate random RFID topologies in which tags areuniformly distributed in a grid of 20×20 m2. A single reader islocated at the centre of the grid. The reader has a maximumtransmission range of 10 m. For the PDC scheme, we havea stepping value of 0.5 m. For the OD-PDC, we have 30

500

1000

1500

2000

2500

3000

3500

4000

100 200 300 400 500 600 700 800 900

Num

ber o

f Que

ries

Number of Tags

QTPDC(QT)

O-PDC(QT)OD-PDC(QT)

Fig. 3. Total number of queries with the QT scheme.

transmission ranges. Tags have randomly generated IDs. Ineach RFID network, an anti-collision scheme is applied andits operation continues until all tags are successfully identifiedby the reader. We use several performance metrics to evaluateand compare different schemes. The main metric is the totalnumber of queries (cycles) needed to identify all tags, which isa direct indicator to the performance of different schemes. Theresults are averaged over 20 randomly generated topologies.

Our PDC schemes are compared with three existingschemes, namely the QT scheme, the EBST scheme, and theQ-Algorithm scheme. To compare the performance of our PDCschemes with these existing schemes, we show the result ofthe existing scheme without any clustering and the results ofthe PDC schemes when integrated with that particular existingscheme. This clearly shows the improvements that can beachieved using different PDC schemes. The O-PDC scheme iscompared with the QT and EBST schemes only because theymeet the assumptions made for the O-PDC scheme. On theother hand, the PDC scheme and the OD-PDC scheme do nothave any assumptions on the classical anti-collision schemebeing used to resolve collisions in single clusters and, hence,they are compared with the QT scheme, the EBST scheme,and the Q-Algorithm scheme.

The main indicator for the efficiency of an anti-collisionscheme is the total number of queries needed to identifyall tags. Fig. 3 shows the improvements made by the PDCschemes when integrated with the QT scheme as comparedto the pure QT scheme. The O-PDC scheme saves around35% of the queries. The OD-PDC scheme saves around 20%of th queries. Fig. 4 shows the results of the PDC schemeswhen integrated with the EBST scheme. The O-PDC schemesaves around 25% of the queries. The OD-PDC scheme savesaround 10% of the queries. Fig. 5 shows the results of thePDC schemes when integrated with the Q-Algorithm scheme.The OD-PDC scheme saves around 25% of the queries. Thestatic PDC scheme saves around 10% of the queries.

13th Annual IEEE Workshop on Wireless Local Networks 2013

872

200

400

600

800

1000

1200

1400

1600

1800

100 200 300 400 500 600 700 800 900

Num

ber o

f Que

ries

Number of Tags

EBSTPDC(EBST)

O-PDC(EBST)OD-PDC(EBST)

Fig. 4. Total number of queries with the EBST scheme.

0

500

1000

1500

2000

2500

3000

100 200 300 400 500 600 700 800 900

Num

ber o

f Que

ries

Number of Tags

Q-AlgoPDC(Q-Algo)

OD-PDC(Q-Algo)

Fig. 5. Total number of queries with the Q-Algorithm scheme.

V. CONCLUSION

In this paper we introduce the optimal discrete power-baseddistance clustering scheme for tag collision resolution in RFIDsystems. The main idea in our approach is to divide tagsinto clusters based on their distance to the reader, and toread tags cluster by cluster. Since the number of tags in asingle cluster is less than that in the whole interrogation zone,the likelihood of a collision is reduced. However, the numberof clusters has to be selected carefully as having too manyclusters results in having many empty clusters, which causesa significant number of idle cycles. Assuming a practical RFIDreader with discrete transmission power control, we devise analgorithm that finds a balance between reducing the likelihoodof collisions and reducing the number of empty clusters byfinding an optimal number of clusters. Theoretical analysis andsimulations are presented to verify performance improvementsof our scheme. Moreover, the proposed approach can beintegrated with any existing anti-collision scheme to improveits performance.

ACKNOWLEDGMENT

This research is supported by the National Plan for Scienceand Technology at King Saud University (Project number: 11-INF1500-2).

REFERENCES

[1] K. Ali, H. Hassanein, and A. M. Taha, “RFID anti-collision protocolfor dense passive tag environments,” LCN ’07: Proceedings of the 32ndIEEE Conference on Local Computer Networks, pp. 819-824, 2007.

[2] J. Cha and J. Kim, “Dynamic framed slotted ALOHA algorithms usingfast tag estimation method for RFID system,” CCNC 2006: 3rd IEEEConsumer Communications and Networking Conference, vol. 2, pp. 768-772, 2006.

[3] J. Choi, D. Lee, and H. Lee, “Bi-slotted tree based anti-collision proto-cols for fast tag identification in RFID systems,” IEEE CommunicationsLetters, vol. 10, pp. 861-863, 2006.

[4] D. M. Dobkin, “The RF in RFID: Passive UHF RFID in Practice,”Newnes, 2005.

[5] EPC Radio-Frequency Identity Protocols Class-1 Generation-2 UHFRFID Protocol for Communications at 860 MHz 960 MHz Version1.2.0, EPCglobal, 2008.

[6] K. Finkenzeller, “RFID Handbook: Fundamentals and Applications inContactless Smart Cards and Identification,” John Wiley & Sons, Inc.,2003.

[7] C. Floerkemeier, “Bayesian Transmission Strategy for Framed ALOHABased RFID Protocols,” IEEE International Conference on RFID 2007,pp. 228-235, 2007.

[8] G. Khandelwal, A. Yener, K. Lee, and S. Serbetli, “ASAP : A MACProtocol for Dense and Time Constrained RFID Systems,” ICC ’06:IEEE International Conference on Communications, vol. 9, pp. 4028-4033, 2006.

[9] C. Law, K. Lee, and K. Siu, “Efficient memoryless protocol for tagidentification,” DIALM ’00: Proceedings of the 4th international work-shop on Discrete algorithms and methods for mobile computing andcommunications, pp. 75-84, 2000.

[10] C. Law, K. Lee, and K. Siu, “Scanning-Based Pre-Processing forEnhanced RFID Tag Anti-Collision Protocols,” Proceedings of the2006 International Symposium on Communications and InformationTechnologies, pp. 1207-1211, 2006.

[11] S-R Lee, S-D Joo, and C-W Lee, “An Enhanced Dynamic FramedSlotted ALOHA Algorithm for RFID Tag Identification,” Proceedings ofthe Second Annual International Conference on Mobile and UbiquitousSystems: Networking and Services, pp. 166-174, 2005.

[12] J. Park, M. Y. Chung, and T. Lee, “Identification of RFID Tags inFramed-Slotted ALOHA with Robust Estimation and Binary Selection,”IEEE Communications Letters, vol. 11, pp. 452-454, 2007.

[13] J. Park, M. Y. Chung, and T. Lee, “Identification of RFID Tags inFramed-Slotted ALOHA with Tag Estimation and Binary Splitting,”ICCE ’06: First International Conference on Communications andElectronics, pp. 368-372, 2006.

[14] X.-L. Shi, X.-W. Shi, Q.-L. Huang, and F. Wei, “An Enhanced BinaryAnti-collision Algorithm of Backtracking in RFID System,” Progress InElectromagnetic Research, Vol. 4, pp.263-271, 2000.

[15] D. Shih, P.L. Sun, D.C.Yen and S. M.Huang, “Taxonomy and Survey ofRFID Anti-collision protocols, ” Computer and communications, vol.29,pp. 2150-2166, 2006.

[16] H. Vogt, “Efficient Object Identification with Passive RFID Tags,” Pro-ceedings of the First International Conference on Pervasive Computing,pp. 98-113, 2002.

[17] F. Zhou, C. Chen, D. Jin, C. Huang, and H. Min, “Evaluating and Opti-mizing Power Consumption of Anti-Collision Protocols for Applicationsin RFID Systems,” ISLPED’04: Proceedings of the 2004 InternationalSymposium on Low Power Electronics and Design, pp. 357-362, 2004.

[18] F. Zhou, D. Jin, C. Huang, and M. Hao, “Optimize the power consump-tion of passive electronic tags for anti-collision schemes,” Proceedingsof 5th International Conference on ASIC, vol. 2, pp. 1213-1217 Vol.2,2003.

[19] The network simulator ns-2, Available at http://www.isi.edu/nsnam/ns/[20] www.skyetek.com, 2011.[21] W. Alsalih, K. Ali, and H. Hassanein, “Optimal distance-based clustering

for tag anti-collision in RFID,” LCN ’08: Proceedings of the 33rd IEEEConference on Local Computer Networks, pp. 266-273, 2008.

13th Annual IEEE Workshop on Wireless Local Networks 2013

873