Probabilistic Optimal Tree Hopping for RFID Identification Muhammad

Probabilistic Optimal Tree Hopping for RFID Identification Muhammad Shahzad Alex X. Liu Dept. of Computer Science and Engineering Michigan State University East Lansing, Michigan, 48824, USA

RFID is everywhere Muhammad Shahzad 2

Radio Frequency Identification 0000 0011 Muhammad Shahzad 0101 0110 1001 1010 1011 1101 3

Tree Walking (EPCGlobal Standard) 0 1 1 8 3 001 010 4 6 5 9 011 7 16 100 11 11 10 12 14 13 Number of queries: 16 Muhammad Shahzad 1011 000 10 1001 1010 2 01 1000 00 15 4

Optimizing Tree Walking § Total queries = successful + collisions + empty § Minimize total queries Muhammad Shahzad 5

Limitations of Prior Art § All prior work proposes heuristics to reduce identification time ─ Mobi. Hoc’ 06, Per. Com’ 07, INFOCOM’ 09, ICDCS’ 10 § No formal model of the Tree Walking process ─ No optimality results Muhammad Shahzad 6

Our Modeling of Tree Walking Position p § Level l § m=4 § n=16 § (Hypergeometric distribution) § Muhammad Shahzad 7

Proposed Approach 1. Estimate unidentified tag population size 2. Find optimal level and the first unvisited node 3. Perform Tree Walking. Go to step 1 Muhammad Shahzad 8

Population Size Estimation § First time estimation: rough, but fast ─ We adapt a fast scheme proposed by Flajolet and Martin in the database community in 1985. ─ Did not use accurate RFID estimation schemes § Subsequent estimation = estimated tags - identified tags Muhammad Shahzad 9

Calculating Optimal Level § § § Muhammad Shahzad 10

Muhammad Shahzad 11

Tree Hopping vs. Tree Walking Muhammad Shahzad 12

Tree Hopping Example 11 11 2 3 011 4 100 6 5 101 7 9 8 1011 010 1001 1010 1 001 1000 10 Number of queries: 11 (compared to 16 of TW) Muhammad Shahzad 13

Experimental Evaluation § Implemented 8 protocols in addition to TH 1. 2. 3. 4. 5. 6. 7. 8. BS (IEEE Trans. on Information Theory , 1979) ABS (Mobi. Hoc, 2006) TW (DIAL-M 2000) ATW (Tanenbaum, 2002) STT (Infocom, 2009) MAS (Per. Com, 2007) ASAP (ICDCS 2010) Frame Slotted Aloha (IEEE Transactions on Communications, 2005) Muhammad Shahzad 14

Improvement of TH over prior art § Uniformly distributed populations ─ Total number of queries: 50% ─ Identification time: 10% ─ Average responses per tag: 30% § Non-uniformly distributed populations ─ Total number of queries: 26% ─ Identification time: 37% ─ Average responses per tag: 26% Muhammad Shahzad 15

Normalized Queries Muhammad Shahzad 16

Identification Speed Muhammad Shahzad 17

Normalized Collisions Muhammad Shahzad 18

Normalized Empty Reads Muhammad Shahzad 19

Conclusion § First effort towards modeling the Tree Walking process § Proposed a method to minimize the expected number of queries § More in the paper ─ Method to make TH reliable in the presence of communication errors ─ Continuous scanning of dynamically changing tag populations ─ Multiple readers environment with overlapping regions § Comprehensive side-by-side comparison of TH with 8 major prior tag identification protocols Muhammad Shahzad 20

Questions? Muhammad Shahzad 21