session number.Each tag only replies the query command ▣ with same session number as it holds.In this way,the readers can only collect non-contentious tags during Phase- I.In Phase-II,the reader broadcasts query commands with 口 a unified session number of zero.Because the contentious ◇ 回 tag's session number has not changed during Phase-I.they will reply to the query command.Another consequence produced Fig.5.Unbalanced loads of readers by unbalanced loads of readers is that the readers with lower loads will wait for the ends of the readers with higher loads during Phase-I.We can simply switch Phase-I and Phase-II round. to shorten such delay.Namely,readers first jointly identify The role of readers may change during the scheduling contentious tags and then identify their own non-contentious round.Assume the scheduling sequence of active readers is tags. (A1,A2),where A1 {r2,r6,r7}and A2 (rs}as 2)Source Sensitive and Insensitive:RFID application can illustrated in the example shown in Fig.4.Reader r5 is a be summarized into two categories.One is source-insensitive, passive reader in the first round but it becomes an active reader in which the source,namely the ID of reader that detected the in the second round.Once a reader becomes an active reader tag,is not concerned.The user may only want to confirm that in one round,its weight will become zero and then finishes all tags can be collected,for example in warehouse monitoring. identification process. Another is source-sensitive,in which tags must be exactly To illustrate execution of Season,we give an example shown reported multiple times,for example the object tracking.In in Fig.4.At the beginning of the first round shown in Fig. the second type of applications,a tag can be approximately 4(a),active reader r2 estimate the number of its contentious located by recording the readers that collect the tag.Duplicate tags n3+5+8+9=25.At the same time,the passive reports of a tag from neighboring readers can also help the reader rs estimates the number of its contentious tags in the administrator to re-deploy readers for better coverage.Season current round n9+5+8=22.Reader r2 continues to can well support both the source insensitive and sensitive power tags until it collects the 25 tags and also receives the applications.For source-insensitive application,Season allows "FINISH"messages from r1.r3,r4 and r5.Concurrently,rs passive readers to immediately send a "FINISH"message listens to the tags'replies.After successfully collecting its 22 to their neighboring active readers without identifying its tags,it sends a "FINISH"message to r2,r6 and r7.At the contentious tags. end of the first round,all of the readers adjust their weights. Reader ri,r2.r3.r6,and r7 set their weights to zero and IV.PERFORMANCE EVALUATION report their collections.In the second round as shown in Fig. We now evaluate Season using real-world logistics and 4(b),there are only ra and rs's weights not equaling to zero tracking traces. in RCG.Reader r5 is selected as the active reader.It starts to power tags and r4 listens to the tags'replies.The procedure A.Evaluation Methodology ends when r4 sends a "FINISH"message to rs. 1)Testbed and deployment:To validate the feasibility of joint identification,we use a NI PXI-1044 RFID testing tool F Discussion with PXI 5600 receiver as our passive reader.We uniformly 1)Unbalanced Loads of Readers:The load of reader is set the power of antenna as 20 dBm which supports around defined as the number of tags located in its integration range. an interrogation range of 2m.We also deploy five readers In Season,neighboring readers may have unbalanced loads. in a logistics enterprise,Xi'an postal processing center in For example in Fig.5,reader r2 have more tags in its inter- Shaanxi,China.The center is the one of the seven largest rogating regions than reader r1.At the beginning of Phase-I, postal processing centers in China.It covers an area of about two readers cannot collect tag to due to the reader collision. 16,128m2 and contains 30 importing/exporting gates.Fig.7 However,reader ri complete running Season-I earlier than r2 shows the architectural plans of the center.We attach more and then stopping interrogating.Then the reader r2 is able than 100 passive tags into pouches and find that the percentage to collect to since it is still running Season-I.In this case, of contentious tags is less than 10%for a stable and full some contentious tags may be collected in Phase-I and cause coverage. confusion to the joint identification in Phase-II. 2)Simulating Real RFID Applications:For simulation,we We introduce session number to solve this problem.Each use two typical application scenarios and three random reader tag contains a session number with the initialized value as zero. topologies described as follows. At the beginning of Season,each reader randomly generates Warehouse:According to our measurement results in Xi'an a non-zero session number and broadcasts it.If a tag can postal center,we simulate a total of 12 *6=72 readers for resolve a session number,it must be non-contentious.Then covering the entire center in a square-grid formation.Each the tag changes its session number to what it receives.In reader is located at one vertex in the grid.Each reader has an Phase-I,the reader sends query commands with the non-zero interrogating range of 7m,which has 126 contentious regions.r1 r2 t0 Fig. 5. Unbalanced loads of readers round. The role of readers may change during the scheduling round. Assume the scheduling sequence of active readers is {A1, A2}, where A1 = {r2, r6, r7} and A2 = {r5} as illustrated in the example shown in Fig. 4. Reader r5 is a passive reader in the first round but it becomes an active reader in the second round. Once a reader becomes an active reader in one round, its weight will become zero and then finishes identification process. To illustrate execution of Season, we give an example shown in Fig. 4. At the beginning of the first round shown in Fig. 4(a), active reader r2 estimate the number of its contentious tags n 1 2 ≈ 3 + 5 + 8 + 9 = 25. At the same time, the passive reader r5 estimates the number of its contentious tags in the current round n 1 5 ≈ 9 + 5 + 8 = 22. Reader r2 continues to power tags until it collects the 25 tags and also receives the “FINISH” messages from r1, r3, r4 and r5. Concurrently, r5 listens to the tags’ replies. After successfully collecting its 22 tags, it sends a “FINISH” message to r2, r6 and r7. At the end of the first round, all of the readers adjust their weights. Reader r1, r2, r3, r6, and r7 set their weights to zero and report their collections. In the second round as shown in Fig. 4(b), there are only r4 and r5’s weights not equaling to zero in RCG. Reader r5 is selected as the active reader. It starts to power tags and r4 listens to the tags’ replies. The procedure ends when r4 sends a “FINISH” message to r5. F. Discussion 1) Unbalanced Loads of Readers: The load of reader is defined as the number of tags located in its integration range. In Season, neighboring readers may have unbalanced loads. For example in Fig.5, reader r2 have more tags in its inter￾rogating regions than reader r1. At the beginning of Phase-I, two readers cannot collect tag t0 due to the reader collision. However, reader r1 complete running Season-I earlier than r2 and then stopping interrogating. Then the reader r2 is able to collect t0 since it is still running Season-I. In this case, some contentious tags may be collected in Phase-I and cause confusion to the joint identification in Phase-II. We introduce session number to solve this problem. Each tag contains a session number with the initialized value as zero. At the beginning of Season, each reader randomly generates a non-zero session number and broadcasts it. If a tag can resolve a session number, it must be non-contentious. Then the tag changes its session number to what it receives. In Phase-I, the reader sends query commands with the non-zero session number. Each tag only replies the query command with same session number as it holds. In this way, the readers can only collect non-contentious tags during Phase￾I. In Phase-II, the reader broadcasts query commands with a unified session number of zero. Because the contentious tag’s session number has not changed during Phase-I, they will reply to the query command. Another consequence produced by unbalanced loads of readers is that the readers with lower loads will wait for the ends of the readers with higher loads during Phase-I. We can simply switch Phase-I and Phase-II to shorten such delay. Namely, readers first jointly identify contentious tags and then identify their own non-contentious tags. 2) Source Sensitive and Insensitive: RFID application can be summarized into two categories. One is source-insensitive, in which the source, namely the ID of reader that detected the tag, is not concerned. The user may only want to confirm that all tags can be collected, for example in warehouse monitoring. Another is source-sensitive, in which tags must be exactly reported multiple times, for example the object tracking. In the second type of applications, a tag can be approximately located by recording the readers that collect the tag. Duplicate reports of a tag from neighboring readers can also help the administrator to re-deploy readers for better coverage. Season can well support both the source insensitive and sensitive applications. For source-insensitive application, Season allows passive readers to immediately send a “FINISH” message to their neighboring active readers without identifying its contentious tags. IV. PERFORMANCE EVALUATION We now evaluate Season using real-world logistics and tracking traces. A. Evaluation Methodology 1) Testbed and deployment: To validate the feasibility of joint identification, we use a NI PXI-1044 RFID testing tool with PXI 5600 receiver as our passive reader. We uniformly set the power of antenna as 20 dBm which supports around an interrogation range of 2m. We also deploy five readers in a logistics enterprise, Xi’an postal processing center in Shaanxi, China. The center is the one of the seven largest postal processing centers in China. It covers an area of about 16,128m2 and contains 30 importing/exporting gates. Fig.7 shows the architectural plans of the center. We attach more than 100 passive tags into pouches and find that the percentage of contentious tags is less than 10% for a stable and full coverage. 2) Simulating Real RFID Applications: For simulation, we use two typical application scenarios and three random reader topologies described as follows. Warehouse: According to our measurement results in Xi’an postal center, we simulate a total of 12 ∗ 6 = 72 readers for covering the entire center in a square-grid formation. Each reader is located at one vertex in the grid. Each reader has an interrogating range of 7m, which has 126 contentious regions