1. Expecting the Unexpected: Fast and Reliable Detection of Missing RFID Tags in the Wild
Muhammad Shahzad Alex X. Liu
Dept. of Computer Science and Engineering
Michigan State University
East Lansing, Michigan, 48824, USA

2. 2011: Retailers lost 34.5 billion USD
Motivation
Shop Lifting
Employee Theft
2011: Retailers lost 34.5 billion USD

3. Problem Statement Input Objective Set of IDs of expected tags
RFID tag population containing:
some or all expected tags
unexpected tags
Threshold on number of missing tags, T
Required reliability, α ∈ [0,1)
Objective
Detect the event: missing tags ≥T
Event detection probability ≥α
Minimize detection time

4. Limitations of Prior Art
Assume there are no unexpected tags
ICDCS 2008: How to monitor for missing RFID tags; Tan, Sheng, and Li
MobiHoc 2010: Identifying the missing tags in a large RFID system; Li, Sheng, and Lin
SECON 2011: Fast identification of the missing tags in a large RFID system; Zhang, Liu, and Sun
IEEE ToC 2013: Completely pinpointing the missing RFID tags in a time-efficient way; Liu et. al.
However, in reality, there are unexpected tags
Airline baggage
Multi-tenant warehouse

5. Naïve Solutions Identification protocol Estimation protocol
Slow: fastest RFID identification protocol is 14.3 times slower compared to our scheme
SIGMETRICS 2013: Probabilistic Optimal Tree Hopping for RFID Identification; Shahzad and Liu
Estimation protocol
Inaccurate: if new tags join, can not tell whether some tags went missing

6. Communication Protocol Overview1 2 3 4 5 6 7 1 C 1 1 C 1 1 3 2 6 4 7 4
Frame size 𝑓=7
Seed 𝑅
Faster to distinguish between empty and non-empty slots
Singleton and collision » non-empty
At the end of frame, reader gets a sequence of 0s and 1s
011C011 becomes

7. RUN: Missing Tags Detection1 1 4 11 5 6
Expected tags to be monitored 1 2 3 4 5 6 7 8 9 10 11
Frame size 𝑓=11
Seed 𝑅 1
Pre-computed frame 1
Executed frame
Unexpected false
positive
Unexpected tag
detected
Missing tag event
detected
Unexpected tags 8 4 10 10

8. RUN: Handling Unexpected FPs
Repeat frame 𝑛 times 1 1 1 1 1 1

9. RUN: Parameter Selection
Three unknown parameters
Frame size 𝑓
Number of frames 𝑛
Persistence probability 𝑝
Two equations
𝑓= 𝑝(𝑇−|𝐸|−|𝑈|) ln {1−𝑞 }− ln {𝑝}
where 𝑞= (1−α) 1/𝑛𝑇
obtained using the expression of false positive probability
|𝐸| : number of expected tags
|𝑈| : number of unexpected tags
𝑝=(1−𝑞) (1−α) 𝑞 𝑛𝑇(𝑞−1)
Obtained using the required reliability condition
Need the number of unexpected tags |𝑈|

10. RUN: Estimating Unexpected Tags1 1 4 11 5 6 1 2 3 4 5 6 7 8 9 10 11 8 4 10 10
Number of total slots in frame 𝑓
Number of grey slots in frame 𝑘
Number of white slots that become green slots: 𝑁 01
𝑈 =− 𝑓 𝑝 ln 1− 𝑁 01 𝑓−𝑘

11. RUN: Experimental Evaluation
Implemented 4 protocols in addition to RUN
TRP (ICDCS, 2008)
IIP (MobiHoc, 2010)
MTI (SECON, 2011)
SFMTI (IEEE ToC, 2013)
TH (SIGMETRICS, 2013)

12. Actual Reliability vs. Missing Tags
Number of expected tags = 1,000
Number of unexpected tags = 10,000

13. Actual Reliability vs. Unexpected Tags
Number of expected tags = 1,000
Number of missing tags = 200

14. Effect of Threshold T Number of expected tags = 1,000
Number of unexpected tags = 10,000
Threshold = 200
Required reliability = 0.99

15. RUN vs. RFID Identification
Compared RUN with TH (SIGMETRICS 2013)
RUN is 14.3 times faster than TH for
Number of expected tags = 1,000
Number of unexpected tags = 10,000
Threshold = 200
Required reliability = 0.99
TH is faster than RUN when
Required reliability > , OR
Threshold < tags, which is impossible

16. Conclusion
Proposed a protocol to reliably detect missing tags in presence of unexpected tags
Reliable
Fast
C1G2 compliant
Handles multiple readers

17. Questions?