1. Every Bit Counts – Fast and Scalable RFID Estimation Muhammad Shahzad and Alex X. Liu Dept. of Computer Science and Engineering Michigan State University East Lansing, Michigan, 48824 USA

2. 2 Radio Frequency Identification Mobicom 2012 Chip Antenna ActivePassive

3. 3 Radio Frequency Identification Mobicom 2012

4. 4 RFID Estimation Exact IDs can not be read due to privacy requirements Mobicom 2012 Exact IDs are not required but only a count Identification protocols can use the count to speed up identification process

5. 5 Problem Statement  Input ─ Confidence interval β ∈ (0,1] ─ Required Reliability α ∈ [0,1)  Output ─ An estimate t e of tag population size t such that ● 1- β ≤ t e / t ≤ 1+ β ● P{ 1- β ≤ t e / t ≤ 1+ β } ≥ α Mobicom 2012

6. 6 Additional Requirements  Single Reader environment  Multiple reader environment with overlapping regions  C1G2 standard compliant tags  Active tags and Passive tags  Scalable Mobicom 2012

7. 7 Why do we need a new protocol?  Non compliance with C1G2 standard  Non-scalable  Inability to achieve required reliability  Room for improvement in speed Mobicom 2012

8. 8 Communication Protocol Overview Mobicom 2012 011C011 frame size f =7 326447  Faster to distinguish between empty and non-empty slots  Slower to distinguish between empty, singleton, and collision  Singleton and collision » non-empty  At the end of frame, reader gets a sequence of 0s and 1s ─ 011C011 becomes 0111011 1234567 011C011

9. 9 Estimation  Any measure which is a monotonous function of t can be used for estimation ─ Number of 1s in a frame ─ Number of 0s in a frame  Any measure which is a monotonous function of t can be used for estimation ─ Number of runs of 1s ─ Number of runs of 0s  Any measure which is a monotonous function of t can be used for estimation ─ Average run size of 1s ─ Average run size of 0s Mobicom 2012  011100  0  111  00

10. 10 Useable Measures Mobicom 2012  Average run size of 1s  Number of 1s  Number of 0s  Number of runs of 1s  Number of runs of 0s  Average run size of 1s  Average run size of 0s

11. 11 ART Protocol Mobicom 2012 0111011 frame size f = 7 326447 1234567 0111011 Repeat frames n times Calculate avg. run size of 1s from n frames Obtain the estimate

12. 12 Scalability Problem Mobicom 2012 0001111CCCCCCCC

13. 13 Scalability Problem Addressed Mobicom 2012 001C  Use persistence probability p frame size f = 4/p = 16 = 0.25 83161211 2531279 5  Obtain the estimate using information from this frame  Tags follow a uniform distribution  Extrapolate with the factor of p

14. 14 Optimization  The expression for number of rounds n depends on ─ Confidence interval β ─ Required Reliability α ─ Frame size f  n = func( α, β, f ) Mobicom 2012  Estimation time ∝ f × n ─ d/df (f ×n ) = 0  Two equations 1. n = func( α, β, f ) 2. d/df (f ×n ) = 0  Two unknowns 1. Number of rounds n 2. Frame size f

15. 15 Multiple Readers Environment  First proposed by Kodialam et. al. in “Anonymous tracking using RFID tags” Mobicom 2012 frame size f = 4 f =4 R f =4 R 2 2 3 1 1 10101110 Seed R Logical OR 101011101110

16. 16 Advantages of ART over prior art  Speed: ─ 7 times faster than fastest ● β = 0.1%, α = 99.9%  Deployability ─ Does NOT require modifications to ● tags ● communication protocol Mobicom 2012

17. 17 Performance Evaluation Mobicom 2012

18. 18 Performance Evaluation Mobicom 2012

19. 19 Performance Evaluation Mobicom 2012

20. 20 Conclusion  New estimator: the average run size of 1s  Faster than existing estimation schemes ─ smaller variance  Single and multiple reader environment  C1G2 standard compliant tags  Active tags and Passive tags  Scalable: independent of tag population size Mobicom 2012

21. 21 Questions? Mobicom 2012