Kuo-Hui Yeh1, N.W. Lo2, Yingjiu Li3 and Enrico Winata4 Department of Information Management
National Taiwan University of Science and Technology, Taipei, Taiwan R.O.C.
[email protected]1; [email protected]2;[email protected]4 School of Information Systems
Singapore Management University, Singapore 178902 [email protected]3
Abstract—In this paper, an Adaptive n-Resolution scheme (AnR) is presented for RFID tag identification. The proposed protocol first divides target tags into two categories: staying tags and new arriving tags. Then two distinct identification processes are in-voked in sequence to recognize tags in the two tag groups. Based on our theoretical analyses and simulation experiments, AnR reduces at least 10% on communication overhead and 10% to 40% identification delay in comparison with Pair-Resolution Blocking algorithm [1-2].
Keywords-RFID, anti-collision, tag identification, resolution.
I. INTRODUCTION
For an RFID tag identification operation, how to improve the identification throughput is the main concern when design-ing the corresponddesign-ing anti-collision algorithm [1-4]. Recently, Myung et al. [3-4] proposed two tree-based anti-collision algo-rithms, called Adaptive Query Splitting (AQS) and Adaptive Binary Splitting (ABS), to improve tag identification efficiency of the Query Tree and Binary Tree protocols, respectively. By maintaining information of staying tags from the last identifica-tion session, AQS and ABS can quickly identify the staying tags and achieve higher throughput when identifying the new arriving tags at the current identification session. Later, a per-formance-enhanced version of ABS, named as Pair-Resolution Blocking algorithm (PRB), is introduced by Lai et al. [1-2] to increase tag identification throughput. In PRB, a blocking tech-nique is used to divide the responding sequence (phase) by tag groups. In every session, each tag has to compare the received inquiry reader ID with its stored reader ID in the previous ses-sion to divide tags into two groups. If both IDs are identical with each other, this tag belongs to the group of staying tags.
Otherwise, it is one of the new arriving tags. This design pre-vents RFID signals of staying tags from being collided with new arriving tags when executing a tag identification operation.
In addition, to improve identification throughput, PRB adopts a pair resolution scheme to identify one pair of tags instead of one tag at each query timeslot. However, in PRB the required number of timeslots to identify all staying tags is proportional to the number of staying tags which are recognized in the
pre-vious identification session. In this study, we develop an Adap-tive n-Resolution (AnR) anti-collision protocol to further in-crease the identification performance by identifying any n stay-ing tags in only three timeslots. Our performance analyses and simulation results show that, in general, AnR significantly out-performs PRB by achieving about 10% to 40% reduction of the total time delay on total identification delay and about 10%
decrease in the number of transmitted bits.
II. THE PROPOSED TAG IDENTIFICATION PROTOCOL The proposed protocol is targeted for a single RFID reader system to efficiently recognize all passive RFID tags within its interrogation range in an identification session. We assume that the communication channel between the reader and all involved tags is synchronized during each query timeslot. The reader maintains a two dimensional array to store the timeslot numbers in sequence and corresponding identified tag IDs in the previous session. In addition, a unique reader ID, denoted as rRID, is associated with the reader. In AnR, tag identifica-tion in a session is divided into two phases by adopting the blocking technique. Each tag tj has two counters: Progressed Slot Counter (PSCj) and Allocated Slot Counter (ASCj), where PSC denotes the current number of tags identified by the read-er in the ongoing session and ASCj indicates the assigned time-slot for tj to send its ID as the response. In addition, each tag has a variable tRIDj to store the corresponding reader ID in the current session. A bit indicator, p2, is used for a tag to imply the start of Phase 2 in an identification session when it is set as 1. The reader uses three counters: Progressed Slot Counter (PSC), Terminated Slot Counter (TSC), and New Tag Counter (NTC). The definition of PSCj is the same as the one in the tag.
TSC is used to determine the number of timeslots required during an identification session; therefore, in the beginning of a new session Si, TSC stores the largest value of ASC sent by identified tags in the previous session Si-1. NTC indicates the number of new arriving tags in Si-1. Note that during system initialization all counters and variables of both tags and the reader are set to zero or null.
1 is invoked when the reader broadcasts command phase1 along with TSC and rRID to all existing tags. Once receiving this command, each tag tj verifies whether its tRIDj is equal to the received rRID. For a tag tj, if its tRIDj received rRID, the tag will set p2 = 1 and tRIDj = rRID to mark itself as a new arriving tag; otherwise, the tag is a staying tag and already possesses an appropriate ASCj value determined in the pre-vious session Si-1. Note that all values of PSC and PSCj in the reader and tags are reset to zero while broadcasting and receiv-ing the command phase1, respectively.
In Phase 1, only staying tags will respond to the reader. As-sume the decimal value of ASCj for a staying tag is Mj. A dis-tinct Response Bit String (ResBSj) is generated by each staying tag and sent to the reader at the same timeslot as the tag re-sponse. A bit sequence generator, which can be implemented with simple circuit logics and a bit counter, is utilized by a tag to construct the ResBSj with TSC+1 bits-long. Every time the generator produces and sends out a bit, the bit counter is in-creased by 1. The bit generator only generates bit 1 when the value of its counter is Mj. Otherwise, bit 0 is generated by the bit generator. We assume that Manchester encoding technique is adopted for all data transmission. As each timeslot is syn-chronized during an identification session, the reader can easily detect the positions of collided bits among received ResBSj bit strings from all staying tags. By computing the length Mj be-tween the first received bit and each collision bit, the reader identifies all staying tags in the current session through search-ing at its 2D array with each calculated timeslot Mj as the index.
The reader also sets its PSC as the total number of identified staying tags. A TSC+1 bits-long Acknowledgement Bit String (AckBS) is constructed at the reader by generating bit 0 against the corresponding positions of bit collision of received ResBSj
strings and bit 1 in other positions. In the next timeslot, the reader broadcasts the bit string AckBS to all tags. Among all tags, only each staying tag tj will react and compute the number Nj of occurrence of bit 1 between the first bit and the (Mjʳˀʳ1)-th bit of the received string AckBS. Next, each staying tag sets its ASCj=ASCjʳˀʳNj. Note that this step allows all staying tags in the current session to compute and assign new timeslot numbers from 0 to PSCˀ1 to themselves without collision and the reader will do the same computation to re-sort the positions of identi-fied staying tags at its 2D array. At the same time a me-tricNA=z×NA+(1−z)×NTC used by PRB is adopted by the reader to estimate the number of new arriving tags in the current session Si, where NA denotes the estimated number of timeslots required for new arriving tags and the factor z is a pre-defined weighting number, 0 ʳ z < 1. In addition, the value of TSC in the reader is set to PSC+ NA −1.
In the next timeslot, the reader broadcasts command phase2 along with TSC and PSC to activate Phase 2 in which only new arriving tags will respond to the queries from the reader.
Once receiving phase2 command, each new arriving tag sets its PSCj value as the received PSC and randomly selects a val-ue between PSC and TSC as its new ASCj value. Then, each new arriving tag tj will repeatedly transmit its ID to the reader when its PSCj is equal to ASCj in a distinct and synchronized timeslot until all new arriving tags are identified in Phase 2
current session (i.e., when PSC > TSC at the reader side).
According to the responses sent from new arriving tags, the reader acts as follows.
• Collision: Once detecting signal collision occurred, the reader increases TSC by 1 and sends a Collision feedback.
• One tag response: Once recognizing one tag, the reader adds 1 to both NTC and PSC, stores this tag ID into its 2D array and transmits a Readable feedback.
• No tag response: Once detecting no response, the reader subtracts 1 from TSC and transmits an Idle feedback.
Based on the feedback command sent from the reader, the new arriving tags act as follows.
• Readable: Tag tj increases its PSCj by 1.
• Idle: Tag tj with ASCj > PSCj, decreases its ASCj by 1.
• Collision: Tag tj with ASCj = PSCj, randomly generates a binary number (0 or 1) and adds this number to its ASCj. Other tags with ASCj > PSCj, add 1 to their ASCj.
Table 1. An example of AnR protocol Slot Command
*ResBSi is sent by each staying tag; other commands are sent by the reader.
Table 1 shows a normal tag identification process with AnR.
Suppose tag A, B, C and D have been identified in Si-1 and all of them possess a unique ASCj value such as ASCA=0, ASCB=1, ASCC=2 and ASCD=3. The reader records TSC=3 and NTC=4.
Before the session Si starts, tags B and C left while tags E, F and G arrived.
At Phase 1 of Si, the reader sends the command phase1 with TSC=3 and rRID to all tags and only tags A and D with tRIDj=rRID will respond. Tags A and D calculate their own ResBSj and send ResBSj back to the reader simultaneously.
Tag A: ASCA =0 and TSC=3 Æ ResBSA 0001 Tag D: ASCD =3 and TSC=3 Æ ResBSD 1000
Once the reader receives the x00x bit string where x de-notes the bit collision position, it recognizes that only tags A and D still exist, and rearranges their stored ID positions in its 2D array. Next, the reader sends an AckBS 0110 to all tags.
Based on this AckBS, tag A sets its ASCA=ASCA-0=0 and tag D sets its ASCD=ASCD-2=1. Then, the reader sends a phase2 command with PSC=2 and TSC=PSC+0.5×2+0.5×4−1=4
by assuming z=0.5 and NA=2. Suppose that tags E and F select the 3rd timeslot by setting ASCE and ASCF to 3, respectively.
reader will not detect any response at the current timeslot 2;
therefore, it sets TSC=TSC-1=3 and sends an idle feedback to tags E, F and G. After receiving the idle feedback, tags E, F and G decrease their own ASC values by 1 as the ASCE=ASCF
=3 and ASCG=4 are all larger than PSC=2. In the next timeslot, since ASCE=ASCF =2 are equal to PSC=2, tags E and F will send out their IDs at the same time and result in a signal colli-sion. After detecting the occurrence of collision, the reader sets TSC=TSC+1=4 and sends a Collision feedback to tags E, F and G. Then, tags E and F, with ASCE=ASCF=2 and PSC=2, ran-domly add a binary number to their ASC values. Tag G with ASCG=3>PSC=2 sets its timeslot counter ASCG=3+1=4. Sup-pose that tag E selects a binary number 0 and sets ACSE= ASCE+0=2, and tag F selects a binary number 1 and sets ACSF= ASCF+1=3. Consequently, the tags E, F and G are identified in sequence at the next three timeslots. Finally, the reader sends a termination command as the PSC=5 is larger than TSC=4 at this moment.
III. PERFORMANCE ANALYSES
In this section, we analyze the average identification delay of AnR in terms of consumed timeslots [1-4]. Let Sr,i be the set of all the tags recognized by the reader r in the i-th identifica-tion session Si. DAnR(Sr,i|Sr,i-1) indicates the consumed timeslots of AnR in Si with the identified tag information at the previous session Si-1.
Theorem 1: The total timeslots required for identifying n tags under Binary Tree protocol is shown in [3-4]. Note that p(k)=1-2-k in which k means the currently involved depth of the target binary tree.
∞Theorem 2: For new arriving tags, estimated slots and arbitrary number of staying tags, the total consumed timeslots of AnR in Si is as follows.
, when the ASCj values chosen by new arriving tags are un-iformly distributed over the interval [PSC+0, PSC+(-1)].
(2)
the ASCj values chosen by new arriving tags are not uniformly distributed over the interval [PSC+0, PSC+(-1)].
Proof: In Si, AnR utilizes two phases to identify staying tags and new arriving tags, respectively. Obviously, the number of timeslots consumed in Phase 1 is near optimal, i.e. only three timeslots are required to identify all staying tags. In Phase 2, new arriving tags are identified with the same procedure as Binary Tree protocol. Assume that the ASCj values chosen by new arriving tags are uniformly distributed over the interval [PSC+0, PSC+(-1)]. Then, tags are uniformly divided into groups and each group (with -1 tags) is assigned to each tags-set l at the beginning of the second phase, where 0 l
-1. Thus, the total number of consumed timeslots by AnR in Si
is listed as follows.
Moreover, suppose that the ASCj values chosen by new ar-riving tags are not uniformly distributed over the interval [PSC+0, PSC+(-1)]. Under this assumption, the total number of consumed timeslots by AnR in Si is computed as follows.
=
Note that tl denotes the expected (average) number of con-sumed timeslots for each tags-set l, where 0 l -1. The probability for p new arriving tags ( p ) to select the same timeslot is computed as p p
p
In this section, we evaluate the identification delay and communication overhead of AnR in terms of the number of timeslots and the total transmitted bits for tag identification in Si [1-4]. Our simulation was written with C# under Visual Stu-dio .NET environment. Fig.1 and Fig.2 show the simulation results of AnR and PRB against various numbers of tags and different values of ratio w which is the number of staying tags in Si to the number of identified tags in Si-1. In Fig.1, AnR sig-nificantly outperforms PRB by reducing 10% (w=0) to 40%
(w=0.8) identification delay time. In terms of communication overhead, AnR reduces more than 10% amount of bit transmis-sion compared to PRB when w>0.4 as depicted in Fig.2. As the ratio w increases, the rate of overhead reduction becomes larger.
With w=1, AnR has the best performance for tag identification;
only three timeslots are required to complete the identification session with the smallest amount of bit transmission.
˃
Figure 1. Comparison between AnR and PRB on identification delay.
˃
Figure 2. Comparison between AnR and PRB on communication overhead.
V. CONCLUSIONS
Efficient anti-collision algorithm is a major factor for throughput improvement of RFID tag identification. In this study, we develop a novel anti-collision protocol, called Adap-tive n-Resolution algorithm (AnR), to gain better system throughput. The performance analyses and simulation results show that AnR significantly reduces identification delay and communication overhead in comparison with PRB.
ACKNOWLDGEMENT
The authors gratefully acknowledge the support from projects sponsored by the National Science Council, Taiwan, under the Grants NSC E-011-018 and NSC 98-2218-E-011-020.
REFERENCES
[1] Y.-C. Lai and C.-C. Lin, “Two blocking algorithms on adaptive binary splitting: single and pair resolutions for RFID tag identification,”
IEEE/ACM Transactions on Networking, Vol. 17, No. 13, pp.962-975, June 2009.
[2] Y.-C. Lai and C.-C. Lin, “A Pair-Resolution Blocking Algorithm on Adaptive Binary Splitting for RFID Tag Identification,” IEEE communications letters, Vol. 12, No. 6, pp.432-434, June 2008.
[3] J. Myung, W. Lee and J. Srivastava, “Adaptive Binary Splitting for Efficient RFID Tag Anti-Collision,” IEEE Communications Letters, Vol.
10, No. 3, pp.144-146, Mar. 2006.
[4] J. Myung, W. Lee, J. Srivastava, and T. K. Shih, “Tag-Splitting:
Adaptive Collision Arbitration Protocols for RFID Tag Identification,”
IEEE Trans. Parallel and Distributed Systems, Vol. 18, No. 6, pp.763-775, June 2007.
計畫主持人：吳宗成 計畫編號：98-2218-E-011-018-
研討會論文 6 1 600%
教材 0
舉辦之活動/競賽 0
研討會/工作坊 0
電子報、網站 0
計 畫 加 填 項
目 計畫成果推廣之參與（閱聽）人數 0