A Low Latency Scheme for Bulk RFID Tag Reading 1

Erik F. Golen1, Nirmala Shenoy1, and Xiaojun Cao2 Rochester Institute of Technology, 2 Georgia State University

Abstract—Passive RFID tags transmit their ID information to a tag reader upon energization by the reader. These transmitted signals, or IDs, may collide if multiple tags transmit their ID simultaneously, requiring repeated energizations by the reader to read all tag IDs successfully. The number of energizations increases with the number of ID bits in the tag and the number of tags to be read. An increase in energizations results in a greater incurred latency to read the tags, and additionally, more transmissions. This is undesirable in an industry environment especially, and with this comes higher expended energy by the reading system, leading to higher system costs. In this paper, a new proposed scheme is evaluated, the Shortcut Bisected Countdown Scheme (SBCS), in terms of the number of energizations by a reader. Three different techniques are progressively combined into three schemes in order to cut down on the number of energizations. The final scheme, SBCS, which is a combination of the three techniques, stabilizes on the number of energizations irrespective of the tag ID length and the number of tags to be read, resulting in a distinct advantage for bulk reading. The three schemes are memoryless meaning that the tags are unable to remember if they were read or not; hence the performance of each of the schemes is compared with the Query Tree protocol since it is also memoryless.

T

I. INTRODUCTION

HE utility of RFID tags has been established for years. Drivers on many major highways have relied upon them to go through toll booths quickly by using a transponder (i.e., RFID tag) mounted on the car’s windshield. When driving through a special lane, the transponder is energized by a reader and will send out information about the driver, including an account number to charge the toll. This same technology has been put to use by Wal-Mart to track inventory [1]. Analogous to the transponder on a car, passive tags are attached to palettes or individual boxes of merchandise and remain latent until a special RFID reader activates the tag, at which point, it will report its identification number associated with the product. Unlike active RFID tags, passive tags require much less circuitry and no built-in battery; thus they are more inexpensive and more preferred [1]. Typically, an RFID reader will be able to emit a signal with a range of tens of meters [1], and any passive tag within that region will wake up and send out its ID information to the reader. What must be taken into account with passive tags is that these tags have just enough complexity to send out their ID, but have

no mechanism for communicating with other tags, resulting in a medium access problem. In other words, the massive response from the RFID tags can cause a plethora of collisions if a proper bulk reading technique is not employed. Bulk tag reading is very important for inventory tracking and control as a company may like to read a truckload of merchandise or more at a time and the need to reduce collisions, with its accompanying reduction in latency, is pivotal. With more collisions, come more transmissions by tags, which are undesirable in an industry environment since more energizations by the reader, results in higher energy consumption by the system [2] as well as higher latency in the reading of the tags. A scheme for reading in multiple passive tags, known as Query Tree (QT), was first proposed in [3] and was later slightly improved in [2] to reduce power consumption. However, as to be described in Section II.A, QT is a means for reducing collisions, not preventing them. Other collision avoidance and reduction techniques have been attempted in [5], using an adaptive splitting protocol. While a hybrid randomized protocol is used in [6], the authors in [7] suggest a medium access scheme for using multiple readers within the vicinity of one another. A second reading approach is along the lines of the Aloha and slotted Aloha based schemes discussed in [4]. In these protocols, the tags are randomly assigned time slots and if a collision occurs during a specified time slot, the reader will at least know at which time slot the collision occurred in. One of the major drawbacks to this type of scheme is the high incidence of collisions during time slots when a large number of tags are being read in. Furthermore, once these collisions have occurred, the probability that these tags will ever be read decreases, causing reading starvation of these collided tags. In this paper, a set of three RFID tag reading schemes is presented with each one showing better performance than the previous one. The focus is to reduce the number of energizations by the reader in order to read in all tags within its vicinity, reduce latency, and the number of transmissions. The Countdown Scheme (CS) uses a countdown technique on the tag’s bits. The second scheme, Bisected Countdown Scheme (BCS), enhances the Countdown Scheme by bisecting the tag’s ID. Lastly, a shortcutting technique known as restricted transmission is combined with the previous two techniques, leading to the Shortcut Bisected Countdown Scheme (SBCS), which has the unique advantage of stabilizing on the number of energizations irrespective of the tag length and the number of tags to be read. Each technique has been investigated for various

applications, however, this is the first time that these schemes are being applied to RFID tag reading and the novelty lies in the effective combination of the three techniques to achieve the unique advantages previously explained. The simulation and analysis results show that the proposed SBCS protocol is more favorable than CS and BCS, and all three schemes significantly outperform the Query Tree protocol in [3]. There is no attempt to make a comparison with Aloha, slotted Aloha, or other variants, as these proposed schemes are memoryless, while Aloha assumes tags know they have been read. The rest of this paper is organized as follows. A discussion of QT, as well as the three proposed schemes will be presented in Section II. Section III shows a series of results, which are analyzed and discussed, while Section ! IV includes a further discussion of CSSBS. Finally, Section V states the conclusions along with a summary of the major contributions. II. RFID BULK READING PROTOCOLS This section begins with a review of the Query Tree protocol [3]. The three schemes, namely the Countdown scheme (CS), Bisected Countdown Scheme (BCS), and Shortcut Bisected Countdown Scheme (SBCS) are presented to demonstrate how each of the schemes performs in terms of the total number of energizations required by the RFID reader to read in a series of tags. A. Query Tree Protocol The Query Tree (QT) protocol is a memoryless protocol wherein tags have no notion as to whether or not they have been read in by a reader [3]. Tags only know that if their tag ID matches the prefix being queried by a reader, they should respond by transmitting their full tag ID in a bit by bit TABLE I PROCESS OF READING IN TAGS WITH ID {010, 011, 110, 111} Energization

Reader Prefix

Resulting Tag Response

1 2 3 4 5 6 7 8 9 10

0 1 00 01 10 11 010 011 110 111

010 and 011 collide 110 and 111 collide No tags respond 010 and 011 collide No tags respond 110 and 111 collide Read tag 010 Read tag 011 Read tag 110 Read tag 111

fashion. Take the example presented in Table I, where the RFID tag length is 3 bits and each tag has a unique ID. Notice that the tags compare the prefix bits sent by the reader from the most significant bit (MSB) to the least significant bit (LSB). Hence, as indicated in the first and second rows of Table I, tags 010 and 011 collide on a prefix of ‘0’ during Energization 1, while tags 110 and 111 collide on the ‘1’ prefix during Energization 2. On the other hand, if none of the tags match the prefix, the reader will receive no responses. None of the tags are successfully read until the prefix has been extended to the complete 3 bits, at which

point, each of the tags will match exactly one of the prefixes. It is important to note, however, that shortcuts are possible in QT. Since no tags were read using prefixes ‘00’ and ‘10’, all other possible prefixes subsequent to those may be skipped by the reader, resulting in a savings in energizations. With a tag length of n bits and a possible savings δ, the number of reader energizations ε required can be represented by Equation 1. n

" = # 2 i $ % = 2 n +1 $ 2 $ %

(1)

i=1

As the number of bits in the tag ID increases, the number of reader energizations required by QT can become unmanageable. For instance, to read in a set of tags with an ID 20 bits long, it may take more than 2.1 million energizations to read in every tag if no shortcuts are possible. This will be the case if the distribution of tag IDs is extremely unfavorable wherein nearly every branch in the query tree is followed to its leaf nodes. B. Countdown Scheme In the CS scheme, each tag does a countdown based on the numerical value of its tag ID. For example, a tag with an ID of ‘1110’ would do a countdown starting from 14 and once the count reaches zero, the tag would send a single ‘1’, indicating to the reader that within that time slot, a tag has that particular ID. Each tick in the countdown is equal to the time it takes to send a bit plus a tiny guard period to allow for any discrepancies in synchronization. Most convenient with this protocol is that the tag need only send a single bit to the reader since the reader will know the time slot number. Assuming that each tag in the vicinity of the reader has a unique ID, tag collisions are completely avoided if the tag is able to retain energy until its turn. This scheme is based on the well-known countdown schemes used by protocols such as Token Ring. The scheme, however, depends on the capacitor’s capability to retain charge. If the numerical value of a tag’s ID is large, the capacitor may lose charge before the tag has counted down to zero. CS begins with an initial energization by the reader, which sends a “start count” signal comprising of n 0’s. Upon this initial energization, every tag begins counting down and the reader begins its slot counter at 0. The reader waits for a tag to respond for the same tick time that the tags use to count down. When a tag responds, the reader records the ID as being read. If a tag does not respond during that tick time, the reader will increase its slot number and waits for the next tick time. The process continues for a predefined number of time ticks. If the tags IDs are long and exhaust its capacitor’s charge, the reader will have to reenergize. In this subsequent reenergization, the reader will send the value of the last slot that was read correctly or the last slot number that the reader understands to be the time for which the capacitor retains its energy. Only tags that have a higher slot number than the announced slot number will count down. Equation 2 shows that the number of reader energizations required by CSS

with tag length n, may be less than that of QT by a factor of nearly 2 in addition to the effect of the average capacitor life, CL, of the tags.

!=

2n CL

(2)

For example, with a tag length of 20 bits and a capacitor life of 5 readings, which is a highly conservative value, CSS requires around 210,000 energizations to read all possible tags, a 10-fold improvement over the 2.1 million energizations possibly required by QT for the same tag length. C. Bisected Countdown Scheme The second scheme, BCS, has two stages and combines the ideas of QT and CS in a novel way and was introduced to leverage the positive aspects of the two schemes. Rather than counting down on the entire length of the tag ID, BCS bisects the tag ID into an upper and lower half. In Stage 1, the tags will count down on the numerical value of the lower half of their respective IDs using the same method as the CS scheme. However, multiple tags can end up in the same slot and therefore, instead of transmitting a single ‘1’, as in the CSS, each tag will sequentially send the upper half bits of its ID starting from the MSB. If two or more tags respond, the reader will detect a collision and will record the value of the slot at which the collision occurred. Once the time slots have been cycled through, including the need for additional reader energizations due to capacitor life restrictions, the reader will go into Stage 2 where it will execute the QT protocol on the upper half of the bits. In this stage, the reader will go through its list of collided slots and populate bits in the upper half of the tag starting from the LSB. The reader sends out the suffix in addition to the prefix generated by QT until all tags have been read. Only tags within a collided slot will have this unique suffix value and thus will compare the prefix sent by the reader to its upper half tag bits, starting from the LSB. Any time a tag matches this unique prefix and suffix combination, it will respond with its remaining upper half bits. Further collisions may occur during this reading and there exists the distinct possibility that a tag may match multiple prefix and suffix combinations. Thus, when a tag is uniquely identified during this stage, the reader must go through its list of received tags in order to make sure it is not reading the tag multiple times. Equation 3 shows the number of reader energizations where nslots is the number of slots that had collisions. n n

2 2 nslots 2 +1 "= + % [(2 # 2) # $i ] CL i=1

!

(3)

!

The first term of Equation 3 refers to the reduced number of energizations over CS since the slots now being counted down on are only half of the total tag length. A simplified version of QT is used in the second term where the number of bits being considered is approximately half of what it was

before since only the upper half bits of the tag are used. Using the example of a 20 bit tag with a capacitor life of 20 readings and assuming that 50 slots have collided, the number of reader energizations is approximately 103,000 without any QT shortcuts. D. Shortcut Bisected Countdown Scheme The SBCS scheme enhances BCS by restricting the upper half bits transmitted to the reader. Rather than explicitly sending out 0’s and 1’s during a tag’s slot, or when replying to a prefix and suffix combination, the tags will send out a ‘1’ when the bit is a ‘1’ and remain silent for a ‘0’ bit. Using this method of restricted transmission, the reader may be TABLE II RESPONSES OF TAGS WITH UPPER HALF BITS {0100, 1000, 1100, 1101}

Bit

Tag 1

Tag 2

Tag 3

Tag 4

Resulting Knowledge

0 1 2 3

0 1 0 0

1 0 0 0

1 1 0 0

1 1 0 1

1’s collided, none 1’s collided, none This bit is 0 One tag has a 1, none

able to ascertain the contents of some upper half bits, allowing for reduced transmission in Stage 2 of the scheme. Using the example in Table II, Bit 2 can be assumed to be a ‘0’ for this particular prefix since all tags were silent during the send time for this bit. During Bits 0 and 1, multiple tags send a ‘1’, thus no new knowledge is ascertained by the reader since it has no idea how many tags sent a ‘1’. Finally, in the case of Bit 3, the reader knows that exactly one tag has a ‘1’ in that bit position, however, this does not rule out that a tag has a ‘0’ in that position since it cannot know how many tags may have been silent during that time, thus both of these possibilities must be considered by Stage 2 of the scheme. A significant reduction in energizations can be made per bit if the bit were known to be ‘0’. For example, if the upper half of a tag is 10 bits long and Bits 2 and 9 are known to be ‘0’, when the reader gets to Bit 10, instead of having to do a maximum of 1024 energizations, the reader will only have to do a maximum of 256 energizations, a savings of 768 energizations. Equation 4 is essentially the same as Equation 3 except that each collided slot may further incur a saving ϕ in transmissions due to this scheme. n n

2 2 nslots 2 +1 "= + & [(2 # 2) # $i # % i ] CL i=1

(4)

The summation term of Equation 4 shows that savings will be dependent upon the tag IDs of each individual collided slot, which will result in savings possibilities that vary from no savings whatsoever to the ability to shortcut the upper half bits significantly, as previously described. III. PERFORMANCE ANALYSIS The required number of reader energizations for QT and the three schemes CS, BCS, and SBCS, was measured for

varying capacitor life and tag length. For each of the following experiments, 1000 unique tag IDs were uniformly distributed across the tag length. Tag lengths of 20 to 38 bits were considered despite the fact that most RFID tags are 48 or 96 bits in length since it may be possible for the reader to be preprogrammed to know a number of the bits in the tags of products that a company regularly purchases. This in effect could reduce the tag lengths of these products to one of these shorter tag lengths used in the following analyses. A. The Effects of Varying Capacitor Life In the first experiment, the tag length was set at 20 bits, while the capacitor life was varied from 1 tick time up to 30 in increments of 1 to see the effect on the four schemes. Each capacitor life value was used to read in 1000 tags a total of 10 times, the average of these runs is shown in Figure 1.

With BCS and SBCS, the number of energizations is not heavily dependent on the capacitor life and hence the curves show stabilization. The reason is that ratio of the number of energizations required during collisions to the number of energizations to simply cycle through the slots is high. For example, with a 20 bit tag length and a capacitor life of 30, the first term of Equations 3 and 4 results in only 34 energizations. However, the average number of energizations is about 100,000 for SBCS and 260,000 for BCS, meaning that significant penalties are being paid in the collided slots. B. Effect of Varying Tag Length This study was further extended to investigate the dependency on the tag length of the three schemes as well as QT, while maintaining the capacitor charge hold time to 30 time ticks. The tag length was varied form 20 to 38 bits in increments of 2 bits and the plot is shown in Figure 2.

Figure 1. Energization dependency on capacitor life

It can be seen in Figure 1 that SBCS takes the fewest energizations to read the 1000 tags up to a capacitor life of 11, remaining stable irrespective of the capacitor life, while BCS also exhibits stabilization at around 260,000 energizations. For CS, the number of energizations drops considerably, based on the capacitor life, and exhibits better performance than BCS when the capacitor charge hold time exceeds 4 tick times. It also surpasses SBCS in performance if the capacitor life exceeds 11 time ticks. All of the proposed schemes require significantly fewer energizations than QT, which is pegged at approximately 2.1 million energizations. The total number of energizations for QT depends upon the length of the tag and any shortcuts and not on capacitor life. Energizations using the CS scheme decrease with capacitor life since it is inversely proportional to it, as seen in Equation 2. However, in BCS and SBCS, the inverse proportionality only effects one term in Equations 3 and 4, respectively. The second term of each of these equations is governed by the number of slots that resulted in collisions, where the QT protocol had to be employed on the upper half bits of the tags. Figure 1 also shows direct evidence of the benefits of the “restricted transmission” as SBCS takes advantage of a considerable number of additional transmission reduction opportunities, resulting in an improvement in the number of required energizations of nearly 2.6 over BCS, regardless of the capacitor life.

Figure 2. Energization dependency on tag length

Figure 2 shows that by a tag length of 22, SBCS emerges as the most efficient of the three schemes. When the tag length reaches 24 bits CS requires more energizations than BCS. By tag length 28, CS has exceeded 5 million energizations and continues to increase exponentially thereafter. As expected, the number of energizations for QT quickly spikes and goes off of the chart by tag length 22. Most interesting in this experiment is that BCS and SBCS remain relatively flat regardless of the bit length with BCS ranging roughly from 270,000 energizations to 610,000 while SBCS ranges from between 42,000 and 102,000 energizations. To understand the stability achieved by these schemes the function of bisecting the tag must be understood. As the tag length increases, the number of slots will also increase. With an increased number of slots, the chances that a slot will collide is decreased since there is more variability among the lower half bits when the tag length is increased. As a result, more tags will be successfully read during Stage 1 (i.e., reading the lower half bits) of these schemes, resulting in fewer collided slots that are rectified during the second stage. This indicates that with an increased tag length, the first term of Equations 3 and 4 contribute more profoundly to the overall number of energizations than with shorter tag lengths.

The effect of reduced capacitor life on SBCS and BCS tends to be relatively low except during longer tag lengths when the first term of their respective equations becomes more prevalent. Overall, the minimum and maximum number of energizations at each tag length was insignificantly affected as compared with this same effect on CS and QT.

Figure 3. Log ratio of energizations using SBCS over QT

As a final exercise, the relative improvement in the number of energizations when using SBCS over QT was measured for a capacitor life of 5 time ticks, with a tag length range of 20 to 38 bits. Figure 3 illustrates the overall advantage of the SBCS scheme with its combination of countdown, bisection of the tag, and shortcutting due to restricted transmission. As shown in Figure 3, even at tag length 20, SBCS had a greater than 21 fold improvement over QT, with this value reaching 25,000 by tag length 30, and finally topping out at an improvement of greater than 4.5 million at a tag length of 38 bits. IV. FURTHER DISCUSSION The improvements in SBCS are achieved at a price of slightly increased circuitry in the tag and complexity in the reader. First, the tags have to implement a countdown timer, which can be done using the clock that is already present in most passive tags [2]. Tags will also need the ability to take advantage of “restricted transmission”, which is simply a logical operation deciding if the tag should send a bit or not. In addition, tags must be able to differentiate between the stages of the scheme so that the tag knows when to use restricted transmission or to do a countdown. Doing so will offer a savings in power consumption since less bit transmissions will occur and will require minimal changes in the tag. Finally, the comparator circuit in the upper half of the tag bits must be variable, depending on the length of the prefix sent by the reader during the second stage of SBCS. These aforementioned changes should make little, if any, impact on the price of the tags. The reader will need additional software to implement the scheme, but considering that the reader has essentially unlimited resources as compared with the tags, this should not be a significant issue. With minor modifications to the tags and the reader, SBCS will afford substantial savings on

the number of energizations, leading to reduced bulk tag reading latency, a reduction in the number of transmissions, which is highly desirable in industrial environments, and reduced expended energy in the systems, leading to an overall reduction in system cost. V. CONCLUSION This work represents the proposal of three schemes, Countdown Scheme (CS), Bisected Countdown Scheme (BCS), and Shortcut Bisected Countdown Scheme (SBCS), that may be employed to improve upon Query Tree, the most prevalent of the currently existing memoryless RFID bulk reading protocols. CS, BCS, and SBCS aim to reduce the number of tag collisions and the number of energizations required by the reader, thus reducing the reading latency of the tags. It was shown that each of these schemes requires significantly fewer energizations than their preceding scheme. CS avoids collisions completely by giving each tag its own unique slot time in which to announce its presence by transmitting a single bit; it could be a highly efficient scheme if the capacitors are able to hold charge for a long time. Given that a long capacitor life may not be possible, the scheme was enhanced through tag bisection and applying CS only to the lower half bits while using the Query Tree protocol for the upper half bits when collisions are detected. A further improvement in SBCS is then achieved by restricting the transmission on some upper half bits to further enhance the use of Query Tree during Stage 2 of the scheme. Future work will include further enhancements of SBCS so that it can handle more realistic RFID tag sizes of 48 and 96 bits as well as reading latency comparisons with more contemporary RFID tag reading protocols such as Frame Slotted Aloha. Each technique used is simple, but when combined, provides a significant reduction in energizations with its accompanying advantages, which has been proven through extensive simulations and analysis. REFERENCES [1] [2]

[3]

[4]

[5] [6] [7]

R. Weinstein, “RFID: A Technical Overview and Its Application to the Enterprise,” IEEE IT Pro, May/June 2005, p.27-33. F. Zhou et al, “Evaluating and Optimizing Power Consumption of Anti-Collision Protocols for Applications in RFID Systems,” ISLPED’04, Newport Beach, CA, USA, August 9-11, 2004, p.357362. C. Law, K. Lee, and K. Siu, “Efficient Memoryless Protocol for Tag Identification,” Proceedings of the 4th International Workshop on Discrete Algorithms and Methods for Mobile Computing and Communications, Boston, MA, August 2000, p.75-84. M. A. Bonuccelli, F. Lonetti, and F. Martelli, “Tree Slotted Aloha: A New Protocol for Tag Identification in RFID Networks,” Proceedings of the 2006 International Symposium on a World of Wireless, Mobile and Multimedia Networks, 2006. J. Myung and W. Lee, “Adaptive Splitting Protocols for RFID Tag Collision Arbitration,” MobiHoc’06, Florence, Italy, May 22-25, 2006, p.202-213. D. Simplot-Ryl et al, “A Hybrid Randomized Protocol for RFID Tag Identification,” Sensor Review, Issue 26, Volume 2, 2006, p.147-154. S. Jain and S. R. Das, “Collision Avoidance in a Dense RFID Network”, WiNTECH’06, Los Angeles, CA, USA, September 29, 2006, p.49-56.