33 CHAPTER 3 ANTI-COLLISION PROTOCOLS IN RFID BASED HUMAN TRACKING SYSTEMS (A BRIEF OVERVIEW) In a RFID based communication system the reader activates a set of tags, and the tags respond back. As outlined in the previous chapter the reply back signals of some tags do hit the reader simultaneously causing collisions. The reasons are many which cause collisions. Exploring the reasons for collisions requires in depth study of a number of phenomena and parameters like radiation patterns of readers and tags, type of protocols used in the system, dynamics of tag carriers, which is beyond the scope of this study. In fact, a number of collision avoidance strategies and techniques are already available in the literature. Unfortunately, no such technique ensures totally collision free RFID communication system. It is in this context, this thesis introduces a novel technique which guarantees a multiple access RFID communication system with out collision. Before getting into such details, we shall briefly review some of the existing collision avoidance techniques just for the sake of understanding the underlying principle of collision avoidance. Collision avoidance techniques for tags are mainly classified into two types: (a) probabilistic, and (b) deterministic. Alternatively, reader based collision avoidance protocols are mainly classified into four types:
34 (a) Polling based protocol (b) Splitting method based protocol (c) I-code protocol and (d) Contact less protocol. Figure 3.1 shows various multiple access and anti-collision procedures. Figure 3.1.: Types of Multiple Access and Anti-collision procedures The multiple-access techniques are categorized as Space Division Multiple Access (SDMA) technique, Time Division Multiple Access (TDMA) technique, Frequency Division Multiple Access (FDMA) technique and Code Division Multiple Access (CDMA) technique. We shall briefly review these four techniques. (i) Space Division Multiple Access (SDMA) Technique Space Division Multiple Access technique categorizes the available channel capacity into separate areas which are spatial. It minimizes the reading range of readers and forms as an array in space. It uses
35 electronically controlled directional antenna. Various tags will be distinguished based on their angular positions. (ii) Time Division Multiple Access (TDMA) Technique Time Division Multiple Access technique allocates the available channel bandwidth to readers and tags with respect to time. Based on the time allocated by the reader, tags respond in their corresponding time slots with the reader. In this technique, reader talks first, tags listen. Later tags respond to the reader based on the time allocated. Most of the RFID communication systems use TDMA technique. (iii) Frequency Division Multiple Access (FDMA) Technique In this method, the existing resource i.e., channels bandwidth into smaller bandwidths which in turn are dedicated to individual tags until the communication among the tag and the reader is completed. (iv) Code Division Multiple Access (CDMA) Technique This technique works basically on the principle of cellular mobile communication. Readers are identified by their unique codes and works in synchronization with each other. This technique is too complicated to implement for RFID system as the computation time required is more for both tags as well as reader. The research work has been laser focused on Tag anti-collision procedures only. Hence, reader s collision avoidance procedures are not discussed in this thesis.
36 3.1. TAG COLLISION AVOIDANCE PROTOCOLS The collision avoidance protocols of the tags are mainly classified into two types. They are (a) Probabilistic and (b) Deterministic Probabilistic techniques make use of different types of ALOHA protocols. Specifically, the use of Framed Slotted ALOHA [32] makes the reader inform all the tags present in its vicinity about the frame length. In turn, the tags select any one of the time slots available and communicate with the reader. This process continues until all the tags have been identified and communicated at least once. All the active and semi active tags which have successfully communicated with reader are acknowledged at the end of every frame. The tags that have communicated with the reader will not further participate in the subsequent rounds so that the probability of collisions is minimized and thus the overall identification time reduces. On the other hand, in a multiple tag environment, the passive tags transmit their packets incessantly in every frame causing a monotonic increase in the total read time. Thus, we observe that there is a dire necessity for improvised protocols to resolve the collision problem posed by RFID systems that work with passive tags. Deterministic protocols come under the classification of tree-based protocols [30, 40 and 42] where the tags are treated as nodes in a binary tree. The tags are identified based on their unique ID s where the reader scans all the nodes of the tree in every round. Refer to figure 3.2, in
37 which one of the deterministic protocols named Framed Query Tree (FQT) protocol is shown to use a binary tree search approach. Figure 3.2: Binary tree representation of a typical identification process by FQT Protocol These algorithms are quite accurate in identifying the tags. However, more over head is involved in the identification process and the reader gets burdened in specifying every time address ranges of the tags. To summarize, the general requirement for both probabilistic and deterministic collision avoidance protocols is proper tag identification and enumeration of the actual number of tags K in the system. This is very much required in order to (i) optimize the frame size in Framed Slotted ALOHA protocol (probabilistic approach) and (ii) continue tree-based identification process for computing the number of slots (deterministic approach). Fast enumeration of tags is an important requirement which can also be achieved by using non-identifiable information (a string of bits) available with the tags so that the tag density K is calculated. For certain applications where privacy is of prime concern, it may not be
38 possible for the readers to identify the tags using the normal procedure defined. In such situations, tags can transfer non-identifiable encrypted information, which can be accepted to enumerate the cardinality of the tag set. 3.1.1. Probabilistic Protocols. ALOHA PROTOCOL: As outlined in chapter 1, ALOHA protocol is a multiple medium access control protocol developed for a packet radio network which is implemented at the data link layer to permit multiple tags access the medium in a semi-random manner without any interference or collision. Protocol was developed by a scientist by name Norman Abramson and his team at University of Hawaii. A group of stations forms a network among them and starts transmitting their information in a random fashion to one another under the supervision of one of the stations which controls and monitors the transmission of packets. After transmitting, the station checks whether the transmission was successful or lost due to collision by listening to the broadcast. If the data is lost due to collision, the station retransmits the same information packet once again. The same methodology is adopted and implemented in RFID communication system. The success rate or throughput that has achieved with this protocol is around 18%.
39 In what follows, we present different versions of ALOHA protocols that are in use currently. Figure 3.3 shows the classification of Probabilistic Algorithms (ALOHA based protocols). All probabilistic protocols are basically enhanced versions of ALOHA protocol. Figure.3.3 shows a variety of probabilistic algorithms. The main aim of these algorithms is to decrease the probability of collisions since all the tags try to respond in different intervals of time. Figure 3.3: Classification of Probabilistic algorithms 3.1.1.1. Pure ALOHA (Basic ALOHA) Protocol Figure 1.2 (a) represents an example RFID environment where four tags and one reader are engaged with the help of pure ALOHA protocol. Here, the tags-reader engagement is totally random. That is, tags send their packets randomly when electromagnetically activated by the reader without any control. In such a case, collision of packets is inevitable. To be precise, when two or more tags try to transmit their data packets at
40 the same time which is represented by dark rectangles at the reader side, collision is said to occur. Refer to figure 1.2 (a). The efficiency of this algorithm will drop down if there is a large tag density in the vicinity of the reader or if the tags have huge amount of data [34]. One possibility for optimizing the performance of the ALOHA algorithm is the slotted ALOHA algorithm. 3.1.1.2. Slotted ALOHA Protocol Figure 1.2 (b) represents the slotted ALOHA in which the available time is further divided into distinct time intervals called as time slots and a tag is permitted to transmit at the beginning of a slot. In this protocol, collision of the tags will be either completely or not at all, i.e., no chances of occurrence of partial collisions like in the pure ALOHA case. The tags will transmit in predefined and synchronous time slots. The synchronization among the tags is achieved under the supervision of the reader. Thus, this can be called as a reader-driven TDMA anticollision protocol [35]. The Slotted ALOHA algorithm has better performance than the basic ALOHA because of the synchronization between the reader and the tags. Moreover, the defined points of time where the tags start responding decrease the probability of collision. However, the Slotted ALOHA algorithm still has the weak point that it provides a meager performance when the tag density increases [24, 35].
41 3.1.1.3. Basic Framed Slotted ALOHA (BFSA) Protocol Figure 1.2 (c) represents a Basic Framed Slotted ALOHA (BFSA) protocol which is an enhanced version of the above two protocols. This protocol includes further discrimination of time in the form of frames, with N slots per frame. Every tag tries to communicate with the reader at most once in a randomly selected slot. Every tag can have one collision at the most in a given frame. A frame is a time period with some preamble and synchronization bits appended and transmitted by the reader to the tags which contain number of slots. In BFSA protocol the number of slots per frame is fixed and not changed during the identification process. The reader broadcasts the request command which includes the frame size (number of time slots available) and the random number. Each tag in turn uses the random number to select a slot in a frame, and respond in that slot [36]. The process of BFSA algorithm can be shown in Fig.3.4. Figure 3.4: Process of BFSA Algorithm
42 The frame shown in the above example consists of eight slots. In slot 1; Tag 6, Tag 7 and Tag 14 try to communicate with the reader at the same time where a collision occurs. Hence it is identified as unsuccessful slot. In slot 2, no tag responds and is identified as empty slot. In slot 3, Tag 1 and Tag 4 try to send their ID where again collision occurs which will be identified as unsuccessful slot. In slot 4 Tag 3 alone responds and is identified as successful slot with single tag responding. In slot 5, Tag 5 alone responds which is again identified as a successful slot. In slot 6, Tags 8, 9 and 11 try to send their IDs where collision occurs and remains as unsuccessful slot. In slot 7, collision occurs because of Tag 10 and Tag 12 responding at the same time. Hence it turns out to be an unsuccessful slot. Finally, during slot 8, Tag 13 responds with its ID which turns out to be a successful slot. Therefore, the tags which could not communicate with the reader must respond in the next round. Although the Basic Framed Slotted ALOHA protocol is simple to implement, the percentage of tag identification is less because of fixed frame size. This protocol completely fails when there is sudden variation in the tag density. Further wastage of time will happen if large frame size is used for less number of tags [36]. The modifications immediately made to Basic Framed Slotted ALOHA (BFSA) are to adapt the frame size in accordance with available tags in the next successive rounds. The number of slots per frame is
43 decided as per the number of tags present in the reader field. Few of them are discussed below. 3.1.1.4. Dynamic Framed Slotted ALOHA (DFSA) Protocol In Dynamic Framed Slotted ALOHA (DFSA) protocol, the number of slots is adaptable in order to have excellent tag identification. The number of slots per frame will be determined by collecting the information about the number of successful slots, unsuccessful slots and idle slots. Based on this information the frame size will be designed for the successive rounds. If the value of the probability of collisions is greater than the defined threshold value, the frame size is automatically increased in the successive rounds. If the collision probability is less than the defined threshold value, the frame size will be automatically decreased in the next round. Thus the frame size is adaptable i.e., number of slots can be either decreased or increased based on the threshold value maintained at the reader [24]. DFSA algorithm is more efficient than BFSA algorithm because the reader readjusts the frame size based on the tag density. Thus, DFSA algorithm can improve the probability of identification and estimation of the tags when compared to the BFSA algorithm [24, 37, and 38]. To summarize, adjusting the frame size alone cannot decrease the tag collision in case of high tag density because the number of slots per frame cannot be increased indefinitely and takes more time to read the entire frame. This algorithm failed especially when tag set is high.
44 3.1.1.5. Advanced Dynamic Framed Slotted ALOHA (ADFSA) Protocol Advanced Dynamic Framed Slotted ALOHA (ADFSA) protocol adjusts the number of slots per frame based on the enumeration of the tags. It works better than Basic Framed Slotted ALOHA (BFSA) protocol. There are many existing estimation functions. Vogt [26] suggested two approaches to enumerate the tags present in the vicinity of the reader. The first suggested method is Lower Bound (LB) method and the second method is called CIILB (Chebyshev Inequality Improved Lower Bound). ADFSA algorithm has the same problem as DFSA algorithm that the number of slots per frame cannot be increased infinitely as the tag density increases. Therefore, this protocol shows better performance when the tag density is less, however, it fails when the tag density is high [38]. 3.1.1.6. Enhanced Dynamic Framed Slotted ALOHA (EDFSA) Protocol In all the other variants of Framed Slotted ALOHA protocols, increasing the number of slots per frame indefinitely is practically not feasible in order to increase the system efficiency to its maximum value. However, this problem can be handled by avoiding the number of active tags approximately to the same size which is equal to number of slots per frame. Enhanced Dynamic Framed Slotted ALOHA (EDFSA) protocol can handle this problem by grouping unread tags using modulo operation [27].
45 3.1.2. Deterministic Identification Protocols Most Tree based protocols are basically modified forms of Binary or Query Tree protocol. Fig.3.5 shows variety of Tree based algorithms. 3.1.2.1. Binary Tree Protocols Binary Search (BS) algorithm can solve the collisions by gradually decreasing the collided bits in transponder s electronic serial number which is reciprocated by the transponder to the reader. In turn, the reader will detect the position of the collided bit by using efficient bit coding techniques; for example, using Manchester code the collided bit can be traced out. In this bit coding, the value of the bit can be determined based on the way the transition has occurred. In Manchester coding, positive transition is treated as logic 0 and negative transition is considered as logic 1. If there is no transition it means that an error has occurred. Figure 3.5: Tree based deterministic algorithms Figure 3.5 shows the bit coding using Manchester coding scheme.
46 Figure 3.6: Bit coding using Manchester code If at a time more than one transponder transmits bits of different values simultaneously, the positive transitions cancel the negative transitions. If the reader detects no transition, it is clearly understood that a collision has occurred between any two transponders who are simultaneously trying to access the channel. BS algorithm divides the contending transponders into two groups. It permits the transponders which have first collided bit 0 to respond in the next request and transponders with first collided bit 1 are not permitted to respond. Assume that there are four transponders in the vicinity of the reader. At the beginning, the reader sends a request asking all the transponders that have serial numbers less or equal to 11111111 to respond. Since 11111111 is the highest serial number, all transponders will answer to the reader in the starting iteration by sending their own serial numbers. Suppose if a collision is sensed in the received serial number at bit 0, bit 4 and bit 6. It implies that there are eight (2 3 =8) or less transponders in the reading range. Bit 6 is the
47 highest collided bit. This means that there is at least one transponder whose serial number value is less than 10111111. Thus, in order to restrict the number of answering transponders, the reader transmits a new request for transponders that meet the condition of ( 10111111) in the next round. Now let us assume that the received serial number will have collisions at bit 0 and bit 4. The same procedure will be repeated. The reader will request the transponders that have the serial numbers less than or equal 10101111 in the next round. Transponder 2 is only located in this iteration (3rd iteration). The reader then selects Transponder 2 and starts the communication. By repeating these procedures, all transponders can be recognized by the reader [24]. The Adaptive Query Splitting protocol is an enhancement made to that of the Query Tree protocol and the Adaptive Binary Splitting protocol is an improvement made to the Binary Tree protocol are proposed by Myung [23, 25, and 40]. The enhanced protocols use the data retrieved from the final process of transponder identification in order to minimize the collision. 3.1.2.2. Query Tree (QT) Protocols In QT protocol, the reader broadcasts a prefix to all the transponders in the vicinity and the transponders whose ID matches to the prefix will respond. If a collision occurs, the reader increments the prefix by one bit until no collision is observed. The round will
48 automatically gets terminated and a new round with new prefix will be issued by the reader, once a transponder is identified [41]. 3.1.2.3. Tree ALOHA Based Protocols In collision prone environment, performance is usually measured as system efficiency which is defined as the ratio between the numbers of transponders identified to the frame size used in the whole identification process. All the algorithms discussed till now under deterministic approach have shown an average performance well below 50%. In fact the best performing algorithms, namely QT and TSA achieve 40% of system efficiency on an average [28, 41]. Long identification time is the major problem associated with tree based anti-collision algorithms especially with large number of tags and long ID s However, taking into account the advantages of tree based protocols and ALOHA based protocols and combining them together can lead to better protocol which will have excellent performance. The following section presents some of the protocols which combine different algorithms derived from the two methods. 3.1.2.4. Slotted Binary Tree (SBT) Protocols SBT protocol tackles the collisions as soon as it occurs. Whenever a collision happens in a particular slot J and the transponders who have not responded in slot J wait until the reader informs the collided slot number to all the transponders. The reader then directs the
49 transponders which has undergone collision to choose group 0 or group 1 randomly. The tags that opted group 0 respond in slot J+1 while the tags in the other group remain standstill until all the transponders in the group 0 are successfully recognized. If slot J+1 is either a idle slot or successful slot, transponders belonging to group 1 are instructed to respond in slot J+2. Otherwise, there is a chance of collision occurrence; the same procedures are repeated to solve the next collision. 3.1.2.5. Bi-Slotted Tree Based Anti-Collision Protocols Bi-slotted Tree based protocols are broadly classified into two types. They are (i) Bi-slotted query tree (BSQT) protocol and (ii) Bi-slotted collision tracking tree (BSCTT) protocol. The unique feature of these protocols is that the reader transmits two query bits each of length n to all the transponders in its vicinity, which have unique (n-1) bits with different end bit. The following steps explain the collision procedure of the above two protocols [30, 31]: Step 1- Request: In this step, the reader transmits a prefix of length (n-1) to all the transponders. Step 2- Grouping:
50 In this step, transponders which are in the proximity zone of the reader respond if the prefix of length (n-1) matches with the (n-1) bits of their ID s. The transponders answer in any one of the time slots based on the n th bit. If the value of the n th bit is 0, transponders select the initial slot to communicate; otherwise, if the value of the n th bit is 1, transponders select the next slot to communicate. Hence, based on the selected slot the value of the n th bit is known. If BSQT protocol is used, the transponders transmit their IDs starting from (n+1) bit to the last bit. On the other hand, if BSCTT protocol is used, the transponders transmit their IDs starting from (n+1) bit to the last bit and continues until they receive acknowledgement to indicate the reader that collision has occurred. Step 3- Decision: In this step a decision will taken based on the occurrence of collision. The reader stores a new prefix value in (LIFO) memory if a collision occurs (i) In BSQT protocol, the prefix of length (n-1) will be stored in addition to the information about the selected slot. (ii) In BSCTT protocol, the prefix of length (n-1), the data received before the occurrence of collision and the information related to the selected slot will be stored. Step 4- Repeating the steps 1, 2 and 3 until the LIFO memory is empty.
51 By using BSQT protocol and BSCTT protocol, the main advantage is the average prefix overhead required will be reduced to the half of its value when implemented using other tree based protocols like QTA and CTTA. Other advantage of BSQT protocol over QTA is that only few bits are required to identify the transponders when compared to QTA. Furthermore, the performance of the BSQT protocol and Query Tree protocol gets affected as the number of bits required identifying the transponders increases as the transponder density increases. To evaluate the system performance, the iterations taken to identify a transponder is used as an alternate. BSQT protocol requires half the number of iterations when compared to that of Query Tree protocol. Finally, BSQT protocol performs better than Query Tree Protocol with respect to the total number of bits required and secondly the total number of iterations taken for identifying each transponder [31]. 3.1.2.6. Framed Query Tree (FQT) Protocol In this protocol, the transponders which are in the vicinity of the reader are randomly divided into frame units [33]. QT protocol is used to identify transponders which are randomly divided in each unit. The process of Framed Query Tree protocol is explained as follows: In the starting of the identification process, the reader transmits a request command to all the transponders which are in the vicinity asking for their ID. The request command transmitted includes Frames Number, also called as Epoch Size which informs about the total
52 number of the frames available (see fig.3.7). Then each transponder randomly chooses a frame and communicates with the reader only in the frame selected by the transponder. Within each frame, the reader uses Query Tree protocol to identify the transponders belonging to this frame. Thus, the reader sends the Frames Number along with the prefix of ID. If the opted frame is same as the frame to be transmitted, the transponder responds if its ID matches with the prefix (QT protocol). The reader continues to the next frame until all the transponders opted for that frame are identified. The above process is repeated for every frame [33]. Figure 3.7: Identification process in FQT Protocol (Same as Figure 3.2) Figure.3.7. indicates an example of the identification process where Framed Query Tree protocol is used in the RFID system where the transponder density in the vicinity of the reader is eight and the number of frames is 4 [33]. To improve the performance of Framed Query Tree protocol, an appropriate Frames Number (epoch size) is required. Query Tree
53 protocol is best suitable when the depth of the tree does not exceed more than 2. When the transponder density to be identified is N and the Frames Number i.e., epoch size is ES, the most ideal ES can be determined as follow [33]: N = 2 * ES However, it is very difficult to calculate a proper Frames Number in the starting of the identification process because transponder density N is not known. Therefore, the Framed Query Tree protocol uses a test by name First Frame Test. First Frame Test starts the process with small number of frames and if the collision in the first frame exceeds a threshold value, FFT increases the epoch size. As mentioned above all the transponders are randomly divided into multiple frames. If the first frame records many collisions, the successive frames may also face the same problem. [33]. Frame 3 in Figure 3.7 shows the best performance by implementing the above protocol when the transponder count = 2 and tree depth is equal to one. Hence a threshold value is to be defined in order to save the tree depth from not exceeding than the threshold value [11]. 3.1.2.7. Query Tree Aloha (QT-ALOHA) Protocol Query Tree-ALOHA protocol is derived from combining Framed Slotted ALOHA (FS-ALOHA) protocol and Query Tree protocol. Framed Query Tree protocol (FQT) is build using Framed Slotted ALOHA protocol
54 and Query Tree protocol and is used for actual transponder identification process. The process of QT-ALOHA protocol is explained as follows [33]: In the starting of the process, the reader transmits a request command which includes both prefix and frame size. The transponders whose IDs are matched are treated using Framed Slotted ALOHA protocol and transmits in the allocated slots belonging to the transmitted frame. During Framed Slotted-ALOHA protocol phase, if a collision takes place even in a single slot, it is treated as a collision pertaining to Query Tree protocol. Now a new prefix is generated and included in the queue [33]. Making comparison between the Framed Query Tree protocol and other protocols discussed so far, Framed Query Tree protocol exhibits best performance than any other deterministic collision avoidance protocols [33]. So far we had a preliminary review on the current RFID technology and subsequently the problems associated with its use in Human Tracking System. As a special case, the problem of tag collision and certain current solutions to overcome this problem has been briefly reviewed. It is also been outlined that the tag collision avoidance techniques that are used in the present day system does not guarantee absolute or does not provide a complete solution to this problem. In the light of the above argument we present a novel and absolutely reliable tag collision avoidance technique called Adaptive Slot Adaptive Frame (ASAF) ALOHA Protocol in the next chapter.