888 EEE TRANSACTIONS ON COMPUTERS,VOL.65,NO.3.MARCH 2016 Focus and Shoot:Exploring Auto-Focus in RFID Tag ldentification Towards a Specified Area Yafeng Yin,Student Member,IEEE,Lei Xie,Member,IEEE,Jie Wu,Fellow,IEEE,and Sanglu Lu,Member,IEEE Abstract-With the rapid proliferation of RFID technologies,RFID has been introduced into applications such as inventory and sampling inspection.Conventionally,in RFID systems,the reader usually identifies all the RFID tags in the interrogation region with the maximum power.However,some applications may only need to identify the tags in a specified area,which is usually smaller than the reader's default interrogation region.An example could be identifying the tags in a box,while ignoring the tags out of the box.In this paper,we respectively present two solutions to identify the tags in the specified area.The principle of the solutions can be compared to the picture-taking process of an auto-focus camera,which firstly focuses on the target automatically and then takes the picture. Similarly,our solutions first focus on the specified area and then shoot the tags.The design of the two solutions is based on the extensive empirical study on RFID tags.Realistic experiment results show that our solutions can reduce the execution time by 44 percent compared to the baseline solution,which identifies the tags with maximum power.Furthermore,we improve the proposed solutions to make them work well in more complex environments. Index Terms-RFID,tag identification,auto-focus,specified area,experimental study,algorithm design 1 INTRODUCTION ECENTLY,RFID tags have been widely used in various Moreover,identifying the tags out of the area is rather time- Lapplications,such as inventory control,sampling consuming.Due to the large number of tags,time-efficiency inspection,and supply chain management.Each RFID tag is very important.Therefore,it is essential to identify the has a unique ID,thus the reader can recognize the object by tags in the specified area efficiently without moving the tags. identifying its attached tag.Many existing research works on Fortunately,we note that tag identification in the speci- RFID have concentrated on tag identification [1],[21,[3],[4],fied area can be compared to the picture-taking process in [51,[61,[71,[81,aiming to identify a large number of tags as an auto-focus camera.The camera automatically focuses on quickly as possible.While detecting the missing tags and the object before shooting,aiming to lock the target object searching a particular subset of tags only concentrate on the while ignoring the others.In this paper,we propose the part of tags.However,all the literature do not research the photography based identification method,which works in a problem of tag identification in a specified area,which is similar way.It first focuses on the specified area by adjust- rather important in many applications,e.g.,inventory and ing the antenna's angle and the reader's power,and then sampling inspection in warehouse management.Taking the identifies the tags in the area.When the reader's interro- inventory for example,in order to have a knowledge of the gation region is just enough to cover the specified area,it tags in the target area,we may only need to identify the tags achieves the best performance.To solve this problem,we in some specified boxes while ignoring the others,i.e.,identi- respectively propose two solutions working in the realistic fying the tags in the specified area.In regard to a sampling environments.Both solutions conform to the EPC-C1G2 inspection,it also requires focusing on the tags in the current standards. area,while ignoring the others.Sometimes,it is difficult to However,efficiently identifying the tags in realistic move the objects out for tag identification,especially for environments is difficult.There are a few research works the objects obstructed by obstacles.A traditional solution is concentrating on this problem.The reading performance to identify the tags with the maximum power (MaxPw).in the realistic experiments is still unknown,especially for However,this solution will identify the tags out of the area,a large number of tags.Hence,we conduct a series of leading to lower accuracy of the identification process. measurements over RFID tags in realistic settings,inves- ting the factors which affect the reading performance.For- tunately,we have a few important new findings.For .Y.Yin,L.Xie,and S.Lu are with the State Key Laboratory for Novel Soft- example,we find that the tag density affects the effective ware Technology,Nanjing University,Nanjing 210023,China. scanning range,i.e.,the larger the tag density,the smaller E-mail:yyf@dislab.nju.edu.cn,(lxie,sanglu@nju.edu.cn. I.Wu is with the Department of Computer and Information Sciences the effective scanning range.The findings are crucial for Temple UIniversity,Philadelphia,PA 19122.E-mail:jiewu@temple.edu. improving the performance of our solutions.They indicate Manuscript received 28 June 2014;revised 26 Feb.2015;accepted 22 Apr. that the reader should adaptively adjust its interrogation 2015.Date of publication 19 May 2015;date of current version 10 Feb.2016. region,considering the actual situation.We propose the Recommended for acceptance by G.Min. two solutions based on the extensive experimental study, For information on obtaining reprints of this article,please send e-mail to: in order to make the solutions work well in the realistic reprints@ieeeorg,and reference the Digital Object Identifier below. Digital Object Identifier no.10.1109/TC.2015.2435749 environments
Focus and Shoot: Exploring Auto-Focus in RFID Tag Identification Towards a Specified Area Yafeng Yin, Student Member, IEEE, Lei Xie, Member, IEEE, Jie Wu, Fellow, IEEE, and Sanglu Lu, Member, IEEE Abstract—With the rapid proliferation of RFID technologies, RFID has been introduced into applications such as inventory and sampling inspection. Conventionally, in RFID systems, the reader usually identifies all the RFID tags in the interrogation region with the maximum power. However, some applications may only need to identify the tags in a specified area, which is usually smaller than the reader’s default interrogation region. An example could be identifying the tags in a box, while ignoring the tags out of the box. In this paper, we respectively present two solutions to identify the tags in the specified area. The principle of the solutions can be compared to the picture-taking process of an auto-focus camera, which firstly focuses on the target automatically and then takes the picture. Similarly, our solutions first focus on the specified area and then shoot the tags. The design of the two solutions is based on the extensive empirical study on RFID tags. Realistic experiment results show that our solutions can reduce the execution time by 44 percent compared to the baseline solution, which identifies the tags with maximum power. Furthermore, we improve the proposed solutions to make them work well in more complex environments. Index Terms—RFID, tag identification, auto-focus, specified area, experimental study, algorithm design Ç 1 INTRODUCTION RECENTLY, RFID tags have been widely used in various applications, such as inventory control, sampling inspection, and supply chain management. Each RFID tag has a unique ID, thus the reader can recognize the object by identifying its attached tag. Many existing research works on RFID have concentrated on tag identification [1], [2], [3], [4], [5], [6], [7], [8], aiming to identify a large number of tags as quickly as possible. While detecting the missing tags and searching a particular subset of tags only concentrate on the part of tags. However, all the literature do not research the problem of tag identification in a specified area, which is rather important in many applications, e.g., inventory and sampling inspection in warehouse management. Taking the inventory for example, in order to have a knowledge of the tags in the target area, we may only need to identify the tags in some specified boxes while ignoring the others, i.e., identifying the tags in the specified area. In regard to a sampling inspection, it also requires focusing on the tags in the current area, while ignoring the others. Sometimes, it is difficult to move the objects out for tag identification, especially for the objects obstructed by obstacles. A traditional solution is to identify the tags with the maximum power (MaxPw). However, this solution will identify the tags out of the area, leading to lower accuracy of the identification process. Moreover, identifying the tags out of the area is rather timeconsuming. Due to the large number of tags, time-efficiency is very important. Therefore, it is essential to identify the tags in the specified area efficiently without moving the tags. Fortunately, we note that tag identification in the speci- fied area can be compared to the picture-taking process in an auto-focus camera. The camera automatically focuses on the object before shooting, aiming to lock the target object while ignoring the others. In this paper, we propose the photography based identification method, which works in a similar way. It first focuses on the specified area by adjusting the antenna’s angle and the reader’s power, and then identifies the tags in the area. When the reader’s interrogation region is just enough to cover the specified area, it achieves the best performance. To solve this problem, we respectively propose two solutions working in the realistic environments. Both solutions conform to the EPC-C1G2 standards. However, efficiently identifying the tags in realistic environments is difficult. There are a few research works concentrating on this problem. The reading performance in the realistic experiments is still unknown, especially for a large number of tags. Hence, we conduct a series of measurements over RFID tags in realistic settings, investing the factors which affect the reading performance. Fortunately, we have a few important new findings. For example, we find that the tag density affects the effective scanning range, i.e., the larger the tag density, the smaller the effective scanning range. The findings are crucial for improving the performance of our solutions. They indicate that the reader should adaptively adjust its interrogation region, considering the actual situation. We propose the two solutions based on the extensive experimental study, in order to make the solutions work well in the realistic environments. Y. Yin, L. Xie, and S. Lu are with the State Key Laboratory for Novel Software Technology, Nanjing University, Nanjing 210023, China. E-mail: yyf@dislab.nju.edu.cn, {lxie, sanglu}@nju.edu.cn. J. Wu is with the Department of Computer and Information Sciences, Temple University, Philadelphia, PA 19122. E-mail: jiewu@temple.edu. Manuscript received 28 June 2014; revised 26 Feb. 2015; accepted 22 Apr. 2015. Date of publication 19 May 2015; date of current version 10 Feb. 2016. Recommended for acceptance by G. Min. For information on obtaining reprints of this article, please send e-mail to: reprints@ieee.org, and reference the Digital Object Identifier below. Digital Object Identifier no. 10.1109/TC.2015.2435749 888 IEEE TRANSACTIONS ON COMPUTERS, VOL. 65, NO. 3, MARCH 2016 0018-9340 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information
YIN ETAL:FOCUS AND SHOOT:EXPLORING AUTO-FOCUS IN RFID TAG IDENTIFICATION TOWARDS A SPECIFIED AREA 889 We make the following contributions in this paper. We conducted extensive experiments on the com- Specified area modity RFID system in the realistic environments, and investigated the factors affecting the reading performance. ● To the best of our knowledge,this is the first work investigating efficient tag identification in the speci- ☐Tag Interrogation fied area,which is essential for many applications, region such as inventory and sampling inspection.We pro- pose the photography-based identification method, Fig.1.Identify the tags in the specified area. which works in a way similar to a camera.Besides, we respectively propose two solutions to solve realistic environment.We aim to identify as many tags in the problem,and reduce the execution time by the specified area as possible,while minimizing the execu- 44 percent compared to the baseline solution.More- tion time over,we improve the proposed solutions to make them work well in the more complex environments. 3 PROBLEM FORMULATION Our solutions work in the realistic environments with the commercial RFID system,which do not 3.1 System Model require any changes to the protocols and the low- Each object is attached with an RFID tag,which has a level parameters in the system.Both of the two solu- unique ID.In this paper,we use the terms 'object'and 'tag' tions conform to the EPC-CIG2 standards. interchangeably.The number of tags and the distribution of tag IDs are unknown.The reader is statically deployed and 2 RELATED WORK configured with an antenna.The antenna is associated with an interrogation region,within which the reader can iden- Previous research on RFID concentrates on designing anti- tify the tags.The antenna is deployed in a fixed position, collision ID-collection protocols to collect all the tag IDs. e.g,it can be fixed on the wall or the ceiling.It cannot The existing anti-collision protocols can be categorized as change its distance to the objects,but it can be rotated,just being either tree-based [1],[2],[3],[4]or ALOHA-based [5], like the electric fan fixed on the ceiling.This can be a typical [6],[7],[8].Tree-based protocols resolve collisions by mut- setting for the application scenarios like inventory and sam- ing subsets of tags that are involved in a collision.ALOHA- pling inspection in warehouse management.By rotating the based protocols assign a distinct transmission time slot to antenna,we can identify more tags with fewer readers. each tag,and sequentially identify the tags. Besides,the reader can control the interrogation region by Instead of identifying all the RFID tags,protocols for iden- adjusting the power. tifying missing tags monitor a set of tags and detect the miss- The objects are packaged in boxes.The boxes out of the ing-tag event [9],[10],[11],[12],[13],[14].Unknown tag specified area S have reasonable distances between the identification aims to identify the tags (e.g,the new added boxes in S,which means that the area s has a clear bound- tags or the misplaced tags),which appear to be unknown by ary.As shown in Fig.1,the tags in S are called target tags, the reader(s)currently covering them [15],[16].Unknown- while the tags outside s are called interference tags.The target information collection needs to find out the target tags objective of this paper is to identify as many target tags as and read the information from them [17].Fast tag searching possible,while minimizing the execution time. aims to quickly search the particular tag IDs [181.Besides, the polling-based protocols are proposed to collect the infor- mation from RFID tags in a time/energy-efficient approach 3.2 Performance Metrics [19],[20].Rather than identifying the tags,the RFID cardinal- We consider the three performance metrics for evaluating ity estimation protocols count the number of distinct tags the solution's efficiency. [21],[221,[23],[24],which can serve as useful inputs to 1)Coverage ratio p constraint.Let S be the set of tags in S improve the efficiency of tag identification [25],[261. (target tags),s=S].Let M be the set of the tags that are The above research works mainly consider the situation identified in S,m=|Ml.Obviously,M≤S and m≤s. without considering issues like path loss,energy absorp- Then,p=",0a.a is related to the specific scenario:when the environ- [27]examine the performance of the C1G2 RFID system in a ment and the deployment of the RFID system are fixed,the realistic setting.Aroor and Deavours [28]use a simple, value of a can be determined. empirical,experimental approach to identify the state of the 2)Execution time T.It represents the duration of the technical capability of passive UHF RFID systems.Xie et al.whole process.It shows the time efficiency,which is rather [29]conduct an extensive experimental study on the mobile important,especially for the identification of a large number RFID system,and build a model to depict how various of tags.The smaller the time T,the better the time efficiency. parameters affect the reading performance. 3)Misreading ratio A.Let U be the set of tags out of S Different from the related work,our research focuses on (interference tags)that are identified,u=Ul,Uns=0. identifying the tags in the specified area while ignoring the Then,=The smaller the value of the lower the mis- tags outside the area.Besides,our solutions work in the reading ratio
We make the following contributions in this paper. We conducted extensive experiments on the commodity RFID system in the realistic environments, and investigated the factors affecting the reading performance. To the best of our knowledge, this is the first work investigating efficient tag identification in the speci- fied area, which is essential for many applications, such as inventory and sampling inspection. We propose the photography-based identification method, which works in a way similar to a camera. Besides, we respectively propose two solutions to solve the problem, and reduce the execution time by 44 percent compared to the baseline solution. Moreover, we improve the proposed solutions to make them work well in the more complex environments. Our solutions work in the realistic environments with the commercial RFID system, which do not require any changes to the protocols and the lowlevel parameters in the system. Both of the two solutions conform to the EPC-C1G2 standards. 2 RELATED WORK Previous research on RFID concentrates on designing anticollision ID-collection protocols to collect all the tag IDs. The existing anti-collision protocols can be categorized as being either tree-based [1], [2], [3], [4] or ALOHA-based [5], [6], [7], [8]. Tree-based protocols resolve collisions by muting subsets of tags that are involved in a collision. ALOHAbased protocols assign a distinct transmission time slot to each tag, and sequentially identify the tags. Instead of identifying all the RFID tags, protocols for identifying missing tags monitor a set of tags and detect the missing-tag event [9], [10], [11], [12], [13], [14]. Unknown tag identification aims to identify the tags (e.g., the new added tags or the misplaced tags), which appear to be unknown by the reader(s) currently covering them [15], [16]. Unknowntarget information collection needs to find out the target tags and read the information from them [17]. Fast tag searching aims to quickly search the particular tag IDs [18]. Besides, the polling-based protocols are proposed to collect the information from RFID tags in a time/energy-efficient approach [19], [20]. Rather than identifying the tags, the RFID cardinality estimation protocols count the number of distinct tags [21], [22], [23], [24], which can serve as useful inputs to improve the efficiency of tag identification [25], [26]. The above research works mainly consider the situation without considering issues like path loss, energy absorption, multipath effect, etc. While considering the impact of the physical layer’s unreliability, Buettner and Wetherall [27] examine the performance of the C1G2 RFID system in a realistic setting. Aroor and Deavours [28] use a simple, empirical, experimental approach to identify the state of the technical capability of passive UHF RFID systems. Xie et al. [29] conduct an extensive experimental study on the mobile RFID system, and build a model to depict how various parameters affect the reading performance. Different from the related work, our research focuses on identifying the tags in the specified area while ignoring the tags outside the area. Besides, our solutions work in the realistic environment. We aim to identify as many tags in the specified area as possible, while minimizing the execution time. 3 PROBLEM FORMULATION 3.1 System Model Each object is attached with an RFID tag, which has a unique ID. In this paper, we use the terms ‘object’ and ‘tag’ interchangeably. The number of tags and the distribution of tag IDs are unknown. The reader is statically deployed and configured with an antenna. The antenna is associated with an interrogation region, within which the reader can identify the tags. The antenna is deployed in a fixed position, e.g., it can be fixed on the wall or the ceiling. It cannot change its distance to the objects, but it can be rotated, just like the electric fan fixed on the ceiling. This can be a typical setting for the application scenarios like inventory and sampling inspection in warehouse management. By rotating the antenna, we can identify more tags with fewer readers. Besides, the reader can control the interrogation region by adjusting the power. The objects are packaged in boxes. The boxes out of the specified area S have reasonable distances between the boxes in S, which means that the area S has a clear boundary. As shown in Fig. 1, the tags in S are called target tags, while the tags outside S are called interference tags. The objective of this paper is to identify as many target tags as possible, while minimizing the execution time. 3.2 Performance Metrics We consider the three performance metrics for evaluating the solution’s efficiency. 1) Coverage ratio r constraint. Let S be the set of tags in S (target tags), s ¼ jSj. Let M be the set of the tags that are identified in S, m ¼ jMj. Obviously, M S and m s. Then, r ¼ m s , 0 r 1. The larger the value of r, the better the coverage ratio. Given a constant a, r should satisfy r a. a is related to the specific scenario: when the environment and the deployment of the RFID system are fixed, the value of a can be determined. 2) Execution time T. It represents the duration of the whole process. It shows the time efficiency, which is rather important, especially for the identification of a large number of tags. The smaller the time T, the better the time efficiency. 3) Misreading ratio . Let U be the set of tags out of S (interference tags) that are identified, u ¼ jUj, U \ S ¼ ;. Then, ¼ u uþm. The smaller the value of , the lower the misreading ratio. Fig. 1. Identify the tags in the specified area. YIN ET AL.: FOCUS AND SHOOT: EXPLORING AUTO-FOCUS IN RFID TAG IDENTIFICATION TOWARDS A SPECIFIED AREA 889
890 EEE TRANSACTIONS ON COMPUTERS,VOL.65,NO.3.MARCH 2016 the Alien-9640 tag.Each tag is attached into a distinct book. The antenna and the books are placed on the tablet chairs with a height of 0.5 m.Unless otherwise specified,we make the antenna face towards the center of the objects,set the reader's power P=30.7dBm,and the distance between the tags and the antenna d=1m.For each experiment,the Antenna reader scans the tags for 50 cycles. (a)Rotating the antenna (b)Rotating the tag Fig.2.Identify the tag at different angles. 4.1 Identifying the Tag at Different Angles As the angle between the radiation direction and the surface of the The objective of this paper is to minimize the execution antenna deceases,the reading performance usually deceases. time T,while the coverage ratio satisfies p>a.When p>a, Besides,the placement of the tag may affect the reading perfor- minimizing T means avoiding identifying the interference mance.As shown in Fig.2,we respectively rotate the tags,in order to reduce the identification time.There is no antenna and the tag to observe the minimal power Pmim constraint on A,which is related to T.However,for the needed to activate one tag.Firstly,we rotate the antenna same execution time,the lower the misreading ratio,the bet- while keeping the tag unchanged.We use 0,(see Fig.2a)to ter the performance of a solution. represent the angle between the antenna's radiation direc- tion and the antenna's surface,a,∈[0°,90l.Fig.3 a shows 4 OBSERVATIONS FROM THE REALISTIC that as decreases,P becomes larger.When the EXPERIMENTS antenna faces towards the tag(0,=90),it achieves the best reading performance.Secondly,we rotate the tag while In order to know the factors affecting the reading perfor- keeping the antenna unchanged.We use 6(see Fig.2b)to mance in real environments,we conduct the following represent the angle between the radiation direction and the experiments.We use the Alien-9900+reader and Alien-9611 tag's surface.We respectively rotate the tag along the x-axis, antenna.The reader's maximal power max Pe is 30.7 dBm y-axis,and z-axis,as shown in Fig.2b.Fig.3a shows when and its minimal power min P is 15.7 dBm.The RFID tag is the tag rotates along the x-axis,decreases,Pmin becomes 32 30 ■25.7dBm026.7dBm 6 5 ---凡a1年:间 E28 45 26 -d-Raae1gz改a 耍24 30 922 25 要20 20 818 1写 10 1530Ang0 75 90 16718720762636.7287307 05105283035 (a)Minimal power vs.angles (b)Distribution of identified (c)Identified tag IDs vs.powers (d)Coverage ratio vs.distances (Group 1) tags vs.powers 2 100 32 :157m 30 880 15 gy —Pcw0G1&7日m 1 60 一二 26 2 20 20 Paper Water FAng蓝o0 15.30 75 (e)Scanning range vs.tag densi- (f)Number of identified tags vs. (g)Identification ratio vs.materi- (h)Minimal power vs.angles ties tag cardinality als (Group 2) 32 32 32 30 0 30 E28 2 E28 26 26 号26 24 24 四22 中22 20 20 20 a18 18 16 14 15 75 90 15 Angle 3 45 75 90 1530Ang60 75 90 89Aa.A2.077 (i)Minimal power vs.angles (j)Minimal power vs.angles (k)Minimal power vs.angles (1)Identification ratio vs.mixed (Group 3) (Group 4) (Group 5) tags Fig.3.Observations from the realistic experiments
The objective of this paper is to minimize the execution time T, while the coverage ratio satisfies r a. When r a, minimizing T means avoiding identifying the interference tags, in order to reduce the identification time. There is no constraint on , which is related to T. However, for the same execution time, the lower the misreading ratio, the better the performance of a solution. 4 OBSERVATIONS FROM THE REALISTIC EXPERIMENTS In order to know the factors affecting the reading performance in real environments, we conduct the following experiments. We use the Alien-9900+ reader and Alien-9611 antenna. The reader’s maximal power max Pw is 30.7 dBm and its minimal power min Pw is 15.7 dBm. The RFID tag is the Alien-9640 tag. Each tag is attached into a distinct book. The antenna and the books are placed on the tablet chairs with a height of 0.5 m. Unless otherwise specified, we make the antenna face towards the center of the objects, set the reader’s power Pw ¼ 30:7 dBm, and the distance between the tags and the antenna d ¼ 1m. For each experiment, the reader scans the tags for 50 cycles. 4.1 Identifying the Tag at Different Angles As the angle between the radiation direction and the surface of the antenna deceases, the reading performance usually deceases. Besides, the placement of the tag may affect the reading performance. As shown in Fig. 2, we respectively rotate the antenna and the tag to observe the minimal power Pwmin needed to activate one tag. Firstly, we rotate the antenna while keeping the tag unchanged. We use ur (see Fig. 2a) to represent the angle between the antenna’s radiation direction and the antenna’s surface, ur 2 ½0; 90. Fig. 3a shows that as ur decreases, Pwmin becomes larger. When the antenna faces towards the tag (ur ¼ 90), it achieves the best reading performance. Secondly, we rotate the tag while keeping the antenna unchanged. We use ut (see Fig. 2b) to represent the angle between the radiation direction and the tag’s surface. We respectively rotate the tag along the x-axis, y-axis, and z-axis, as shown in Fig. 2b. Fig. 3a shows when the tag rotates along the x-axis, ut decreases, Pwmin becomes Fig. 2. Identify the tag at different angles. Fig. 3. Observations from the realistic experiments. 890 IEEE TRANSACTIONS ON COMPUTERS, VOL. 65, NO. 3, MARCH 2016
YIN ETAL:FOCUS AND SHOOT:EXPLORING AUTO-FOCUS IN RFID TAG IDENTIFICATION TOWARDS A SPECIFIED AREA 891 TABLE 1 Combinations of Different Readers,Antennas,and Tags Reader Antenna (Polarization mode) Tag(IC type;Antenna technology) Alien-9900+ Alien-9611(Circular) Alien-9640(Alien Higgs-3;Dipole) Alien-9650 Alien-9650(Circular) Alien-9654(Alien Higgs-3;Dipole) 3 Alien-9900+ Alien-9611(Circular) Impinj H47(Monza 4;Dual-differential) 4 Impinj R420 Impinj Threshold (Linear) Impinj H47(Monza 4;Dual-differential) Impinj R420 ImpinjThreshold (Linear) Alien-9640(Alien Higgs-3;Dipole) larger.When the tag rotates along the y-axis,0 keeps the specified area(tag density)is unknown,we can not unchanged,thus P keeps unchanged.When the tag calculate the interrogation region accurately.However,if rotates along the z-axis,although decreases,the tag is eas- we only want to identify a few tags (e.g.,for sampling), ily identified.Even though the tag holds the same angle we can choose an estimated power,because the tag cardi- the different placement of the tag may result in different nality has a little effect on the number of identified tags,as reading performances. shown in Fig.3f. It is best to identify the tags with 0,=90 and 6=90. However,we may have no access to the placement of each 4.5 Effect of the Materials of Objects tag,then we should try to make the antenna face towards the tags(0,=90)to improve the reading performance. The materials of the objects can affect the reading performance, especially for metal materials and water-containing materials.We respectively attach 60 tags to the books,plastic clips,iron 4.2 Adjusting the Reader's Power clips,and milk.The corresponding materials of the objects The larger the reader's power,the larger the interrogation region, are paper,plastic,metal,and water.Fig.3g shows the mate- but the new identified tags may not be located in the interrogation rials can greatly affect the reading performance.The read- region's boundary.However,if a tag can be identified with a low ing performance for the tags attached to the metal materials power,it must be identified with a larger power.We uniformly is not good,due to the reflection of the metal materials.The deploy 72 tags on the wall.The distance between two adja- reading performance for the tags attached to the water-con- cent tags is 20 cm,as shown in Fig.3b.The new identified taining materials is poor,due to the energy absorbtion in tags may not be in the interrogation region's boundary.We the water.In this case,we can use the functional tags tai- cannot distinguish a tag's position by only adjusting the lored to specific materials (such as anti-metal tags against power.However,Fig.3c shows that if a tag can be identified metal)to tackle this issue. with a low power,then it definitely can be identified by a larger power.Usually,the large power can increase the number of identified tags. 4.6 Different Types of Tags and Antennas The types of the tags and the antennas can affect the reading 4.3 Varying the Distance between the Tags performance.However,the tags and the antennas belonging to and the Antenna the same series may have the similar reading performance.As shown in Table 1,we use five different groups of RFID As the distance between the tags and the antenna increases,the systems to repeat the experiments in Section 4.1.The reading performance decreases.Besides,when the distance is fixed, experiment result in group 1 is shown in Fig.3a.While the maximum coverage ratio has an upper bound,whatever the the experiment results in group 2,3,4,5 are shown in reader's power is.We vary the distance d from 0.5 to 3.5 m. Figs.3h,3i,3j,and 3k,respectively.By comparing Fig.3a Fig.3d shows that as d increases,the identified tag cardinal- and Figs.3i,3j,and 3k,we can find that different types of ity decreases.When d is small (e.g.,d<1.5m),the reading antennas and tags will result in different reading perform- performance is relatively good.However,when the distance ances.This is mainly caused by the polarization mode of and the number of tags are fixed,the coverage ratio has an the antenna,the IC type and the antenna technology of upper bound.For example,when d=1.5m and n=55, the tag,and the matching way between the antenna the maximum coverage ratio is 78 percent.Fortunately, and the tag.While comparing Figs.3a and 3h,we can find some applications (e.g,sampling inspection)just need the that the two RFID systems have the similar reading per- coverage ratio to meet the constraint instead of achieving formance,because the components in group 1 and group 100 percent.However,when considering the high coverage 2 belong to the same series. ratio,the antenna should not be placed far away from the tags. In addition to this,we conduct another experiment by mixing different types of tags.The 60 tags are consisted of 4.4 Effect of the Tag Cardinality Aline-9640 tags,Alien-9650 tags and Impinj H47 tags. The tag cardinality can affect the effective interrogation region. Becuase Alien-9640 tags and Alien-9654 tags have the simi- However,it has a little effect on the number of identified tags. lar performance,we vary the number of Impinj H47 tags to We uniformly deploy the tags in a row with length 4 m observe the reading performance.Impinj H47 is the omni- and vary the number of tags (tag cardinality)as 20,40,60,directional tag,which is insensitive to the placement of tag 80.As shown in Fig.3e,given a fixed power (30.7 dBm),as When the proportion of H47 increases,the reading perfor- the tag cardinality increases,the effective interrogation mance becomes better,as shown in Fig.31.However,the region decreases.Therefore,when the tag cardinality in combinations of different types of tags have a little effect on
larger. When the tag rotates along the y-axis, ut keeps unchanged, thus Pwmin keeps unchanged. When the tag rotates along the z-axis, although ut decreases, the tag is easily identified. Even though the tag holds the same angle ut, the different placement of the tag may result in different reading performances. It is best to identify the tags with ur ¼ 90 and ut ¼ 90. However, we may have no access to the placement of each tag, then we should try to make the antenna face towards the tags (ur ¼ 90) to improve the reading performance. 4.2 Adjusting the Reader’s Power The larger the reader’s power, the larger the interrogation region, but the new identified tags may not be located in the interrogation region’s boundary. However, if a tag can be identified with a low power, it must be identified with a larger power. We uniformly deploy 72 tags on the wall. The distance between two adjacent tags is 20 cm, as shown in Fig. 3b. The new identified tags may not be in the interrogation region’s boundary. We cannot distinguish a tag’s position by only adjusting the power. However, Fig. 3c shows that if a tag can be identified with a low power, then it definitely can be identified by a larger power. Usually, the large power can increase the number of identified tags. 4.3 Varying the Distance between the Tags and the Antenna As the distance between the tags and the antenna increases, the reading performance decreases. Besides, when the distance is fixed, the maximum coverage ratio has an upper bound, whatever the reader’s power is. We vary the distance d from 0.5 to 3.5 m. Fig. 3d shows that as d increases, the identified tag cardinality decreases. When d is small (e.g., d 1:5 m), the reading performance is relatively good. However, when the distance and the number of tags are fixed, the coverage ratio has an upper bound. For example, when d ¼ 1:5 m and n ¼ 55, the maximum coverage ratio is 78 percent. Fortunately, some applications (e.g., sampling inspection) just need the coverage ratio to meet the constraint instead of achieving 100 percent. However, when considering the high coverage ratio, the antenna should not be placed far away from the tags. 4.4 Effect of the Tag Cardinality The tag cardinality can affect the effective interrogation region. However, it has a little effect on the number of identified tags. We uniformly deploy the tags in a row with length 4 m and vary the number of tags (tag cardinality) as 20, 40, 60, 80. As shown in Fig. 3e, given a fixed power (30:7 dBm), as the tag cardinality increases, the effective interrogation region decreases. Therefore, when the tag cardinality in the specified area (tag density) is unknown, we can not calculate the interrogation region accurately. However, if we only want to identify a few tags (e.g., for sampling), we can choose an estimated power, because the tag cardinality has a little effect on the number of identified tags, as shown in Fig. 3f. 4.5 Effect of the Materials of Objects The materials of the objects can affect the reading performance, especially for metal materials and water-containing materials. We respectively attach 60 tags to the books, plastic clips, iron clips, and milk. The corresponding materials of the objects are paper, plastic, metal, and water. Fig. 3g shows the materials can greatly affect the reading performance. The reading performance for the tags attached to the metal materials is not good, due to the reflection of the metal materials. The reading performance for the tags attached to the water-containing materials is poor, due to the energy absorbtion in the water. In this case, we can use the functional tags tailored to specific materials (such as anti-metal tags against metal) to tackle this issue. 4.6 Different Types of Tags and Antennas The types of the tags and the antennas can affect the reading performance. However, the tags and the antennas belonging to the same series may have the similar reading performance. As shown in Table 1, we use five different groups of RFID systems to repeat the experiments in Section 4.1. The experiment result in group 1 is shown in Fig. 3a. While the experiment results in group 2, 3, 4, 5 are shown in Figs. 3h, 3i, 3j, and 3k, respectively. By comparing Fig. 3a and Figs. 3i, 3j, and 3k, we can find that different types of antennas and tags will result in different reading performances. This is mainly caused by the polarization mode of the antenna, the IC type and the antenna technology of the tag, and the matching way between the antenna and the tag. While comparing Figs. 3a and 3h, we can find that the two RFID systems have the similar reading performance, because the components in group 1 and group 2 belong to the same series. In addition to this, we conduct another experiment by mixing different types of tags. The 60 tags are consisted of Aline-9640 tags, Alien-9650 tags and Impinj H47 tags. Becuase Alien-9640 tags and Alien-9654 tags have the similar performance, we vary the number of Impinj H47 tags to observe the reading performance. Impinj H47 is the omnidirectional tag, which is insensitive to the placement of tag. When the proportion of H47 increases, the reading performance becomes better, as shown in Fig. 3l. However, the combinations of different types of tags have a little effect on TABLE 1 Combinations of Different Readers, Antennas, and Tags Reader Antenna (Polarization mode) Tag (IC type; Antenna technology) 1 Alien-9900+ Alien-9611 (Circular) Alien-9640 (Alien Higgs-3; Dipole) 2 Alien-9650 Alien-9650 (Circular) Alien-9654 (Alien Higgs-3; Dipole) 3 Alien-9900+ Alien-9611 (Circular) Impinj H47 (Monza 4; Dual-differential) 4 Impinj R420 Impinj Threshold (Linear) Impinj H47 (Monza 4; Dual-differential) 5 Impinj R420 Impinj Threshold (Linear) Alien-9640 (Alien Higgs-3; Dipole) YIN ET AL.: FOCUS AND SHOOT: EXPLORING AUTO-FOCUS IN RFID TAG IDENTIFICATION TOWARDS A SPECIFIED AREA 891
892 IEEE TRANSACTIONS ON COMPUTERS,VOL.65,NO.3,MARCH 2016 Tag Identification in the Specified Aren /知tenna Minor 00 Application Sampling Inspection Inventory Other similar applications interrogation region 00 下oc8I0dule Shoot Module Major Focus on the specified area) interrogation region Antenna 00 Adjustable components RFID tags Minor RFID System A口teia Reader Target tags (Rotate to the xpecifed area) (Power stepping) interrogation region 0 (Focus the target tags in the specified area) Interference tags ag Fig.5.The framework of PID. (a)3D region (b)Major Minor region Fig.4.The model of an antenna's interrogation region. in order to reduce the probability of identifying the interference tags. the reading performance.When the tags are located in the center area of the interrogation region,the tags usually can 5 BASELINE SOLUTIONS be identified easily. In order to identify the target tags in the specified area S, while ignoring the interference tags,we should focus on S 4.7 Analysis and identify as many target tags as possible.As mentioned Based on the above observations,we can use the model in 4.2,the larger the reader's power,the larger the interro- shown in Fig.4 to describe the interrogation region of an gation region.If we only want to focus on the area s,we antenna.In the three-dimensional space,the interrogation should use a lower power.On the contrary,if we want to region is like an ellipsoid,as shown in Fig.4a.We further identify more tags,we should use a larger power.Therefore, divide the interrogation region into major interrogation scanning with the minimal power min P and the maximal region and minor interrogation region,as shown in Fig.4b. power max P are two baseline solutions,which are respec- In the major interrogation region,the reader usually has tively called as MinPw and MaxPw. good reading performance.While in the minor interro- However,if the reader's power is too small,the inter- gation region,the reading performance is usually poor. rogation region cannot cover the specified area,leading to Based on Fig.3,the range of major interrogation region the low coverage ratio.If the reader's power is too large, for Alien-9611 antenna is about 0major=60,which is the interrogation region may be too large,leading to the almost consistent to the 3 dB beamwidth (65 degree)of identification of the interference tags.It increases the time Alien-9611 antenna. cost and the misreading ratio.Therefore,it is important However,Fig.3 illustrates that the radiation angle 0, to use a reasonable power to identify the tags in the speci- the placement of the tag,reader's power Pe,the distance fied area. d between the tags and the antenna,the tag density,the tag cardinality,the material the tag is attached to,and 6 PHOTOGRAPHY BASED IDENTIFICATION WITH the types of antennas and tags all affect the reading per- DISTANCE MEASUREMENT formance.Therefore,we use the above model as a guide. We will not depend on the model to calculate the param- In this section,we propose a solution called Photography based tag Identification with Distance measurement(PID), eters in our algorithms.Fortunately,when the tags are located in the major interrogation region,they usually which works with a 3D camera (e.g.,a Kinect).The process can be identified easily.As a result,instead of concentrat- of PID can be compared to the picture-taking process in a camera.It focuses on the area and shoots the objects,as ing on the influence from tags,we concentrate on adjust- ing the antenna's radiation angle 0,and the reader's shown in Fig.5.The application appoints the specified area power P to focus on the specified area.Unless otherwise S and the middleware collects the tag IDs in S by the RFID specified,we use Alien-9900+reader,Alien-9611 antenna, systems.It consists of the focus module and the shoot mod- ule.The focus module adjusts the reader's power and and Aline-9640 tag in this paper. According to the above analysis,we conclude the follow- rotates the antenna to make the interrogation region focus ing clues to design our solutions. on S.The shoot module collects tag IDs.The two corre- sponding processes are respectively called Focusing Process Rotating the antenna.If the objects are placed with and Shooting Process. a reasonable distance,we can distinguish them by rotating the antenna,based on the different read- 6.1 Focusing Process ing performances in major/minor interrogation The focusing process aims to make the interrogation region regions. focus on the specified area s by adjusting 0,Pe,while Facing towards the target tags.Due to the good perfor- ignoring the tags outside S.It contains three phases,select- mance in major interrogation region,we rotate the ing the initial power,establishing the boundary,and power antenna to face towards the target tags. stepping.The objective of this process is to get the optimal Power stepping.We can adjust the power to make the power P,whose corresponding interrogation region is just interrogation region be just cover the specified area,enough to cover the specified area s
the reading performance. When the tags are located in the center area of the interrogation region, the tags usually can be identified easily. 4.7 Analysis Based on the above observations, we can use the model shown in Fig. 4 to describe the interrogation region of an antenna. In the three-dimensional space, the interrogation region is like an ellipsoid, as shown in Fig. 4a. We further divide the interrogation region into major interrogation region and minor interrogation region, as shown in Fig. 4b. In the major interrogation region, the reader usually has good reading performance. While in the minor interrogation region, the reading performance is usually poor. Based on Fig. 3, the range of major interrogation region for Alien-9611 antenna is about umajor ¼ 60, which is almost consistent to the 3 dB beamwidth (65 degree) of Alien-9611 antenna. However, Fig. 3 illustrates that the radiation angle ur, the placement of the tag, reader’s power Pw, the distance d between the tags and the antenna, the tag density, the tag cardinality, the material the tag is attached to, and the types of antennas and tags all affect the reading performance. Therefore, we use the above model as a guide. We will not depend on the model to calculate the parameters in our algorithms. Fortunately, when the tags are located in the major interrogation region, they usually can be identified easily. As a result, instead of concentrating on the influence from tags, we concentrate on adjusting the antenna’s radiation angle ur and the reader’s power Pw to focus on the specified area. Unless otherwise specified, we use Alien-9900+ reader, Alien-9611 antenna, and Aline-9640 tag in this paper. According to the above analysis, we conclude the following clues to design our solutions. Rotating the antenna. If the objects are placed with a reasonable distance, we can distinguish them by rotating the antenna, based on the different reading performances in major/minor interrogation regions. Facing towards the target tags. Due to the good performance in major interrogation region, we rotate the antenna to face towards the target tags. Power stepping. We can adjust the power to make the interrogation region be just cover the specified area, in order to reduce the probability of identifying the interference tags. 5 BASELINE SOLUTIONS In order to identify the target tags in the specified area S, while ignoring the interference tags, we should focus on S and identify as many target tags as possible. As mentioned in 4.2, the larger the reader’s power, the larger the interrogation region. If we only want to focus on the area S, we should use a lower power. On the contrary, if we want to identify more tags, we should use a larger power. Therefore, scanning with the minimal power min Pw and the maximal power max Pw are two baseline solutions, which are respectively called as MinPw and MaxPw. However, if the reader’s power is too small, the interrogation region cannot cover the specified area, leading to the low coverage ratio. If the reader’s power is too large, the interrogation region may be too large, leading to the identification of the interference tags. It increases the time cost and the misreading ratio. Therefore, it is important to use a reasonable power to identify the tags in the speci- fied area. 6 PHOTOGRAPHY BASED IDENTIFICATION WITH DISTANCE MEASUREMENT In this section, we propose a solution called Photography based tag Identification with Distance measurement (PID), which works with a 3D camera (e.g., a Kinect). The process of PID can be compared to the picture-taking process in a camera. It focuses on the area and shoots the objects, as shown in Fig. 5. The application appoints the specified area S and the middleware collects the tag IDs in S by the RFID systems. It consists of the focus module and the shoot module. The focus module adjusts the reader’s power and rotates the antenna to make the interrogation region focus on S. The shoot module collects tag IDs. The two corresponding processes are respectively called Focusing Process and Shooting Process. 6.1 Focusing Process The focusing process aims to make the interrogation region focus on the specified area S by adjusting ur, Pw, while ignoring the tags outside S. It contains three phases, selecting the initial power, establishing the boundary, and power stepping. The objective of this process is to get the optimal power P w, whose corresponding interrogation region is just enough to cover the specified area S. Fig. 4. The model of an antenna’s interrogation region. Fig. 5. The framework of PID. 892 IEEE TRANSACTIONS ON COMPUTERS, VOL. 65, NO. 3, MARCH 2016
YIN ETAL:FOCUS AND SHOOT:EXPLORING AUTO-FOCUS IN RFID TAG IDENTIFICATION TOWARDS A SPECIFIED AREA 893 Specified area Paccording to the distance d,P=P(,d).If the power P is not large enough,the reader increases the power by APe and identifies no tags,as shown in Algorithm 1.It repeats the above process until n>ne,which means that it has collected enough tag IDs N =[ID1,ID2,...,ID from the boundary.However,if the reader's power has achieved Tag to the maximum value max Pn is still less than n,which indicates that most of the interference tags are far away from S. Interrogation Antenna Then,the reader stops the process and gets the optimal region power P=max P.After that,the antenna rotates towards the center of S for power stepping and tag identification. Fig.6.Identify the tags in S with a 3D camera Algorithm 1.PID:Establishing the Boundary 6.1.1 Selecting the Initial Power Input:The specified area S Before the reader identifies the tags,it selects the initial Determine the boundary S of S by the 3D camera,and calcu- power instead of the default(maximum)one to control the late d,and d.. interrogation region.In RFID systems,the reader's interro- The antenna rotates to So with=arccos() gation region of an antenna is like an ellipsoid.The larger the Pw=Penn(90°,do,Pe=Pa,n6,=0. angle 6,between the radiation direction and the antenna's while n ne and Pe max Pu do surface,the longer the reader's scanning range.However,in Collect tag IDs with Pe and get n responses. the realistic environment,the tag size,the reader's power Pr if Pe=max P and n$then ever,the reader can hardly find the boundary of S,due to while Pe>min P do the unknown distribution of tag IDs.Therefore,PID first Pu max(Pu -APu;min P). establishes the boundary S of the area S based on the inter- Check IDs in Ne,get Ane responses, ference tags located around S,as shown in Fig.6.PID uses ne=△nc. the 3D camera to calculate the minimum distance d if 8 then P Pu,Return. Then,the antenna rotates degree to face the interference if Pi=maxP then P=max Pe,Return. tags in S for identification.The identified tags are used as Output:The optimal power P reference tags to describe S. In PID,the antenna always faces towards the center of the In the commercial RFID systems,the reader (e.g.,Alien- objects,0,=90.Then,the reader selects the initial power 9900+)selects a specified tag by setting the mask equal to
6.1.1 Selecting the Initial Power Before the reader identifies the tags, it selects the initial power instead of the default (maximum) one to control the interrogation region. In RFID systems, the reader’s interrogation region of an antenna is like an ellipsoid. The larger the angle ur between the radiation direction and the antenna’s surface, the longer the reader’s scanning range. However, in the realistic environment, the tag size, the reader’s power Pw, the radiation angle ur, and the distance d all affect the effective interrogation region, as mentioned in Section 4. Therefore, in the realistic environments, we measure the minimum power (MinPw) Pwmin based on ur and d, and use them to calculate the initial power. In this paper, we measure Pwmin ður; dÞ with the distances dj ¼ 0:5 m j; j 2 ½1; 7 and the angles ui ¼ 90 15 i; i 2 ½0; 6. For example, we get Pwmin ð90; 1:0Þ ¼ 15:7 dBm, Pwmin ð75; 1:5Þ ¼ 18:8 dBm, Pwmin ð60; 2:0Þ ¼ 23:4 dBm. The reader first selects the reference angle ui closest to ur, jur uijjur ukj (k 2 ½0; 6 and k 6¼ i). Then, it uses d to calculate the initial power Pwmin ður; dÞ Pwmin ðui; djÞ if d ¼ dj Pwmin ðui;djÞþPwmin ðui;djþ1Þ 2 if d 2 ½dj; djþ1: ( (1) However, the power is only used as the initial power. In order to identify more tags, the reader can repeatedly increase the power by DPw. We set DPw ¼ 1 dBm, which is achievable by most of the commercial readers [30]. 6.1.2 Establishing the Boundary The 3D camera can recognize the specified area by RGB camera and measure distance by 3D depth sensors. However, the reader can hardly find the boundary of S, due to the unknown distribution of tag IDs. Therefore, PID first establishes the boundary Sb of the area S based on the interference tags located around S, as shown in Fig. 6. PID uses the 3D camera to calculate the minimum distance db between the interference tags in Sb and the antenna, and the distance ds between the center of S and the antenna. Furthermore, it calculates the rotation angle ’ as follows: ’ ¼ arccos ds db ; ’ 2 ð0 ; 90 Þ: (2) Then, the antenna rotates ’ degree to face the interference tags in Sb for identification. The identified tags are used as reference tags to describe Sb. In PID, the antenna always faces towards the center of the objects, ur ¼ 90. Then, the reader selects the initial power Pwb according to the distance d, Pwb ¼ Pwmin ð90; dÞ. If the power Pwb is not large enough, the reader increases the power by DPw and identifies nb tags, as shown in Algorithm 1. It repeats the above process until nb n", which means that it has collected enough tag IDs Nb ¼ fID1; ID2; ... ; IDnbg from the boundary. However, if the reader’s power has achieved to the maximum value max Pw, nb is still less than n", which indicates that most of the interference tags are far away from S. Then, the reader stops the process and gets the optimal power P w ¼ max Pw. After that, the antenna rotates towards the center of S for power stepping and tag identification. Algorithm 1. PID: Establishing the Boundary Input: The specified area S Determine the boundary Sb of S by the 3D camera, and calculate db and ds. The antenna rotates to Sb with ’ ¼ arccosð ds db Þ. Pwb ¼ Pwmin ð90; dbÞ, Pw ¼ Pwb, nb ¼ 0. while nb d then while Pw > min Pw do Pw ¼ maxðPw DPw; min PwÞ. Check IDs in Nc, get Dnc responses, nc ¼ Dnc. if nc nb d then P w ¼ Pw, Return. if Pw ¼ min Pw then P w ¼ min Pw, Return. if nc nb < d then while Pw < max Pw do Pw ¼ min(Pw þ DPw, max Pw). Check IDs in Nb Nc, get Dnc responses, nc ¼ nc þ Dnc. if nc nb d then P w ¼ Pw, Return. if Pw ¼ maxPw then P w ¼ maxPw, Return. Output: The optimal power P w In the commercial RFID systems, the reader (e.g., Alien- 9900+) selects a specified tag by setting the mask equal to Fig. 6. Identify the tags in S with a 3D camera. YIN ET AL.: FOCUS AND SHOOT: EXPLORING AUTO-FOCUS IN RFID TAG IDENTIFICATION TOWARDS A SPECIFIED AREA 893
894 EEE TRANSACTIONS ON COMPUTERS,VOL.65,NO.3.MARCH 2016 TABLE 2 Specified area Upper Bound Tu N 20 60 100 140 180 220 nu 2 4 9 17 12 the tag ID.If the tag gives response,the reader gets a non- Tag empty slot.Otherwise,it gets an empty slot.The reader Interrogation checks all the IDs in N and gets ne responses Ne.Obviously, Antenna region nea.At this time,the reader gets the optimal power POn the contrary,ifn.That is to say,ne is the minimum number of interfer- ANGLE ROTATION ence tags needed to establish the boundary of the area S.In In PID,a 3D camera is used in the focusing process.How- the realistic environment,some tags may not be identified ever,in some environments,the 3D camera can not work by the reader steadily.If the power arriving at a tag is well(e.g.,in a dark space).Besides,considering the cost sav- approximate to the minimum power to activate it,the tag ings,it may not be used.Therefore,identifying the target may respond to the reader or keep silent at random,which fags without the auxiliary equipment is important.For this will affect the judgement of the boundary.Therefore,we problem,we propose a solution called Photography based measure the value of ne with a different tag size N]in the tag Identification with Angle rotation(PIA).It also consists realistic environments,as shown in Table 2.Based on of the Focusing Process and Shooting Process.The only differ Table 2,we can conclude that tag size N]has a little effect ence between PID and PIA is how to determine the bound- on ne,which is usually very small.Therefore,in order to ary of S.Therefore,we only describe how to find the definitely get enough tag IDs in S,we set na=15 by boundary in PIA,while ignoring the others. default,while considering the stability and time efficiency. In regard to 8,it affects the misreading ratio.The smaller 7.1 PIA the value of 8,the lower the misreading ratio,the smaller the Without the 3D camera,PIA cannot calculate any distance; execution time.However,the larger the value of 8,the larger it explores the boundary by rotating the antenna,as shown the value of coverage ratio.Considering the constraint of cov-in Fig.7.Firstly,the application appoints S and the antenna erage ratio p and time efficiency,we set 8=a.When rotates towards S.Then the reader sets its initial power
the tag ID. If the tag gives response, the reader gets a nonempty slot. Otherwise, it gets an empty slot. The reader checks all the IDs in Nb and gets nc responses Nc. Obviously, nc nb. When nc nb ¼ d, the interrogation region just achieves the boundary of S. The corresponding power is the optimal power P w. However, if nc nb > d, the reader reduces the power by DPw and checks the verified tag IDs in Nc. If a tag does not give response, the reader removes it from Nc. It repeats the above process until nc nb d and gets the optimal power P w. On the contrary, if nc nb < d, the reader increases Pw by DPw and checks the unverified tag IDs in Nb Nc ¼ fIDi j IDi 2 Nb and IDi 2= Ncg. If the tag gives response, the reader adds the ID into Nc. It repeats the process until nc nb d and gets the optimal power P w. In the following process, the reader uses P w to identify the target tags. 6.2 Shooting Process In this process, the reader collects the tag IDs in S. The reader’s power is equal to P w and we use frame slotted ALOHA (FSA) protocol to identify the tags. FSA is a popular anti-collision protocol. In FSA, the reader first broadcasts a number f, which specifies the following frame size. After receiving f, each tag selects hðIDÞ mod f as its slot number, where h is a hash function. If none of the tags respond in a slot, the reader closes the slot immediately. If only one tag responds in a slot, the reader successfully receives the tag ID. If multiple tags respond simultaneously, a collision occurs, and the involved tags will be acknowledged to restart in the next frame. The similar process repeats until no tags respond in the frame. The collected IDs are considered as the target tag IDs. 6.3 Performance Analysis In order to definitely describe the boundary Sb, PID needs to steadily get at least n" interference tag IDs, and nb satisfies nb n". That is to say, n" is the minimum number of interference tags needed to establish the boundary of the area S. In the realistic environment, some tags may not be identified by the reader steadily. If the power arriving at a tag is approximate to the minimum power to activate it, the tag may respond to the reader or keep silent at random, which will affect the judgement of the boundary. Therefore, we measure the value of n" with a different tag size jNj in the realistic environments, as shown in Table 2. Based on Table 2, we can conclude that tag size jNj has a little effect on n", which is usually very small. Therefore, in order to definitely get enough tag IDs in Sb, we set n" ¼ 15 by default, while considering the stability and time efficiency. In regard to d, it affects the misreading ratio. The smaller the value of d, the lower the misreading ratio, the smaller the execution time. However, the larger the value of d, the larger the value of coverage ratio. Considering the constraint of coverage ratio r and time efficiency, we set d ¼ a. When nc nb ¼ d ¼ a, we consider that the interrogation region only achieves the boundary, and the coverage ratio r satisfies r a. At this time, the reader gets the optimal power P w ¼ Pws þ kc DPw; kc 2 Z: (3) Here, we set DPw ¼ 1 dBm, which is achievable by most of the commercial readers [30]. kc represents the number of steps needed to update the reader’s power, and kc is adaptively determined in the identification process. In PID, when the rotation angle is determined, the antenna rotates to the target direction immediately. The time for rotating the antenna can be neglected when compared to the identification time. Based on the above analysis, we can estimate the lower bound of the execution time Tl for PID as follows: Tl ¼ i X¼kb i¼0 ðe nbðiÞ tÞ þ j X¼kc j¼0 ðncðjÞ tÞ þ e n t: (4) Here, nbðiÞ is the number of tags identified from the boundary in the ith step, ncðjÞ is the number of tags identified in Sb in jth step, n is the number of target tags checked in the shooting process, t is the time for a slot, and e is the base of the natural logarithm. From Eq. (4), we can find that the numbers kb, kc, n affect Tl, which is related to the reader’s power. Thus, choosing the optimal power P w is essential to the problem, as shown in Algorithm 2. 7 PHOTOGRAPHY BASED IDENTIFICATION WITH ANGLE ROTATION In PID, a 3D camera is used in the focusing process. However, in some environments, the 3D camera can not work well (e.g., in a dark space). Besides, considering the cost savings, it may not be used. Therefore, identifying the target tags without the auxiliary equipment is important. For this problem, we propose a solution called Photography based tag Identification with Angle rotation (PIA). It also consists of the Focusing Process and Shooting Process. The only difference between PID and PIA is how to determine the boundary of S. Therefore, we only describe how to find the boundary in PIA, while ignoring the others. 7.1 PIA Without the 3D camera, PIA cannot calculate any distance; it explores the boundary by rotating the antenna, as shown in Fig. 7. Firstly, the application appoints S and the antenna rotates towards S. Then the reader sets its initial power TABLE 2 Upper Bound nu N 20 60 100 140 180 220 nu 2 4 7 9 11 12 Fig. 7. Identify the tags in S without any auxiliary equipment. 894 IEEE TRANSACTIONS ON COMPUTERS, VOL. 65, NO. 3, MARCH 2016
YIN ETAL:FOCUS AND SHOOT:EXPLORING AUTO-FOCUS IN RFID TAG IDENTIFICATION TOWARDS A SPECIFIED AREA 895 equal to the minimum power minP and identifies n,tags in power stepping.If p has not been determined,PIA utilizes S.If n,land△nP,then No=Nr PIA will not be affected by the surroundings and can work well in different environments. else if.△an>△9,then N%=Nr Output:Tag IDs in the boundary:N 8 EXTENSION FOR TAG IDENTIFICATION TOWARDS When the antenna rotates to another direction (called TWO-DIMENSIONAL SPACE left),the identified tags in s decreases.As shown in In Section 3,the target tags and the interference tags are Algorithm 3,the radiation angle decreases by Aor.In the ith mainly located in one-dimensional space (Our previous step,Ans;tags disappear from S,the number of identified conference version [31]focuses on this problem.).However, tags in S is n At the same time,the reader gets n tag IDs in some applications,the objects may be located around the out of S,and they are considered as the tag IDs from the farget tags in two-dimensional space,as shown in Fig.8. boundary.If n ne,the reader collects enough tag IDs In this case,PID can still use the 3D camera to distinguish N=[ID,ID2,...,ID}from the boundary.Otherwise,it the target tags and interference tags.Unfortunately,PIA increases the power by AP.Each time,it should make sure may not work well,because it mainly identifies the interfer- that nne.Then,the antenna has rotated Identification with Angle rotation(EPIA)to identify the tar- A0r degrees.At this time,the ending power of the reader is get tags in two-dimensional space. P After that,the antenna rotates to the opposite direction In EPIA,the reader identifies some target tags N,as PIA (right)and works in the same way.It rotates A0r degrees to does.Then,it begins to identify the interference tags in the the right side and the ending power of the reader is P.If boundary So around the specified area s,as shown in Fig.8. Aor>Ao,it indicates that the boundary on the right side The antenna first randomly selects a direction to identify is farther than that of the left one,then the reader terminates some tags NI from the boundary,as shown in Algorithm 4. Then,it rotates Aor in clockwise direction in order to iden- the process.Otherwise,it obtains N=[ID,ID2,..., tify other tags around the boundary,aiming to find the inter- ID.The reader compares Aen Aerr and P Pr to find ference tags closest to the target tags,as shown in Fig.8.In the the nearer boundary,and gets the new set N of interference newly-selected direction,it identifies the interference tags as tags.If o=A0rr and Pir=Pur,N=NUN,.If PIA does.When the antenna has finished identifying the on=A0r,while PPr,N =N,.Besides,ifon>tags around the specified S,it will select the interference tags Ae,N=N.Otherwise,N=N.Here,N is used for with the smallest rotation angle value[1
equal to the minimum power minPw and identifies ns tags in S. If ns 1 and Dnsi Pwr then Nb ¼ Nr. else if Durl > Durr then Nb ¼ Nr. Output: Tag IDs in the boundary :Nb When the antenna rotates to another direction (called left), the identified tags in S decreases. As shown in Algorithm 3, the radiation angle decreases by Dur. In the ith step, Dnsi tags disappear from S, the number of identified tags in S is nsi . At the same time, the reader gets nl tag IDs out of S, and they are considered as the tag IDs from the boundary. If nl n", the reader collects enough tag IDs Nl ¼ fID0 1; ID0 2; ... ; ID0 nl g from the boundary. Otherwise, it increases the power by DPw. Each time, it should make sure that nsi Durl , it indicates that the boundary on the right side is farther than that of the left one, then the reader terminates the process. Otherwise, it obtains Nr ¼ fID00 1; ID00 2; ... ; ID00 nr g. The reader compares Durl , Durr and Pwl , Pwr to find the nearer boundary, and gets the new set Nb of interference tags. If Durl ¼ Durr and Pwl ¼ Pwr , Nb ¼ Nl [ Nr. If Durl ¼ Durr , while Pwl > Pwr , Nb ¼ Nr. Besides, if Durl > Durr , Nb ¼ Nr. Otherwise, Nb ¼ Nl. Here, Nb is used for power stepping. If P w has not been determined, PIA utilizes the ending power Pwl , Pwr to determine the initial power Pws. If Nb ¼ Nl, Pws ¼ Pwl . Otherwise, Pws ¼ Pwr . Pws is used for power stepping, which is described in Algorithm 2. The values of parameters in PIA are equal to those in PID. In regard to Dur in PIA, we set Dur ¼ 30. Based on Fig. 3a, when ur 2 ½75; 90, the reader undoubtedly has good performance. Therefore, when Dur ¼ 30, each tag can be requested in the center of the interrogation region with ur 2 ½75; 90. PIA does not identify the tags while the antenna is rotating. This is because the reader cannot determine where the identified tags are located, when the antenna is rotating. Therefore, PIA rotates to the next direction immediately as PID does, then it identifies the tags. 7.2 Comparison of PID and PIA We compare PID and PIA in the following aspects: System equipment. PID uses an auxiliary equipment (i.e., a 3D camera), while PIA does not need any auxiliary equipment. Performance comparison. With the 3D camera, PID can quickly focus on the specified area and reduce the execution time. PIA needs to rotate the antenna to find the specified area, it often needs more execution time and identifies more interference tags. Application environment. By using the 3D camera, PID recognizes the specified area by the RGB camera and measures distance by the 3D depth sensor. In the environments like dark spaces, the RGB camera can not work well, thus PID can not work well. While PIA will not be affected by the surroundings and can work well in different environments. 8 EXTENSION FOR TAG IDENTIFICATION TOWARDS TWO-DIMENSIONAL SPACE In Section 3, the target tags and the interference tags are mainly located in one-dimensional space (Our previous conference version [31] focuses on this problem.). However, in some applications, the objects may be located around the target tags in two-dimensional space, as shown in Fig. 8. In this case, PID can still use the 3D camera to distinguish the target tags and interference tags. Unfortunately, PIA may not work well, because it mainly identifies the interference tags in one-dimensional space (see Fig. 7). It may not find the boundary appropriately in two-dimensional space. Thus we improve PIA as the Enhanced Photography based Identification with Angle rotation (EPIA) to identify the target tags in two-dimensional space. In EPIA, the reader identifies some target tags Ns as PIA does. Then, it begins to identify the interference tags in the boundary Sb around the specified area S, as shown in Fig. 8. The antenna first randomly selects a direction to identify some tags Nb1 from the boundary, as shown in Algorithm 4. Then, it rotates Durc in clockwise direction in order to identify other tags around the boundary, aiming to find the interference tags closest to the target tags, as shown in Fig. 8. In the newly-selected direction, it identifies the interference tags as PIA does. When the antenna has finished identifying the tags around the specified S, it will select the interference tags with the smallest rotation angle value Durk, k 2 ½1; d360 Durce, YIN ET AL.: FOCUS AND SHOOT: EXPLORING AUTO-FOCUS IN RFID TAG IDENTIFICATION TOWARDS A SPECIFIED AREA 895
896 IEEE TRANSACTIONS ON COMPUTERS,VOL.65,NO.3.MARCH 2016 Interference tags 7 Tag Specified area Interference tags Interference tags Interference tags Antenna Fig.9.Tags located at different depths. Fig.8.Identify the tags in the specified area in two-dimensional space. which means the identified tag sets {Nk}are closest to the In order to find the area containing interference tags, target tags.Then,the reader selects one set Nk which is iden- EPIA rotates the antenna around the specified area S.We tified with the smallest power Pk from {Nok},and uses Nol use Aore to divide the boundary into different regions,as as the interference tags,which will be used for power step- shown in Fig.8.According to Fig.3a in Section 4,when the ping as PIA does.Similarly,EPIA selects this smallest end- rotation angle 0,∈[60°,90],the reader has better perfor- ing power Pok as the initial power Ps for power steeping, mance to identify the tags.Therefore,we set△ore=6O°,in which is described in Algorithm 2. order to make sure that each region of the boundary can be located in the reader's effective scanning range.However, Algorithm 4.EPIA:Exploring the Boundary in Two- when the direction is fixed,the reader identifies the tags as Dimensional Space PIA does.When identifying the tags in one direction,the Input:The specified area S antenna rotates away from the target tags by A0,=30 to Pw min Pu,na=0,0e=0 identify the interference tags as PIA does.In this way, while ns1and△ns,<△ns-then diri(i [1,4)may be different. Break. At this time,the rotation angle o of the antenna cannot be Pu min(Pu +APu;max Pu). calculated by 3D camera in the way described in Section 6.1.2. else Break. For example,the 3D camera can calculate the distances dL i=i+1. Puk=Pu. and dis,whilearccos(),because ddr.PID can if△9k<△Omin then△Bmit=△9rk. not work well in such an environment.Fortunately,we can 0e=0e+△0e,k=k+1. use EPIA to explore the boundary of the specified area S by For the sets {Nu}satisfying Aork=Aomin,select one Nok with rotating the antenna.However,if the di(i[1,4)is much the smallest Pick. smaller than d,more interference tags will be identified.At Ni=Nik. this time,we can try to adjust the placement of the objects or Output:Tag IDs in the boundary:No antenna,aiming to reduce the depth difference among the tags.Then,the reader uses EPIA to identify the target tags
which means the identified tag sets fNbkg are closest to the target tags. Then, the reader selects one set Nbk which is identified with the smallest power Pwk from fNbkg, and uses Nbk as the interference tags, which will be used for power stepping as PIA does. Similarly, EPIA selects this smallest ending power Pwk as the initial power Pws for power steeping, which is described in Algorithm 2. Algorithm 4. EPIA: Exploring the Boundary in TwoDimensional Space Input: The specified area S Pw ¼ min Pw, ns ¼ 0, uc ¼ 0. while ns 1 and Dnsi < Dnsi 1 then Break. Pw ¼ minðPw þ DPw; max PwÞ. else Break. i ¼ i þ 1. Pwk ¼ Pw. if Durk < Dumin then Dumin ¼ Durk. uc ¼ uc þ Durc, k ¼ k þ 1. For the sets fNbkg satisfying Durk ¼ Dumin, select one Nbk with the smallest Pwk. Nb ¼ Nbk. Output: Tag IDs in the boundary :Nb In order to find the area containing interference tags, EPIA rotates the antenna around the specified area S. We use Durc to divide the boundary into different regions, as shown in Fig. 8. According to Fig. 3a in Section 4, when the rotation angle ur 2 ½60; 90, the reader has better performance to identify the tags. Therefore, we set Durc ¼ 60, in order to make sure that each region of the boundary can be located in the reader’s effective scanning range. However, when the direction is fixed, the reader identifies the tags as PIA does. When identifying the tags in one direction, the antenna rotates away from the target tags by Dur ¼ 30 to identify the interference tags as PIA does. In this way, the tags to be identified are always located in the center of the interrogation region. It means that ur 2 ½75; 90, where the reader will undoubtedly have good performance for tag identification. The difference between the value of Durc and Dur lies in that Durc is only used to divide the boundary, while Dur is used to identify the tags. 9 DISCUSSION FOR TAG IDENTIFICATION TOWARDS MORE COMPLEX ENVIRONMENTS In the above sections, the target tags and the interference tags are mainly located at the same depth. However, in more complex environments, the objects may be located at different depths. As shown in Fig. 9, the vertical distance between the target tags and the antenna is d?s, while the vertical distance between the interference tags and the antenna are d?r1; d?r2; d?r3; d?r4, respectively. The value of d?s and d?riði 2 ½1; 4) may be different. At this time, the rotation angle ’ of the antenna cannot be calculated by 3D camera in the way described in Section 6.1.2. For example, the 3D camera can calculate the distances d?s and db3, while ’3 6¼ arccosð d?s db3 Þ, because d?s 6¼ d?r3. PID can not work well in such an environment. Fortunately, we can use EPIA to explore the boundary of the specified area S by rotating the antenna. However, if the d?riði 2 ½1; 4Þ is much smaller than d?s, more interference tags will be identified. At this time, we can try to adjust the placement of the objects or antenna, aiming to reduce the depth difference among the tags. Then, the reader uses EPIA to identify the target tags. Fig. 9. Tags located at different depths. Fig. 8. Identify the tags in the specified area in two-dimensional space. 896 IEEE TRANSACTIONS ON COMPUTERS, VOL. 65, NO. 3, MARCH 2016
YIN ETAL:FOCUS AND SHOOT:EXPLORING AUTO-FOCUS IN RFID TAG IDENTIFICATION TOWARDS A SPECIFIED AREA 897 nten Kinect Identifying the tags Identifying the tags (3D Cameral Smart Car Car (a)PID (b)PIA Fig.10.System prototypes work in the realistic environments. 10 PERFORMANCE EVALUATION than 60 percent (p >a),which satisfies the requirement.As We evaluate the performance of each solution in the realistic mentioned in Section 3.2,the coverage ratio must be satis- environments.The experimental facilities are the same as fied.Therefore,the solution MinPw is invalid and we ignore those used in the observations.The execution time,coverage it in the following comparisons. ratio,and misreading ratio are used for performance metrics. In the experiments,each book is attached with an RFID 10.3 Execution Time T tag,and the tag ID is 96 bits.The books are randomly Fig.12 shows the execution time of each solution.Our solu- deployed in three boxes and the distribution of the tag IDs tions PID and PIA have better performances than MaxPw. are unknown.Each box is placed on a tablet chair with a This is because PID and PIA only focus on the target tags in height of 0.5 m,as shown in Fig.10.PID uses a 3D camera, S.MaxPw identifies all the tags in the interrogation region, while PIA does not.The antenna is deployed on the smart including a lot of interference tags.Usually,PID has a better car,which is controlled by the program and can rotate with performance than PIA,due to the use of a 3D camera.In the antenna flexibly.The antenna faces towards the tags to Figs.12a and 12b,the difference in execution time between be identified.The specified area here is the center box, PID,PIA and MaxPw is small.This is because the tag size is which is the target box,while the other two boxes are interfer- relatively small.When the tag size becomes large,our pro- ence boxes.The distance between the target box and the posed solutions become more efficient.When s=120,PID antenna is d.The minimum distance between the interfer- reduces the execution time by 44 percent compared to ence box and the target box is l.s and u respectively repre- MaxPw,as shown in Fig.12c.When u 270,PID even can sent the number of target tags (in target box)and the reduce the execution time by 85.6 percent compared to number of interference tags(in interference boxes).We vary MaxPw,as shown in Fig.12d. the values of d,l,s,u to evaluate the performance of each solution.We set d 1m,I=1 m,s =80,u=70 by default. 10.4 Misreading Ratio In Section 3.2,we analyze that the execution time is related 10.1 Upper Bound of a to the misreading ratio.Fig.13 shows the misreading ratio As mentioned in Section 3.2,when the distance d,and the of each solution.Our solutions PID and PIA have lower mis- number of tags n are fixed,we can determine the value of a. reading ratios than MaxPw.This is because PID and PIA In Table 3,we give the upper bound of a under different use the optimal powers instead of the maximum one conditions.We set a=60%for the following experiments (30.7 dBm).PID and PIA mainly focus on the target tags, by default. while avoiding identifying the interference tags. When we change a,our solutions can also work well.For example,when d=1m,l=1m,s=80,u=70,we set 10.2 Coverage Ratio p Constraint a=80%.The coverage ratio of MaxPw,PID,PIA is respec- We first investigate the coverage ratio p of each solution,as tively equal to 89,82.5,86 percent,which satisfy p>a.The shown in Fig.11.We can observe that scanning with the execution time of MaxPw,PID,PIA is respectively equal to minimum power cannot achieve the requirement of cover- 2.2,1.45,2.0s.Our solutions outperform the baseline age ratio (a =60%).This is because the power is too small solutions. to activate the majority of the tags.When we identify the tags with the maximum power(MaxPw)or our proposed 10.5 Using Different Types of Tags solutions (PID and PIA),the coverage ratios are all larger We conduct the experiment with Alien-9640 tags,Alien- TABLE 3 9654 tags,and Impinj H47 tags.The tag cardinality of each Upper Bound of a type is almost the same.We randomly mix these tags together in interference tags and target tags.We vary the 会 d (m) 0.5 1.0 1.5 minimal distance l between target tags and interference tags 40 100% 100% 90% in the experiments.As shown in Fig.14,the mixed tags 80 95% 85% 65% have a little effect on the reading performance,which is 120 89% 81% 63% coincident with Fig.131.The performance of each solution is similar to that shown in Figs.11b,12b,13b.When l=1m
10 PERFORMANCE EVALUATION We evaluate the performance of each solution in the realistic environments. The experimental facilities are the same as those used in the observations. The execution time, coverage ratio, and misreading ratio are used for performance metrics. In the experiments, each book is attached with an RFID tag, and the tag ID is 96 bits. The books are randomly deployed in three boxes and the distribution of the tag IDs are unknown. Each box is placed on a tablet chair with a height of 0.5 m, as shown in Fig. 10. PID uses a 3D camera, while PIA does not. The antenna is deployed on the smart car, which is controlled by the program and can rotate with the antenna flexibly. The antenna faces towards the tags to be identified. The specified area here is the center box, which is the target box, while the other two boxes are interference boxes. The distance between the target box and the antenna is d. The minimum distance between the interference box and the target box is l. s and u respectively represent the number of target tags (in target box) and the number of interference tags (in interference boxes). We vary the values of d, l, s, u to evaluate the performance of each solution. We set d ¼ 1 m, l ¼ 1 m, s ¼ 80, u ¼ 70 by default. 10.1 Upper Bound of a As mentioned in Section 3.2, when the distance d, and the number of tags n are fixed, we can determine the value of a. In Table 3, we give the upper bound of a under different conditions. We set a ¼ 60% for the following experiments by default. 10.2 Coverage Ratio r Constraint We first investigate the coverage ratio r of each solution, as shown in Fig. 11. We can observe that scanning with the minimum power cannot achieve the requirement of coverage ratio (a ¼ 60%). This is because the power is too small to activate the majority of the tags. When we identify the tags with the maximum power (MaxPw) or our proposed solutions (PID and PIA), the coverage ratios are all larger than 60 percent (r a), which satisfies the requirement. As mentioned in Section 3.2, the coverage ratio must be satis- fied. Therefore, the solution MinPw is invalid and we ignore it in the following comparisons. 10.3 Execution Time T Fig. 12 shows the execution time of each solution. Our solutions PID and PIA have better performances than MaxPw. This is because PID and PIA only focus on the target tags in S. MaxPw identifies all the tags in the interrogation region, including a lot of interference tags. Usually, PID has a better performance than PIA, due to the use of a 3D camera. In Figs. 12a and 12b, the difference in execution time between PID, PIA and MaxPw is small. This is because the tag size is relatively small. When the tag size becomes large, our proposed solutions become more efficient. When s ¼ 120, PID reduces the execution time by 44 percent compared to MaxPw, as shown in Fig. 12c. When u ¼ 270, PID even can reduce the execution time by 85:6 percent compared to MaxPw, as shown in Fig. 12d. 10.4 Misreading Ratio In Section 3.2, we analyze that the execution time is related to the misreading ratio. Fig. 13 shows the misreading ratio of each solution. Our solutions PID and PIA have lower misreading ratios than MaxPw. This is because PID and PIA use the optimal powers instead of the maximum one (30.7 dBm). PID and PIA mainly focus on the target tags, while avoiding identifying the interference tags. When we change a, our solutions can also work well. For example, when d ¼ 1 m, l ¼ 1 m, s ¼ 80, u ¼ 70, we set a ¼ 80%. The coverage ratio of MaxPw, PID, PIA is respectively equal to 89, 82:5, 86 percent, which satisfy r a. The execution time of MaxPw, PID, PIA is respectively equal to 2:2, 1:45, 2:0 s. Our solutions outperform the baseline solutions. 10.5 Using Different Types of Tags We conduct the experiment with Alien-9640 tags, Alien- 9654 tags, and Impinj H47 tags. The tag cardinality of each type is almost the same. We randomly mix these tags together in interference tags and target tags. We vary the minimal distance l between target tags and interference tags in the experiments. As shown in Fig. 14, the mixed tags have a little effect on the reading performance, which is coincident with Fig. 13l. The performance of each solution is similar to that shown in Figs. 11b, 12b, 13b. When l ¼ 1m, Fig. 10. System prototypes work in the realistic environments. TABLE 3 Upper Bound of a n d (m) 0.5 1.0 1.5 40 100% 100% 90% 80 95% 85% 65% 120 89% 81% 63% YIN ET AL.: FOCUS AND SHOOT: EXPLORING AUTO-FOCUS IN RFID TAG IDENTIFICATION TOWARDS A SPECIFIED AREA 897