Analysis of energy consumption for multiple object identification system with active RFID tags

Transcription

1 WIRELESS COMMUNICATIONS AND MOBILE COMPUTING Wirel. Commun. Mob. Comput. 2008; 8: Published online 18 September 2007 in Wiley InterScience ( Analysis of energy consumption for multiple object identification system with active RFID tags Xu Su 1 and Yang Xiao 2, 1 Computer Science Department, The University of Texas at San Antonio, TX, U.S.A. 2 Computer Science Department, The University of Alabama, AL, U.S.A. Summary Radio frequency identification (RFID) systems are very effective for identifying objects. Existing published works focus on designing efficient collision resolution protocols for the tag identification problem in RFID systems with passive RFID tags. However, advances in low-cost and low-power sensing technologies will make active RFID tags more popular and affordable in the near future. In multiple object identification systems with active tags, the tags are designed for extremely low-cost large-scale applications such that battery replacement is not feasible. This imposes a critical energy-constraint on the communication protocols used in these systems. In this paper, we analyze energy consumption and identification times for several protocols. The objective is to decrease energy consumption of tags by reducing both the total identification time and the total active time. Copyright 2007 John Wiley & Sons, Ltd. KEY WORDS: radio frequency identification; anti-collision algorithms; conflict resolution; energy consumption; identification time 1. Introduction Radio Frequency Identification (RFID) systems can identify an object or a person using wireless transmission. An RFID system consists of RFID tags (also called transponders) and readers (also called interrogators). Readers broadcast queries to tags in their wireless transmission ranges for information contained in tags, and tags reply with required information such as identification (ID) numbers. RFID systems have been gaining more popularity in areas such as supply chain management, automated identification systems, and any place requiring identification of products or people. RFID systems can be divided into three classes: passive RFID systems, active RFID systems, and hybrid RFID systems. In a passive RFID system, all tags are passive tags which have no battery. While in an active RFID system, all tags are active tags which have their own batteries. Hybrid RFID systems consist of both passive and active tags. In all RFID systems, the reader must be able to identify tags as quickly as possible. However, signals of readers or tags collide because they communicate over the shared wireless channel and collisions interfere with fast identification. Therefore, anti-collision protocols which reduce collisions and identify tags regardless of the occurrence of collisions are required. *Correspondence to: Yang Xiao, Department of Computer Science, The University of Alabama, 101 Houser Hall, Box , Tuscaloosa, AL U.S.A. yangxiao@ieee.org Copyright 2007 John Wiley & Sons, Ltd.

2 954 X. SU AND Y. XIAO Collision can be divided into reader collisions and tag collisions [1]. Reader collisions occur when close readers communicate at the same time on the same frequency, or when neighboring readers attempt to communicate with the same tag at the same time. Tag collisions occur when more than one tag tries to respond to a reader at the same time and make the reader recognize no tags. Reader collisions can be easily resolved because RFID readers can detect collisions and communicate with one another. In Reference [2], Engels and Sarma define Reader Collision Problem and present several graph coloring formulations. Waldrop et al. [3] propose a simple, distributed, on-line algorithm: Colorwave algorithm for the Reader Collision Problem. Kim et al. [4] propose an adaptive and dynamic localized algorithm for hierarchical clustering in RFID networks. It minimizes the overlapping areas among clusters and the reader collision by regulating an RFID reader s cluster radius dynamically. In passive RFID systems, current tag anti-collision protocols can be categorized into ALOHA based protocols and tree based protocols. ALOHA based tag anti-collision protocols such as pure ALOHA, slotted ALOHA, and framed ALOHA [5 7] reduce the probability of the occurrence of tag collisions, in which tags respond at a distinct time. In a pure ALOHA protocol, upon receiving a query from the reader, each tag randomly selects its response time and sends its response message to the reader. In a slotted ALOHA protocol, time is slotted and each tag can respond only at the beginning of a time slot. A framed ALOHA protocol is a variant of slotted ALOHA protocol. After the reader has sent a query to tags, it waits a certain amount of time for their answers. This time frame is divided into a number of slots that can be occupied by tags and used for sending their answers so that it reduces collisions. Framed ALOHA shows the best performance among the ALOHA based protocols, followed by slotted ALOHA. However, ALOHA based protocols cannot perfectly prevent collisions. In addition, they have the serious problem that a tag may not be identified for a long time the so called tag starvation. Tree based tag anti-collision protocols such as the binary tree protocol [5] and its variant the query tree protocol [8,9] do not cause tag starvation though they have relatively long identification delay. They split the group of colliding tags into two or more subgroups until the reader recognizes the IDs of tags without collisions. In a passive RFID system, the objective is to reduce the identification time and simplify the implementation. While in the active RFID system, energy conservation for active tags is another critical system parameter for communications. Active tags are designed for extremely low-cost large-scale applications, such that battery replacement is not feasible. This imposes a critical energy-constraint on the communication protocols used in these systems, so that the total time a tag needs to be active for transmitting information should be minimized. At the same time, identification delay must be small too. Thus this gives rise to a trade-off between energy consumption and identification time. To the knowledge of the authors, existing works focus only on passive RFID systems. The major advantages of an active RFID tag over a passive tag are: (i) it has longer read range; (ii) it has larger storage capacity; (iii) data can be sent at designated times. Active RFID tags may have all or some of the following features: the capability to perform independent monitoring and control, the capability of initiating communications; the capability of performing diagnostics; the highest data bandwidth; active RFID tags may even be equipped with autonomous networking and the tags autonomously determine the best communication path. Active tags have been used in many different areas such as manufacturing, transportation and automotive, air industry, healthcare, military sector, logistics, and retails. For example, active tags are used for tracking of cargo, proof of tamper evidence on containers, long range asset tracking, peer to peer networks for sensing, real time location of assets and people, and remote identification in difficult environments. Advances in low-cost and low-power sensing technologies will make active RFID tags more popular and affordable in the near future. Therefore, in this paper, we focus on RFID systems with active tags. Since the main difference between active tags and passive ones is that active RFID tags have their own internal power source, and existing anti-collision protocols in passive RFID systems can still be used in active RFID systems. In this paper, we apply existing anti-collision protocols in passive RFID systems to active RFID systems, and analyze their energy-consumption and identification times. The rest of the paper is organized as follows. Section 2 gives the model for multiple object identification in active RFID systems. Section 3 introduces different

3 ANALYSIS OF ENERGY CONSUMPTION 955 anti-collision protocols for active RFID systems and analyzes their power consumption and identification times. Section 4 presents performance analysis of these protocols. Finally, we conclude the paper in Section Model Description In a multiple object identification system with active RFID tags, there are one reader and n active tags. The reader is a powerful entity with abundant memory and computational power. Active tags are limited in memory and computation power. The process, starting from time that a reader sends a query to the time that the reader receives messages from all tags within its radio range is called an identification process. Before formulating the problem, we introduce some notations to be used, as listed in Figure 1. There is a single communication channel between the reader and tags. However, tags are not able to exchange messages among each other. The reader can broadcast messages to tags. After receiving a message, each tag can send a response back to the reader optionally. If only one tag responds, the reader receives the message successfully. But if more than one tag responds, their messages would collide on the communication channel. Thus the reader detects a collision on the channel and cannot receive message from any of them. We will not take into consideration the capture effect by which a tag s message may be able to be recognized by the reader inspite of a collision. The capture effect is quite common if tags are placed close to each other. This means practically that data, which would normally be lost due to the collision, can be read, and thus the system performance rises. However, it seems that whenever such a weak collision between the same two tags occurs, one of them always wins and the data from the other one is lost. Therefore, the influence of the capture effect is only minimal and seems not to have great impact on the performance. Each active tag can be in one of the two states: listening state and transmitting state. Normally an active tag is in listening state and listens the query from the reader. When it is required to send a response back to the reader, it switches to transmitting state. Listening state consists of two states: receiving state and idle state. Since active tags consume much more power in transmitting state than in receiving and idle states, we do not differentiate the receiving state and idle state. For simplicity, we consider them as a single state listening state. We can reduce the power consumption of an active tag in an identification process by decreasing the number of messages that it transmits. This problem is equivalent to the collision minimization problem. At the same time, the identification time for all the tags within the radio range of the reader should be as short as possible. Otherwise, the system is useless if the identification time is too long. Without loss of generality, we assume that each message sent by tags has the same size, and the time needed for a tag to send a message is the same, denoted as T trans. The power consumption rate for sending each message is fixed and represented by P trans as given in Figure 1. Also we assume that the process time for reader and tags are negligible. In a multiple object identification system with one reader and n active RFID tags, the identification time, T total, is the time elapsed from the time the query is sent from the reader to the time all tags are identified by the reader. The total power consumption of all tags during transmitting state in an identification process, P total, is given by n P total = P trans T trans r i (1) i=1 Fig. 1. Notations used in the problem formulation. where r i is the number of messages sent by tag i in an identification process. The purpose of the anti-collision protocols for passive RFID systems is to minimize T total. However, in the active RFID system, energy conservation for active tags is another critical system parameter. If the objective is to minimize P total, the solution can be

4 956 X. SU AND Y. XIAO obtained easily, using min{p total }=np trans T trans (2) min{p total } can be obtained using protocols similar to query tree protocol (see Section 3). The reader broadcasts all possible tags IDs one by one in the queries to all tags such that no collision occurs, i.e., only 0 or 1 tag will respond at each time. However, identification time T total = (D t2r + D r2t )2 k is too long, where k is the length of the ID string. Therefore, the protocol is useless. As shown above, minimization of P total only is not desired, so we consider reduction of both P total and T total at the same time and define a protocol effectiveness function, E, as given in Equation (3). The objective is to find the most effective protocol which minimizes E: E = αp total + (1 α)t total P idle, where 0 α 1 (3) α is used to adjust weights on P total and T total. There are three special cases. When α = 0, the objective is to find a protocol which minimizes the identification time T total only. When α = 1, the objective is to find a protocol which minimizes the P total only (the solution is given in Equation (2)). When α = 0.5, the objective is to find a protocol which minimizes the total energy-consumption for all tags in an identification process, including both energy-consumption during transmitting state and during listening state. 3. Protocol Analysis In tag anti-collision protocols, at the beginning of an identification process, all tags are in listening state. When a tag needs to send a response message to the reader, it switches to the transmitting state. After the tag finishes sending response message, it goes back to the listening state ALOHA-based Protocol In ALOHA-based protocols, if a tag receives the confirmation message back from the reader which says that the reader has received the message from the tag successfully, it will not send messages any more during current identification process until a new identification process begins. If a tag does not receive the confirmation message from the reader, it has to resend the response message at different times in the following different protocols Pure ALOHA protocol In the pure ALOHA protocol, after a tag transmits its ID, it waits to see whether transmission is acknowledged by the reader; no response within a specified period of time (D r2t + D t2r ) indicates a collision with other transmission, or the reader will send an explicit message indicating the collision. If the presence of a collision is determined by the tag, it retransmits after some random wait time. The random time is different for different tags, thereby avoiding collision during the retransmission cycle. The worst case collision period is two times the length of each message (2 D t2r ), assuming all messages from the tags to the reader to be of equal length Slotted ALOHA protocol Slotted ALOHA improves on pure ALOHA protocol by cutting the vulnerable period for message collision by half. Time is slotted and packets must be transmitted within a slot (D t2r ). If the tag needs to send a message, it waits until the beginning of the next slot before sending. The tag listens to the broadcast and checks if the message was transmitted successfully. If there was a collision, indicated by explicit collision indication message or the absence of acknowledgment message from the reader within a pre-specified period of time (D r2t + D t2r ), the tag waits a random number of slots and attempts to send it again Framed ALOHA protocol In the framed ALOHA protocol [7], after the reader has sent its query to the tags, it waits a certain amount of time for their answers. This time frame is divided into a number of time slots that can be occupied by tags and used for sending their answers. When multiple tags use the same slot, a collision occurs and data get lost. The tag reading cycle consists of two steps. In the first step, the reader broadcasts a query for data, which contains the frame size m and a random value Rnd. In the second step, tags compute a random response slot number (say k, where 1 k m), using the frame size m and the random value Rnd as parameters, and correspond at slot k to avoid collisions occurring repeatedly. If the tag determines that it has sent the response message to the reader successfully, it will switch to listening state

5 ANALYSIS OF ENERGY CONSUMPTION 957 and will not respond during the current identification process. If there was a collision, in the next time frame the tag will compute a random response slot number and attempt to send it again Contention Tree Protocol In the contention tree protocol, every contention frame is divided into a number of time slots. Let us assume that there are n tags which want to send response messages after they receive the reader s query message. During the first contention frame, i.e., the frame at the root of the tree, each of the n tags picks at random a number (say k) between 1 and m with equal probabilities and transmits its response message during the kth contention slot, where m is the frame size (number of slots in the frame). After completion of the contention frame and pre-specified period of time, each tag knows whether its message has been successfully received by the reader. If not, a new contention frame is assigned to all tags that caused the collision during the particular slot. Therefore, if there were collisions in all contention slots, m new contention frames would become available. This leads to the formation of a tree with nodal degree m. The expansion of the tree stops at either empty or successful slots. Upon completion of the tree protocol, all the n tags have successfully sent their response messages. Let L n denote the number of contention frames required to complete the tree. Let d n denote the number of levels (number of time frames) required for a random tag to send response message successfully. Let D n denote the number of levels required to complete the tree. The expectation of L n can be expressed recursively according to References [10,11] L n = 1 (n = 1) m m 1 (n = 2) (1 m 1 n ) 1 [ 1 + n 1 k=2 (n 3) ( n k) (m 1) n k m n 1 L k ] (4) According to formulas given in Equation (5), we can compute the expectation value of the total identification time, T total. T total = L n (md t2r + D r2t ) nm (md t2r + D r2t ) (8) Let P tag denote the power consumption of a random tag (during transmitting state) to have successful contention. The expectation value of P tag can be written as P tag = d n P trans T trans log m (n 1)P trans T trans (9) Let max(p tag ) denote the maximum power consumption of the tag among all tags (during transmitting state) to have successful contention. The expectation value of max(p tag )isgivenby max(p tag ) = D n P trans T trans 2 log m np trans T trans (10) And the expected power consumption of all tags during transmitting state in one identification process can be denoted as P total = np tag n log m (n 1)P trans T trans (11) The expected value of effectiveness (E) of the protocol can be written as E = αp total + (1 α)t total P idle αn log m (n 1)P trans T trans + (1 α) nm (md t2r + D r2t )P idle = αn ln(n 1)P transt trans + (1 α)nm(md t2r + D r2t )P idle where 0 α 1 (12) The expectation values for L n, d n, and D n can be calculated [12] as follows: L n n (5) d n log m (n 1) (6) Given the total number of tags n and value of α, we can find a value for m which minimizes E. Let us rewrite E as a function of m as follows: E(m) = αn ln(n 1)P transt trans D n 2 log m n (7) + (1 α)nm(md t2r + D r2t )P idle (13)

6 958 X. SU AND Y. XIAO Let m be the value for m that minimizes E(m), as given by E(m ) = min {E(m)} (14) m Z,m 2 Both P total and T total are monotonic functions. When m increases, P total decreases and T total increases more dramatically. There exists a unique value m = m such that E(m) has the minimal value. Since m is an integer and m 2, intuitively we can find m by searching from m = 2 and increasing the m by 1 each time. We stop searching when E(m + 1) > E(m) holds at the first time, then set m = m and E(m ) = E(m) isthe minimal value. Search space of m from 2 to + is too huge. We limit the search space of m and give a small search space [2,m B ] such that m [2,m B ], as given in Lemma 1. We can also start searching from m = m B and decrease m by 1 each time to find the optimal value m = m which minimizes E(m). Lemma 1. Given the total number of tags (n) and α, m, which minimizes E(m) is in the range [2,m B ], where the upper bound ( ln(n 1)αP trans T trans m B = max 10, (1 α)d t2r P idle (2 ln 10 1) Proof is given in Appendix A Query-Tree Protocol The Query-Tree (QT) protocol [8] consists of queries and responses. In each round, the reader asks tags whether any of their IDs contains a certain prefix. If more than one tag answers, then the reader knows that there are at least two tags that have the same prefix. When a prefix matches a tag uniquely, that tag is identified. Therefore, by extending the prefixes until only one tag s ID matches, the algorithm can identify all tags. The query tree describes the complete dialog between the reader and the tags in an execution of the QT protocol, as follows. Suppose in an execution of the QT protocol, the reader has sent the set Q of query strings. The query tree of the QT protocol execution is defined recursively. If a tag receives the confirmation message back from the reader which says that the reader has received the message from the tag successfully, it will not send messages any more during the current identification process until a new identification process is initiated. If a tag does not receive the confirmation message from the reader, it has to resend the response message at different times determined by the query tree. The root of the query tree corresponds to the query string ε (empty string). If the query tree node x corresponds to the query string q, and both q0,q1 Q, then leftchild(x) and rightchild(x) are query tree nodes that correspond to the query strings q0 and q1 respectively. Otherwise, both leftchild(x) and rightchild(x) equal NIL. The above definition implies that an internal node of a query tree corresponds to a query string that results in a collision. A leaf corresponds to a query string that results in either no reply or a response from exactly one tag. The query tree is similar to a special case of contention tree protocol with frame size m = 2. But the sizes of messages sent from both the reader and the tags are different from those in the contention tree protocol. Each tag randomly chooses one of the two slots in contention tree protocol, and tags are assigned to different sub-trees according to their prefix bit value in QT protocol. Another difference is that every time after the reader sends query, it waits for (D r2t + D t2r ) time and sends another query with new prefix until all tags are identified. For simplicity, we also assume the time needed for a tag to send a message is same, denoted as D trans. Then, the expected total identification time is obtained as follows: T total = L n 2(D t2r + D r2t ) m=2 n 2(D t2r + D r2t ) m=2 = 2n ln 2 (D t2r + D r2t ) (15) The expected power consumption for a random tag in the transmitting state is given by P tag = d n P trans D trans m=2 log m (n 1)P trans T trans m=2 = log 2 (n 1)P trans T trans (16) The expected power consumption of all tags during transmitting state in one identification process can be denoted as P total = np tag n log 2 (n 1)P trans T trans (17)

7 ANALYSIS OF ENERGY CONSUMPTION 959 The expectation of total identification time (D) QT m = 2 m = 6 m = 10 m = Total number of tags: n Fig. 2. The expectation of total identification time T total. The expected maximum power consumption for a tag among all the tags in the transmitting state is given by max(p tag ) = D n P trans T trans m=2 2 log m np trans T trans m=2 4. Performance Evaluation = 2 log 2 np trans T trans (18) In this section, we give performance analysis of contention tree protocols. Without loss of generality, we assume the time for a message sent from the reader to a tag is the same as that for a message sent from a tag to the reader, i.e., D r2t = D t2r = D. Also we let P trans T trans = J. Figure 2 gives the expected total identification time (T total ) for n tags in an identification process. When the total number of tags increases, T total increases too. Given the total number of tags, when frame size (m) increases, T total decreases accordingly because less collisions occur. The expected total identification time for QT protocol is slightly less than that for contention tree protocol with frame size m = 2. Figures 3 and 4 give the expectation value of power consumption of a random tag and the expected maximum power consumption of a tag among all tags (during transmitting state) to have successful contention in an identification process, i.e., P tag and max(p tag ), respectively. When the number of tags n increases, both P tag and max(p tag ) increase because more tags cause more collisions and the average The expectation of total power consumption of all tags (J) m = 2, QT m = 6 m = 10 m = Total number of tags: n Fig. 3. The expectation of power consumption of a random tag P tag. number of messages sent by a tag to have a successful contention increases correspondingly. The P tag and max(p tag ) for Query-Tree protocol are same as those for contention tree protocol with frame size m = 2. Figure 5 gives the expectation of total power consumption of all tags (during transmitting state) in an identification process. The more tags are in the system, the more is the power is consumed correspondingly. Given a fixed number of tags in a system, when the frame size (m) increases, the power consumption of all tags during transmitting state decreases due to less collision. When the frame size (m) changes from 2 to 6, the total power consumption of all tags decreases a lot. Total power consumption of all tags does not change too much when the frame size is large enough (e.g., m from 6 to 14). The total power consumption of all tags The expectation of max power consumption of a tag (J) Total number of tags: n m = 2, QT m = 6 m = 10 m = 14 Fig. 4. The expectation of maximum power consumption of a tag max(p tag ).

8 960 X. SU AND Y. XIAO Probability that at least one tag sends message successfully in one frame n = 15 n = 25 n = 45 n = Frame size: m Fig. 5. The expectation of total power consumption of all tags P total. P total for Query-Tree protocol is the same as that for contention tree protocol with frame size m = 2. Given the total number of tags (n) and weight α, we can find m that minimizes E(m) by searching m. Figure 6 gives the expectation of effectiveness of protocols E(m), where α = 0.5 and DP idle = 0.01P trans D trans = 0.01J. When the total number of tags is small (e.g., 100), frame size m does not affect the performance too much. When the total number of tags is large (e.g., 2000), increasing the frame size does improve performance. However, frame size does not help any more if it is large enough. The optimal frame size m and the upper bound m B are shown in the legend of Figure 6. We can see that the upper bound given in Lemma 1 is very close to m. In order to find the optimal frame size m, we can either search from m = 2 and increase m by 1 each time, or search from m = m B and decrease m by 1 each time. Given the upper bound The expectation of effectiveness of protocols (J) x 10 4 n = 100 (m * = 10, m B = 12) n = 400 (m * = 11, m B = 13) n = 800 (m * = 12, m B = 14) n = 2000 (m * = 12, m B = 15) Frame size: m Fig. 6. The expectation of effectiveness of protocols: E. m B, it will save search time a lot by searching from m = m B back to m = 2. Figure 6 shows that we only need to search maximum 5 times (n = 2000) to find m starting from frame size m = m B back to m = Conclusions Radio Frequency Identification (RFID) system is an automatic identification system. Tag anti-collision protocols such as ALOHA, slotted ALOHA, framed ALOHA, content tree protocol, and query tree protocols are well studied in passive RFID systems. However, till now, anti-collision protocols for active RFID systems are not investigated. We believe that advances in low-cost, low-power sensing technologies will make active tags more popular and affordable in the near future. In the active RFID system, energy conservation for active tags is another critical system parameter besides identification time for communications. The active tags are designed for extremely lowcost large-scale applications, such that the replacement of batteries is not feasible. This imposes a critical energy-constraint on the communications protocols used in these systems, so that the total time a tag needs to be active for transmitting information should be as small as possible. At the same time identification delay must be small too. Thus this gives rise to a trade-off between energy consumption and delay. In this paper, we applied existing protocols in passive RFID systems into active RFID systems. We defined a new objective function, effectiveness of a protocol (E), which considers reduction of both total power consumption of all tags and total identification time at the same time. We compare and analyze the power consumption and identification time of these protocols in active RFID systems. Acknowledgment This work was partially supported by the US National Science Foundation (NSF) under grants CNS and CNS Appendix A: Proof of Lemma 1 Let a = αnp trans T trans, b = (1 α)nd t2r P idle, and c = (1 α)nd r2t P idle. then ( ) a ln(n 1) m B = max 10, b(2 ln 10 1)

9 ANALYSIS OF ENERGY CONSUMPTION 961 E(m) = a ln(n 1) + bm2 + cm Let m be the value of m which minimizes E(m), then m is the minimal value such that E(m) E(m ) > 0 We have to show that m B m. m >m if m 10, m B m holds obviously. if m > 10, according to the m s definition, it suffices to prove E(m) > E(m B ) m >m B 10 If m>m B holds, then a ln(n 1) m>m B = b(2 ln 10 1) and it can be rephrased as 2b ln 10m 2 bm 2 >aln(n 1) since >ln 10 and bm +cm cm > 0, it holds that 2b m 2 + bm + cm (bm 2 + cm) >aln(n 1) using the fact 0 < ln (1 + 1 m )m < 1, we have We can rephrase it as 2b m 2 + bm + cm ( (bm 2 + cm)ln ) m m ( >aln(n 1) ln ) m m 2b m 2 + bm + cm (bm 2 + cm)m [ln(m + 1) ] >aln(n 1)m [ln(m + 1) ] Divide m on both sides of the inequality, then 2b m+ b + c (bm 2 + cm) [ln(m + 1) ] >aln(n 1) [ln(m + 1) ] It can be rephrased as [ a ln(n 1) + b(m + 1) 2 + c(m + 1) ] [ a ln(n 1) + bm 2 ln(m + 1) + cm ] ln(m + 1) > 0 Divide ln(m)ln(m + 1) on both sides of the inequality, gives a ln(n 1) + b(m + 1) 2 + c(m + 1) ln(m + 1) a ln(n 1) + bm2 + cm > 0 then we have m B m. References E(m + 1) E(m) > 0 1. Sarma S, Brock D, Engels D. Radio frequency identification and the electronic product code. IEEE Micro 2001; 21(6): Engels DW, Sarma SE. The reader collision problem. In IEEE International Conference on Systems, Man and Cybernetics, Vol. 3, October Waldrop J, Engels DW, Sarma SE. Colorwave: an anticollision algorithms for the reader collision problem. IEEE International Conference on Communications, Vol. 2, May 2003, pp Kim J, Lee W, Yu J, Myung J, Kim E, Lee C. Effect of localized optimal clustering for reader anti-collision in rfid networks: fairness aspects to the readers, In Proceedings of International Conference on Computer Communications and Networks (ICCCN), 2005, pp Hush DR, Wood C. Analysis of tree algorithm for rfid arbitration. In IEEE International Symposium on Information Theory, Chan S, Hinrich H, Kandlur D, Krishna A. U.S. patent: multiple item radio frequency tag identification protocol. Patent number: US , Vogt H. Multiple object identification with passive rfid tags. IEEE International Conference on Systems, Man and Cybernetics, Vol. 3, October Law C, Lee K, Siu K-Y. Efficient memoryless protocol for tag identification. In Proceedings of the 4th International Workshop on Discrete Algorithm and Methods for Mobile Computing and Communications, August 2000, pp Zhou F, Chen C, Jin D, Huang C, Min H. Evaluating and optimizing power consumption of anticollision protocols for applications in rfid systems. In Proceedings of the International Symposium on Low Power Electronics and Design, August 2004, pp Massey JL. Collision resolution algorithms and random access algorithms. In Multi-User Communication Systems, Number 265 in CISM Courses and Lectures, Longo G (ed.). 1981; Huang J-C, Berger T. Delay analysis of interval searching contention resolution algorithms. IEEE Transactions on Information Theory 1985; 31(2): Janssen AJEM, de Jong MJM. Analysis of contention tree algorithms. IEEE Transactions on Information Theory 2000; 46(6):

10 962 X. SU AND Y. XIAO Authors Biographies Xu Su is currently pursuing the Ph.D. degree in Computer Science at the University of Texas at San Antonio. He received the B.E. degree in Computer Science and Application from Yanshan University, Qinhuangdao, China, in 1997, and the M.S. degree in Computer Science from Wright State University, Dayton, Ohio, USA in He was a research assistant from 2001 to 2002 at Wright State University. He was a teaching assistant from 2003 to 2005, and has been a research assistant since 2005, at the University of Texas at San Antonio. He received National Achievement Award in Defense Science and Technology by Commission of Science Technology and Industry for National Defense, Second Class, China, in He also received National Award in Science and Technology Progress by Ministry of Education, Second Class, China, in His research interests includes mobile ad hoc networks, wireless sensor networks, wireless network security, and networked multimedia. He is a student member of IEEE and ACM. Yang Xiao worked at Micro Linear as a Medium Access Control (MAC) architect involving the IEEE standard enhancement work before he joined Department of Computer Science at The University of Memphis in Dr. Xiao is currently with Department of Computer Science at The University of Alabama. He was a voting member of IEEE Working Group from 2001 to He is an IEEE senior member. He is a member of American Telemedicine Association. He currently serves as Editor-in-Chief for International Journal of Security and Networks (IJSN), International Journal of Sensor Networks (IJSNet), and International Journal of Telemedicine and Applications (IJTA). He serves as a referee/reviewer for many funding agencies, as well as a panelist for NSF and a member of Canada Foundation for Innovation (CFI) s Telecommunications expert committee. He serves as TPC for more than 90 conferences such as INFOCOM, ICDCS, ICC, GLOBECOM, WCNC, etc. His research areas are security, telemedicine, sensor networks, and wireless networks. He serves as an associate editor or on editorial boards for several journals (such as IEEE Transactions on Vehicular Technology, etc.). He has published more than 200 papers in major journals (more than 50 in various IEEE Journals/magazines), refereed conference proceedings, book chapters related to these research areas. Dr. Xiao s research has been supported by the US National Science Foundation (NSF).