1. Anti-tag collision algorithms of RFID
Chun-Fu Lin

2. Agenda Taxonomy of tag collision protocols
Bi-slotted tree based anti-collision protocols
Adaptive binary splitting
I-code

3. Reader collision Tag collision Centralized Probability-based
Distributed
Probability-based
Deterministic-based (Prefix-based)

4. RFID research topics Hardware Collision Security & privacy Application
circuit, antenna design, etc.
Collision
Reader collision
Tag collision
Security & privacy
Application
CSI/CTPAT/WCO SAFE
Supply chain management, medical care, etc.

5. Taxonomy of tag anti-collision protocols by Dong-Her Shih et. al
Taxonomy of tag anti-collision protocols by Dong-Her Shih et. al., published in Computer Communications, 2006

6. SDMA (Space Division Multiple Access)
Reuse a certain resource, such as channel capacity in spatially separated area.
Reduce the reading range of readers and forms as an array in space.
Electronically controlled directional antenna
Various tags can be distinguished by their angular positions.

7. FDMA (Frequency Division Multiple Access)
Several transmission channels on various carrier frequencies are simultaneously available.
Tags respond on one of several frequencies.

8. CDMA (Code Division Multiple Access)
Too complicate and too computationally intense for RFID tags as well

9. TDMA (Time Division Multiple Access)
The largest group of RFID anti-collision protocols
Tag driven (tag talk first, TTF)
Tag transmits as it is ready
Aloha
SuperTag
Tags keep quiet and retransmit until reader acknowledges
Reader driven (reader talk first, RTF)
Polling, splitting, I-code, contactless

10. Polling Reader must have the complete knowledge (database) of tags
Reader interrogates the RFID tags by polling ‘‘whose serial number starts with a 1 in the first position?’’
Those tags meet this test reply “yes” while others remain
Slow, inflexible

11. Splitting method Tree algorithm Based on binary search tree algorithm
Each collided tag generates a random number by flipping an unbiased B-sided coin
B = 2, each collided tag would generate a number 0 or 1
The reader always sends a feedback informing the tags whether 0 packet, 1 packet, or more than 1 packet is transmitted in the previous slot.
Each tag needs to keep track of its position in the binary tree according to the reader’s feedback

12. R set responds first
L: set generates 1
R: set generates 0
S: single reply
Z: zero reply
C: collision

13. Query Tree
Prefix based
Tags match the prefix respond

14. I-Code
stochastic passive tag identification protocol based on the framed-slotted Aloha concept.
Each tag transmits its information in a slot that it chooses randomly based on the seed sent by the reader.
The reader can vary the frame size N, the actual size of a slot is chosen according to the amount of data requested

15. Approximation of N
The reader detects the number of slots by a triple of numbers c = (c0, c1, ck), where c0 stands for the number of slots in the read cycle in which 0 tags have transmitted, c1 denotes the number of slots in which a single tag transmitted and ck stands for the number of slots in which multiple tags are transmitted.
Lower bound method
Minimum Distance method: distance between read result c and the expected value vector of n

16. Various N values corresponding to specific ranges have
been found from experiments and tabulated
If n  [17, 27], both 32 and 64 are appropriate choices for N

17. Contact-less
Is based on the tree splitting methodology to identify one bit of the ID in every arbitration step
The tag uses the modulation scheme which identifies “0” in the specified bit position with 00ZZ (Z stands for no modulation) and “1” as “ZZ00”. In this way, the reader can recognize the responses from all the tags and divide the unidentified tags into 2 groups.

18. 1 1
Identified 1101

19. Related papers published in IEEE communications letters (I/F 1
Related papers published in IEEE communications letters (I/F 1.196) from 2006 ~ now
MARCH 2006, Adaptive Binary Splitting for Efficient RFID Tag Anti-Collision, Jihoon Myung, Student Member, IEEE, Wonjun Lee, Senior Member, IEEE, and Jaideep Srivastava, Fellow, IEEE
APRIL 2006, Enhanced Binary Search with Cut-Through Operation for Anti-Collision in RFID Systems, Tsan-Pin Wang
DECEMBER 2006, Bi-Slotted Tree based Anti-Collision Protocols for Fast Tag Identification in RFID Systems, Ji Hwan Choi, Student Member, IEEE, Dongwook Lee, and Hyuckjae Lee, Member, IEEE
JANUARY 2007, Optimized Transmission Power Control of Interrogators for Collision Arbitration in UHF RFID Systems, Joongheon Kim, Member, IEEE, Wonjun Lee, Senior Member, IEEE, Eunkyo Kim, Dongshin Kim, Student Member, IEEE, and Kyoungwon Suh, Student Member, IEEE
JANUARY 2007, Query Tree-Based Reservation for Efficient RFID Tag Anti-Collision, Ji Hwan Choi, Student Member, IEEE, Dongwook Lee, and Hyuckjae Lee, Member, IEEE
Tag collision
Reader collision

20. Ji-Hwan Choi, Dongwook Lee, Hyuckjae Lee, “Bi-slotted tree based anti-collision protocols for fast tag identification in RFID systems,” IEEE communication letter, December 2006

21. BSQTA
Reader sends n-1 bits prefix.
Tag that match the prefix will
Send ID from n+1 bit to end bit if its n-th bit is 0.
Wait for LENGTH-n bit duration if its n-th bit is 1.
If reader detects any collision occurs, it pushes the prefixes (prefix+0, prefix+1) into stack and repeats the exploring process.
BSCTTA (variation from BSQTA, tags send their ID from n+1 bit to the time that ACK signal received)

22. P:101
Reader 1
Tag1 1
Tag2
Wait |N-prefix| bits

25. Simulation One reader Number of tags increases from 2 to 65536
Tag IDs are randomly generated
ID length is not clear (believe to be 250)
Compared to QTA and CTTA

29. In another way of thinking
The tag collision problem can be referred to as a distributed systems problem : how to reach a consensus agreement (transmission slot) for every distributed node (tag) without interaction to each other?
The constraints will be
Tag is unable to detect collision
Tag has limited or no calculation capability
Tag has limited or no memory space
Tags can be moved in and out dynamically
Reader has no information of the number of tags
There could be not just one reader

30. Jihoon Myung, Wonjun Lee, Jaideep Srivastava, “Adaptive binary splitting for efficient RFID tag anti-collision,” IEEE communication letter, March 2006

31. Every tag maintains two local variables: Pc and Ac(i).
Pc, progressed-slot counter; number of tags recognized by the reader so far.
Ac(i), allocated-slot counter, time slot for tag i’s transmission.
Reader sends feedback (readable, idle, collision) to tags
According to reader’s feedback, each tag decides its transmission time of slot.
If feedback=readable, tag adds 1 to Pc.
If feedback=idle and Pc < Ac(i), tag decreases Ac(i) by 1
If feedback=collision and Pc < Ac(i), tag generates a random number of 0 and 1 and adds to Ac(i).

34. Binary tree protocol and query tree protocol are compared
Simulation results
Identification delay
Number of time slots required for all tags
Tag communication overhead
Average number of bits transmitted by a tag for identification
n: number of tags recognized in the last process
γ: number of staying tags (tsSr,i∩Sr,i-1)
α: number of arriving tags (taSr,i+1-Sr,i)
β: number of leaving tags (tlSr,i-Sr,i+1)
w: given, w=γ/n, α/n=β/n=1-w
Binary tree protocol and query tree protocol are compared
Sr,i: the set of all the tags recognized by reader r in the ith
Identification process

35. Stable for
static tags
w=0, No staying tags, arriving tags=leaving tags
w=1, static tags, no arriving and leaving tags

38. Harald Vogt, “Efficient Object Identification with Passive RFID Tags,” Proceedings of the International Conference on Pervasive Computing, April 2002, pp

39. I-Code Reader is attached to the serial interface of a host (PC)
Communication and power transmission between the readers and tags takes places by inductive coupling
All tags within reading range will answer requests from the reader
An I-Code tag provides 64 bytes memory
It employs a variant of slotted Aloha for access to the shared communication medium

40. Tag reading cycle I: what data is requested in memory
Rnd: random value  {0, 31} for tag’s function sT of randomization
N: frame size  {1, 4, 8, 16, 32, 64, 128, 256}
s: tag sends in slot s, 0≤s≤N
The result of a read cycle can be viewed as a triple of numbers <c0, c1, ck>

41. Mathematical Preliminaries Occupancy
Given N slots and n tags, the number r of tags in one slot is
The expected value of the number of slots with occupancy number r is given by

42. μr: the number of slots being filled with exactly r tags
Remaining arrangement

43. Tags reading as a Markov process1 2 3 n

44. The matrix Q is used to compute a lower bound of the number of reading steps necessary to identify all tags with a given probability.

45. How to estimate n?
Based on the results of read cycles c=<c0, c1, ck>, and the current value of N, the function that compute estimations of n is:
Error function: sums up the weighted errors over all possible outcomes of the read cycle

46. An problem to optimal value for the number of cycles
Small frame size  high collisions
Large frame size  high response time
Stochastic nature of reading process (frame slotted Aloha) can not guarantee 100% probability of identifying all tags
Compute the time to achieve a given assurance level α : values were obtained by performing read cycles for 1 min. and computing the average consumed time
tN : cycle time, also depends
on the connection speed
between reader and host

47. TN is nearly linear in N
For a fixed frame size N, the time Tα required to achieve an assurance level α is
S satisfies
If the optimal frame size is known, e.g. if n can be estimated correctly, then the identification time that meet the threshold α increase linearly with the number of tags
Min number of read cycles
Probability of identify k tags after s read cycles,
K = 1, 2, …, n. Choose its nth component and
Compare it to α

49. Two estimation functions
Lower bound  a collision involves at least 2 different tags
Distance between read result c and the expected value vector of n

50. Lower-bound is accurate for small n but grows fast with larger n.
e-dist is more steady

51. Due to the inaccuracy of the estimation functions and the jitter as shown in Fig.3, it is free to choose the actual frame size for a given estimate; ex. if n{17,27}, both 32 and 64 are appropriate choice for N.

52. Thank you