中 華 大 學

中 華 大 學

碩 士 論 文

中文題目:

改善以樹狀為基礎之 RFID 標籤防碰撞技術

系 別 所: 資訊工程學系碩士班 學號姓名: E09402022 薛仁豪 指導教授: 許 慶 賢 博士

中華民國 九十七 年 七 月

On Improving Tag Anti-Collision

Techniques towards High Throughput and Reliable RFID Services

By

Ethen Shai

Advisor: Prof. Ching-Hsien Hsu

Department of Computer Science and Information Engineering Chung-Hua University

Hsin-Chu, 30067, Taiwan

July 2008

中文摘要

隨著無線 RFID 技術的出現，防碰撞(Anti-Collision)的問題已逐漸被大家所 重視，也促使大家針對讓 RFID 系統用更有效率的方式運作，提出不同的啟發 性方法(heuristic algorithms)。然而在提升系統處理能力及穩定度上仍然有一些 挑 戰 ， 也 就 是 由 於 當 網 狀 系 統 (network) 密 度 高 時 ， 根 本 的 技 術 (underlying technologies)在系統執行層面上面會有不同的限制。在此份研究中，我們提出 了” 門檻跳躍(Threshold Jumping，TJ)”及”迂迴環繞掃描(Warp-Around Scan，

WAS)”技術，目的用來整合高密度 RFID 環境下同時發生的通訊(simultaneous communications)、加速 tag 的識別以及增加無線 RFID 系統的整體讀取率(read rate)。

門檻跳躍(Threshold Jumping，TJ)的主要想法是在資料庫搜尋樹(query tree) 的不同階層限制碰撞的次數。當碰撞的次數超過預先所定義可接受的比率時，

其 所 透 露 的 是 在 RF 範 圍 內 的 密 度 過 高 。 為 了 避 免 不 必 要 的 識 別 (identification)，先前比對(prefix matching)會移到資料庫搜尋樹(query tree)的較 下 層 ， 以 便 緩 和 碰 撞 的 問 題 。 設 定 頻 率 範 圍 的 方 法 (The method of setting frequency bound)的確改善了高密度及任意分佈之 RFID 系統(randomly deployed RFID systems)的效率。然而在不規則或是不平衡的 RFID 網狀系統中，無效率 的情況仍有可能發生。

問題在於每個識別層級中，先前比對(prefix matching)會從左邊的次級樹進 行到右邊的次級樹，而這可能會導致不平衡的資料庫搜尋樹(query tree)，在這 種情況下，如果識別跳到下一階層的話，右邊的次級樹就不會被辨識。而將”

迂 迴 環 繞 掃 描 (Warp-Around Scan ， WAS)” 的 概 念 及 ” 設 定 頻 率 範 圍 (setting frequency limitation)”的想法放在一起可以大大改善這個問題。為了評估所提出

建議之技術的表現，

TJ

WAS

實驗結果顯示，所建議的技術不論在低或高密度的環境下都提供了較優的表 現。因此顯示出來的是，由增加系統生產率、效率及易於執行的觀點而言，此

結果說明了

TJ

WAS

關鍵字 : 碰撞,防碰撞, 標籤, 資料庫搜尋樹, 門檻跳躍, 迂迴環繞掃描

II

Abstract

With the emergence of wireless RFID technologies, the problem of Anti-Collision has been arousing attention and instigated researchers to propose different heuristic algorithms for advancing RFID systems operated in more efficient manner. However, there still have challenges on enhancing the system throughput and stability due to the underlying technologies had faced different limitation in system performance when network density is high. In this thesis, we present a Threshold Jumping (TJ) and a Warp-Around Scan (WAS) techniques aim to coordinate simultaneous communications in high density RFID environments, to speedup tag identification and to increase the overall read rate and throughput over wireless RFID systems. The main idea of Threshold Jumping is to limit number of collisions in different level of the query tree.

When number of collisions over than the predefined acceptable rate, it reveals that density in RF field is too high. To avoid unnecessary identification, the prefix matching will be moved to lower level of the query tree, alleviating the collision problems. The method of setting frequency bound indeed improves the efficiency in high density and randomly deployed RFID systems. However, in irregular or imbalanced RFID networks, inefficient situation may happen. The problem is that the prefix matching proceeds from left sub-tree to right sub-tree in every identification level, this may result an imbalance query tree on which the right sub-tree was not examined if the identification is jumped to next level. Putting the concept of Warp-Around Scan and the idea of setting frequency limitation together can largely ameliorate the problem.

To evaluate the performance of proposed techniques, we have implemented the TJ and the WAS method along with other Tag identification approaches. Our preliminary experimental results show that the proposed techniques provide superior performance in both low and high density environments. It is shown that the TJ and WAS is effective in terms of increasing system throughput, efficiency and easy implementation.

Keywords: Tag Anti-Collision, Hidden Terminal, RFID, Query Tree, Warp-Around

Acknowledgements

First of all, I would like to thank Prof. Ching-Hsien Hsu and all oral committee, for their valuable suggestions on improving the quality of this thesis.

Prof. Ching-Hsien Hsu is one’s adviser. He is a conscientious and careful scholar.

He also gives lots of suggestions not only for the thesis but also for my life of graduate.

One is fortunate to be one of Prof. Hsu’s graduate student. I would also like to thank members of P.D. Lab, they always give me support when one works on the thesis.

Finally, I would like to thank my family for giving me great constant support, show tolerance for me and give me power to get through these 3 years.

IV

Table of Contents

Chienese Abstract ... I English Abstract... II Acknowledgements ...III Table of Contents ... IV List of Figures ...V List of Tables………VI

1 Introduction...1

1.1 Motivation ...1

1.2 Objective ...3

1.3 Thesis Organization ...4

2 Related Work...5

3 Preliminary ...8

3.1 Anti-Collision Problem ...8

3.2 Tree Base Tag-Collision Protocol...9

3.3 Research Architecture ...11

4 Threshold Jumping (TJ) ...12

5 Warp-Around Scan (WAS)...16

6 Performance Evaluation and Results ...18

6.1 Simulator and Comparison Metrics ...19

6.2 Experimental Results ...20

7 Conclusions and Future work ...24

Reference ...25

List of Figures

Figure 1: An example of Query Tree Algorithm ...9

Figure 2: An example of Binary Tree Algorithm ...9

Figure 3: Concept of Threshold Jumping...11

Figure 4: Example of Threshold Jumping ...13

Figure 5: Concept of Warp-Around Scan...16

Figure 6: Example of Warp-Around Scan...16

Figure 7: Performance comparison of the TJ, WAS and QT with 8 bits(a) and 12bits(b) RFID tag...19

Figure 8: Performance comparison of the TJ, WAS and QT with 16 bits RFID tag. ...20

Figure 9: Performance comparison of the TJ, WAS and QT with different threshold ratio (a) results of density = 80% (b) results of density = 50%...21

Figure 10: Performance comparison of the TJ, WAS and QT in imbalanced tree. ...22

VI

List of Tables

Table 1: QT and TJ Identification of two different schemes...14 Table 2: Tag Identification of three different schemes ...17

CHAPTER 1

Introduction

1.1 Motivation

Radio Frequency Identifier (RFID) System is an automatic technology aids machines or computers to identify objects, record metadata or control individual target through radio waves. The RFID system is composed by two components, tags and readers. An RFID tag is comprised of integrated circuit with an antenna for storing information and communication, respectively. An RFID reader is capable of reading the information stored at tags located in its sensing range. The electronics in the RFID reader use an outside power resource to generate signal to drives the reader’s antenna and turn into radio wave. The radio wave will be received by RFID tag which will reflect the energy in the way of signaling its identification and other related information. In matured RFID systems, the reader’s RF can also instruct the memory to be read or written from which the tag contained.

Many applications, such as supply chain automation, identification of products at check-out points, security and access control, localization, and object tracking have been developed to take the primary function of RFID systems. Advantages of RFID technologies, such as price efficiency, fast deployment, reusable and accuracy of stock management also broaden the scope of applications of RFID systems. Advanced characteristics of recent RFID readers, like size miniaturization and capabilities of Wi-Fi or cellular also motivate the development of large scale RFID systems.

In recent RFID technologies, it is motivated that an RFID system can be integrated with wireless sensor network by interfacing RFID tags with external sensing capabilities, such as light, temperature or shock sensors; forming a hybrid infrastructure

2

that combines advantages of both techniques, such as accurate identification, monitoring of objects and efficient deployment. Similar to wireless sensor network, RFID tags can be deployed in an ad-hoc fashion instead of pre-installed statically. As a result, RFID has gradually been applied to our daily lives so that some identification problems happen. When readers try to identify multiple tags, signals collision will appear because the RFID system is through radio waves to carry signals. If one tag follows another one, the read rate is naturally high, but if multiple tags simultaneously pass through readers, the collision may happen. Therefore, efficient methods for identifying multiple tags simultaneously are of great importance for the development of large-scale wireless RFID systems.

1.2 Objective

At present, there are anti-collision technologies and tree-based technologies for identifying multiple tags. Anti-collision technologies are based on collision rates, when signals collide, tags will repeatedly send signals in next interval of time to make sure completed read. For the complete read of multiple tags, the read time would be extended. The problem of signal collision can be avoided by the binary search technology to raise correct read rates and to shorten the read time when multi-tags are read at the same time. Regarding that the query tree technology can raise correct read rates, that may be true, but there still have challenges in long identification latency.

In this thesis, we present a Threshold Jumping (TJ) and a Warp-Around Scan (WAS) techniques aim to coordinate simultaneous communications in high density

RFID environments, to speedup tag identification and to increase the overall read rate and throughput in large-scale RFID systems. The main idea of the proposed technique is to limit number of collisions during the identification phase. When number of collisions larger than the predefined acceptable rate, it reveals that the density in RF field is too high. In order to minimize unnecessary inquiry, the prefix matching will be moved to lower level of the query tree, alleviating the collision problems. The method of setting collision bound indeed improves the efficiency of large-scale RFID tag identification. Together with the concept of Warp-Around Scan, the tag anti-collision problem could be significantly ameliorated. To evaluate the performance of proposed techniques, we have implemented the TJ and the WAS method along with other tag identification approaches. The experimental results show that the proposed techniques present significant improvement in most circumstance.

4

1.3 Thesis Organization

The rest of this thesis is organized as follows: In Chapter 2, a brief survey of related work will be presented. Chapter 3 introduces the Binary search problem.

The Binary search optimization for tag identification will be discussed in Chapters 4 and 5. Performance analysis and simulation comparisons will be given in Chapter 6.

Finally, in Chapter 7, some concluding remarks are made.

CHAPTER 2

Related Work

Many research results have been proposed in literature. Security and privacy related literatures [19] focused on methods of preserving and protecting privacy of RFID tags; the RFID reader collision avoidance and hidden terminal problems were first addressed in [5] aiming to enhance accuracy of RFID systems; the energy saving and coverage problem [13, 21] was extensively studied in order to improve lifetime of wireless RFID networks.

Research efforts for collision avoidance have been presented in literature.

Frequency Division Multiple Access (FDMA), Code Division Multiple Access (CDMA), Time Division Multiple Access (TDMA) and Carrier Sense Multiple Access (CSMA) [6] are four basic access methods to categorize MAC-layer protocols. Standard collision avoidance protocols like RTS-CTS [17] cannot be directly applied in RFID systems due to the reason, in traditional wireless networks, the CTS are sent back to the sender. Similar situation in RFID system, when a reader broadcasts an RTS, all tags in the read range need to send back CTS to the reader. It then requires another collision avoidance mechanism for CTS, and it will make the protocol more complicated.

Techniques for resolving RFID reader collision problems are usually proposed as reader anti-collision techniques or tag anti-collision solutions. The Colorwave [20] is a scheduling-based approach prevents RFID readers from simultaneously transmitting signal to an RFID tag. The Colorwave is used as a distributed anti-collision system based on TDMA in RFID network. Pulse protocol [1] is referred as beacon broadcast and CSMA mechanism. Readers periodically in separated control channels send a

“beacon” during communication with tags. The contend_back-off and the delay_before_beaconing in the protocol are similar in wireless networks. If the reader receives a beacon, the residual back-off timer will be stored and kept till the next coming chance. This process is expected to achieve the fairness among all readers. A coverage-based RFID reader anti-collision mechanism was proposed in [7]. Kim et al.

6

[7] presented a localized clustering coverage protocol for solving reader collision problems occurring among homogeneous RFID readers. In [3], Cha et al. proposed two ALOHA-based algorithms with a Tag Estimation Method (TEM) for speedup object identification in RFID systems.

The existing tag identification approaches can be classified into two main categories, the Aloha-based [8, 9, 14, 18, 22] anti-collision scheme and the tree-based scheme [2, 12, 15, 24]. RFID readers in the former scheme create a frame with a certain number of time slots, and then add the frame length into the inquiry message sending tags in its vicinity. Tags response the interrogation based on a random time slot. Because collisions may happen at the time slot when two or more tag response simultaneously, making those tags could not be recognized. Therefore, the readers have to send inquiries contiguously until all tags are identified. As a result, Aloha-based scheme might have long processing latency in identifying large-scale RFID systems [9]. In [18], Vogt et al. investigated how to recognize multiple RFID tags withinthe reader's interrogation ranges without knowing the numberof tags in advance by using framed Aloha. A similar research is also presented in [22] by Zhen et al. In [8], Klair et al. also presented a detailed analytical methodology and an in-depth qualitative energy consumption analysis of pure and slotted Aloha anti-collision protocols. Another anti-collision algorithm called enhanced dynamic framed slotted aloha (EDFSA) is proposed in [10]. EDFSA estimates the number of unread tags first and adjusts the number of responding tags or the frame size to give the optimal system efficiency.

In tree-based scheme, such as ABS [12], IBBT [4] and IQT [16], RFID readers split the set of tags into two subsets and labeled them by binary numbers. The reader repeats such process until each subset has only one tag. Thus the reader is able to identify all tags. The adaptive memoryless tag anti-collision protocol proposed by Myung et al. [11] is an extended technique based on the query tree protocol. Choi et al. also proposed the IBBT (Improved Bit-by-bit Binary-Tree) algorithm [4] in Ubiquitous ID system and evaluate the performance along three other old schemes.

The IQT protocol [16] is a similar work approach by exploiting specific prefix patterns in the tags to make the entire identification process. Recently, Zhou et al. [23]

consider the problem of slotted scheduled access of RFID tags in a multiple reader environment. They developed centralized algorithms in a slotted time model to read all the tags. With the fact of NP-hard, they also designed approximation algorithms for

the single channel and heuristic algorithms for the multiple channel cases.

Although tree based schemes have advantage of implementation simplicity and better response time compare with the Aloha based ones, they still have challenges in decreasing the identification latency. In this thesis, we present two enhanced tree based tag identification techniques aim to coordinate simultaneous communications in large-scale RFID systems, to speedup minimize tag identification latency and to increase the overall read rate and throughput.

8

CHAPTER 3

Preliminary

3.1 Anti-Collision Problem

Radio Frequency IDentification (RFID) systems proceed with the identification function through radio frequency, but usually, not only one tag is applied to the identification, that means RFID reader needs to identify multiple tags at the same time.

However, collision problems may happen in the process of identifying multiple tags.

At present, the binary splitting is commonly applied in solving the tag collision problems; nevertheless, when tag density is higher, search duplication may occur, and that wastes the time and frequency of identification. In Figure 1, we can obviously find that if the depth of a query tree becomes deeper, almost each layer has collision situation when reader starts the identification from the first layer of the query tree in order. Once collision happens, it’s a waste of identification time. Due to this, the frequency of collision can be set a limitation if the frequency is over the limitation, we can predict that tag density has a higher level, and then skip the identification at the layer to save unnecessary time of identification. This idea motivates the design of tag identification methods that has a limitation of collisions in each layer of the query tree.

00 00 01 01 10 10 11 11

Figure 1: An example of Query Tree Algorithm

3.2 Tree Based Tag-Collision Protocol

Radio Frequency IDentification (RFID) systems proceed with the identification function through radio frequency, but usually, not only one tag is applied to the identification, that means RFID reader needs to identify multiple tags at the same time.

However, collision problems may happen in the process of identifying multiple tags.

At present, the Binary Tree (BT) is commonly applied in solving the tag collision problems; nevertheless, when tag density is higher, search duplication may occur, and that wastes the time and frequency of identification. Figure 2 shows an example of identified four tags, using tree scan scheme. In the tree scan schem, tags and reader are communicated through RF. An RFID reader will send a prefix bit 1 or 0 to all Tags in its vicinity. Tags have the same prefix code will reflect its ID to the corresponding reader. There are three sorts of node in tree scan scheme, the identified node, the collision node and the idle node. The identified node means that only one Tag replied reader’s prefix query and transmit its ID code to the reader successfully. The collision node represents that more than one tag match the same prefix code and replies their ID

10

to the corresponding RFID reader. The Idle node shows that no tags with the same prefix code. Tree search protocol will cause a serious problem that is tag transmit ID code to reader repeatedly and increase the difficulty of identification so that the all system efficiency gradually decrease.

We can observe that if the depth of a query tree becomes deeper, almost each layer has collisions. Once collision happens, it’s a waste of identification time. As a result, the standard binary tree protocol [10] and the query tree protocol [9] have such drawbacks of long identification time.

Figure 2: An example of Binary Tree Algorithm

0 1

00 01 10 11

Tag A Tag B Tag C Tag D

3.3 Research Architecture

If the depth of Tree becomes deeper, almost each layer has collision situation when Reader starts the identification from the first layer of Tree in order. Once collision happens, it’s a waste of identification time. Due to this, the frequency of collision can be set a limitation if the frequency is over the limitation, we can predict that tag density creaches to a threshold representing high degree, and then jump the identification of the layer to save unnecessary time of identification.

The method of setting frequency limitation indeed improves the efficiency in high Tag density, but in low Tag density, inefficient cases may happen. The main problem is that each search begins from LSB, the left sub-tree, leading to that the right sub-tree can not start the identification until the last moment. Therefore, putting the concept of left and right scan and the idea of setting frequency limitation together can largely ameliorate the problem.

We have implemented the TJ and the WAS method along with other Tag identification approaches. Our preliminary experimental results show that the proposed techniques provide superior performance in both low and high density environments. It is shown that the TJ and WAS is effective in terms of increasing system throughput, efficiency and easy implementation.

12

CHAPTER 4

Threshold Jumping (TJ)

When readers stay in the environment of high tags density, collision may happen very frequently, and owing to this, a lot of query time will be wasted. In Threshold Jumping algorithm, a threshold of collisions is mainly used as the criterion of moving the identification to next level of a query tree. By means of the criterion to judge the level of tag density, if the density is high, the corresponding layer in the query tree will be jumped over. As shown in Figure 3, when the number of collisions is over the predefined threshold, the layer can be omitted to save the query time in the right sub-tree. Through the step, the number of broadcast could be significantly reduced.

Figure 3: Concept of Threshold Jumping

Taking an example, where tag id is equivalent to 4-bits represented by full binary tree as shown in Figure 4. In this example, the threshold is set as 2^I/M, where I is the layer of a query tree and M is a pre-defined constant. If the collisions of layer I is over 2^I/M, jumping to the next layer for a new search directly. Figure 4 shows an example of identifying 9 RFID tags using TJ with M=3. According to the definition above, when I=1, the identification will move to layer 2 if 2/3 collision is occurred in layer 1;

when I = 2, the identification will move to layer 3 if more than 4/3 collisions are occurred in layer2; similarly, when I = 3, the identification will move to layer 4 if more than 8/3 collisions are raised in layer 3, and so on. Table 1 summarizes the identification of the TJ for 9 tags with totally 16 inquiries. One thing worthy to mention is that although 16 inquiries is more than the number of leaves in a query tree with depth=4, but to scan all leaves using full tags’ bits is not a feasible approach.

This is because for RFID tags having more than 16 bits id, the process of scanning leave will lead significant large amount of inquiries, i.e., 65536. Therefore, the query tree based mechanism with prefix matching is usually considered as the feasible solution in tag anti-collision problem.

Recall the example described above, it’s clear that there are a lot of unnecessary broadcast in the method of query tree with binary splitting. For the TJ technique, when collision exceeds a certain proportion, it reveals that tag density is too high so there is no need to continuously launch broadcast to current layer to save identification time. Therefore, in the Threshold Jumping method, the threshold is taken to count the collision numbers. Once collision is over the proportion, there is no need to continue the query to the current layer, instead, jumping to the next layer. Based on the reduction of broadcast number, the efficiency of identification will speed.

14

Figure 4: Example of Threshold Jumping

Table 1: QT and TJ Identification of two different schemes

Query Tree Threshold Jumping

Step.

Broadcast Identification Status Identification Tag Broadcast Identification Status Identification Tag

1 0 Collision 0 Collision

2 1 Collision 00 Collision

3 00 Collision 01 Collision

4 01 Collision 000 No Response

5 10 Identification 1001 001 Collision

6 11 Collision 010 Identification 0101

7 000 No Response 011 Identification 0110

8 001 Collision 100 Identification 1001

9 010 Identification 0101 101 No Response

10 011 Identification 0110 110 Collision

11 110 Collision 0010 Identification 0010

12 111 Collision 0011 Identification 0011

13 0010 Identification 0010 1100 Identification 1100

14 0011 Identification 0011 1101 Identification 1101

15 1100 Identification 1100 1110 Identification 1110

16 1101 Identification 1101 1111 Identification 1111

17 1110 Identification 1110

18 1111 Identification 1111

16

CHAPTER 5

Warp-Around Scan (WAS)

The method of setting frequency limitation in TJ (Threshold Jumping) is adopted to improve identification efficiency when tag density is high. While searching is always from MSB (Most Significant Bit), the left sub-tree, it will lead to the identification of right sub-tree starting in the last moment, and it may waste some identification time in the process. The main reason is that if threshold is fulfilled in TJ (Threshold Jumping), right sub-tree will be skipped over so that inefficiency may happen if tag density is low or meets the situation of unbalanced tree. The problem can be solved through the Warp-Around concept, the so-called “WAS (Warp-Around Scan)”. In WAS, the identification of tags in the right sub-tree could be advanced by reversing the direction of scan in each layer alternatively. Figure 5 demonstrates the concept of such mechanism. In other words, the level ordered scan starts from MSB to LSB in all “odd” levels and starts from LSB to MSB in all “even” levels. Having the prediction in WAS method, the identification numbers can be decreased due to the earlier finding of the positions of no response.

Figure 5: Concept of Warp-Around Scan

Taking the same example that illustrated in section 4.1. Figure 6 gives the identification flow of the WAS method. It’s clear that the efficiency of identification is enhanced by means of the addition of WAS (Warp-Around Scan) in TJ (Threshold Jumping). The reason is that WAS (Warp-Around Scan) has identified the tags of right sub-tree and got the exact information of no response and collisions. The advanced information can diminish identification numbers. It concludes that WAS (Warp-Around Scan) can speed identification efficiency, especially in unbalanced tree.

Figure 6 : Example of Warp-Around Scan

18

Table 2: Tag Identification of three different schemes

Query Tree Threshold Jumping Warp-Around Scan

Step.

Broadcast Status

Identified Tag

Broadcast Status

Identified Tag

Broadcast Status

Identified Tag

1 0 Collision 0 Collision 0 Collision

2 1 Collision 00 Collision 11 Collision

3 00 Collision 01 Collision 10 No Response

4 01 Collision 000 Identified 0001 01 Collision

5 10 No Response 001 Identified 0010 000 Identified 0001

6 11 Collision 010 Identified 0100 001 Identified 0010

7 000 Identified 0001 011 Identified 0111 010 Identified 0100

8 001 Identified 0010 100 No Response 011 Identified 0111

9 010 Identified 0100 101 No Response 110 Identified 1100

10 011 Identified 0111 110 Identified 1100 111 Identified 1111

11 110 Identified 1100 111 Identified 1111

12 111 Identified 1111

CHAPTER 6

Performance Evaluation and Results

In this chapter, we first introduce the simulator, a TJ(Threshold Jumping) and WAS(Warp-Around Scan) for performance comparison. Then, we will show the simulation results.

6.1 Simulator and Comparison Metrics

To evaluate performance of the proposed optimization technique, we have implemented a RFID identification system to simulate circumstances with different characteristics. It is shown that the TJ and WAS is effective in terms of increasing system throughput, efficiency and easy implementation. The proposed QT method was compared along with the TJ and the composite algorithm, TJ+WAS.

20

6.2 Experiment Results

To evaluate the performance of our proposed techniques, we have implemented the TJ and the WAS schemes along with the query tree protocol (QT).

Figure 7 compares the number of inquiries to identify different number of RFID tags. Because the tag id is set 8 bits, density = 10% means that there are 2⁸ 10%

= 26 tags and density = 20% means that there are 2⁸ 20% = 51 tags, and so on.

All tags are randomly generated in uniform distribution. This is so called balance tree, as entitled in the figure. In this test, the threshold is set 1/3. We observe that the TJ and WAS methods do not improve the inquiry messages very much in low density network. When density increases, both of the proposed methods present significant improvements to the traditional query tree protocol.

Balance Tree (8-Bits)

0 100 200 300 400 500 600

10 20 30 40 50 60 70 80 90 100

Density(%)

Read Inquire

QT TJ WAS

(a)

B a la nc e Tre e (12-B it s)

1000 2000 3000 4000 5000 6000 7000

20 30 40 50 60 70 80

Density(%)

Read Inquire

QT TJ WAS

(b)

Figure 7: Performance comparison of the TJ, WAS and QT with 8 bits(a) and 12bits(b) RFID

tag.

Figure 8 uses the same configuration as that of Figure 7 but increases the length of RFID tag. Because of the long simulation time for high density network in 16 bits length, we only report the results from density = 20% to density = 80%.

Namely, the number of tags in this test is set from 2¹⁶20% ~ 2¹⁶80%, i.e., 13107

~ 52428. This simulation has similar observations as those of Figure 7.

Balance Tree(16-Bits)

20000 40000 60000 80000 100000

20 30 40 50 60 70 80

Density(%)

Read Inquire

QT TJ WAS

Figure 8: Performance comparison of the TJ, WAS and QT with 16 bits RFID tag.

As the TJ and WAS define a collision threshold to limit unnecessary inquiries.

Therefore, the value of the threshold should be critical to the process of tag identification through a binary tree. Figure 9 examines the effects of different threshold values to the performance of the two schemes. Figure 9(a) is the results for density = 80% and Figure 9(b) is the results for density = 50%. For high density network, i.e., Figure 9(a), we observed that lower threshold resulting less inquiry cost. Both of the TJ and WAS could outperform the traditional query tree protocol (QT). For low density RFID network, it is obvious that TJ and WAS will reach a lower bound of the number of inquiry when threshold is less than 1/5.

22

Balance Tree (8-Bits), density = 80%

250 270 290 310 330 350 370 390 410 430

3 4 5 6 7 8

Threshold Ratio(1/TR)

Read Inquire

QT TJ WAS

(a)

Balance Tree (8-Bits), density = 50%

200 210 220 230 240 250 260 270 280 290 300

3 4 5 6 7 8

Threshold Ratio(1/TR)

Read Inquire

QT TJ WAS

(b)

Figure 9: Performance comparison of the TJ, WAS and QT with different threshold ratio (a) results of density = 80% (b) results of density = 50%

The last simulation is conducted with different distribution of tags’ id. The term Tree Balance Factor (TBF) defined in Figure 10 is used to indicate the percentage of tags distributed in left-subtree and right-subtree of a complete query tree. TBF = 10% means that number of tags in left-subtree and right-subtree are with the ration 1:9; similarly, TBF = 20% represents the number of tags in left-subtree and right-subtree are with the ration 2:8. Since the distribution of tags in a query tree is not a uniform distribution, this is so called imbalance tree, as

titled in Figure 10. The density is set 80% in this test. From the experimental results, the WAS scheme outperforms the TJ and the query tree methods. For imbalanced instances, the TJ performs worse than the query tree (QT) protocol.

This is because the TJ method does not scan righ-subtree in most of levels in a query tree. As a result, there is no information of “collision” or “no response” in upper levels of the right-subtree. Consequently, when the identification goes to lower levels, it has to match more and more tags’ id. This situation usually happened in imbalanced tree. Such phenomenon matches the results shown in Figure 10. For the cases of TBF=40% and 50%, the TJ scheme also presents noticeable improvement.

Imbalanced Tree (8-Bits)

200 250 300 350

10 20 30 40 50

Tree Balance Factor (%)

Read Inquire

QT TJ WAS

Figure 10: Performance comparison of the TJ, WAS and QT in imbalanced tree.

24

CHAPTER 7

Conclusions and Future work

With the emergence of wireless RFID technologies, identifying high density RFID tags is a crucial task in developing large-scale RFID systems. In this thesis, we have presented two enhanced tree-based tag identification techniques for minimizing tag identification cost. By setting a collision bound, the prefix matching will be jumped to lower level of the query tree in order to alleviate the collision problems. Together with the warp-around scan method, the efficiency of tag identification can be significantly improved. To evaluate the performance of proposed techniques, we have implemented the TJ and the WAS techniques along with the query tree protocol (QT).

We have shown the effectiveness of setting collision threshold in scanning a query tree.

The experimental results show that the proposed techniques provide considerable improvements on the latency of tag identification. It is also shown that the TJ and WAS are effective in terms of increasing system throughput, efficiency and easy

implementation. There is also room for improvement of the identification technique in order to eliminate runtime peaks. It remains challenging, however, to find an optimal approach that would take environmental effects into account in order to adapt to them.

We hope that the work done so far helps implementing pervasive computing environments that employ RFID systems.

Reference

[1] Shailesh M. Birari and Sridhar Iyer, “Mitigating the Reader Collision Problem in RFID Networks with mobile readers”, International Conference on Networks (ICON), 2005.

[2] J. I. Capetanakis, “Tree algorithms for packet broadcast channels,” IEEE Transactions on Information Theory, vol. 25, pp. 505-515, Sept. 1979.

[3] J. R. Cha and J. H. Kim, “Dynamic Framed Slotted ALOHA Algorithm using Fast Tag Estimation method for RFID System,” in proceedings CCNC’06, Las Vegas, USA, Jan. 8-10, 2006.

[4] H. S. Choi, J. R. Cha and J. H. Kim, “Improved Bit-by-bit Binary Tree Algorithm in Ubiquitous ID System,” in proceedings of the IEEE PCM 2004, Tokyo, Japan, Nov. 29-Dec. 3, pp. 696-703, 2004.

[5] D. W. Engels and S. E. Sarma, “The reader collision problem,” in proceedings of the IEEE International Conference on Systems, Man and Cybernetics, 2002.

[6] S. Jain and S. R. Das, “Collision avoidance in multi-reader RFID networks”, in proceedings of the WiNTECH 2006, Los Angeles, California, 2006.

[7] Joongheon Kim, Sunhyoung Kim, Dongshin Kim, Wonjun Lee and Eunkyo Kim,

“Low-Energy Localized Clustering: An Adaptive Cluster Radius Configuration Scheme for Topology Control in Wireless Sensor Networks,” proceedings of the IEEE Vehicular Technology Conference (VTC), pp. 2546-2550, 2005.

[8] D. K. Klair, K.-W. Chin and R. Raad, “An investigation into the energy efficiency of pure and slotted aloha based RFID anti-collision protocols,” in proceedings of the IEEE WoWMoM’07, June 18-21, Helsinki, Finland, 2007.

[9] C. Law, K. Lee and K.-Y. Siu, “Efficient Memoryless Protocol for Tag Identification.” in proceedings of the International Workshop on Discrete Algorithms and Methods for Mobile Computing and Communications, August 2000.

[10] S. Lee, S.D. Joo, and C.W. Lee, “An enhanced dynamic framed slotted aloha algorithm for RFID tag identification,” in ACM Mobiquitous, July 2005.

[11] J. Myung and W. Lee, “An adaptive memoryless tag anti-collision protocol for RFID networks,” IEEE INFOCOM’05, Poster Session, Mar. 2005.

[12] J. Myung, W. Lee and J. Srivastava, “Adaptive binary splitting for efficient RFID

26

tag anti-collision,” IEEE Communication Letter, vol. 10, no. 3, pp. 144-146, Mar.

2006.

[13] Vinod Namboodiri and Lixin Gao, ”Energy-Aware Tag Anti-Collision Protocols for RFID Systems”, in proceedings of the Fifth IEEE International Conference on Pervasive Computing and Communications, pp. 23-36, 2007.

[14] L.G.Roberts, ”Aloha Packet System with and without Slots and Capture,” ACM SIGCOMM Computer Communication Review, vol. 5, pp. 28-42, 1975.

[15] Jiho Ryu, Hojin Lee, Yongho Seok, Taekyoung Kwon and Yanghee Choi, “A Hybrid Query Tree Protocol for Tag Collision Arbitration in RFID systems”, IEEE International Conference on Communications (ICC-07), 24-28 June 2007 pp. 5981 – 5986, 2007.

[16] A. Sahoo, S. Iyer and N. Bhandari, “Improving RFID System to Read Tags Efficiently”, KRSIT Technical Report, IIT Bombay, June 2006.

[17] Joao Luıs Sobrinho, Roland de Haan and Jose Manuel Brazio, “Why RTS-CTS is not your ideal wireless LAN multiple access protocol,” in proceedings of the IEEE Wireless Communications and Networking Conference, pp. 81-87, 2005.

[18] Harald Vogt, ”Efficient Object Identification with Passive RFID Tags”, in proceedings of the 1st International Conference on Pervasive Computing, LNCS 2414, pp. 98-113, 2002.

[19] Stephen Weis, Sanjay E. Sarma, Ronald L. Rivest and Daniel W. Engels, ”Security and privacy aspects of low-cost radio frequency identification systems,” Security in Pervasive Computing, LNCS 2803, pp. 201-212, 2004.

[20] James Waldrop, Daniel W. Engels and Sanjay E. Sarma, “Colorwave: an anticollision algorithm for the reader collision problem,” proceedings of the IEEE International Conference on Communications, pp. 1206-1210, 2003.

[21] Honghai Zhang and Jennifer C. Hou, “Maintaining sensing coverage and connectivity in large sensor networks,” Wireless Ad Hoc and Sensor Networks:

Vol. 1, No. 1-2, pp. 89-124, March 2005.

[22] B. Zhen, M. Kobayashi and M. Shimizui, ”Framed aloha for multiple RFID objects Identification,” IEICE Trans. on Comm., E88-B(3):991–999, 2005.

[23] Zongheng Zhou, Himanshu Gupta, Samir R. Das and Xianjin Zhu. ”Slotted Scheduled Tag Access in Multi-Reader RFID Systems”, in proceedings of the IEEE International Conference on Networks Protocols (ICNP), pp. 61-70, 2007.

[24] F. Zhou, C. Chen, D. Jin, C. Huang and H. Min, ”Evaluating and Optimizing Power Consumption of Anti-Collision Protocols for Applications in RFID systems,” in proceedings of the International Symposium on Low Power Electronics and Design, Newport Beach, California, USA, August 2004.