• Share
  • Email
  • Embed
  • Like
  • Save
  • Private Content
Mimo rfid
 

Mimo rfid

on

  • 429 views

 

Statistics

Views

Total Views
429
Views on SlideShare
429
Embed Views
0

Actions

Likes
0
Downloads
11
Comments
0

0 Embeds 0

No embeds

Accessibility

Categories

Upload Details

Uploaded via as Adobe PDF

Usage Rights

© All Rights Reserved

Report content

Flagged as inappropriate Flag as inappropriate
Flag as inappropriate

Select your reason for flagging this presentation as inappropriate.

Cancel
  • Full Name Full Name Comment goes here.
    Are you sure you want to
    Your message goes here
    Processing…
Post Comment
Edit your comment

    Mimo rfid Mimo rfid Document Transcript

    • 1 A Passive Multiple-Antenna UHF RFID System Shaoyuan Chen, Xiaodong Wang Abstract A passive ultra-high frequency (UHF) radio frequency identification (RFID) system is proposed that employs multiple antennas at the reader and single antenna at each tag. The gain due to multiple antennas in terms of the maximum reader interrogation range is first quantified. Then a blind adaptive beamforming algorithm is proposed to improve the data transmission performance of the system. Moreover, to enhance the system throughput, the number of tags within the reader interrogation range needs to be estimated. An estimator for such purpose is proposed that is based on the received collided signals from multiple tags. Simulation results show that both the interrogation range and the packet error rate performance for data transmission can be improved significantly by using multiple antennas at the reader. And the proposed tag quantity estimator helps to improve the system throughput. Index Terms Ultra-high frequency (UHF), radio frequency identification (RFID), multiple antennas, interrogation range, beamforming, tag quantity estimator. I. I NTRODUCTION The ultra-high frequency (UHF) radio frequency identification (RFID) technology, whichoperates in the frequency range of 860 - 960 MHz, has recently received significant atten-tion in many application areas such as supply chain management (e.g., automated inventory),logistics, automatic toll collection, asset tracking (e.g., books in libraries, animals), intelligenttransportation, etc., due to its much higher efficiency than the traditional barcode [1]-[4]. The authors are with the Department of Electrical Engineering, Columbia University, New York, NY 10032 (e-mail:wangx@ee.columbia.edu).December 20, 2012 DRAFT
    • 2 RFID systems consist of readers (a.k.a. interrogators), tags (a.k.a. transponders) and dataprocessing center. In passive UHF RFID systems, tags are not battery-powered. Instead, theyabsorb energy from the RF field generated by the signals transmitted by the reader to be poweredup. Therefore, tags feature rather low cost and small size. However, the interrogation range andread reliability of the passive UHF RFID system are limited due to the lack of built-in powersource of the tag, especially in fading environments. Several works have addressed the issue of interrogation range or transmission performanceimprovement of passive UHF RFID with single antenna [9][10] and with multiple antennas [11]-[13]. In particular, in [11], the reverse link interrogation range of the UHF RFID is increased byemploying multiple antennas at the reader, where maximal ratio combining (MRC) is adopted toachieve the optimal range improvement. Note that MRC requires the channel state information(CSI), so channel estimation should be performed before applying MRC. If the distance betweenthe reader and tag is the maximum range that can be achieved by applying MRC, since at thebegining the reader cannot apply MRC due to the lack of CSI, the delivered power from the readermay not be sufficient to power up the tag. If the tag is not powered up, channel estimation cannotbe performed and hence MRC cannot be applied. In [12], multiple RF antennas are equippedat the tag while the reader has single antenna. However, tags are supposed to be as simple andlow-cost as possible, so employing multiple antennas at the tag may not be practically feasible.Bistatic RFID is considered in [13] where the reader employs multiple receive antennas and themulti-antenna receiver algorithm requires complex processing and channel estimation. Within the interrogation zone of a reader, there may exist many tags, which are ready tocommunicate with the reader. The framed slotted ALOHA [5]-[8] protocol is widely used incurrent RFID systems for multiple access. A reader starts by issuing a frame consisting F timeslots. Then each tag selects a slot at random from [0, F-1]. If more than one tags select the sameslot, these tags will collide, so no tag could be read successfully. It is known that the throughputis maximized if the frame size is set equal to the number of tags in the range. But the numberof tags is unknown to the reader at the beginning, this motivates the estimation of the numberof tags in the interrogation range of the reader. A number of works have addressed the estimation of the tag quantity. A collision ratioestimation (CRE) algorithm is proposed in [15] by searching the number of tags to make theactual collision ratio equal to the expected one. Three Bayesian methods are proposed in [16]December 20, 2012 DRAFT
    • 3to estimate the tag quantity with reduced complexity. In [17] the number of tags in a collisionslot is estimated according to the tag’s radar cross-section (RCS) plot. In this paper, we consider an RFID system with the reader equipped with multiple antennas,whereas each tag has a single antenna, which is an effective way to reduce the overall cost,since for typical RFID applications such as objects identification in warehouses, there are usuallyhundreds or thousands of tags. Instead of bistatic RFID considered in [13], monostatic one isconsidered in this paper. The interrogation range and data transmission performance are bothinvestigated under the new configuration. A blind transmit and receive adaptive beamformingalgorithm is proposed for data transmission. Following the basic idea in [17], a tag quantityestimator is proposed. Multiple antennas are shown to improve the estimation performance.Simulation results show that without increasing the total transmit power, the interrogation rangeand the data transmission performance are significantly improved thanks to the multiple antennas.The proposed simple adaptive beamforming scheme makes the data transmission performancenear-optimal even without channel estimation. The estimation of the tag quantity helps to increasethe system throughput, and the achievable throughput with multiple antennas gets much closerto the optimal one. The remainder of the paper is organized as follows. Reader interrogation range and data trans-mission performance along with an adaptive beamforming scheme are investigated in Sections IIand III, respectively. Section IV proposes a tag quantity estimator. Simulation results are givenin Section V. Finally, Section VI concludes the paper. II. R EADER I NTERROGATION R ANGE An RFID system mainly consists of a reader and a set of tags. The block diagram of the systemis shown in Fig. 1, where M transmit/receive antennas are employed at the reader whereas eachtag has a single antenna. Each antenna of the reader is used for both transmitting and receivingsignals, and the transmitter and receiver are connected via a circulator or a coupler [1]. In this paper, a Rician channel model with the Rician factor K = 2.8 dB reported in [18] isadopted. We assume that the forward link and the corresponding reverse link observe the samechannel coefficient (see Fig. 1), i.e., a time-division duplex (TDD) system. Before analyzing the interrogation range of the multiple-antenna RFID system, we first con-sider the single-antenna case.December 20, 2012 DRAFT
    • 4 Reader Wireless channel Tags RF 1 h1 wH h1 ... TX ... ... hM ... M hM RX wH RF ... Forward link Reverse linkFig. 1. System model of a passive multiple-antenna UHF RFID system.A. Single-antenna Case At the very beginning, the reader transmits a continuous wave (CW) to power up the passivetag in the forward link. The received power by a tag can be written as PRX (d) = PT X Gr Gt PL (d) |h|2 , tag (1)where PT X is the transmit power of the reader, Gr is the reader antenna gain, Gt is the tagantenna gain, h is the channel coefficient, PL (d) is the path loss given by ( )2 λ PL (d) = , (2) 4πdwhere λ is the wavelength of the carrier and d is the distance between the reader and the tag. In order to activate the tag, the received signal power at the tag should be no less than thetag sensitivity PT S , i.e., PT X Gr Gt PL (d) |h|2 ≥ PT S , (3)which is the constraint of the forward link. In the reverse link, after the tag is powered up, it scatters the signal back to the reader bymodulating the received CW. The backscattered power received by the reader is given by PRX (d) = ηPRX (d) Gr Gt PL (d) |h|2 reader tag = ηPT X G2 G2 PL (d) |h|4 , r t 2 (4)December 20, 2012 DRAFT
    • 5where η is the backscattering modulation efficiency of the tag. In order to successfully demodulate the backscattered signal, the received backscattered powershould be no less than the reader sensitivity PRS , i.e., ηPT X G2 G2 PL (d) |h|4 ≥ PRS , r t 2 (5)which is the constraint of the reverse link. It is clear that both constraints (3) and (5) should be satisfied to determine the reader inter-rogation range d. From (3) and (5), we have ( )−1 PL (d) ≥ PT X Gr Gt |h|2 PT S (6) (√ )−1 √ and PL (d) ≥ ηPT X Gr Gt |h| 2 PRS . (7) ( )−1 (√ )−1 √ Denote α PT X Gr Gt |h|2 PT S , β ηPT X Gr Gt |h|2 PRS , and √ α/β = (PT X PRS )− 2 1 δ ηPT S . (8)If δ > 1, we have α > β, so once (6) is satisfied, (7) is always satisfied. In this case, the systemis forward-link-limited (FLL) which means the interrogation range is the forward link range √determined by (6). From δ > 1 and (8), it can be easily derived that PT S > η −1 PT X PRS .The interrogation range is determined by (6). By substituting (2) into (6), we have (λ/4πd)2 ≥( )−1 PT X Gr Gt |h|2 PT S , so it follows that the maximum interrogation range is √ d F LL = (16π 2 PT S )−1 PT X Gr Gt λ2 |h|2 . (9) If δ < 1, we have α < β. In this case, the system is reverse-link-limited (RLL) which meansthe interrogation range is determined by the reverse link range. From δ < 1 and (8), it can √be easily derived that PT S < η −1 PT X PRS . The interrogation range is determined by (7). By (√ )−1 √substituting (2) into (7), we have (λ/4πd)2 ≥ ηPT X Gr Gt |h|2 PRS , so it follows thatthe maximum interrogation range is √( )−1 √ √ dRLL = 16π 2 PRS ηPT X Gr Gt λ2 |h|2 . (10)December 20, 2012 DRAFT
    • 6B. Multiple-antenna Case In the forward link, for the multiple-antenna case, the received power at the tag is given by tag 2 PRX (d) = PT X Gr Gt PL (d) wH h , (11)where h = [h1 , . . . hM ]T is channel vector with hi being the channel coefficient from the ithreader antenna to the tag antenna, M is the number of antennas, w = [w1 , . . . , wM ]T is theantenna weigh vector, i.e., beamformer, with ∥w∥ = 1. Thus the constraint of the forward linkfor the multiple-antenna case can be expressed as 2 PT X Gr Gt PL (d) wH h ≥ PT S . (12) In the reverse link, the received backscattered power at the reader is given by 2 PRX (d) = ηPRX (d) Gr Gt PL (d) wH h reader tag 4 = ηPT X G2 G2 PL (d) wH h . r t 2 (13)So we have the constraint of the reverse link for the multiple-antenna case as 4 ηPT X G2 G2 PL (d) wH h r t 2 ≥ PRS . (14) Similarly to the single-antenna case, we have that the system is forward-link-limited if PT S >√ η −1 PT X PRS . The maximum interrogation range is √ dF LL = (16π 2 PT S )−1 PT X Gr Gt λ2 |wH h|2 . MA (15) √And the system is reverse-link-limited if PT S < η −1 PT X PRS . The maximum interrogationrange is √( √ )−1 √ dRLL MA = 16π 2 PRS ηPT X Gr Gt λ2 |wH h|2 . (16) From (9), (10), (15), and (16), it can be derived that the interrogation range gain of the |wH h|multiple-antenna case over the single-antenna case is |h| for both FLL and RLL systems. The proper choice of w increases the range gain. The optimal beamforming (OBF) is achievedby choosing w = h/ ∥h∥. However, OBF requires perfect knowledge of the channel state whichis not available at the reader at the startup of the system but is typically obtained by estimatingthe channel based on the reply signal from the tag. In other words, the tag must have beenpowered up before the reader is about to estimate the channel. In this sense, it is clear that thereDecember 20, 2012 DRAFT
    • 7 Reader CW Select CW Query CW ACK CW QueryRep Tag RN16 PC/XPC+EPC+CRCFig. 2. Interactions between the reader and the tag.is no need to estimate the channel in order to improve the interrogation range since the tag hasbeen powered up and the distance between the reader and tag has been physically determined. Soin terms of the interrogation range, the reader has only two beamforming options: equal-weightbeamforming (EBF) and random beamforming (RBF). For EBF, w is a normalized all-onevector and for RBF, w is generated randomly following certain distribution, e.g., w ∼ Nc (0, I).A comparison among OBF, EBF, and RBF is given in Section V. III. DATA BACKSCATTERING T RANSMISSION After the tag has been powered up by the CW, the reader sends commands (e.g., Query) tocommunicate with the tag. The tag replies to the reader based on the backscattering modulation ofthe received CW. Fig. 2 illustrates the interaction between the reader and the replying tag duringthe inventory process. The Select command is used to select a tag population. An inventoryprocess starts by the reader sending a Query command to the tag, which broadcasts a frameconsisting of F time slots. After receiving the Query command, each tag randomly selects aslot. The tag that picks the 0th slot replies to the reader with RN16, which is a sequenceof 16 bits randomly generated. Then, the reader decodes the received RN16, and sends thedecoded 16 bits as the ACK to the tag. Next, the tag extracts the 16 bits from the ACK. If theextracted 16 bits are the same as the originally generated RN16, the tag then sends its ID, i.e.,the electronic product code (EPC) to the reader. After receiving the ID of the current tag, thereader will send the QueryRep command to read the next tag. In this section, we assume that theprocess before the tag replying its EPC is perfectly done, and we focus on the EPC transmissionperformance of the tag. Note that the EPC is incorporated in a packet during the transmissionand besides the EPC, the whole packet, which is 128-bit long, also includes protocol control /extended protocol control (PC/XPC) and cyclic redundancy check (CRC) [14]. In the following,December 20, 2012 DRAFT
    • 8 FM0 Symbols 0 0 00 1 1 01 10 (a) FM0 Sequences 00 00 01 01 10 10 11 11 Figure 6.9 – FM0 symbols and sequences (b)Fig. 3. (a) FM0 symbols and (b) FM0 sequences.PC/XPC+EPC+CRC is denoted as ID for simplicity. In the baseband, a tag encodes the backscattered data using the FM0 encoding schemeillustrated in Fig. 3. For FM0 encoding, there is always one phase inverse at every symbolboundary, and an extra phase inverse appears in the middle of symbol 0. Thus, the FM0 sequenceis decoded by judging whether there is a phase inverse in the middle of each symbol.A. Single-antenna Case Let r(ID) (t) denote the received complex baseband FM0 encoded ID signal by the reader afterpassing through the direct-conversion receiver filter: r(ID) (t) = hs(ID) (t) + nr (t), (17)where s(ID) (t) is the complex baseband FM0 encoded ID signal replied from the tag and nr (t)is the complex Gaussian noise of the reverse link. Since the tag replies the signal to the readerDecember 20, 2012 DRAFT
    • 9by modulating the received CW sent by the reader, in the baseband, s(ID) (t) could be expressedas √ (ID) ( ) s(ID) (t) = ηf (t) h + nf (t) , (18)where f (ID) (t) is the FM0 encoded ID and nf (t) is the complex Gaussian noise of the forwardlink. 2 Suppose the timing synchronization is perfect and the sampling rate is T . To decode the kthbit replied by the tag, the reader receiver performs the following operation: { } y (ID) (k) = ℜ r(ID) (kT + T /4)[r(ID) (kT + 3T /4)]∗ , (19)where ℜ {x} denotes the real part of x, * is the conjugation operator, k is the symbol index,and T is the symbol duration of the backscattered signal. Then the symbol a(k) is decoded according to the following:   1, if y (ID) (k) > 0 a (k) = ˆ . (20)  0, if y (ID) (k) ≤ 0B. Multiple-antenna Case For the multiple-antenna case, (18) can be rewritten as √ (ID) ( H ) s(ID) (t) = ηf (t) w h + nf (t) . (21)Thus, the received complex baseband signal at the ith antenna of the reader after passing throughthe filter is represented as (ID) ri (t) = hi s(ID) (t) + nr (t), i (22)where nr (t) is the complex Gaussian noise at the ith antenna of the reverse link. Similar to (19), ifor the multiple-antenna case, the receiver computes {M } ∑ [ (ID) ][ (ID) ]∗ y (ID) (k) = ℜ wi ri (kT + T /4) wi ri (kT + 3T /4) i=1 {M } ∑ [ ]∗ (ID) (ID) = ℜ |wi |2 ri (kT + T /4) ri (kT + 3T /4) , (23) i=1where wi denotes the receive beamforming weight. Finally, the symbol a(k) is decoded accordingto (20).December 20, 2012 DRAFT
    • 10 We now consider the choice of the antenna weight wi . Although the OBF would be thebest choice, it requires channel estimation which is hard to implement under the current RFIDstandard. Hence, we propose a blind adaptive beamforming (BABF) scheme [20][21] for oursystem. As can be seen in Fig. 1, the transmit and receive antenna weight vectors are the same.The beamformer vector w should be chosen to maximize the received SNR at the reader receiver,or equivalently, to maximize the received backscattered power (13) from the tag. The BABF scheme starts by the reader sending the CW for probing the tag, i.e., evaluatingthe backscattered power from the tag. At the nth iteration, given the weight vector w(n−1) , Kpperturbation vectors pi are generated where pi ∼ Nc (0, I), i = 1, . . . , Kp to form Kp new weightvectors w(n−1) + βpi wi ⇐ ˜ , i = 1, . . . , Kp (24) ∥w(n−1) + βpi ∥where β is the weight adaptation step size. Then for each of these Kp generated weight vectors,the corresponding received backscattered power (13) is measured at the reader. Finally, the weightvector is updated as the one that has the largest backscattered power among the Kp vectors in(24). The iteration terminates when the received backscattered power fluctuates below a tolerancethreshold. The algorithm is summarized as Algorithm 1.Algorithm 1 Proposed blind adaptive beamforming algorithm 1: Initialize n ⇐ 0 and w(0) ∼ Nc (0, I). 2: repeat 3: n ⇐ n + 1. 4: Generate Kp perturbation vectors pi ∼ Nc (0, I), i = 1, . . . , Kp . (n−1) 5: Form Kp new weight vectors wi ⇐ ˜ w +βpi , i = 1, . . . , Kp . ∥w(n−1) +βpi ∥ reader 4 6: Measure the received power PRX,i = ηPT X G2 G2 PL wi h , i = 1, . . . , Kp . r t 2 ˜H 7: Update w(n) ⇐ wI , where I = arg max PRX,i . ˜ reader ( (n) ) ( (n−1) ) i 8: reader until PRX w − PRX reader w < ε, where ε is the threshold. Note that the above weight adaptation can be performed in the time period after the tagis powered up and before the reader sends the Query command for initial reading; and forsubsequent readings, the weight adaptations can be performed before sending the QueryRepDecember 20, 2012 DRAFT
    • 11commands. We assume that only the tags that will respond to the Query or QueryRep commandrespond to the probing of the reader. IV. TAG Q UANTITY E STIMATION Upon interrogating the tags, the reader starts by broadcasting an initial frame consisting Ftime slots. Then each tag selects a slot at random from [0, F -1]. If more than one tag select thesame slot, these tags will collide, so no tag could be read successfully. Since the number of tagsto be identified is unknown to the reader in the initial interrogation round, if the issued framesize (i.e., the total number of slots) is much larger than the tag quantity, more slots of the framewill be empty which wastes the limited channel resource and decreases the system throughput;if the frame size is much smaller than the tag quantity, more collisions will occur which alsodecreases the system throughput. It is known that the system achieves the optimal throughputwhen the assigned frame size equals to the number of tags in the interrogation range. After the initial interrogation round, the reader may experience three kinds of time slots:empty slot where there is no tag replying, single-tag slot where there is only one tag replying,and collision slot where there are more than one tag replying simultaneously. Since thoseunsuccessfully read tags due to the erroneous transmission in the single-tag slot or due to thecollision in the collision slot will participate in the next interrogation round, if the number ofthose unsuccessfully-read tags in the current frame could be estimated, the frame size of the nextinterrogation round could be determined according to the estimation result so as to maximizethe system throughput. Suppose that the current frame contains N unsuccessfully-read slots dueto the erroneous transmission or the collision. At each slot, the reader estimates the numberof tags replying, say ni for the ith slot. Then the total number of the tags participating in thenext interrogation round could be estimated as n1 + . . . + nN , and the frame size of the nextinterrogation round could be set accordingly. In the following, we assume that the reader knows there is one tag in each unsuccessfully-readsingle-tag slot and we focus on the estimation of the number of tags in each collision slot basedon the collided signal. Instead of the tag ID, it is a sequence of FM0 encoded random 16 bits (RN16) that is firstlyreplied to the reader by the tag upon receiving the Query command. The reader should be ableto successfully receive and decode this sequence to enable the subsequent communication. ButDecember 20, 2012 DRAFT
    • 12 15 10 5 Quadrature 0 −5 −10 −15 −15 −10 −5 0 5 10 15 In−Phase (a) 15 10 5 Quadrature 0 −5 −10 −15 −15 −10 −5 0 5 10 15 In−Phase (b)Fig. 4. RCS plots based on the reception of the replied RN16s of tags: (a) two tags replying simultaneously resulting in fourclusters and (b) three tags replying simultaneously resulting in eight clusters.December 20, 2012 DRAFT
    • 13if collision happens, which means multiple RN16s from different tags are sent to the readersimultaneously, the reader is unable to decode the received overlapped RN16s. Thus the collidedtags cannot be read successfully and will reply in the next available interrogation round. It is noticed that one tag’s replied RN16 signal contributes two clusters in the RCS plot [17],and R simultaneously replying tags produce 2R clusters in the plot ideally, as illustrated in Fig.4, where the sampling rate is two samples per bit. Consequently, if the reader is able to estimatethe number of clusters, the number of the collided tags can be easily derived. Based on thisidea, we next develop a tag quantity estimator.A. Single-antenna Case In this subsection, we propose a clustering algorithm to estimate the number of tags involvedin the collision slots. We focus on the single receive antenna case first. The clustering algorithmis described as follows. We consider a specific collision slot. Let r(RN 16) (t) denote the received overlapped complexbaseband RN16 signal in this slot after passing through the filter, which can be expressed as ∑ Ntag (RN 16) r (t) = hn s(RN 16) (t) + nr (t), n (25) n=1where Ntag ≥ 2 is the number of tags replying in the slot, hn is the channel coefficient from (RN 16)the nth tag to the reader and sn (t) is the replied RN16 signal from the nth tag, which isgiven as √ (RN 16) ( ) s(RN 16) (t) = n ηfn (t) hn + nf (t) , (26) (RN 16)where fn (t) is the FM0 encoded RN16 generated by the nth tag. 2 Suppose the timing synchronization is perfect and the sampling rate is T at the reader receiver.Then the received signal sample set in this collision slot is { ( ) } 2k + 1 S= r (RN 16) T , k = 0, ..., 31 . (27) 4 After obtaining S, the reader starts clustering the samples by first selecting one element in S atrandom. Then the distances between the selected element and all other elements are calculated.Those elements with distances from the selected element being no larger than r form a clustertogether with the selected element, where r = ρσ and ρ is a parameter. Next, the reader repeatsthe above clustering steps for the elements which have not been clustered until all elements inDecember 20, 2012 DRAFT
    • 14S are clustered. Finally, the number of tags in this slot can be derived by counting the numberof clusters Nc in S. The proposed tag quantity estimator is summerized in Algorithm 2.Algorithm 2 Proposed tag quantity estimator 1: Specify ρ (see Section V). 2: for j = 1, ..., Mcs , where Mcs is the number of collision slots in the current frame do 3: Nc ⇐ 0. 4: Obtain S according to (27) in the jth slot. 5: while S is not empty do 6: Randomly pick one element sk ∈ S. 7: Calculate dh = |sk − sh |, where h = 1, ..., |S|. 8: Choose elements {sg : dg ≤ ρσ} to form a cluster and remove them from S. 9: Nc ⇐ Nc + 1.10: end while11: Estimate the number of tags in the jth collision slot as ⌈log2 Nc ⌉.12: end forB. Multiple-antenna Case Although the reader could observe 2R clusters with R tags involved in a collision slot ideally,it is possible that the actual number of observed clusters is less than 2R . The reason is that theclusters may overlap with each other due to the impact of channel and noise. For example, thereader may observe only four clusters due to the overlapping effect, albeit there are three tagsreplying simultaneously and 23 = 8 clusters are supposed to be observed ideally. Multiple receive antennas may help overcome the overlapping effect. Supposing that multipleantennas are spatially well separated, while one antenna observes overlapped and undistinguish-able clusters, other antennas may observe well-separated clusters. For the multiple-antenna case, (26) can be rewritten as √ (RN 16) ( H ) s(RN 16) (t) = n ηfn (t) w hn + nf (t) , (28)where hn = [h1n , ..., hM n ]T is the channel vector with hin being the channel coefficient fromDecember 20, 2012 DRAFT
    • 15the ith reader antenna to the nth tag. And (25) now becomes (RN 16) ∑ Ntag ri (t) = hin s(RN 16) (t) + nr (t), n i (29) n=1which is the received overlapped complex baseband RN16 signal in a collision slot from the ithreceive antenna after passing through the filter. Stacking the received RN16 signals (29) of eachantenna, we have the vector r(RN 16) (t) = Hs(RN 16) (t) + nr (t), (30) (RN 16) (RN 16) (RN 16) (RN 16)where r(RN 16) (t) = [r1 (t), ..., rM (t)]T , s(RN 16) (t) = [s1 (t), ..., sNtag (t)]T , nr (t) =[nr (t), ..., nr (t)]T , and 1 M   h11 ··· h1Ntag    . ... .  H= . . . . . (31)   hM 1 · · · hM NtagThen, the signal sample set in (27) becomes { ( ) } 2k + 1 S= r (RN 16) T , k = 0, ..., 31 . (32) 4 The clustering algorithm is similar to Algorithm 2, except that it is now applied to the vectorsamples in (32) and the corresponding distances between vectors are used. V. S IMULATION R ESULTS In this section, simulation results are presented. The system parameters are set as follows.Carrier frequency fc = 915 MHz, the total transmit power PT X = 1W (30 dBm), reader antennagain Gr = is 2 dBi, tag antenna gain Gt = 0 dBi, reader sensitivity PRS = 3.16 × 10−8 mW(-75 dBm), modulation efficiency η = 0.25, weight adaptation step size β = 0.05, the number oftags in the reader interrogation range is uniformly distributed in [50, 500] and the initial framesize Finit = 256. Finally, the Rician channel with the Rician factor K = 2.8 dB is adopted asindicated in Section II.December 20, 2012 DRAFT
    • 16 11 OBF 10 EBF RBF 9 Average range (m) 8 7 6 5 4 1 2 3 4 Number of antennasFig. 5. Average interrogation range versus number of antennas in FLL systems.A. Interrogation Range Performance The tag sensitivity varies from tag to tag. In the simulations, we fix the reader sensitivity andvary the tags with different sensitivities. √ First, we choose the tag sensitivity PT S = 0.04 mW (-14 dBm) so that PT S > η −1 PT X PRSand the system is FLL. Fig. 5 shows the average interrogation range versus the number ofantennas under this setup. Although only EBF and RBF can be chosen in practice as discussed inSection II, as a performance upper bound, we also present OBF performace assuming the readerhas the CSI at the startup of the system. It can be observed that the EBF scheme outperforms RBF,and the interrogation range increases with the number of antennas. Specifically, the interrogationrange is increased from 4.74 m (M = 1) to 6.31 m (M = 2) and to 8.67 m (M = 4) for EBF;and for RBF, the interrogation range is increased from 4.74 m (M = 1) to 5.73 m (M = 2) andto 7.41 m (M = 4). Clearly, the EBF scheme is a better choice. √ Next, we choose the tag sensitivity PT S = 0.01 mW (-20 dBm) so that PT S < η −1 PT X PRSDecember 20, 2012 DRAFT
    • 17 20 OBF EBF 18 RBF 16 Average range (m) 14 12 10 8 1 2 3 4 Number of antennasFig. 6. Average interrogation range versus number of antennas in RLL systems.and the system is RLL. Fig. 6 shows the average interrogation range versus the number ofantennas under this setup. We observe similar range improvements as in FLL systems. Moreover,under the same antenna configuration, the interrogation range of the RLL system is larger thanthat of the FLL system (see also Table I). This is due to the improvement of the tag sensitivityin RLL systems, which enables the tags to detect weaker signals.B. Data Transmission Performance Once a tag successfully receives the ACK command sent by the reader, it will reply to thereader using a 128-bit packet that includes PC/XPC, EPC, and CRC. If the packet is not receivedsuccessfully by the reader, the tag will then enter the arbitration state to wait for replying tothe reader in the next interrogation round. In this subsection, instead of the bit error rate (BER)performance (e.g., [12] [13]), the more appropriate packet error rate (PER) performance of thesystem is evaluated and the performance gain of employing multiple antennas is examined.December 20, 2012 DRAFT
    • 18 TABLE I AVERAGE INTERROGATION RANGE FOR DIFFERENT NUMBERS OF ANTENNAS AND BF SCHEMES IN FLL AND RLL SYSTEMS . Number of antennas BF scheme FLL (m) RLL (m) M=1 - 4.74 9.05 OBF 7.01 13.29 M =2 EBF 6.31 11.92 RBF 5.73 10.88 OBF 10.17 19.2 M=4 EBF 8.67 16.37 RBF 7.41 14.03 2.8 2.6 OBF Kp=8 2.4 Kp=16 2.2 Kp=4 |wHh|4 2 Kp=2 1.8 1.6 BABF 1.4 0 10 20 30 40 50 Iteration number,nFig. 7. The convergence of the proposed BABF algorithm with different values of Kp .December 20, 2012 DRAFT
    • 19 0 10 −1 10 M=2 PER −2 10 −3 M=1 10 RBF EBF BABF OBF w/ perfect CSI −4 10 5 10 15 20 25 30 35 40 45 50 55 SNR (dB)Fig. 8. PER performance in a Rician fading channel (K = 2.8 dB) for different BF schemes. First we illustrate the performance of the proposed BABF algorithm. Fig. 7 shows the received 4power metric wH h as in (13) versus the iteration number in one simulation. The performanceof OBF with ideal CSI is also plotted as a benchmark. It can be observed that the performanceof the proposed BABF approaches that of OBF, and the convergence rate is increased withKp . Fig. 8 shows the PER performance versus the transmit signal-to-noise ratio (SNR) for datatransmission. For BABF, we set Kp = 8 and the number of iterations is 30. It can be observedthat the PER performance is significantly improved as the number of antennas M increases.With M = 2, BABF has better PER performance than RBF and EBF. The performance offeredby BABF is very close to the optimum, i.e., OBF with perfect CSI. One can also observe thatthe proposed BABF scheme offers about 21 dB gain over the single-antenna case at the PER of10−2 with M = 2 .December 20, 2012 DRAFT
    • 20 TABLE II O PTIMAL VALUES OF THE PARAMETER ρ UNDER DIFFERENT VALUES OF SNR WITH DIFFERENT ANTENNA CONFIGURATIONS . ρ SNR (dB) M=1 M=2 10 2.3 2.6 15 2.7 3.1 20 3.3 3.6 25 3.8 3.9 30 4.0 4.2 35 4.3 4.3 40 4.6 4.8 −2 10 −3 10 Normalized MSE −4 10 M=1 M=2 −5 10 10 15 20 25 30 35 40 SNR (dB)Fig. 9. Normalized MSE of the estimated number of tags versus SNR.December 20, 2012 DRAFT
    • 21 0.5 0.45 0.3984 Average throughput 0.4 0.3886 0.3405 0.35 0.3 0.25 0.2 0.15 1 M=1 2 M=2 3 OptimalFig. 10. Average system throughput with single antenna and multiple antennas (M = 2) at SNR = 30 dB.C. Tag Quantity Estimation Performance We suppose that the reader has the knowledge of the noise variance σ 2 and the collided tagshave the same distance from the reader. In Algorithm 2, ρ should be specified in advance. Weresort to simulations to find the optimal value of ρ by making 5000 simulation runs to evaluatethe performance for each ρ value in the interval of [1.5, 5.5] with step size 0.1. Table II showsthe optimal values of ρ in the sense of minimizing the normalized mean-square error (MSE) ofthe estimated number of tags under different SNR. It is seen that as SNR increases the optimalρ increases accordingly. Fig. 9 shows the normalized MSE of the estimation versus SNR, wherefor each SNR, ρ is set according to Table II. It is seen that the multiple-antenna configuration(M = 2) leads to much more accurate estimation than the single-antenna one. Fig. 10 showsthe average throughput of the system at SNR = 30 dB and with ρ = 4.0 and 4.2 for the casesof M = 1 and M = 2, respectively. The specification of ρ is determined according to Table II,and the average throughput is calculated according toDecember 20, 2012 DRAFT
    • 22 Reader CW Select CW Query CW ACK CW QueryRep Tag RN16 PC/XPC+EPC+CRC FLL RLL Interrogation Tag quantity Data transmission BABF BABF range estimation performanceFig. 11. A brief summary of this work. ( ) 1 ∑ 1 ∑ ns ni pij , (33) ns i=1 ni j=1where ns is the number of simulations which is set as 10000 (For each run, the total number oftags in the reader interrogation range is uniformly distributed in [50, 500].), ni is the numberof interrogation rounds required to complete reading a set of tags in the ith simulation run,pij = (1 − PER) Sij /Fij is the throughput in the jth interrogation round of the ith simulationrun, where PER is given in Fig. 8, Sij and Fij are the number of single-tag slots and frame sizein the jth interrogation round of the ith simulation run, respectively. The estimator sets Fij as(note that the frame size should be a power of two according to the standard [14])   Finit , for j = 1 Fij = , (34)  2⌈log2 (Nij )⌉ , ˆ for j > 1 ˆwhere Nij is the estimated number of tags to be identified in the jth interrogation round of theith simulation run Nij . Note that for data transmission and reception in the single-tag slot, BABFis applied; and for clustering the received raw signal vectors in the multiple-antenna case, nobeamformer is applied. From Fig. 10, it can be observed that the multiple-antenna case (M =2) offers a significant throughput improvement compared with the single-antenna case, and thethroughput offered by M = 2 is close to the optimal one, and the latter is obtained by settingPER = 0 and Fij = 2⌈log2 (Nij )⌉ .December 20, 2012 DRAFT
    • 23 VI. S UMMARY AND C ONCLUSIONS We have considered a passive multiple-antenna UHF RFID system with the reader equippedwith multiple antennas and each tag equipped with a single antenna. The reader interrogationrange, data transmission performance and the estimation of tag quantity have been investigated.As shown in Fig. 11, the improvement of the interrogation range is achieved during the transmis-sion of CW before the Select command is sent for the FLL system and during the backscatteringtransmission of the tag for the RLL system. The transmission performance of tag replying thepacket of PC/XPC+EPC+CRC is then evaluated. The proposed BABF algorithm is performedbefore the reader sends the Query command for the initial reading or the QueryRep command forsubsequent readings. Finally, the tag quantity is estimated by utilizing the received overlappedRN16 signals. Our results indicate that under the multiple-antenna configuration, the interrogation range isincreased substantially and the PER performance of the data transmission approaches the optimalbeamforming performance with the proposed BABF algorithm without channel estimation. Theproposed tag quantity estimator increases the throughput of the system and approaches theoptimal throughput under the proposed multiple-antenna setup. Finally, we note that the schemesproposed in this paper all comply with the current RFID standard and hence they can be readilyimplemented in existing systems. R EFERENCES[1] D. M. Dobkin, the RF in RFID: Passive UHF RFID in Practice. Elsevier, 2008.[2] K. Finkenzeller, RFID Handbook: Fundamentals and Applications in Contactless Smart Cards and Identification. 2nd Ed. New York: Wiley, 2003.[3] H. Vogt, “Efficient object identification with passive RFID tags,” in Proc. Int. Conf. Pervasive Comput., 2002, pp. 98–113.[4] R. Want, “An introduction to RFID technology,” IEEE Pervasive Comput., vol. 5, no. 1, Jan.-Mar. 2006.[5] F. C. Schoute, “Dynamic frame length ALOHA,” IEEE Trans. Commun., vol. Com-31, no. 4, pp. 565–568, Apr. 1983.[6] L. Zhu, T.-S. P. Yum, “The optimal reading strategy for EPC Gen-2 RFID anti-collision systems,” IEEE Trans. Commun., vol. 58, no. 9, pp. 2725-2733, Sep. 2010.[7] C. Wang, M. Daneshmand, K. Sohraby, B. Li, “Performance analysis of RFID Generation-2 protocol,” IEEE Trans. Wireless Commun., vol. 8, no. 5, pp. 2592-2601, May 2009.[8] W. -T. Chen, “An accurate tag estimate method for improving the performance of an RFID anticollision algorithm based on dynamic frame length ALOHA,” IEEE Trans. Autom. Sci. Eng., vol. 6, no. 1, pp. 9-15, Jan. 2009.[9] J. -S. Park, et al. “Extending the interrogation range of a passive UHF RFID system with an external continuous wave transmitter,” IEEE Trans. Instrum. Meas., vol. 59, no. 8, pp. 2191-2197, Aug. 2010.December 20, 2012 DRAFT
    • 24[10] P.V. Nikitin, D. D. Arumugam, M. J. Chabalko, B. E. Henty, and D. D. Stancil, “Long range passive UHF RFID system using HVAC ducts,” Proc. IEEE , vol. 98, no. 9, pp. 1629-1635, Sep. 2010.[11] D. -Y. Kim, H. -S. Jo, H. Yoon, C. Mun, B. -J. Jang, and J. -G. Yook, “Reverse-link interrogation range of a UHF MIMO-RFID system in Nakagami-m fading channels,” IEEE Trans. Ind. Electron., vol. 57, no. 4, pp. 1468-1477, Apr. 2010.[12] J. D. Griffin and G. D. Durgin, “Gains for RF tags using multiple antennas,” IEEE Trans. Antennas Propag., vol. 56, no. 2, pp. 563–570, Feb. 2008.[13] C. Angerer, R. Langwieser, and M. Rupp, “RFID reader receivers for physical layer collision recovery,” IEEE Trans. Commun., vol. 58, no. 12, pp. 3526-3537, Dec. 2010.[14] EPCglobal Inc., “EPC radio-frequency identity protocols Class-1 Generation-2 UHF RFID protocol for communications at 860 MHz – 960 MHz Version 1.2.0.” [Online]. Available: http://www.epcglobalinc.org.[15] J.-R. Cha and J.-H. Kim, “Novel anti-collision algorithms for fast object identification in RFID systems,” in Proc. Int. Conf. Parallel and Distrib. Syst. Comput., 2005, vol. 2, pp. 63–67.[16] H. Wu and Y. Zeng, “Bayesian tag estimate and optimal frame length for anti-collision Aloha RFID system,” IEEE Trans. Autom. Sci. Eng., vol. 7, no. 4, pp. 963-969, Oct. 2010.[17] R. S. Khasgiwale, R. U. Adyanthaya, and D. W. Engels, “Extracting information from tag collisions,” in Proc. IEEE Int. Conf. RFID, Apr. 2009, pp. 131-138.[18] D. Kim, M. A. Ingram, and W. W. Smith, Jr., “Measurements of small-scale fading and path loss for long range RF tags,” IEEE Trans. Antennas Propag., vol. 51, no. 8, pp. 1740-1749, Aug. 2003.[19] S. Chen, L. Wang, and G. Chen, “Data-aided timing synchronization for FM-DCSK UWB communication systems,” IEEE Trans. Ind. Electron., vol. 57, no. 5, pp. 1538-1545, May 2010.[20] B. C. Banister and J. R. Zeidler, “A simple gradient sign algorithm for transmit antenna weight adaptation with feedback,” IEEE Trans. Sig. Proc. vol. 51, no. 5, pp. 1156-1171, May 2003.[21] K. Dong, N. Prasad, X. Wang, and S. Zhu, “Adaptive antenna selection and Tx/Rx beamforming for large-scale MIMO systems in 60 GHz channels,” EURASIP J. Wireless Commun. and Network., vol. 59, Aug. 2011.December 20, 2012 DRAFT