16. multiple rfid tags access algorithm


Published on

Published in: Business, Technology
  • Be the first to comment

  • Be the first to like this

No Downloads
Total views
On SlideShare
From Embeds
Number of Embeds
Embeds 0
No embeds

No notes for slide

16. multiple rfid tags access algorithm

  1. 1. 174 IEEE TRANSACTIONS ON MOBILE COMPUTING, VOL. 9, NO. 2, FEBRUARY 2010 Multiple RFID Tags Access Algorithm Weilian Su, Senior Member, IEEE, Nikolaos V. Alchazidis, and Tri T. Ha, Fellow, IEEE Abstract—One of the main problems that affect the data integrity of passive RFID systems is the collision between the tags. A popular anticollision algorithm which dominates the standards in HF and UHF passive RFID systems is Framed Slotted Aloha (FSA) and some variations of FSA. Throughput and average time delay of the RFID system which determines the performance/efficiency of the system are reduced rapidly when the number of tags inside the interrogation zone is increased. Using larger frame sizes is not always the solution. This paper discusses and compares the existing protocols, and proposes a variation of FSA, called Progressing Scanning (PS) algorithm. The PS algorithm divides the tags in the interrogation zone into smaller groups and gives the reader the ability to communicate with each of them. For performance analysis, the PS algorithm was evaluated with the parameters of a typical passive RFID system at 2.5 GHz. The results showed that the PS algorithm can improve the efficiency of the RFID system and provide a reliable solution for cases with a high density of tags in the area (over 800 tags). Index Terms—Passive RFID systems, tags, framed slotted aloha, collisions, data integrity, progressing scanning algorithm. Ç1 INTRODUCTIONC URRENTLY, a revolution is occurring in Radio Frequency Identification (RFID) technology, and many companiescreate new implementations of RFID systems and new . RFID systems can be used to track people and animals in real time, while this cannot be done with bar codes.products related to this technology daily. The main advan- . A bar code is the same for all similar items, whiletage of RFID technology is the automated identification and with RFID technology, the same items can havedata capture that promises wholesale changes across a broad different data, such as a different expiration date.spectrum of business activities and aims to reduce the cost of A disadvantage of RFID technology is that the manu-the already used systems such as bar codes. For this reason, facturing cost of the main components is still not cheaperalthough RFID technology was discovered many years ago, it than simple bar codes. Therefore, bar codes will coexisthas advanced and evolved only during the last decade since with RFID systems in some applications. Due to the relativecost has been the main limitation in all implementations. high data rates and the long tracking distance, RFID The main advantages of RFID systems compared to bar systems are being examined to ascertain whether theycodes are the following: could be employed for tracking people and supplies in military operations. The most important issues to be solved . In RFID applications intended to replace bar codes, are the integrity of data collection and the security of data contact with the item to be identified is not transferred in the RFID system. In general, RFID technology necessary, and even the line-of-sight (LOS) is often has vulnerabilities in securing the data between the main not necessary. Thus, it is no longer necessary to open components of the RFID system. shipping boxes and scan their contents. This paper proposes to investigate RFID technology and . RFID systems work over long distances. evaluate the performance/effectiveness of the RFID systems . RFID provides full automation of the supply chain in collecting data. It intends to discuss and evaluate the and can reduce the cost of the vendor using it. performance of the different protocols used today for . It can be implemented in different environmental communication between the main components of the RFID conditions, such as in rain or with dust and dirt and system: the tags and the reader. It focuses on RFID systems still operate extremely well. which work in the microwave frequency band of 2.45 GHz, . Also, while data stored in bar codes are fixed and without the use of a battery supply for the tags. The goal of cannot be changed, in most RFID systems, this is this research is to discover ways to increase the performance possible by changing the data inside their electronic of data collection for such systems under the constraints of memory. time delay, throughput, and finally, the working distance. . RFID systems are capable of multiple simultaneous The paper is organized into the following sections: Section 2 presents the existing anticollision protocols used scans of items which reduce the time needed to in RFID systems, and it evaluates the performance of each collect the data. one and compares them in terms of throughput and time delay. Section 3 presents the proposed hybrid anticollision. The authors are with the Department of Electrical and Computer protocol, and Section 4 discusses the performance of the Engineering, Naval Postgraduate School, 833 Dyer Road, Rm. 452, proposed protocol. Finally, we conclude in Section 5. Spanagel Bldg. 232, Monterey, CA 93943-5121. E-mail: {weilian, nalchazi, ha}@nps.edu.Manuscript received 24 Mar. 2008; revised 28 Oct. 2008; accepted 6 May 2 DATA INTEGRITY AND ANTICOLLISIONS2009; published online 29 May 2009. PROTOCOLSFor information on obtaining reprints of this article, please send e-mail to:tmc@computer.org, and reference IEEECS Log Number TMC-2008-03-0108. This paper focuses on passive RFID systems using back-Digital Object Identifier no. 10.1109/TMC.2009.106. scattering modulation in microwave (2.4 GHz) and 1536-1233/10/$26.00 ß 2010 IEEE Published by the IEEE CS, CASS, ComSoc, IES, & SPS
  2. 2. SU ET AL.: MULTIPLE RFID TAGS ACCESS ALGORITHM 175ultrahigh-frequency (UHF) bands, where the transmission and controlling collisions. This algorithm must be simplerate is very high. Therefore, the following discussion refers without numerous computations in the transponders. Into those RFID systems. order to increase the low efficiency of the PURE ALOHA The main disadvantage in RFID communications is the algorithm, a variation of it is used which is called Slottedmultiaccess that occurs in uplink, where multiple tags Aloha. In a Slotted Aloha algorithm, the reader controls andrespond to the reader’s signal command. Multiple re- synchronizes the tags before they respond.sponses at the same time on the Radio Frequency (RF) This section focuses on the description of those algo-communication channel means that the reader cannot rithms and an evaluation of their performance in terms ofidentify the data transmitted from the tag. This event is throughput efficiency and time delay, which is the timecalled tag collision and is responsible for the low tag needed by the reader to read all the tags in the area.identification efficiency [1] in RFID systems. Data integrity in RFID systems also depends on the 2.1 ALOHA-Based Algorithmsfollowing parameters: 2.1.1 Basic ALOHA Procedures The Aloha procedure or algorithm is a probabilistic . Power received from the tag. This power should be procedure, which can be used in the multiaccess uplink efficient to energize the circuit of the tag and also communication from the tag to the reader to avoid collision. transmit the data to the channel. It is very simple to implement, and thus currently, it is very . Signal-to-Noise Ratio (SNR) in the reader. The common in passive RFID systems with read-only tags. received signal in the reader should be high enough In RFID systems using Aloha anticollision algorithms, so the reader will be able to verify the data sequence. the time each tag uses for the transmission of data is a small Some kinds of Error Detection (usually Cyclic Re- dundancy Check (CRC)) need to be used in the reader. fraction of the repetition time and long pauses between transmissions from the same tag occurs. Thus, the commu-In the following discussion, it is assumed that the above nication between the reader and the tags is not continuous.parameters are satisfied in the development of an efficient In addition, each tag occupies, in general, a different periodanticollision algorithm. of time to transmit the data [1], which depends on the Space Division Multiple Access (SDMA) and Frequency amount of data to be transmitted.Division Multiple Access (FDMA) are not generally used in As a result of the simplicity of the Aloha algorithm, thereRFID systems because of the high cost to implement them. is a high possibility of collision between tags because tagsBoth of the above anticollision techniques are limited to “a can transmit their data to the reader randomly at any time.few specialized applications” and therefore, are not suitable This possibility increases while the offered load G isfor passive RFID systems, which need to have a low increased. The average utilization or throughput of theimplementation cost and complexity. channel is given by (1). Time Division Multiple Access (TDMA) is an excellentchoice for RFID systems with a small number of tags. If the S ¼ G Á eðÀ2GÞ : ð1Þnumber of tags is large and not known during the process The Aloha algorithm has a maximum of 18.4 percentof identification, the Packet Radio (PR) technique is by far utilization at G ¼ 0:5, and for this reason, the Alohathe best anticollision technique used in RFID systems. The algorithm has been modified to improve efficiency up toadvantage of PR technique is that it allows the reader to 36.8 percent. This modification is called Slotted Aloha. Theidentify a large number of tags with a small amount of average utilization in Slotted Aloha is given by (2) and has aoverhead. Also, it is not a requirement for the reader to maximum value of G ¼ 1.know the number of tags in the RFID system. Two main procedures of PR are used: asynchronous and S ¼ G Á eðÀGÞ : ð2Þsynchronous procedures [1]. In the asynchronous proce- In Slotted Aloha, the time of the channel is divided intodures, the reader does not control the tags inside the uniform slots with size equal to the transmission time.interrogation zone. Asynchronous procedures are also Currently, tags transmit the data packets only at thecalled transponder drive; the most important of this kind beginning of each slot [1], [2]. Consequently, synchronizationof PR is the PURE ALOHA procedure which is used in is necessary in the Slotted Aloha algorithm. The necessarypassive RFID systems for its simplicity. On the other hand, synchronization is provided by the reader, and therefore,in the synchronous procedures, which are reader driven, Slotted Aloha is a reader-driven TDMA procedure [1].the reader identifies all the tags inside the interrogator area According to (2), it can be shown that as the offered loadby a unique serial number assigned to each tag. Reader- (G) increases, throughput (S) increases until the maximumdriven procedures are divided into polling and binary value of 36.8 percent and then falls rapidly for values of Gsearch procedures. Most of the standards for the UHF RFID greater than 1. This is a main disadvantage of the Slottedsystems propose Aloha-based anticollision algorithms that Aloha algorithm because the system is unstable, and withare probabilistic and binary tree search anticollision algo- low efficiency.rithms that are deterministic [3]. A variation of Slotted Aloha is used in RFID systems and In passive RFID systems, tags are powerless and stateless has been proposed by the International Organization fordevices that cannot sense the channel. They do not know Standardization (ISO) and the Electronic Product Codeabout the existence of other tags in the neighboring area and (EPC). This variation of Slotted Aloha in RFID systems iscannot detect when a collision occurs. Thus, the reader is called Framed Slotted Aloha (FSA). In the following section,responsible for implementing an anticollision algorithm the FSA and dynamic FSA are reviewed and examined.
  3. 3. 176 IEEE TRANSACTIONS ON MOBILE COMPUTING, VOL. 9, NO. 2, FEBRUARY 20102.1.2 Framed Slotted Aloha (FSA) Algorithm again. If this assumption is not valid, as it is for tagsThe FSA algorithm is a Slotted Aloha in which the available that need to be read more than once, the efficiency oftimeslots where the tags can respond to the reader the DFSA algorithm is less than the followingcommands are organized into time frames. Each frame is calculations. It will be assumed that tags after readingdivided into a number of slots (usually powers of 2) and will not respond to future reader requests [3].each timeslot is long enough for the tags to transmit their . In the estimation of the number of tags in thedata. Those time frames have duration equal to the time reader’s field, the Capture Effect, is assumed to bebetween two REQUEST commands of the reader [1], [3]. negligible. This effect helps tags near the reader toThus, the efficiency of the FSA remains the same as in transmit their data although collision had occurredSlotted Aloha. in the timeslot they used. This happens because their A modification of FSA is the Dynamic Framed Slotted signal is stronger than the farthest tags due toAloha (DFSA). The DFSA algorithm dynamically changes channel attenuation. Capture effect increases thethe frame size to increase tag identification, and thus, throughput of the DFSA algorithm and thus in-increases the efficiency in collecting the data from the tags. creases the overall efficiency of the RFID system [1]. . The communication channel, both uplink and down-2.1.3 Dynamic Framed Slotted Aloha (DFSA) link, is assumed to be noise free. Increasing noiseThe DFSA algorithm was first introduced by Schoute [5] for decreases the ability of the reader to read the datamultiuser channel environments and proven to increase the from the tags, and thus decreases the performance ofupper bound of the FSA algorithm to 42.6 percent; it also the system.increases the stability of the multiuser channel. In [6], [7], In DFSA, the reader estimates the total number of tags inDFSA was introduced for passive RFID systems where the the interrogation zone by using the received informationnumber of tags is unknown. from the slots in each frame as a feedback control. Thus, a In specific applications where the number of tags is controller is necessary. This is not a problem because allknown, constant, and not too big, FSA can be used; modern readers have an onboard controller.otherwise, DFSA is the solution. In DFSA, the reader hasthe flexibility to vary the frame size. Hence, it varies the 2.2.1 Estimation of Frame Size and Number of Tags innumber of available slots for the tags. If the reader does not RFID Area Using the DFSA Algorithmdetect tags, which means that collisions occur, it increases In the description of the DFSA algorithm that follows, thethe frame size until an efficient number of tags can be approach and methodology of [3] is used. First, thedetected. As long as tags are detected, it decreases the frame performance of FSA is discussed and then it is applied tosize and so on [3]. DFSA since DFSA just enhances FSA by dynamically The reader tries to identify all the tags in the interrogation changing the frame size. Let N be the random variablezone in multiple read cycles. The amount of time in one read representing the number of slots that have been used by thecycle is equal to the time elapsed between two REQUEST reader at the previous read cycle. Also, let R and C be the random variables representing the number of slots of thecommands sent by the reader. The subject of this research is frame, which are selected by only one tag and by more thanto investigate how the reader can detect and read the one tag, respectively. Finally, n denotes the number of tagsmaximum number of tags with a minimum number of read in the interrogation zone, which is also a random variable.cycles and a maximum probability of detection. Slots that have been selected by only one tag are the slots Much research has been done about the criterion which that the reader can read the data from the tag and identify it.must be used in order to change the frame size. In passive In slots that have been selected by more than one tag,RFID systems, the reader waits for the tags to respond and collision has occurred and the reader cannot identify the tags.changes the frame size after each read cycle according to the The probability p of a tag selecting a specific slot is given bynumber of tags in the interrogation zone. Thus, the criterionin DFSA is that the reader needs to estimate the number of 1 p¼ ; ð3Þtags in the previous read cycle and then adjusts the frame Nsize accordingly. In the following section, the system and the probability for a tag to transmit its data (Pread ) inefficiency of DFSA and FSA is studied. this slot is given by2.2 System Efficiency in RFID Systems with DFSA 1 1 nÀ1 and FSA Algorithms Pread ¼ Á 1À ; ð4Þ N NIn order to measure the efficiency of DFSA in a passiveRFID system, the following assumptions must be made to because the selection of the slot from the tags is random,decrease the complexity of the problem in the next sections: and every tag selects a slot independent of the rest of the n À 1 tags with the same probability. . Tags that have been read once from the reader in a When the reader uses a frame size of N slots, the previous read cycle, after activation, will not send probability that k out of n tags to select a specific slot and their data again (Identification Number) if they not the others is binomially distributed [5]. The number of reenter the reader’s field. This is like flipping an the tags k that occupy a specific frame slot is known as the “inventoried flag” from A ! B or B ! A during an occupancy number of the slots [3]. inventory round according to EPCGlobal Gen 2 UHF The time delay T which is necessary for each tag to RFID [4], so the inventoried RFID does not reply transmit its Identification (ID) successfully is T ¼ N Á i,
  4. 4. SU ET AL.: MULTIPLE RFID TAGS ACCESS ALGORITHM 177 Fig. 2. Throughput S of DFSA with optimal frame size selection.Fig. 1. Throughput S of FSA for different frame sizes. of both subplots because when there is only one tag in thewhere i is the number of read cycles or number of area, the probability for a tag to collide in practice is equalretransmissions. Let Pempty be the probability that a slot is to zero and the reader always read this tag and so S ¼ 1.empty, and Pc is the probability that a collision has occurred After some tags are added, throughput falls very quickly toin a slot. Then, Pempty ¼ ð1 À N Þn and Pc ¼ 1 À Pempty À Pread . 1 its final value. The optimal frame size Noptimal for the next read cycle can Fig. 3 illustrates the time delay (T ) in both FSA forbe calculated in two ways [3], as given below: different frame sizes and DFSA for n ¼ 0 À 512 tags. In DFSA, it is assumed that the reader’s initial frame size is 1. Maximizing the throughput S of the RFID system, equal to the optimal, which in practice is not always true. which is defined as Fig. 3 shows that the DFSA algorithm increases the P robability of Reading a T ag performance of the RFID system in terms of the time delay. S¼ T is greater if the FSA algorithm is used. The performance Ptotal Pread of the FSA algorithm decreases as the difference between n ¼ and N increases. The FSA with N ¼ 512 and DFSA seem to Pc þ Pread þ Pempty have the same behavior; this is only true when the number Pread of tags is near 512 (actually for n ! 350 tags). For a small ¼ ð1 À Pempty À Pread Þ þ Pread þ Pempty number of tags (n 220 tags), FSA with N ¼ 512 gives the ¼ Pread worst performance. 1 1 nÀ1 2.2.2 Developments in DFSA Algorithm therefore S ¼ Á 1À : N N In [8], Lee et al. proposed an alternative probabilistic Aloha- ð5Þ type anticollision protocol for RFID systems. It is called Enhanced Dynamic Framed Slotted ALOHA (EDFSA), which is similar to the DFSA algorithm with one difference. After the 2. Minimizing the time delay T which is given in [3] as estimation of the number of tags: N T¼ : ð6Þ ð1 À N ÞnÀ1 1 Fig. 1 shows the behavior of the RFID system if the FSAalgorithm and (5) are used. It is obvious that when the numberof tags is greater than the selected frame size, the system’sefficiency decreases and the RFID system is unstable. Both of the above methods provide the same Noptimalsolution [3], which is n. Therefore, p ¼ 1=n. Thus, in DFSA,if the reader allocates a frame size equal to the number oftags in the interrogation area, the efficiency of the RFIDsystem increases. The only problem now is that the readerneeds to estimate this value n before it transmits the nextREQUEST command. Fig. 2 illustrates the stability in throughput efficiencyprovided that the DFSA algorithm is used; so, Noptimal ¼ n isselected at the reader at every duty cycle. The figure showsthat throughput is always equal to the maximum theoreticalvalue of 36.8 percent of the Slotted Aloha protocolregardless of n. Moreover, there is a spike at the beginning Fig. 3. Performance of RFID system in terms of time delay.
  5. 5. 178 IEEE TRANSACTIONS ON MOBILE COMPUTING, VOL. 9, NO. 2, FEBRUARY 2010 . if the number of tags n is greater or equal to 177, it Moreover, it will be assumed that the chips on the tags are divides the tags into smaller groups (called “Modulo using the current semiconductor technology, which de- operation”); the number of smaller groups is creases the power consumption in the tag, and a minimum dnumber ofN unread tags e, and thus, a smaller frame size is power received (Ptag;min ) equals to 50 watts is enough [1] used (always N ¼ 256 slots) or, for the tag to transmit the data stored in its memory. . if the number of tags is smaller than 177, then the In the following sections, a detail understanding of the DFSA algorithm is used. maximum distance of the RFID system, the effects of Although the author’s simulations in [8] showed im- passive backscattering modulation on the frame size, andprovement in system efficiency versus frame size, this the benefits of capture effect on FSA type algorithms aremethod needs extended calculations in the reader and presented. These in-depth understanding allows us toincreases the tag’s complexity and power consumption, develop the proposed PS algorithm. For example, thesomething which is critical in passive RFID systems, maximum distance constraints how much power can bebecause tags are powerless. For this reason, the next section transmitted, and the frame size affects the tag collisionproposes a new type of DFSA anticollision protocol based probability, which dictates the number of tags within a scan.on simplicity and the constraint of minimizing power 3.1 Maximum Distance of the RFID Systemconsumption in the transponder (tag). First, it is important to evaluate the maximum distance (rmax ) Moreover, in both DFSA and EDFSA algorithms, the of a typical passive RFID system in the microwave band. Thereader can only calculate the number of tags that responds distance rmax refers to the maximum distance a tag can bewithout collision which is represented by the random placed to receive the necessary power from the reader. Avariable R. Also, the reader can calculate the number of slots part of this power supplies the inner circuit of the tag towith collision (given by random variable C) but how it perform all the necessary operations to wake up the tag, andestimates the number of tags that collides is unknown. It is another part is used to transmit the data back to the reader.known that for every slot that collision occurs at least two tags The maximum distance is smaller than the actual readinghave transmitted their data to it. Thus, the number of slots N distance due to the attenuation in the reverse link.that the reader must use in the next frame is given by [7] The Friis free space equation capturing the relationship N ¼ R þ 2C: ð7Þ between the transmitted power from the reader (Pr ) and the power received by the antenna of the tag (Ptag ) is as follows: 3 A PROGRESSING SCANNING TECHNIQUE 2 Ptag ¼ Pr Gr Gtag ; ð8ÞThis section presents an alternative hybrid anticollision 4rprotocol based on the FSA algorithm for microwave RFID where Gr is the transmitter antenna gain, Gtag is the receivercommunication systems at 2.45 GHz. The parameters of thecommunication system will be according to those estab- antenna gain, is the wavelength in meters, and r is thelished in [9] by the ISO and the International Electrotechni- distance between the transmitter and antenna; (8) assumescal Commission (IEC) for MODE 1 systems, which are the that there is no loss from the transmitter and receiverpassive backscatter RFID systems of interest. equipments. By using the Friis relationship for the free The proposed protocol takes into account the physical space path loss, the one way (reader to tag) path loss aF islink and Media Access Control (MAC) parameters. The Pr Ptag . Substituting Ptag with Ptag;min , (9) gives the maximummost important of these parameters are the following [9]: allowable path loss that the RFID system can experience and . The maximum transmitted power (Pr;max ) measured is still capable of transmitting the stored data to the reader. at the reader’s antenna is 4 Watts (36 dBm) EIRP. Pr . The modulation, which is used in the forward link aF ¼ : ð9Þ Ptag;min (reader to tag), is Amplitude Shift Keying (ASK). ASK is the simplest waveform and can be detected From (9), the maximum distance for a tag with a dipole easily by the tag with a simple circuit. antenna and a reader with maximum EIRP of 4 Watts at the . The modulation, which is used in the reverse link output of the reader’s antenna is calculated and plotted in (tag to reader), is Backscatter Modulation (BM). Fig. 4. Fig. 4 illustrates the reading distance versus the . The data coding is Manchester for the forward link transmitted power from a reader. The maximum distance and FM0 (Bi-Phase Space) for the reverse link. FM0 rmax is 3.5 m, and this result agrees with [9], which specifies is actually a differential Manchester technique. 4 m for a typical passive RFID system with data rates up to . Both the tag and the reader have error detection 30 Kbps. Also, Finkenzeller [1] specifies 3 m as the lower capability by using a CRC with 16 bits. bound for backscattering RFID systems (UHF and Micro- . The data bit rate should be between 30 and 40 Kbps wave bands). in both directions. The results in Fig. 4 are for LOS communication between . Tags use the reader’s signal for synchronization. reader and tags, which is almost a prerequisite for . The maximum occupied channel Bandwidth (BW) is microwave RFID systems. If the RFID system is inside a 500 KHz. building, for a line-of-sight communication, the path loss . The memory size of the tags varies from 8 bytes to exponent is less than 2. Usually, a value of 1.6-1.8 is used for 64 bytes. this case [10], and therefore, the maximum reading distance
  6. 6. SU ET AL.: MULTIPLE RFID TAGS ACCESS ALGORITHM 179 TABLE 1 Rayleigh Distance for Different Antenna Types at 2.45 GHz Âð10ÁlogPtag;min ÞÀaF ðro ÞÃ Pr r ¼ ro Á 10 10Án ; ð12Þ where n is the path loss exponent. Fig. 5 shows the reading distance of two cases: 1) an indoor environment with path loss exponent n ¼ 1:6, and 2) an urban environment with n ¼ 3. For applications, such as scanning products in a store, n ¼ 1:6 is usually the case;Fig. 4. Reading distance of a typical passive RFID system. thus, a reading distance of up to 5.5 m can be achieved. 3.2 Selection of Backscattering Modulationcan be higher than 3.5 m. To determine the maximumdistance in an indoor environment, it is first necessary to In passive RFID systems at the microwave band, the readercompute the path loss aF at a reference distance ro from the transmits the ASK modulated carrier to the shared wirelessreader. The selection of the reference distance ro is not channel. This carrier provides the tag with enough power toarbitrary for the RFID system and is determined in the energize it and is also used by the tag as a carrier forfollowing sections. transmitting its ID in the reverse link by using the backscatter modulation. The tag reflects the reader’s signal3.1.1 Evaluation of Reference Distance (ro ) by changing the impedance of its antenna according to theThe reference distance ro could be any distance in the far- bits that are transmitting, or in other words, it changes thefield region of the reader’s antenna, which is much smaller gain of the antenna [11], [12]. As a result, small fluctuationsthan the maximum distance of this specific system [10]. It is occur in the amplitude of the carrier’s signal.known from antenna theory and design that the far-field When the signal returns, the reader needs to “peak-detect”distance or Rayleigh distance of the antenna is given by the modulation of the tag in the carrier and then decode it. A 2D2 high value in the envelope of the carrier is represented by a rff ¼ ; ð10Þ binary one “1” and a low value of a binary zero “0.” If this is the only change that occurs in the reader’s signal, this type ofwhere D is the maximum size of the antenna in meters and backscatter modulation is called direct modulation and it is is the wavelength of the operating frequency. Anydistance r that meets the following requirements lies in simply an ASK modulation. In addition, the tag can alsothe far-field region of the antenna. change the phase or the frequency of the carrier signal, and thus, create a PSK or FSK modulated signal. . r rff , . r ) D, and finally . r ) . Table 1 gives the Rayleigh distance for different types ofreader antennas at 2.45 GHz, where is 0.1224 m. For anantenna in the microwave band, the far-field region isdetermined by the wavelength. The selection of ro ¼ 0:5 m(more than four times ) satisfies the previous require-ments and is eligible to be selected as a reference distancefor the evaluation of the reference path loss from the Friisequation. The reference path loss at reference distance ro ,aF ðro Þ, is 32 dB.3.1.2 Maximum Distance for RFID Systems in Different EnvironmentsFrom the theory of large-scale path loss at distance r, aF isgiven in [10] by aF ðdBÞ ¼ aF ðro Þ þ 10 Á n Á log10 ðr=ro Þ: ð11ÞUsing (9) and (11) and solving for distance r, the following Fig. 5. Maximum reading distance for different values of path lossrelationship is found: exponent.
  7. 7. 180 IEEE TRANSACTIONS ON MOBILE COMPUTING, VOL. 9, NO. 2, FEBRUARY 2010Fig. 6. Backscatter amplitude modulation signal. Fig. 8. Probability of tag ID error for binary ASK backscattering (direct Modulation). rffiffiffiffiffiffi Eb Pb ¼ Q ; ð13Þ No E where Nb is the average signal-to-noise ratio per bit and QðxÞ o is the Q-function. In the above expression, an Additive White Gaussian Noise (AWGN) channel with no fading is assumed.Fig. 7. BER with ASK-backscatter modulation. The probability for a tag to transmit its ID without having any bit in error is given by Fig. 6 illustrates a direct modulated backscatteringcarrier signal from the tag to the reader. A carrier signal PF ¼ ð1 À Pb ÞL : ð14Þis an ASK modulated sine wave with amplitude 100 V. In the above equation, L is the length in bits of the frameThe tag creates a drop of 100 mV in the amplitude of the that the tag is using to transmit its ID and is assumed to becarrier for each transmitted binary zero “0” [13]. The the same for all tags. Finally, the probability for a frame toreader peak-detects this signal, decodes it, and thus be in error is given asidentifies the tag. For the case of passive RFID tags in the microwave band, Pe;F ¼ 1 À PF ¼ 1 À ð1 À Pb ÞL : ð15Þthis kind of ASK-backscattering modulation is selected Fig. 7 illustrates the Pb for different values of . Note Eb Nobecause of the simplicity either in detection from the reader E that Pb becomes negligible for Nb greater than 7 dB (less than oand in computations inside the tags, which, as a result, 1 percent error).reduces the cost of the tag. In addition, direct modulation Equation (15) shows that shorter tag messages (smaller(ASK) provides higher data rates (up to 40 Kbps) than PSK L) result in lower probability of frame error Pe;F , or in otherand FSK backscatter modulation [11], [13]. General passive words, a lower frame error rate. Fig. 8 plots (15) for threetags must have as simple functions as possible to reduce the frame lengths:power consumption, and thus, to increase the maximum . L ¼ 64 bits, which is the minimum frame lengthworking distance of the RFID system [14]. from the ISO standards, and also is the length of the Also, the messages transmitted from the tags must be as Unique Identification (UID) of the tag,short as possible for two reasons: . L ¼ 144 bits, which is the recommended frame length from ISO standards, and, . Shorter messages mean less power consumption in . L ¼ 512 bits, which refers to the maximum frame the tag. length used in typical passive RFID systems. . Shorter messages have lower probability of error in It shows that for smaller tag messages, the frame error rate transmitting the tag’s ID as discussed in the decreases, and moreover, for a 7 dB signal-to-noise ratio, the following section. probability a frame is received in error is now significant3.2.1 Probability of Error in Transmitting Tag’s ID with (8 Â 10À1 for L ¼ 144 bits). Since only a 16 bit CRC is used for error detection and ASK Backscattering forward error correction coding is not applicable, it is veryWhen ASK is the only modulation (backscattering) used in important to use short messages and for the signal receivedthe reverse link, then the Bit Error Probability (BER) Pb is in the reader to be as strong as possible to reducegiven by (for coherent detection): retransmissions and overhead in the reverse link.
  8. 8. SU ET AL.: MULTIPLE RFID TAGS ACCESS ALGORITHM 181 Fig. 10. Throughput of FSA with the use of capture effect.Fig. 9. Probability of tag error for 96 bits frame length. where T is the selected threshold at the reader, which is3.2.2 Selection of Tags Frame Length called capture ratio and corresponds to how many timesThe frame that the tag transmits to the reader (response) greater the received power in reader should be from aconsists of the following fields [9]: specific tag as compared to the summation of the received powers from the remaining tags that occupy the same slot . Quiet. The tag does not transmit for a specific period in order to be identified by the reader. The throughput of time determined from the protocol. reaches its maximum value of 1þT when G is 1 þ T . The 1 eT . Return Preamble. This consists of 16 bits in a specific value T in (16) is in decimal form instead of dB. sequence, which enables the reader to lock the data Fig. 10 shows that a higher throughput can be achieved from the tag and start decoding the message. in the RFID system by “filtering” the weak signals with the . The data field with at least 64 bits for the UID plus the rest data bits to transmit other kinds of threshold in the same slot and keeping only the survivor information stored in the tag’s memory. signal. It can be seen from this figure, that for a threshold . A 16 bit CRC for error detection. value of 3 dB, maximum throughput increases to 0.55, and for 6 dB, it increases to 0.46. Since it is important as previously mentioned to keep thetag’s message as short as possible, but on the other hand, as So as the threshold T decreases, the performance of theerror detection capability and the return preamble are RFID system increases, and as this takes place for highernecessary, a frame length of 96 bits is selected. However, values of G, so does the stability of the FSA algorithm. Onethe memory size of the tags can be larger (144 bits is the could say that using a low threshold in the reader wouldrecommended standard), and thus it could be compatible eliminate the collision problem, but this cannot happen ineven with the EPC (96 data bits). Fig. 9 shows the frame error practice, because T is determined by the reader’s sensitivity.rate for L ¼ 96 bits. For a signal-to-noise ratio less than 11 The typical reader’s sensitivity requires 6 dB differencedB, the frame error rate is still significant (7 Â 10À1 for 7 dB). between the signal from the tag of interest and the channelA signal-to-noise ratio greater than 11 dB is necessary to noise (white noise plus interference from other tags) inachieve low frame error probability (lower than 10À2 ). order to identify the tag [1]. Therefore, only an increase up to 46 percent in performance in terms of throughput can3.3 Increasing the Efficiency of FSA with Capture be achieved. Effect Table 2 shows the maximum theoretical throughput ofThe tag’s responses in the reverse link can be identified the FSA algorithm for different threshold values. It isfrom the reader even if they collide (occupy the same slot). obvious that theoretically, even for very high thresholdThis can happen if the strength of one signal is higher than values, such as 20 dB (a cheap reader with very lowthe rest of the signals in the same slot. This is known as the sensitivity), the performance is still better than the max-capture effect [1], [10]. The capture effect is used very often imum of the FSA algorithm without using filtering in thein most common cellular systems. For this paper, it is receiver of the reader.possible to take advantage of the capture effect and increasethe throughput S of the FSA algorithm by choosing the 3.4 Progressing Scanning (PS) Algorithmappropriate threshold in the reader, which acts as a filter for The proposed PS algorithm considers two constraints of thethe weak signals. As a result, the reader identifies the tageven if collision has occurred in the specific slot. RFID system: The throughput S in the Slotted Aloha algorithm with . the power consumption in tags must be minimum inthe capture effect is given [15] by order for the RFID system to achieve the maximum distance, and S ¼ G Á eÀð1þT Þ ; TG ð16Þ . the RFID system must be simple, especially the tags.
  9. 9. 182 IEEE TRANSACTIONS ON MOBILE COMPUTING, VOL. 9, NO. 2, FEBRUARY 2010 TABLE 2 Maximum Theoretical Throughput for Different ThresholdsIn the PS algorithm, the reader takes advantage of the range Fig. 11. Number of tags on each cycle (uniform distribution around thedifference between the tags and the reader’s antenna. Tags reader).that are near the reader receive more power from the readerthan those which are further away. . Finally, in the last scanning, the reader transmits In the PS algorithm, the reader starts transmitting from a with Pr ¼ Pr;max .minimum EIR power level Pr;min until the maximum Pr;max . This is the end of the first cycle of the PS algorithm, P ÀPthat is permitted by the regulations. Tags that are further which consists of d r;max k r;min e transmissions. Afterfrom the reader do not receive enough power and thus this point, a new cycle with multiple scans beginscannot transmit their IDs. In each retransmission, the reader and the whole procedure repeats until there are noincreases the transmitted power by an increment k and the more tags in the interrogation zone.tags that are further in distance reply. This continues until . After each scan, the frame size changes according tothe transmitted power reaches Pr;max . Then, a new cycle the frame size estimation algorithm if dynamicbegins and the procedure is repeated from the beginning. frame size is used. At the beginning of each cycle, The PS algorithm divides the number of tags n in the the first scan should use the minimum frame sizearea into smaller groups as in the EDFSA algorithm selected from a set of available frame sizes; this is tointroduced in [8], and therefore, this method has the avoid using a large frame size estimated at previousbenefits of EDFSA since the reader does not use large scan, which is the last scan of the previous cycleframe sizes that reduce the efficiency of Aloha-based using the maximum transmit power thus covering aalgorithms [8]. Also, it does not have the complexity of large area.EDFSA to set up the groups. The PS algorithm is an alternative and simplex method A detail description of the PS algorithm is as follows: to divide the tags in the interrogation zone into smaller groups like in EDFSA, without any involvement from the . At first, the reader transmits with Pr ¼ Pr;min . The tags. As a result, PS algorithm decreases the complexity of tags at distance rt rmax;pi , where rmax;pi is the the tags. Thus, PS has all the benefits of EDFSA in the maximum distance from Fig. 4 that corresponds to performance of the Framed Slotted Aloha. this transmitted power, become energized, and reply Fig. 11 shows that the PS algorithm successfully divides using the FSA protocol. . Next, the reader increases the power level by k, and the tags in the interrogation zone into groups with a fewer the aforementioned procedure repeats, but now number of tags. In the simulation, 1,000 tags were randomly with transmitting power Pr ¼ Pr;min þ k. All new generated and uniformly placed around the reader with tags that entered the interrogator zone of the reader distances of 0-3.5 m. In addition, two different minimum reply. Tags from the previous scanning do not reply transmitted power values Pr;min were used. The reason that to the reader’s command. This can be accomplished more tags are in the first group for both cases is the result of if the reader transmits a command in the header that the reverse proportional relationship between Pr and r. For informs the tags, which have already transmitted example, for Pr;min ¼ 1 Watt, this corresponds to r ¼ 1:7 m, once, not to reply until the next cycle. Of course, the which is almost half the maximum distance. By increasing tags need to have been programmed to do so. This the transmitting power by a factor of 0.2 or 0.5 Watts, the programming can be done from the manufacture by increase in parameter r is almost insignificant; thus, fewer using a flash memory in tags for quick loading to tags are included in the groups following the first one. compare its state, or by using tags with Read/Write As Fig. 11 illustrates, the number of tags in each group memory. For example, this is like flipping an “inventoried flag” from A ! B or B ! A during a decreases as the minimum transmitted power from the inventory round according to EPCGlobal Gen 2 reader and the increment k decreases. However, both smaller UHF RFID [4], so the inventoried RFID does not Pr;min and k must, as a result, increase the times the reader reply again. needs to scan to identify all the tags in the interrogator zone, . This aforementioned procedure continues with Pr ¼ which thus, negatively affects performance in the PS Pr;min þ i Á k (i ¼ 1; 2; 3; . . . ). algorithm as will later be demonstrated. On the contrary, if
  10. 10. SU ET AL.: MULTIPLE RFID TAGS ACCESS ALGORITHM 183 at 3.54 m, which is the maximum distance for an outdoor line-of-sight RFID system at a frequency of 2.45 GHz. The number of tags n in the interrogation zone is 1,000 and the simulation is ran 1,000 times to increase the accuracy of the results. The scope of the simulation is to evaluate the performance in terms of the time delay needed for the identification of all the tags. The time delay T for FSA in terms of the number of slots is given by (6). Thus, the delay in units of seconds is given by the following equation: N L½bits=slotŠ TF SA ¼ À Á ½slotsŠ Á ð17Þ 1 nÀ1 R½bits=secŠ 1ÀN N LFig. 12. Performance comparison of PS algorithm using constant power ¼À ÁnÀ1 Á ½secŠ; ð18Þstep with variable one. 1 1ÀN Rthe number of tags is too high, small values of Pr;min and k are whereneeded to decrease the number of tags in each group. . L is the length of the tag response in bits (L ¼ 96 bits If instead of a constant value for the increment k, a is selected).variable one kv is used, this will decrease the number of . R is the bit rate from the ISO standards (a minimumtimes the reader needs to transmit in one cycle. The only 30 Kbps is selected).requirement is that the increment should increase when the To calculate the delay from (18) for the PS algorithm, it istransmitted power is increased, or in other words, smaller important to understand that the PS algorithm is just a FSAvalues of kv should be used with the first scans where more procedure with the only difference being that the number oftags are involved in the identification process and higher tags in the interrogation zone is divided into multiplevalues when the transmitted power reaches Pr;max . groups. The delay for each group which contains ni tags is Fig. 12 compares the performance of the PS algorithm that then given asuses constant power step increase (k ¼ 0:2) with a variable one (k ¼ f0; 0:3; 0:9; 1:5; 2:5; 3:1; 3:6g). The tags are uniformly N Ldistributed around the reader. The solid line corresponds to TP Si ¼ 1 ni À1 Á ½secŠ; ð19Þ ð1 À N Þ Rthe constant step increase and divides the tags into19 groups, while the dashed line represents the variable step. where i indicates the group number, and ni is the number of As seen in Fig. 12, an increasing variable step decreases tags in this group that the PS algorithm has created.the number of scans of the PS algorithm, which has a Thus, the total delay in seconds of the PS algorithm is thesimilar effect if a higher constant k value was used. summation of the individual delay for each group and isGenerally, a variable step can be avoided because it given byincreases the complexity of the system. However, when X imax !the distribution of tags is not uniform, it might be useful. N L TP S ¼ Á ½secŠ: ð20ÞFor example, if the tag’s distance from the reader follow a i¼0 ð1 À N Þni À1 1 RGaussian distribution, smaller increments can be used nearthe mean distance and larger ones for tags that are many This average total delay does not include the time instandard deviations greater. which the reader needs to temporally inactivate the tags after identification. Such omission is also done for FSA. Since the time to inactivate the tags for both FSA and PS4 PERFORMANCE EVALUATION algorithms is the same, comparing the performance of PS toPerformance comparison between FSA and progressing FSA using (18) and (20) is fair.scanning algorithm is discussed in Section 4.1. Afterward, a Table 3 summarizes the simulation results of the PS anddetail analysis of the PS algorithm and comparison of the PS FSA algorithms for Pr;min of 0.2 Watt and step size k ofand DFSA algorithms with dynamic frame size estimation 0.2 Watt; thus, imax is 18 for this case. The third column of thisare given in Sections 4.2 and 4.3, respectively. table shows how many times (power levels) the PS algorithm needs to transmit to complete one cycle, while the fourth4.1 Comparison between FSA and Progressing column shows how many tags were identified in the first Scanning Algorithm cycle. Finally, the last column gives the average time delayIn order to compare the performance of the PS algorithm calculated from (18) and (20) for each case, but except for FSAwith FSA, it is necessary to simulate in discrete time using with N ¼ 64 and N ¼ 128 slots, which is from the simulation.Matlab as the frame transmitted from the reader and also The results indicate that when a small frame size is usedthe random selection of a slot from the tags. A fixed frame (N ¼ 64 or N ¼ 128 slots), the performance of the PSsize is used both for the FSA and PS algorithms. The algorithm is much better than that of FSA, which is unabledistance of the tags from the reader is assumed to follow a to identify such a large number of tags due to collision.uniform distribution with minimum at 0 m and maximum Moreover, the time delay introduced by the PS algorithm is
  11. 11. 184 IEEE TRANSACTIONS ON MOBILE COMPUTING, VOL. 9, NO. 2, FEBRUARY 2010 TABLE 3 FSA Algorithm versus PS Algorithm with 1,000 Tagsreasonable, which is between 7 and 10 seconds for of tags increases. This was expected because the PS1,000 tags. When a larger frame size is used (N ¼ 256 algorithm uses 19 power levels and thus divides the tagsslots), the PS algorithm is able to read more tags in the first into 19 groups. Therefore, an increase of 10-20 tags does notcycle than can FSA with a greater frame size, but more affect the performance since the tags are mapped intotransmissions occur than in FSA. In addition, the average 19 groups and not just one group (as in FSA).time delay is much lower in the PS algorithm than in FSA According to Fig. 14, the PS algorithm has high initial(56 percent lower), and thus, the PS algorithm rapidly delay of around 15.8 sec for scanning 100 tags. This occursincreases the performance of Slotted Aloha in terms of because the PS algorithm transmits 19 times instead of onlyidentification time. It is important to notice from Table 3 1 time, and also after the first cycle, the reader needs tothat FSA is ineffective in identifying the entire number of transmit again in all those different power levels corre-tags in the area when the frame size is small. On the other sponding to different distances or tag groups, even if somehand, the PS method with the same frame size requires a of those groups do not contain any more tags. To minimizevery low average identification time. the effect of the initial delay, a small frame size should be In Fig. 13, the average time delay T is plotted for both the used. This is shown in Table 3, where the delay is smallerFSA and PS algorithms. It is evaluated for tags ranging from for small frame size when reading 1,000 tags.100 to 4,000. For a better presentation of the results, a linear Having said that, Fig. 15 illustrates the performance ofscale was used for the horizontal axis (T ), while a two different frame sizes, 64 and 256 slots, according tologarithmic scale was used for the vertical axis (number of Table 3. The results show that the frame size of 64 slots has atags). According to Fig. 13, the PS algorithm performs much lower delay than 256 slots for number of tags smaller thanbetter than the FSA algorithm when the number of tags in 2,760. However, 2,760 tags is a big number. Normally, inthe areas is too high; for example, for 4,000 tags, the PS most applications, the number of tags does not reach thisalgorithm requires less than 30 seconds while the FSA number. For this reason, in most of the simulations, thealgorithm requires more than 120 seconds. However, if the authors use up to 1,000 tags. As a result, a smaller framenumber of tags is less than 790, the FSA algorithm has lower size is the best choice when the PS algorithm is used.average delay; thus, the FSA performs better. In addition, a smaller frame size in the PS algorithm is Fig. 14 more distinctly shows the variation of the average not only superior than using a larger frame size in the PSdelay with the number of tags for the PS algorithm. This algorithm, but it is also the best choice as compared to FSAfigure shows that T increases very slowly while the number with a larger frame size. However, because of the initialFig. 13. Comparison of performance (delay) between FSA and PS. Fig. 14. Delay incurred when using PS algorithm.
  12. 12. SU ET AL.: MULTIPLE RFID TAGS ACCESS ALGORITHM 185 Fig. 17. Performance of the PS algorithm with frame size (N ¼ 32; 64; 128, and 256 slots) and constant step (k ¼ 0:2 Watts).Fig. 15. Comparison of delay for PS and FSA. . the minimum transmitted power Pread;min , whichdelay that the PS algorithm introduces (due to multiple determines the number of scans in each transmittedscans), it has a lower performance than the FSA for a small cycle as well as the effectiveness of the algorithm innumber of tags in the interrogation zone (less than 480 in the first scan, andthis case as shown in Fig. 15). . the number of tags n in the interrogation zone. Furthermore, the PS algorithm performs better when a To evaluate the performance of the PS algorithm in termslarger step size is used, i.e., 0.4 watts as shown in Fig. 16. In of the average time delay and to observe how thisthis case, the tags are divided into fewer groups than before. performance is affected by the number of tags in theBoth frame sizes of the PS algorithm have a smaller initial interrogation zone, the step size, and the frame size, the PSdelay and exceed the performance of FSA sooner, e.g., at algorithm is simulated and three-dimensional mesh plots338 tags for frame size of 64 slots. are created. In the simulation, the minimum transmitted power is4.2 Detail Analysis of PS assumed constant and equals to 0.4 Watts. Also, either theAs observed from the plots described in the previous section, frame size or the step size is kept constant, depending onthe effectiveness of the PS algorithm against the FSA is not which plot is referenced. Fig. 17 shows the performance ofalways the same but depends on the following variables: the proposed algorithm for different frame sizes of 32, 64, . the selected frame size N of the reader, 128, and 256 slots. A constant step size of 0.2 Watts is used, . the increment (step) k in watts used by the reader, and the number of the tags is increased from 100 to 2,000. which determines the number of scans in each The horizontal y-axis represents the number of tags in the transmitted cycle, interrogation zone, while the x-axis represents the selected frame size. Finally, perpendicular to the xy plane, the z-axis gives the average time delay for the PS algorithm. As Fig. 17 shows, the slope of the mesh plot with respect to the y-axis is smaller for a smaller frame size and increases while the frame size increases. That means the initial delay is higher when a large frame size is used by the reader, and the tags in the area are just a few. This was expected since the reader needs more time to transmit the larger frame and it also waits longer for the response from the tags. Moreover, the average time delay is increased at a higher rate for the smaller frame size than the larger while the number of tags is increased. This was expected as well, due to the increasing number of collisions which occurs when a small frame size is used. It is easy to observe that for 2,000 tags, the frame size of 32 slots has a delay of over 40 seconds, which is twice that of 256 slots frame size. However, with 100 tags, this frame size is the most efficient. The above results show that the performance of the PS algorithm is closely related to the existing number of tags as in the FSA algorithm. Therefore, it is very important for the reader to estimate the number ofFig. 16. Comparison of delay for PS and FSA with larger step size. tags before the next cycle, as in DFSA.
  13. 13. 186 IEEE TRANSACTIONS ON MOBILE COMPUTING, VOL. 9, NO. 2, FEBRUARY 2010Fig. 18. Performance of the PS algorithm with step size (k ¼ 0:2; 0:4; 0:9, Fig. 19. Total time to scan tags: PS versus DFSA. (Choices of N are 64,and 1.8 Watts) and constant frame size of 128 slots. 128, 256, and 512.) Fig. 18 shows the performance of the proposed algorithm identify them using the Aloha algorithm. The exact value ofwhen the power transmission step size is not constant. The the step size depends on the number of tags and also thestep sizes are 0.2, 0.4, 0.9, and 1.8 Watts, and the frame size selected frame size.is constant and equals to 128 slots. In this three-dimensional For the case of the right frame size, it is essential for themesh plot, the y-axis represents the number of the tags, and reader to have a good estimation of the number of tags. Thisthe x-axis the different step sizes used. The vertical z-axis is the main problem from which both DFSA [3] and EDFSAonce again gives the average time delay of the algorithm. [8] also suffer. The simplest estimation can be provided by The first subplot has a maximum number of tags in the the lower bound of (7), which is N ¼ R þ 2C.interrogation area equal to 1,000, while in the second After the first cycle of transmissions, the reader calculatessubplot, the maximum number of tags is 2,000. This was the frame size N by counting the number of identified tags,done for better presentation of the results when the number given by R, and the number of collisions, given by C. Thisof tags is increased rapidly. The second subplot of Fig. 18 easily shows that when the bound should work better for the PS algorithm since thenumber of tags is too high, the average time delay is number of collisions in each scanning is less than in FSA orincreased rapidly when a large value for the step size is DFSA using the same frame size, because the number of tagsused; thus, the performance of the PS algorithm decreases. is not as great as in those algorithms.This inverse proportional relationship between the delay 4.3 Comparison of PS and DFSA with Dynamicand the step size is logical, because when a large value for Frame Size Estimationthe step size is used, the PS algorithm is almost the same asthe FSA algorithm; thus, in order to identify a large number Since both PS and DFSA can enhance the performance ofof tags, more time is needed due to collisions, or larger reading the tags by adaptively changing the frame size, theyframe size is used as in DFSA [3], [8]. are simulated and compared. Both of them use (7) to In the first subplot, it is more obvious that the step size estimate the next frame size after each scan. The choices ofaffects the performance of the PS algorithm. If the tags are frame size, N, are 64, 128, 256, and 512. If (7) gives a valueonly a few, a higher step size is better, and actually as shown between two of the choices, the larger frame size is selected.in Figs. 15 and 16, the FSA is even better. However, when the The maximum frame size is 512 and minimum is 64.number of tags is increased, smaller step sizes result in less As shown in Fig. 19, DFSA takes more than 300 secondsdelay. In this subplot, the step size of 0.4 Watts generally while PS takes less than 50 seconds to scan 3,600 tags. Theseems to result in better performance for the PS algorithm. performance of PS improves as the step size k reduces. The performance results indicate that in order for the Having a smaller step size allows the reader to scan a smallerproposed algorithm to be effective, it is important to select region multiple times, thus, reducing the number of colli-the right frame size as in DFSA as well as the right step size. sions. If the step size k is reduced lower than 0.01 Watts, theTo do so, it is essential to know the number of tags inside time needed to read 3,600 tags will be smaller than 50 seconds.the interrogation zone as this mainly affects performance Furthermore, PS outperforms DFSA when the number of tagsand is closely related to the other two variables. is greater than 600. Since the PS algorithm is more effective than the simple In addition, Fig. 20 illustrates the energy consumed by theFSA when the number of tags in the interrogation zone of reader in order to read the tags. DFSA consumes 2,000 Joulesthe reader is high, only in that case it is necessary to while PS with k set to 0.01 Watts consumes aroundimplement it. Thus, the selection of the step size is simple. It 100 Joules; that is around 20 times saving. It is interestingmust be small enough (0.2-0.4 Watts) to divide the tags into to note that the smaller the k value the less energy consumed.many small groups in order for the reader to be able to This corresponds to less time spent reading the tags.
  14. 14. SU ET AL.: MULTIPLE RFID TAGS ACCESS ALGORITHM 187 [8] S. Lee, S. Joo, and C. Lee, “An Enhanced Dynamic Framed Slotted ALOHA Algorithm for RFID Tag Identification,” Proc. Second Ann. Int’l Conf. Mobile and Ubiquitous Systems: Networking and Services (MobiQuitous ’05), pp. 166-172, July 2005. [9] ISO/IEC, “18000 Part 4: Parameters for Air Interface Commu- nications at 2.45 GHz,” ISO, 2004. [10] T.S. Rappaport, Wireless Communications—Principles and Practices, second ed. Prentice Hall, 2002. [11] R. Bridgelall, “Bluetooth/802.11 Protocol Adaptation for RFID Tags,” Symbol Technologies, RFDESIGN, July 2002. [12] G.D. Vita and G. Iannaccone, “Design Criteria for the RF Section of UHF and Microwave Passive RFID Transponders,” IEEE Trans. Microwave Theory and Techniques, vol. 53, no. 9, pp. 2978-2990, Sept. 2005. [13] P. Sorrells, Passive RFID Basics, AN680, DS00680B, Microchip Technology, Inc., pp. 1-5, 1998. [14] F. Zhou, C. Chen, D. Jim, C. Huang, and H. Min, “Evaluation and Optimizing Power Consumption of Anticollision Protocols for Applications in RFID Systems,” Proc. Int’l Symp. Low Power Electronics and Design (ISLPED ’04), pp. 357-362, Aug. 2004. [15] F. Borgonovo and M. Zorzi, “Slotted ALOHA and CDPA: A Comparison of Channel Access,” Wireless Networks, vol. 3, pp. 43- 51, 1997.Fig. 20. Energy consumed to scan tags: PS versus DFSA. (Choices of Nare 64, 128, 256, and 512.) Weilian Su received the BS degree in electrical, computer, and systems engineering (ECSE) from Rensselaer Polytechnic Institute in 1997 If the number of nodes is much greater than 3,600, the with Summa Cum Laude, and the ECSEmaximum frame size should be greater than 512. If not, department’s Lockheed Martin Capstone Designcollisions will occur to a point where all the tags cannot be Award. He also received the MSECE and PhDread. For example, as shown in Fig. 19, DFSA with a frame degrees in electrical and computer engineering from Georgia Institute of Technology in 2001size of 512 seems to approach a scanning limit of 4,000 tags. and 2004. He specializes in sensor and ATM networks under the guidance of Dr. Ian F. Akyildiz in the Broadband and Wireless Networking Laboratory at5 CONCLUSIONS Georgia Institute of Technology. In 2003, he received the 2003 Best Tutorial Paper Award from the IEEE Communications Society.This paper proposes a variation of the FSA called the Currently, he is an assistant professor at the Naval PostgraduateProgressing Scanning algorithm. The Progressing Scanning School. His current research interests are sensor networks, ad hocalgorithm improves the performance of the FSA when the networks, quality of service in Internet, distributed networks, satellitenumber of tags in the area is too high by dividing the tags networks, and cyber warfare. He is a senior member of the IEEE.into groups and dealing with each group individually. The Nikolaos V. Alchazidis graduated from theparameters that control the performance of the PS algorithm Naval Postgraduate School in 2006 with theare the minimum transmitted power level from the reader, master of science degree in electrical engineer-the frame size, and finally the step size of increasing the ing and the master of science degree in systems engineering. His research interest ispower in each cycle. Different values cause the PS algorithm in wireless networks.to perform differently. Generally, the PS algorithm is betterthan FSA when the number of tags in the area is over 1,000.Also, it is better than DFSA if the number of tags is over 600. Furthermore, higher initial frame sizes correspond to ahigher initial average delay, but can also handle more tagsdue to collisions. The most important conclusion for the Tri T. Ha received the BSEE and MSEEperformance of the proposed algorithm is that it can provide degrees from Ohio University and the PhD degree from the University of Maryland. Hea high degree of data integrity in the RFID system, even with joined the Electrical and Computer Engineeringthe use of small frame sizes, while FSA and DFSA cannot. Department at the US Naval Postgraduate School in 1987. He is a fellow of the IEEE and currently holds a joint appointment in theREFERENCES Department of Electrical and Computer Engi- neering and the Department of Systems En-[1] K. Finkenzeller, RFID Handbook, Fundamentals and Applications in gineering. His current research interests include Contactless Smart Cards and Identification, second ed. Wiley, 2004. spatial signal processing, interference cancellation in 2G, 3G, and 4G,[2] S. Lahiri, RFID Sourcebook. IBM Press, 2005. robust detection algorithms for CDMA signals, Doppler correction[3] J. Cha and J. Kim, “Novel Anti-Collision Algorithms for Fast algorithms, equalization for OFDM signals, and signal analysis in the Object Identification in RFID System,” IEEE Proc. 2005 11th Int’l presence of interference. Conf. Parallel and Distributed Systems (ICPADS), vol. 2, pp. 63-67, July 2005.[4] http://www.epcglobalinc.org, 2009.[5] F.C. Schoute, “Dynamic Frame Length ALOHA,” IEEE Trans. . For more information on this or any other computing topic, Comm., vol. 31, no. 4, pp. 565-568, Apr. 1983. please visit our Digital Library at www.computer.org/publications/dlib.[6] H. Vogt, “Efficient Object Identification with Passive RFID Tags,” Proc. Int’l Conf. Pervasive Computing, pp. 98-113, Apr. 2002.[7] H. Vogt, “Multiple Object Identification with Passive RFID Tags,” Proc. IEEE Int’l Conf. Systems, Man and Cybernetics (SMC ’02), vol. 3, pp. 4-9, Oct. 2002.