1. Every Bit Counts – Fast and Scalable RFID Estimation
Title: <catchy word play> - <what the paper is about>
Every Bit Counts – Fast and Scalable RFID Estimation
Muhammad Shahzad and Alex X. Liu
Michigan State University, USA
MobiCom 2012
Course: Advanced Topics on the Internet of Things
Presented by Jeroen Delvaux (student ID )

2. Problem: RFID Cardinality Estimation
= counting tags
ID = …01
We assume 1 stationary reader
RFID tag
range
The overhead of an exact count is fundamentally high
RFID reader
Goal: Estimate population size as fast as possible
Jeroen Delvaux RFID Cardinality Estimation - Introduction
Slide 02/12 (+1)

3. Why Counting RFID Tags? Example use cases
Assumption: no need to identity individual tags
Supermarket – product inventory & restocking
Logistics & Distribution – monitor manufactured, packaged and shipped quantities
Music festival – crowd management
Jeroen Delvaux RFID Cardinality Estimation - Introduction
Slide 03/12 (+1)

4. Presentation Outline Problem + Motivation Proposed solution + Review
Anti-Collision Protocols (preliminaries)
Tags should not send data simultaneously
Problem + Motivation
Proposed solution + Review
“i am 95% sure that the number of tags is between 410 and 430”
Related work + Conclusion
Jeroen Delvaux RFID Cardinality Estimation – Anti-Collision
Slide 04/12 (+1)

5. Anti-Collision Protocols
We are not supposed to all speak at the same time
frequency-based
time-based
location-based
ALOHA + variants
Tree search + variants
Our scope
Narrow down IDs until only one tag replies
Hawaiian: ‘peace’, ‘affection’ English mutation: ‘hello’, ‘goodbye’
ID = 01****** = 011***** = 0110****
Randomize the waiting time for tags to reply
Jeroen Delvaux RFID Cardinality Estimation – Anti-Collision
Slide 05/12 (+1)

6. Framed Slotted ALOHA
Randomized waiting time ∈ {1,2,3,4} for each tag to send 1 data packet
Standardized by EPCGlobal (Class 1 Generation 2 tags)
slot
Electronic Product Code
frame
frame A A
data
data
rand = 3 B B
data
data
rand = 1 C
data
data D D
data
data
rand = 4 1 ≥2 1 1 1 ≥2
all
rand = 3
time C
Reader accepts data if exactly one tag replies;
reject in case of a collision
Jeroen Delvaux RFID Cardinality Estimation – Anti-Collision
Slide 06/12 (+1)

7. Framed Slotted ALOHA (continued)
Random number derivation, at the start of each frame: rand = Hash(seed, ID)
frame
frame … 1 ≥2 1 1 1 ≥2 A
rand = 3 B
Reader-driven protocol (synchronous)
rand = 1 D
‘Democratic’: each tag has an equal probability of getting its data accepted by the reader
broadcast seed
rand = 4
rand = 3
Main parameter: frame size (#slots) C
Jeroen Delvaux RFID Cardinality Estimation – Anti-Collision
Slide 07/12 (+1)

8. Presentation Outline Problem + Motivation
Anti-Collision Protocols (preliminaries)
Tags should not send data simultaneously
Problem + Motivation
ALOHA
Proposed solution (builds upon ALOHA) + Review
“i am 95% sure that the number of tags is between 410 and 430”
Related work + Conclusion
Jeroen Delvaux RFID Cardinality Estimation – Proposed Solution
Slide 08/12 (+1)

9. Proposal for Counting RFID Tags
Averaged over 10 frames. In paper, 100 frames (smoother).
ALOHA Frame with 16 slots
run length
0 tags
saturation
25 tags 1 2+ 1 2+ 1 2+ 2+ 1 2+ 1 2+ 2+ 2+
error
100 tags 2+ 2+ 2+ 2+ 1 2+ 2+ 2+ 2+ 2+ 2+ 2+ 2+ 2+ 2+ 2+
inverse curve = job done!
The more tags, the more non-empty slots
Estimator: the averaged length of sequences of non-empty slots (runs)
error
MATLAB simulation 2+ 1 2+ 2+ 2+
Number of tags
Jeroen Delvaux RFID Cardinality Estimation – Proposed Solution
Slide 09/12 (+1)

10. Estimator seems Crude Does not distinguish ‘1’ and ‘2+’
Not discussed in the paper
Does not distinguish ‘1’ and ‘2+’
Authors provide formal mathematical analysis of their estimator (30 equations) 2+ 2+ 2+ 2+ 1 2+ 2+ 2+ 2+ 2+ 2+ 2+ 2+ 2+ 2+ 2+
more tags 2+ 1 2+ 2+ 1 2+ 2+ 1 1 1 2+ 1 2+ 2+ 1 1
less tags
1 run having 16 slots => same number of tags
Overlooks that slots are independent events
However, the proposed estimator does not reach full potential and seems ‘arbitrarily’ chosen
Average run length = 5 => more tags 1 1 1 1 1
same number of tags 1 1 1 1 1
Average run length = 1 => less tags
Jeroen Delvaux RFID Cardinality Estimation – Proposed Solution Review
Slide 10/12 (+1)

11. Why Not a More Informative Estimator?
averaged over 10 frames
Inversion seems more reliable
Rather shaky curve
Run length (discussed in paper)
Slot type fractions (not in paper) ≥2
Smaller ripple 1
Number of tags
Number of tags
Jeroen Delvaux RFID Cardinality Estimation – Proposed Solution Review
Slide 11/12 (+1)

12. Conclusion & Related Work
[Chen et al., “Understanding RFID Counting Protocols”, MobiCom 2013]
Many RFID counting protocols have been proposed (10+)
Not necessarily build upon ALOHA nor having a stationary reader
The proposed solution does not reach full potential (yet)
The chosen estimator (run length) is rather crude
More attention needed regarding two-phase approach: coarse + refinement
Initial parameter values (frame size, tag reply probability)
More optimal parameter values
Rough estimate # tags
More accurate estimate # tags
Thank you! Questions?
Jeroen Delvaux RFID Cardinality Estimation – Conclusion & Related Work
Slide 12/12 (+1)

13. Scalability Trick
Measure large populations with a small frame size (maintain speed)
16 slots per frame 10 frames averaging
At the expense of accuracy loss (curves become less smooth)
Probability that a tag competes in a given frame ≠1 e.g., 1/4
Count up to 100 tags
Count up to 400 tags
Jeroen Delvaux RFID Cardinality Estimation – Proposed Solution
Slide (+1)