Efficient Monitoring of Dynamic Tag Populations in RFID Systems

Transcription

1 2 2 Ninth IFIP IEEE/IFIP Ninth International Conerence on on Embedded and and Ubiquitous Computing Eicient Monitoring o Dynamic Tag Populations in RFID Systems Qingjun Xiao, Kai Bu, Bin Xiao Department o Computing The Hong Kong Polytechnic University {csqjxiao, cskbu, csbxiao}@comp.polyu.edu.hk Abstract As RFID tags become more ubiquitously available, e.g., in a supermarket, it is necessary to monitor larger-scale tag populations in a dynamic environment to get updated tag inormation. This paper considers the problem o monitoring a dynamic tag population, to identiy both the missing tags and new tags. Traditional approach can solve the problem by collecting all tag IDs in the current population, which could be slow because it ignores the knowledge o the tag population in a previous scan. To be more eicient, this paper presents two protocols: () a baseline protocol with optimized length o random number bits, (2) an improved one-phase protocol with easy labor to identiy only the new and missing tags in ALOHA rames by ully utilizing previous tag population knowledge. Our analysis shows that the one-phase protocol can improve the monitoring accuracy by 25% and improve the time eiciency by 55%, as compared with the two-phase protocol proposed in a recent paper which also identiies population changes. Keywords-Ubiquitous Computing, RFID, Dynamic Tag Population, Tag Population Monitoring I. INTRODUCTION About a decade ago, RFID (radio-requency identiication) was envisioned as one o the enabling techniques towards the uture o ubiquitous computing []. Nowadays, RFID has arguably become one o the most successul technologies in the computing history, which has been adopted by a wide range o applications, e.g. warehouse management, object tracking and inventory control. RFID s success is largely due to its clear advantages over the classical barcode system. RFID extends the operation distance rom inches to tens o eet (or passive tags) or even hundreds o eet (or active tags). RFID also dramatically improves the tag reading speed: barcode attached products can only be checked manually, one at a time, while all RFID tags in the range o a reader can be inventoried as a population, at the speed o several milliseconds per tag. This paper studies an important RFID problem o monitoring a dynamic tag population, which identiies the population changes, including both the missing tags that disappear within the readers range and the new tags that are previously unknown and newly appear. Such a population change identiication ability can bring solid beneits to many industries. For example, imagine a large warehouse with tens o thousands o stocks. Every night, the warehouse manager needs to check the inventories and ind the missing items and the new items, because the existence o missing items probably indicates inventory thie or vendor raud, and the presence o new items may expose management aults, e.g. unregistered stocks and misplaced stocks. However, manual counting is laborious. I each stock item is attached with a tag, then the RFID readers can know the stocks changes automatically. In recent years, researchers also investigated various other RFID problems. Much prior work concentrates on tag identiication problem which is to collect all the tag IDs in a tag population as quickly as possible [2], [3], [4], [5], [6], [7], [8], [9]. Another extensively studied topic is the tag estimation problem which is to give a rough estimate o the tag population size in a time-eicient manner [], []. Missing tag identiication problem also attracted much attention recently, which is to identiy the IDs o missing tags in a tag population [2], [3]. Our tag population monitoring problem is dierent rom the previous research topics. It may appear that, i we can collect all the tag IDs in the current population (i.e. the tag identiication problem), then we will learn the population changes by comparing current population with the population in the previous scan. However, there can be a large number o tags that exist in both the previous scan and the current scan, called remaining tags. Recollecting the IDs o these remaining tags in the current scan is a waste o time, especially when each tag ID can be as long as 96 bits, according to EPC UHF RFID speciication [3]. Can we detect the tag population change by the tag estimation problem? I we cannot detect any change in the number o tags, then we have no need to urther identiy those newlyarrived tags and missing tags. However, the tag number estimate is just a rough estimate with at least % error [], []. I the population change only involves a ew tags, we can hardly assert the change based on the estimated population size that is statistically variant. Finally, our tag population monitoring problem that identiies both missing tags and new tags is also dierent rom the missing tag identiication problem which determines only the IDs o missing tags. The most related is a recent paper [4] which designed a protocol to detect both missing tags and new tags. However, the time eiciency o this protocol can be improved, because it uses only empty slots in the previous scan (or in the current scan) or the detection. To guarantee the inal detec / $26. 2 IEEE DOI.9/EUC.2.3 6

2 tion accuracy, it must rely on the multiple-round execution which is time consuming. Moreover, this protocol uses two separated phases to detect missing tags and new tags. This is time-wasting since both phases need the remaining tags to respond, which is unnecessary and can be avoided. We propose protocols that not only detect the tag population change events but also identiy which tags are new tags or missing tags. The most important perormance criterion is to minimize the identiication time while meeting the accuracy lower bound. Otherwise, i the protocol execution takes too long, the normal operation (e.g. relocating goods in a warehouse) may alter the current population during the scan, and misleads the protocol to report wrong sets o missing tags and new tags, which would cause conusion to warehouse management. We highlight the contributions o this paper to identiying new tags and missing tags. () We describe the protocol that identiies the current tag population as the baseline protocol, and we improve its time eiciency by %, by optimizing the length o the random number bits used or collision detection. (2) We propose a one-phase protocol that identiies only the population changes. This protocol is 7% more eicient than the baseline, when the remaining tag ratio is.5 or above. This protocol is 25% more accurate than state-o-the-art two-phase protocol in [4], by utilizing both empty slots and singleton slots in the previous scan (or in the current scan) to detect population changes. (3) We analyze the tradeo between accuracy and eiciency. We show that, or our one-phase protocol, the increase o accuracy requirement will cause time eiciency decrease, i.e. higher time cost per identiied new tag or missing tag. We also show that our one-phase protocol is 55% more eicient in execution time than the two-phase protocol in [4] at the same accuracy level. The rest o this paper is organized as ollows. We introduce some preliminary knowledge about RFID systems in Section II. We ormulate our tag monitoring problem in Section III. For this problem, we present two monitoring protocols: the baseline protocol in Section IV and the onephase protocol in Section V. We analyze the accuracyeiciency tradeo o the one-phase protocol in Section VI. We review the related work in Section VII and conclude our paper in Section VIII. II. RFID BACKGROUND AND SYSTEM MODEL In this section, we introduce the technical background in RFID systems. An RFID system consists o RFID tags and RFID readers. Each tag stores a unique ID in its memory, and each reader can read the IDs o its surrounding tags by wireless connections. We consider the scenario where there is only one reader covering all the tags. This is the most common scenario adopted by most existing RFID research [5], [6], [8], [9]. I multiple readers are used, we believe that it is easy to extend our protocols using existing reader scheduling protocols [5]. A. Framed Slotted ALOHA Protocol One major challenge in the RFID research is the collision problem, i.e. when multiple tags hear the reader s query and respond, their responses will overlap and the reader may ail to decode the overlapped waveorm. There exist a plethora o RFID anti-collision protocols, which can be classiied into two major categories: tree-traversal algorithms [2], [7] and ramed slotted ALOHA [3], [4], [5], [6], [8], [9], [6]. We adopt the latter because it has higher time-eiciency than the ormer in large RFID systems [3], [9]. The basic idea o the latter is to start an ALOHA rame with many time slots, and distribute the many tags uniormly to these slots to reduce the chance that multiple tags respond in a same slot. This uniorm tag distribution is implemented by each tag selecting its own slot autonomously and pseudo-randomly, using a hash unction h (id, r), where id is the tag ID, is the number o slots in the ALOHA rame, and r is a random seed. The parameters and r are broadcast by the reader when starting the rame. B. Slot States in Framed Slotted ALOHA A slot in a rame has three possible states, according to the number o replying tags in this slot. I there are no tags replies, the slot is empty (noted as number ). I there is one and only one tag reply, it is called a singleton slot (noted as ). I there are at least two tag responses, it is a collision slot (noted as 2). Singleton slots can be urther classiied as singleton-with-id slots and singleton-without- ID slots, according to whether tag ID transmission occurs. The singleton-without-id slots is or the reader to know the presence o tags without ID collection. We denote the probability or a slot to be empty as P, the probability o being singleton as P, and the probability o being collision as P 2. Then we have P = e ρ, P = ρ e ρ, and P 2 = P P, because the number o replying tag in a slot ollows Poisson distribution whose expected number o occurrences equals the rame s load actor ρ, i.e. the number o tags divided by the number o slots. III. TAG POPULATION MONITORING PROBLEM In this section, we ormulate our problem precisely. A tag population is inevitably dynamic because tags can move in or out o the reader s range or various reasons. Such a dynamic tag population can be modelled as a dynamic process [T,..., T t, T t +,... ], where T t is the tag population at discrete time t. A tag population monitoring protocol is, with the knowledge o previous tag population T t, to derive an estimate o the current tag population T t +. Each tag population is represented by a set o tag IDs. I comparing the previous population and the current population, we have three kinds o tags, as depicted in Fig.. 7

3 Missing Tags Fig.. T t T t + T t + T T + t T t + t + New Tags Remaining Tags Missing Tags, Remaining Tags, and New Tags. Missing Tags: the tags that were ound in previous population but no longer exist in current population, which are noted as Tt + = T t T t +. Remaining Tags: the tags that exists in both the previous population and the current population, i.e. T t T t +. New Tags: the tags that are unknown in previous population but appear in current population, which are noted as T t + + = T t + T t. We introduce a vector [β,β,β + ] to ormalize the relation between the previous population T t and the current population T t +. β is the missing tag ratio that equals β is the remaining tag ratiothat equals T t + T t T, t + Tt Tt + T t T, t + T + t + T t T. t + β + is the new tag ratio that equals O course, the sum o the three ratios is equal to one. For example, vector [.,.5,.5] means no missing tags, 5% remaining tags, 5% new tags. Vector [.5,.5,.] means 5% missing tags, 5% remaining tags, no new tags. We consider two perormance metrics or evaluating tag monitoring protocols: accuracy and time eiciency. ) Tag monitoring accuracy α is the degree o similarity between the current population T t + and the generated estimate ˆT t +. We deine α = Tt + ˆT t + T t + ˆT. The maximum t + value o the accuracy α is equal to one, when the estimate ˆT t + is absolutely accurate and identical to T t +.Akey concern o this paper is to guarantee the estimation accuracy α to be above a threshold α. 2) Time Eiciency γ is deined as the total execution time t total divided by the size o identiied population changes, i.e. γ = t total T t ˆT t + + ˆT t + T t, where T t ˆT t+ is the number o identiied missing tags and ˆT t+ T t is the number o identiied new tags. We do not deine γ as the total time cost divided by the size o current population, because we believe that the users o our RFID systems are mainly interested in the changes rather than the current population. IV. BASELINE PROTOCOL The most straightorward protocol to solve this tag monitoring problem is to collect all the IDs in the current population, and then compare the collected IDs with the memory o previous population to identiy the changes. The accuracy o this baseline protocol can be close to one, since the traditional tag identiication methods usually support the multiple-round execution, in which the next round can collect the tag IDs that ails to collect in the previous round. The time eiciency o this baseline protocol is analyzed by the ollowing subsections. A. Time Cost or Dierent Slot States Much previous analysis o protocol time eiciency assumes the time cost o all slots is the same. But in practice the time cost o the our slot states (i.e. empty, collision, singleton-without-id and singleton-with-id) is dierent. We list the our time cost in the ollowing table, where υ is the length o the random number (RAND or short) sent by tags to acilitate the detection o collisions at the reader side. When two tags send dierent RANDs (whose possibility is 2 υ ), the reader can detect the collision at waveorm level. Note that the listed time cost assumes EPC RFID protocol [3], and include both data transmission time and waiting time between transmissions 2. Deinition Value t e time cost o an empty slot 84 μs t s time cost o a singleton-without-id slot υ μs t ID time cost o a singleton-with-id slot υ μs t c time cost o a collision slot, i.e. ( 2 υ ) t s +2 υ t ID Time cost t e o an empty slot is the smallest. Time cost t s o a singleton-without-id slot almost doubles t e. Time cost o a singleton-with-id slot t ID is at least ten times larger than t e, because a tag ID is as long as 96-bit EPC plus 6- bit CRC and each bit requires 6μs to transmit. Thereore, reducing the number o ID transmissions is a key concern or many RFID protocols. The calculation o time cost o a collision slot is complicated, i.e. ( 2 υ ) t s +2 υ t ID. Here, 2 υ is the probability or the reader to detect the collision when receiving RANDs. I two tags send identical RANDs with 2 υ probability, the collision can be detected by the reader only when the two tags send out their IDs, whose time cost is t ID. B. Protocol Time Eiciency ) Tag Identiication Eiciency: For a tag identiication protocol, its time eiciency is usually deined as the time t cost per collected tag ID, i.e. e P + t ID P + t c P 2 P. This is because or any slot i, ithasp chance to be empty whose time cost is t e, P chance to be singleton-with-id whose time cost is t ID, and P 2 chance to be collision whose time cost is t c. This time eiciency can be rewritten as t ID + ρ t e + ( e ρ ( + ρ) ) t c. We plot in Fig. 2 the tag identiication eiciency against load actor ρ. Fig. 2 shows that when the load actor o the rame is between.6 and.8, the time eiciency reaches its peak about 3μs i the RAND length υ is 6. We argue that, although Fig. 2 uses the slot time cost parameters t e, t ID, t c o EPC RFID system in Section IV-A, Note: RT rate=64kbps, TR rate=62.5kbps, QueryRep=4bits, Ack=2+υbits. 2 Assume RT cal =3.25μs, T pri =8μs, T =8μs, T 2 = T 3 =4μs. 8

4 our conclusions can be easily adapted to other RFID systems by adjusting these parameters accordingly. Fig. 2. Time Eiciency o Tag Identiication Protocol against Load Factor. Beside the appropriate choice o load actor, another interesting issue is the optimal coniguration o RAND length υ. It is true that it is not a speciication compliant eature that the RAND length is adjustable. For example, EPC UHF RFID protocol deines RAND to have 6 bits [3], and Philips I-Code system deines RAND to be bits [4]. We investigate what is the optimal RAND length or tag identiication eiciency. Fig. 2 shows that the optimal RAND length is 6. I reducing the RAND length rom 6 to 6, the time eiciency can be improved rom 3μs to 27μs (i.e. % improvement). We regard 6 as the optimal length because a value smaller than 6 will lead to worse perormance in heavy load region where the load actor is larger than.4 (i.e. the υ =4 curve is higher than the υ =6curve in the heavy load region). The explanation is that, when RAND length is 6, collisions can be detected with 2 6 =98.44% probability. I RAND length is reduced to 4, the detection probability will drop to 2 4 =93.75%. Note that an undetected collision will be ollowed by a time-consuming tag ID transmission. Such a trend o perormance degradation in heavy load region is even more prominent when RAND length is reduced to or 2. Thereore, we conigure RAND length to 6 to optimize the perormance in medium load regions and heavy load regions. 2) Tag Monitoring Eiciency: Dierent rom the tag identiication eiciency, tag monitoring eiciency is the total time cost 27μs T t + divided by the size o population changes, i.e. T t T t + + T t + T t. This tag monitoring eiciency can be rewritten as Tt Tt + + Tt + Tt β + β+ γ baseline = T t T t + + T t + T t 27μs = β 27μs. As a summary, the monitoring eiciency o the baseline protocol degrades with the increase o remaining tag ratio β which means the baseline protocol will waste more time in collecting the already-known remaining tag IDs. V. ONE-PHASE PROTOCOL We observe that the major drawback o baseline protocol is the re-collection o remaining tag IDs which are known in the previous population. To avoid such redundant collection and improve time eiciency, we adopt an incremental update strategy that identiies only the IDs o new tags and missing tags. We assume the set o identiied new tag IDs is T t + + and the set o identiied missing tag IDs is Tt +. Then we can generate an estimate o the current population by the equation: ˆT t + =(T t Tt + ) T t + +. Thereore, the pivot is shited to the eicient collection o new tag IDs and the eicient identiication o missing tags whose IDs are already known. We present our solution in this section. A. Detection o New Tags and Missing Tags We propose a one-phase protocol that detects the presence o new tags or missing tags in one rame (note: the new tag ID collection will be described in Section V-C). This one rame is depicted in Fig. 3 as the true rame which invites the participation o all tags in the current population T t +. Besides this true rame, we construct a so-called expected rame that involves the tags in previous population T t. This expected rame is generated by pure computation without any wireless transmission involved, and it uses the same hash unction as the true rame. A remaining tag responds in both the true rame and the expected rame and at the same slot, e.g. tags 3-7. A missing tag replies only in expected rame, e.g. tags -2. A new tag responds only in the true rame, e.g. tags 8-. True Frame s i : Expected Frame ŝ i : Fig. 3. T t T t + One-Phase Protocol to Detect Population Changes by One Frame. We detect the presence o new tags and missing tags by scanning the two rames and comparing the slot states. For the slot i, we denote its state in the true rame by s i and denote its state in the expected rame by ŝ i. I the slot state changes with s i ŝ i, it indicates the presence o either new tags or missing tags, or even both. This is because when a slot contains only remaining tags which are mapped to both rames, its state will have no change. I there is no state change with s i =ŝ i, most probably this slot contains only remaining tags, in which we have no interests. But it is also possible that () this slot contains equal number o missing tags and new tags, which will escape rom our detection, or (2) this slot contains at least two remaining tags, which will shield our detection. We will analyze the impacts o these two abnormal cases on detection accuracy later. We list in the ollowing table all the possibilities or a slot whose state changes with s i ŝ i. For the irst two 9

5 cases, the identiication o new tags and missing tags is easy, because the tags in such slots are either all new tags or all missing tags. I the true state s i is nonempty and the expected state ŝ i is empty, all the tags in this slot are new tags, since there are no remaining tags as indicated by ŝ i = (e.g. slot 8). I the true state s i is empty and the expected state ŝ i is nonempty, all the tags in this slot are missing tags, since there are no remaining tags indicated by s i =(e.g. slot). s i ŝ i Detection Notes + all the tags that respond in slot i are new tags + all the tags that should respond in slot i are missing tags 2 some tags that respond in slot i are new tags; also possibly the old tag is missing and all tags that respond are new tags 2 some tags that should respond disappear; also possibly all tag that should respond disappear and one new tag comes NOTE: is empty, is singleton, 2 is collision, and + is non-empty. For the other two cases, the identiication is diicult since the tags in such slots are a mixture o missing tags, remaining tags and new tags, which needs to dierentiate. We solve this dierentiation problem by a technique called population change recalculation which is detailed in Section V-C. I the true state s i is collision and the expected state ŝ i is singleton, the slot contains a remaining tag and a new tag (e.g. slot 6 which has remaining tag 5 and new tag 9). I the true state s i is singleton and the expected state ŝ i is collision, the slot may contain a remaining tag and a missing tag, or even contain a new tag and two missing tags (e.g. slot 4 with new tag 8 and missing tags,2). B. Accuracy Analysis We analyze the accuracy that can be achieved by utilizing all the our kinds o changed slots. A recent paper that also studies new tag and missing tag detection uses only the irst two cases whose tag identiication is easy [4]. We will highlight the improvement we made in accuracy (i.e. about 25%) by also using the other two cases. We assume that ˆρ is load actor o the expected rame which equals T t /. Thus the probability o empty slots (or singleton slots) in the expected rame is ˆP = e ˆρ (or ˆP =ˆρe ˆρ ). Similar parameters can be assumed or the true rame: load actor ρ = T t + /, empty slot probability P = e ρ, and singleton slot probability P = ρe ρ. Theorem. The accuracy o our one-phase protocol is E ( α ) β +(ˆP + ˆP ) β + (P + P ) β = β +(+ˆρ) e ˆρ β + ( + ρ) e ρ β, () where β is missing tag ratio, β is remaining tag ratio and β + is new tag ratio. In contrast, the accuracy o the twophase protocol in [4] is only E ( α ) = β + e ˆρ β + e ρ β. Proo: Consider an arbitrary new tag in the set T t + T t,ithas ˆP + ˆP probability to be mapped to a noncollision slot in the expected rame (i.e. ŝ i =or ), where it can change the slot state and get identiied. An arbitrary missing tags in the set T t T t + has P + P probability to be mapped to a non-collision slots in the true rame (i.e. s i =or ) and get identiied. Thus, the expected number o identiied new tags is ( ˆP + ˆP ) T t + T t, and the expected number o identiied missing tag is (P + P ) T t T t +. The accuracy o one-phase protocol can be estimated as E ( α ) Tt Tt + +(ˆP + ˆP ) T t + T t T t T t + (P + P ) T t T t +, where the nominator is the number o remaining tags T t T t + plus the number o identiied new tags, and the denominator is the union population size T t T t + with the identiied missing tag removed. This accuracy can be urther rewritten to the orm in Theorem. The accuracy equation in Theorem appears to be a unction with two variables: load actor ρ o the true rame, and load actor ˆρ o the expected rame. In act, it is a unction with only one variable, i.e. the union load actor ρ Tt Tt + =. This is because or the true rame load actor ρ, wehave Tt + Tt Tt + ρ = = (β + β + )=ρ (β + β + ), and or the expected rame load actor ˆρ, wehave ˆρ = Tt = Tt Tt + (β + β) =ρ (β + β), where missing tag ratio β, remaining tag ratio β and new tag ratio β + are all constants. The physical meaning o the union load actor ρ is that e ρ is the expected ratio o slots that are empty both in true rame and in expected rame. We plot in Fig. 4 both the accuracy o our one-phase protocol and the accuracy o the two-phase protocol in [4]. Fig. 4 adopts three typical scenarios with dierent combinations o [β,β,β + ]. The remaining tags ratio β is ixed to 5%. Vector [.25,.5,.25] means 25% missing tags and 25% new tags. Vector [.,.5,.5] means no missing tags and 5% new tags. Vector [.5,.5,.] means 5% missing tags and no new tags. Fig. 4., / Our One-Phase Protocol vs. Two-Phase Protocol [4] in Accuracy. Figure 4 shows that the accuracy o our one-phase protocol is roughly 25% better than the accuracy o twophase protocol in [4]. For example, in the [.25,.5,.25] scenario, when the union load actor ρ equals, the accuracy o the two-phase protocol is 7%, while the accuracy o our one-phase protocol is 89%, which means 27.4%

6 improvement. Fig. 4 also shows that the accuracy o both protocols is poor and lower than.7 in the heavy load region with union load actor ρ above 4. This is because when ρ is above 4 and remaining tag ratio β is.5, the density o remaining tags is larger than 2 remaining tags per slot. In slots with 2 or more remaining tags, we can not detect the presence o new tags and missing tags. However, when the union load actor is lower than.3, the accuracy o our onephase protocol can be higher than 95%, which can satisy the needs o many RFID applications. C. Identiication o New Tags and Missing Tags We present our one-phase protocol in Protocol. This protocol provides high accuracy by utilizing all the our kinds o slot state changes, and it addresses the problem o dierentiating missing tags and new tags when all kinds tags mixed in one slot. The input o the protocol is the prior knowledge o previous population T t, and the output is an estimate o the current population ˆT t+. Note that when the prior knowledge T t is an empty set, our protocol will degrade to the baseline protocol that neglects the prior knowledge. Protocol : One-Phase Population Monitoring Protocol input : the prior knowledge o tag population T t at time t output : estimate ˆT t+ o current tag population at time t + Reader generates rame size and random seed r 2 Reader obtains state belie ŝ i o each slot, by T t and h (id, r) 3 Reader resets the session lags o all tags in T t + to 4 Reader starts a rame by broadcasting and r to all tags 5 or slot i to do 6 i a tag selects slot i by h (id, r) then it replies RAND 7 Reader obtains the state s i o slot i when receiving RAND 8 i s i =ŝ i then Reader closes slot i by a special QueryRep that orces tags in slot i to invert session lags 9 else Reader updates the missing tag set by T t + := T t + { id T t tag id should respond in slot i } i slot i is singleton with s i =then 2 Reader sends ACK, and the tag replies its id 3 Reader updates new tag set by adding id to T + t+ 4 Reader closes slot i by QueryRep (note: the single tag in slot i automatically inverts its session lag [3]) 5 Reader uses a new rame or new rames to collect IDs o tags whose session lags are, and adds the collected IDs to T + t+ 6 Reader re-identiies the missing tags by T t + := T t + T + t + 7 Reader re-identiies the new tags by T + t + := T + t + Tt 8 return tag population estimate ˆT t + := (T t T t + ) T + t + Firstly, the reader detects population changes. The reader starts a rame by broadcasting rame size and random seed r to the current population T t + (see Ln. 4). The tags use hash unction h (id, r) to choose their slots in which they respond with RAND to show their presence (see Ln. 6). The reader, ater receiving the RANDs, can obtain the state s i o each slot i (see Ln. 7). The reader can also establish a prior belie ŝ i about slot i s state (see Ln. 2), using the knowledge o the previous tag population T t and the hash unction h (id, r). Then the reader compares the true state s i with the state belie ŝ i (see Ln. 8). I no change can be ound, the reader closes the current slot i instantly by the QueryRep command which is deined in [3]; otherwise, this slot i is detected as a changed slot which contains new tags or missing tags. Secondly, the reader urther identiies new tags and missing tags in the changed slot i by the ollowing three steps. Step (Missing Tag Identiication). The reader adds all the tags in population T t that should respond in slot i to the missing tag set Tt + (see Ln. ). For example, in Fig. 3, the slots {, 4, 6, 8} are changed slots, and the old tags {,, 2, 5} that should respond in these changed slots are marked as missing tags. It is possible that this set Tt + may contain remaining tags, e.g. tags 5 that should respond in changed slot 6 in Fig. 3. Such remaining tags that are wrongly marked as missing tags won t ruin our inal tag population estimate ˆT t +, because their responses will be heard by the reader in step 2 and be re-identiied as new tags. Step 2 (New Tag Identiication). The reader will add the IDs o all the tags that responded by RAND and showed their presence in slot i to the new tag set T t + +. For example, in Fig. 3, tags {5, 8, 9, } that responded in the changed slots {, 4, 6, 8} are regarded as new tags. It is possible that a tag in the set T t + + is in act a remaining tag, e.g. tags 5. But this won t ruin our inal population estimate ˆT t + at step 3. Dierent rom the missing tags whose IDs are contained in T t, the IDs o new tags are unknown and need to be collected. The reader will use two methods to collect new tag IDs. Firstly, i a changed slot i is singleton in the true rame (e.g. slots 4, 8), then the reader collects the single tag ID in slot i directly (see Ln. -3). This because as deined in [3], the reader can send an ACK command to notiy the tag to propagate back its ID. Secondly, i a changed slot i is collision in the true rame, then the multiple tag IDs cannot be collected in the current slot and should be delayed to a new rame (see Ln. 5). The sole purpose o this new rame is to collect IDs o the new tags mapped to the changed slots that are collision in true rame. But the question is how can we let these tags know they should participate in this new rame and let other tags know they shouldn t. The answer is to use the session lag eature 3 deined in [3]. Initially, the session lags o all tags are zero (see Ln. 3). Then, the lags o the tags in unchanged slots will be be orced to invert (see Ln. 8). The lags o the tags in singleton changed slots will invert automatically (see Ln. 4) according to [3]. Thus only the tags in collision changed slots do not invert their lags and will participate the ID collection rame(s) at Ln By section o [3], each tag has our session lags S-S3. The rame start command Query contains two parameters: a session lag id and a desired value (e.g. session lag S2 and value ). A tag will participate in this rame only i the tag s corresponding session lag matches the desired value.

7 Step 3 (Population Change Recalculation). To remove the remaining tags wrongly contained in the missing tag set Tt +, the reader recalculates the set o missing tags by Tt + = T t + T t + + (see Ln. 6). To remove the old tags wrongly contained in the new tag set T t + +, the reader recalculates the set o new tags by T t + + = T t + + T t (see Ln. 7). Finally, with the recalculated missing tag set and new tag set, we give the population estimate ˆT t + at Ln. 8. VI. TIME-EFFICIENCY TRADEOFF This section analyzes the tradeo between monitoring accuracy and eiciency o the proposed one-phase protocol. A. Time Eiciency Analysis We analyze the time eiciency o our one-phase protocol, which is another perormance metric besides accuracy. We calculate the time eiciency as γ one-phase = tslot, where n slot +n+ slot t slot is the expected time cost o a slot, n slot is the expected number o missing tags that can be identiied in a slot, and n + slot is the expected number o new tags that can be identiied in a slot. This eiciency γ one-phase is a unction o union load actor ρ, which is plotted in Fig. 5. This igure adopts three scenarios: [.,.5,.5], [.25,.5,.25], [.5,.5,.]. We omit the expression to calculate t slot which is complicated. This is because the true rame has ive kinds slots with dierent time cost: empty slots whose cost is t e, singleton unchanged slots whose cost is t s, singleton changed slots whose cost is t ID, collision unchanged slots whose cost is t c, and collision changed slots whose cost is about 27μs per tag (see Section IV-A or time cost deinitions). We need to calculate their corresponding probabilities, and combine them linearly to obtain t slot. The equations or n slot and n+ slot are as ollows: n slot = ρ (P +P )P + + ρ ( e ρ )P P + + P P P 2 +, n + slot = P (P +P )ρ + + P2 P P + + P P ρ + ( e ρ+ ), where P, P, P 2 are the probabilities o,, 2 missing tags in a slot respectively; P, P, P2 are the probabilities o,, 2 remaining tags in a slot respectively; P +,P +,P 2 + are the probabilities o,, 2 new tags in a slot. Figure 5 shows that the time eiciency o our protocol is prominently higher than that o the baseline protocol in the medium load region. For example, in the [.5,.5,.] scenario, the eiciency o baseline protocol is 27μs per changed tag (i.e. γ baseline = 27μs β+β+ β ), while the best eiciency o our one-phase protocol is roughly μs, i.e. 7% reduction in time cost. In the [.,.5,.5] scenario, the eiciency o baseline protocol is 54μs, while the best time eiciency o our protocol is about 4μs, i.e. 26% improvement. Our one-phase protocol has much better perormance in [.5,.5,.] scenario than in [.,.5,.5] scenario, because [.5,.5,.] scenario has only missing tags whose identiication is does not need to transmit tag IDs. Fig. 5. One-Phase Protocol vs. Baseline Protocol in Monitoring Eiciency. Figure 5 also shows that the time eiciency o onephase protocol degrades rapidly in the high-load region with ρ > 4. This is because in the high-load region, the expected number o remaining tags in a slot ρ is larger than 2 (note: ρ =β ρ ). A slot with at least two remaining tags will have its states in the true rame and in the expected rame to be both collision, which can hide the presence o new tags and missing tags. In contrast, in light-load region with ew such slots, our protocol only degrades mildly in time eiciency. B. Accuracy-Eiciency Tradeo The one-phase protocol needs to keep a balance between accuracy and time eiciency. Although the accuracy can be improved by reducing union load actor (see Fig. 4), this meanwhile will degrade the time eiciency when entering the light-load region (see Fig. 5). Thereore, we analyze the unctional relation between accuracy and eiciency, and plot our analysis results in Fig. 6. It adopts three scenarios with dierent remaining tag ratio β: the scenario [.,.8,.] with β =.8, the scenario [.25,.5,.25] with β =.5, and the scenario [.4,.2,.4] with β =.2. Fig. 6 does not show the portion with accuracy lower than 9% since we believe that RFID users have no interests in such poor accuracy. Fig. 6. Tag Monitoring Eiciency γ vs. Accuracy Requirement α. As we have expected, Fig. 6 shows that the time cost o our one-phase protocol increases as the required accuracy goes higher. I the required accuracy exceeds 98%, the time cost inlates nearly straight up with the increase o accuracy. The explanation is that our one-phase protocol is based on a randomized algorithm that detects missing tags and new tags by randomly distributing them to slots where they can be detected. I ultra-high accuracy is needed, we may have to use extra-large rame size to increase the chance o detection. 2

8 However, we believe that the accuracy between 9% and 99% can already satisy the need o most RFID applications. Finally, we show the advantage o our one-phase protocol over the two-phase protocol [4] in Fig. 6. It shows that at the same accuracy level (e.g. 96%), our one-phase protocol can be 55.43% more eicient than the two-phase protocol in [.25,.5,.25] scenario. This is because the two-phase protocol has lower accuracy than our one-phase protocol. I the two-phase protocol needs to achieve the same accuracy level as ours, it must use much larger rame size (or even multiple round execution) to create more empty slots or eective detection o new tags and missing tags, which however reduces protocol eiciency. VII. RELATED WORK RFID technology has been considered in many applications. The most traditional application is the tag identiication problem which collects all tag IDs in a population. The proposed solutions can be classiied into two major categories: tree-based [2], [7] and ALOHA-based [3], [4], [5], [6], [8], [9]. The ormer organizes all tag IDs in a tree o ID preixes, while the latter distributes all tag IDs uniormly in an ALOHA rame. The major diiculty o ALOHA-based protocols is how to choose the optimal rame size which should be roughly equal to the number o tags. Thereore, tag population size estimation problem becomes another hot topic or RFID research [], []. Another important problem that attracts academic interests is, given the prior knowledge o the tag population, to identiy the missing tags [2], [3]. However, in practice, besides missing tags, there may also exist new tags whose IDs are unknown. The problem o identiying both o these tags is called tag population monitoring problem, since we can establish an estimate o the current tag population, with the knowledge o previous tag population, new tags and missing tags. A relevant study on this problem is a recent paper [4] which can detect a new tag when it is mapped to an empty slot in the expected rame, and detect a missing tag when it is mapped to an empty slot in the true rame. However, the accuracy o this protocol can be improved by at least 25%, i we can also make use o the massive singleton slots in the expected rame or true rame to detect o missing tags and new tags. Moreover, the twophase protocol is ineicient, because it uses two separated phases to detect missing tags and new tags. The remaining tags thus need to respond in both two phases, which waste precious execution time. VIII. CONCLUSION This paper ocused on the problem o monitoring a dynamic tag population. We proposed a one-phase solution which is eicient by using only one rame to detect both missing tags and new tags. This solution is also accurate because it uses both empty slots and singleton slots in the true rames (or in the expected rames) to detect the population changes. Another contribution we made is that we derived an optimal coniguration o RAND length or the traditional tag identiication protocol, which is neglected beore. ACKNOWLEDGEMENT The authors would like to thank the reviewers or comments. This work is supported by HK RGC PolyU 534/E. REFERENCES [] M. Weiser, The computer or the 2st century, in Proc. o ACM SIGMOBILE, vol. 3, 999, pp. 3. [2] ISO/IEC 8 inormation technology - RFID or item management - Part 6: Parameters or air interace communications at 86MHz-96MHz, 24. [3] EPC TM radio-requency identity protocols class- generation-2 UHF RFID protocol or communications at 86MHz-96MHz v.2., 28. [4] P. Semiconductors, I-CODE smart label RFID tags, 24. [5] H. Vogt, Eicient object identiication with passive RFID tags, in Proc. o IEEE PERCOM, 22, pp [6] B. Zhen, M. Kobayashi, and M. Shimizu, Framed ALOHA or multiple RFID objects identiication, IEICE Trans. on Communications, 25. [7] J. Myung and W. Lee, Adaptive splitting protocols or RFID tag collision arbitration, in Proc. o ACM MOBIHOC, 26. [8] S.-R. Lee, S.-D. Joo, and C.-W. Lee, An enhanced dynamic ramed slotted ALOHA algorithm or RFID tag identiication, in Proc. o IEEE MOBIQUITOUS, 25, pp [9] C. Qian, Y. Liu, H. Ngan, and L. M. Ni, ASAP: Scalable identiication and counting or contactless RFID systems, in Proc. o IEEE ICDCS, 2, pp [] C. Qian, H. Ngan, and Y. Liu, Cardinality estimation or large-scale RFID systems, in Proc. o IEEE PerCom, 28. [] M. Kodialam and T. Nandagopal, Fast and reliable estimation schemes in RFID systems, in Proc. o ACM MobiCom, 26, pp [2] C. C. Tan, B. Sheng, and Q. Li, How to monitor or missing RFID tags, in Proc. o IEEE ICDCS, 28, pp [3] T. Li, S. Chen, and Y. Ling, Identiying the missing tags in a large RFID system, in Proc. o ACM MOBIHOC, 2. [4] B. Sheng, Q. Li, and W. Mao, Eicient continuous scanning in RFID systems, in Proc. o IEEE INFOCOM, 2. [5] S. Tang, X. Li, G. Chen, Y. Liu, and J. Zhao, Raspberry: A stable reader activation scheduling protocol in multi-reader RFID systems, in Proc. o IEEE ICNP, 29. [6] K. Bu, B. Xiao, Q. Xiao, and S. Chen, Eicient pinpointing o misplaced tags in large RFID systems, in Proc. o IEEE SECON, 2. 3