An RFID based Ubiquitous Architectural Framework for Mobile Object Tracking

1
An RFID based Ubiquitous Architectural Framework
for Mobile Object Tracking
University of Texas at
Arlington

• _at_

• CSE

• UTA

Pradip De Center for Research in Wireless
Mobility and Networking CReWMaN Department of
Computer Science and Engineering The University
of Texas at Arlington Arlington, TX 76019
2
PRESENTATION OUTLINE

• WHAT IS RFID ?
• APPLICATION AREAS
• MULTI-TAG READING COLLISION
• ARCHITECTURE AND TRACKING PROTOCOL
• DELAY ANALYSIS
• STORAGE STRUCTURE
• PRODUCT RECALL
• DISTRIBUTION ANALYSIS
• SIMULATION AND RESULTS
• FUTURE WORK
• REFERENCES

3
What is RFID?

• RFID as a technology..
• RFID TAGS (Transponders)
• Tags consist of an Integrated Circuit (IC) and
  an Antenna
• Tags can be as small as a grain of rice !
• Read Only or Read/Write or a combination of both
• Active or Passive
• Various Frequency ranges of operation
• RFID Readers (Transceivers)
• A Microcontroller based Interrogating Devices
• Equipped with standard interfaces to interact
  with backend
• computing host systems

4
RFID Applications..

• RFID provides a quick, flexible and reliable
  electronic
• means to detect, track and control a variety
  of items
• A Plethora of Application areas
• Pharmaceutical - embedded RFID tags in
  prescription bottles
• Amusement Parks - Location Stations
• Libraries and Video Stores Theft Detection and
  Misplacement
• Security Hands Free Access to secured areas
• Toll Payment at Roads and Bridges
• Logistics and Supply Chain Management is one of
  the major
• areas where this technology is deployed
• 
  

5

• Checked Baggage passes under
• scanner

• RF tags incorporated into retail tickets

6
Multiple Tag Reading - Collision

• What happens when multiple tags are in the
  readers range?
• All tags will become excited at the same time
• Reader cannot detect individual tags from each
  other
• A Collision Avoidance Algorithm necessary
• A Query Tree Protocol has been proposed by the
  MIT-AutoID
• Center
• Basic Idea is to sequentialize the reading of
  multiple tags
• Achieves an O(n) bound for reading n tags
• 2.881n 1 lt ETs lt 2.887n - 1

Ching Law, Kai Lee and Kai-Yeing Siu,
Efficient Memoryless Protocol for Tag
Identification. MIT AutoId
Center, October 2000
7
Architecture

• Objective of the Architecture
• Basically a Distributed Tracking Architecture
  for Objects that are transacted between
  organizations
• To harness RFID technology to provide automatic
  visibility and control over transaction items
• Minimize delay and error incurred due to the
  involvement of human attendance

8
Architecture..

• A Few Terminologies defined
• Electronic Product Code
• 96 bits
• Hierarchical
• Savant A Data Routing Server
• Data Capture, Data Monitor and Data Transmission
• Physical Markup Language (PML)
• Common Language for describing physical Objects
• Based on the syntax of XML
• Every property of the object captured

9
Architecture..

• A Few Terminologies defined
• PML Server
• Repository for all types of information
  regarding the objects
• Homes both static and dynamic information
• SLA Server
• Used to translate queries into decisions
• The execution is done at the PML Server
• Object Naming Service (ONS)
• Similar in structure to the Internet DNS
• Maps the EPC to the IP Address of its Home PML
  Server

10
Tracking ArchitectureA High Level View
11
Object Tracking Protocol Overview

• Reader reports a triplet to the attached Savant
• ltEPC, timeStamp, Reader_Idgt
• Based on ATTACH/DETACH messages accordingly
  destined for the upstream Home PML Server for
  updating location
• On recognition, Savant recieves an ACCEPT
  message back
• Three readings confirming that an EPC has left
  the field makes the Savant generate a DETACH
  message for that EPC
• Unrecognized ACCEPT message can also come back
  from the Home PML Server

12
Protocol Flow
Savant State Transition
Reader State Transition
13
Protocol Flow
PML Server State Transition
14
Protocol Flow
Home PML Server State Transition
15
Delay Analysis

• Delay incurred in sending n EPC tags from a
  reader to the
• Home PML Server
• Assumptions made
• The connection from the Reader onwards to the
  Network is assumed to follow standard Internet
  Protocols
• The Reader uses the Query Tree Protocol to read
  the tags
• The inter-arrival time of the packets from
  various Readers to the connected Savant follow a
  negative exponential distribution
• The size of the packets arriving at the Savant
  and PML Server also follow a negative exponential
  distribution.
• An arriving packet containing all stale tags is
  dropped at the Savant

Ching Law, Kai Lee and Kai-Yeing Siu,
Efficient Memoryless Protocol for Tag
Identification. MIT AutoId
Center, October 2000
16
Delay Analysis

• The total delay equation is given by
• Ttotal ETreader ETRd-Savant
  ETSavant-PML ETremotePML
• ETreader O(n) 2.88 n
• ETRd-Savant Txdelay ETsavant / (1 - ?)
• ? is the average occupancy of the Savant
• ETsavant p Ts1 (1 - p) Ts2
• p is the probability that the packet would not
  be dropped
• Ts1 and Ts2 are the average service times for
  both types of packets

17
Delay Analysis

• ETSavant-PML p (Txdelay ETPML / (1 -
  ?) )
• ETPML q Tp1 (1 - q) Tp2
• q is the probability that this PML server is
  the Home for all the tags
• in the packet
• Tp1 and Tp2 are the service times for the two
  types of packets at the PML Server
• ETremotePML p (1 - q) (avginternetDelay
  ETrPML / (1 - ?) )
• ETrPML TRP
• TRP is the service time at the remote Home PML
  server

18
Delay Analysis Results

• The average Internet delay for a packet varies
  between 10 and 100 msec
• T/TCP protocol assumed to transmit data
• 3 packets sent to complete one transaction
• 96 bits for each of the EPC and Reader ID and a
  32 bit timestamp value
• Assuming a 1 Mbps data rate we get delays
  ranging from 24 to 113 msec for transferring 100
  to 500 tags.
• average time to read a single tag has been found
  to be around 0.47 sec

Andrew Corlett, D. I. Pullin and Stephen
Sargood, Statistics of one-way Internet Packet
Delays. 53rd IETF, Minneapolis,
March 18, 2002 http//www.spec.com/PipelineJu
ly99.pdf Sep 1, 2003
19
Delay Analysis Results
20
Delay Analysis Results
21
Storage Structure

• The need for an efficient storage structure for
  location information about the EPCs cannot be
  overlooked
• Various forms of balanced search trees have been
  used to store information of ordered elements -
  provides access in logarithmic time
• level linked 2-3 trees (for example)
• We use a degree balanced k-ary search tree for
  the EPCs and use Finger Search for retrieval and
  updation

G. S. Brodal. Finger Search Trees with
constant insertion time. In Proc. 9th Annual
ACM-SIAM Symposium on Discrete Algorithms, pages
540-549, 1998. Guy E. Blelloch et al. Space
Efficient Finger Search on Degree-Balanced Search
Trees. Symposium on Discrete Algorithms, 2003.
22
Storage Structure

• EPCs are generally looked up in bulk!
• Assumed to be clustered together if not
  contiguous!
• Finger Search takes O(log d) worst case time
• Finger Search is accomplished using a
  right-parent stack as an extra data structure
  consuming O(log n) extra space

23
Storage Structure
24
Storage Structure
25
Storage Structure

• The information structure associated with each
  EPC
• 96 bit EPC tag
• Pointer to PML file at Home PML Server (H-PML)
• List of location history and Timestamp
• Time Stamp
• PML Server Address
• Savant Address
• Reader ID

26
Product Recall

• An important domain in need for automation
  badly !
• Object Distribution to the destination is not
  the end of the road!
• Retained location information history needs to
  be leveraged!
• A simple strategy is to traverse the footsteps
  of the Distribution

27
Product Recall

• With networked information available we devise a
  distributed scheme for handling the recall of
  objects
• Every PML Server would be equipped with Recall
  Handling procedures!
• The Recall Algorithm at each node comprises of
• Recall Initiation Algorithm (RI)
• Recall Handling Algorithm (RH)
• The H-PML invokes RI and the subsequent PML
  servers to which the recall percolates invoke RH

28
Product Recall
29
Product Recall
30
Product Recall
31
Product Recall
Recall Multicast
Recall Trie Built
32
Distribution Analysis

• A theoretical analysis of the EPC distribution
  in the PML server network The motivation is to
  get a handle on the number of messages generated
  in both a distribution and a Recall
• The Distribution network among the PML servers
  is Scale Free
• We model the distribution dynamics based on the
  way an infection spreads in a population
• The basic model for the number of susceptibles
  and infected gives
• N(t) X(t) Y(t)

33
Distribution Analysis

• The differential equations for the spreading
  dynamics are
• where
• where

34
Distribution Analysis

• The average rate of secondary infections given
  by
• The final fraction of PML servers to which EPCs
  spread is
• given by
• 
  
• where
• 
  

35
Distribution Analysis

• The distribution of the node degree comes to

from the total probability equation

• Similarly we have

• Using the above results we get

• By some mathematical manipulations we can
  express F in terms of the exponential integral as

where
36
Distribution Analysis
37
Distribution Analysis
38
Distribution Analysis
39
Simulation Results

• Two phases
• EPC Distribution Phase
• EPC Recall Phase
• In the first phase we populate fields of the
  k-ary search tree with simulated values
• In the second phase the Recall Trie is built for
  the EPC set to be recalled

40
Simulation Results
41
Simulation Results

• The first plot shows the average number of
  messages/tag to perform a recall against the
  number of tags in each recall. The different
  curves are for different values of tags
  distributed.
• The second plot shows the average number of
  messages/tag needed to perform a recall against
  the number of tags distributed. The different
  curves are for different values of tags recalled.
  
• The results stabilize to a very low value for the
  average number of messages required for
  performing the recall !

42
Future Work

• Extend the protocol to incorporate much needed
  security issues
• Enhancement of the architecture to suit
  scenarios where surveillance and tracking is
  critical and intelligently predict intrusion or
  unwanted activities in such environments

43
References

• The Association for Automatic Identification and
  Data Capture Technologies.
  http//www.aimglobal.org/technologies/rfid/
• RFID Journal. http//www.rfidjournal.com/
• Kai-Yeing Siu, Ching Law and Kayi Lee, Efficient
  Memoryless Protocol for Tag Identification. MIT
  Auto-Id Center, October, 2000
• MIT Auto-ID Center Publications at
  http//www.autoidcenter.org/
• R. Bridgelall. Enabling mobile commerce through
  pervasive communications with ubiquitous RF tags
  WCNC, 2003
• V. Stanford. Pervasive computing goes the last
  hundred feet with RFID systems Pervasive
  Computing, IEEE, Volume 2, Issue 2, April-June
  2003 Pages 9-14
• G. S. Brodal. Finger search trees with constant
  insertion time. In Proc. 9th Annual ACM-SIAM
  Symposium on Discrete Algorithms, pages 540-549,
  1998.
• Guy Blelloch et al. Space Efficient Finger Search
  on Degree- Balanced Search Trees. Symposium on
  Discrete Algorithms, 2003.
• R. Albert, H. Jeong and A. L. Barabasi, Nature
  (London) 401, 130 (1999)
• R. M. Anderson and R. M. May, Infectious Diseases
  of Humans Dynamics and Control(Oxford Univ.
  Press, Oxford, 1991).
• R. M. May and R. M. Anderson, Philos. Trans R.
  Soc. London, Ser. B 321, 565 (1988).
• The Savant - Version 0.1 (Alpha), Technical
  Manual. Oat Systems and MIT Auto-Id Center,
  February, 2002
• David L. Brock, The Physical Markup Language - A
  Universal Language for Physical Objects. MIT
  Auto-Id Center, February, 2001
• Mark Harrison et al, PML Server Developments,
  White Paper. MIT Auto-Id Center, June, 2003