MoteTrack: A Robust, Decentralized Approach to RF Based Location Tracking

1
MoteTrack A Robust, Decentralized Approach to RF
Based Location Tracking

• Paper Presentation
• CSE 535 mobile computing
• Weijia Che
• Phd student, CSE Dept, ASU

2
Paper Selection

• Title MoteTrack A Robust, Decentralized
  Approach to RFBased Location Tracking
• Authors Konnrad Lorincz and matt Welsh
• Published tech. report TR-19-04, Division of
  Eng. and Applied Sciences, Harvard Univ., 2004.

3
Agenda

• Motivation Scenario
• Background and Related Work
• MoteTrack Overview
• Robust Design
• Implementation
• Evaluation
• Novelty and Drawbacks
• Relationship with our Project
• References

4
Motivation Scenario

• Firefighters entering a large building
• Heavy smoke coverage
• No priori notion of building layout
• Indications
• Centralized approaches not suitable (central
  server/users roaming node may be destroyed)
• Approaches require whole-network wireless
  connectivity not suitable (large num of wireless
  access points may have failed)

5
Background and Related Work

• Indoor Localization based on different context
• Infrared
• Ultrasound
• RF-RSSI

6
Indoor Localization based on Infrared

• Eg. Active Badge 1
• Advantage
• suitable for both indoor and outdoor use
• Disadvantage
• Many receiver nodes are required due to short
  range of infrared signals
• Require line-of-sight exposure
• Suffer errors in the presence of strong light

7
Indoor Localization based on ultrasound

• Eg. Cricket 2,3 and Active Bat 4
• Advantage
• Higher accuracy
• Disadvantage
• Requires of accurate synchronization of the
  sensor nodes
• Requires line-of-sight exposure
• Requires careful orientation of the receivers

8
Indoor Localization based on RF

• Eg. RADAR5
• Advantage
• No additional hardware is required except for the
  sensor nodes
• Low power, inexpensive, easy to deploy
• Disadvantage
• Signal strength are generally unstable
• Vary over time
• Affected by other factors (building structure,
  people moving around

9
RF Indoor localization -triangulation

• Model signal propagation together with current
  RSSI to triangulate the position of a sensor node
• advantage
• No requirement of pre-setup database
• disadvantage
• Requires detailed models of RF propagation
• Does not account for variations in receiver
  sensitivity and orientation

10
RF Indoor localization -fingerprinting

• Use empirical measurements of RSSI to set up a
  database and together with current RSSI to
  estimate the position of a sensor node
• advantage
• No need for detailed models of RF propagation
• disadvantage
• An offline calibration to set up the database is
  required

11
MoteTrack Overview
12
Two Phases of Estimate

• Offline collection of reference signatures
• Reference signature format?
• Online location estimation

13
Online location estimation

• Estimation steps
• I, Compute the signature distances
• II, Option 1, take the centroid of the geographic
  location of the k nearest reference signatures
  (weighting with the signature distances).
• II, Take the centroid of the geographic location
  of the nearest reference within some
  ratio(weighting with the signature distances).

(NOTEC is constant, gained from
experiment 1.11.2 works well)
14
Robust Design

• Definition of robustness
• Graceful degradation in location accuracy as base
  stations fail
• Resiliency to information loss (poor antenna
  orientation)
• Work well with perturbations in RF (people moving
  around, movement of furniture, opening or closing
  of doors, solar radiation )
• No single point of failure (no central server)

15
Robust Design

• Challenges
• For decentralization consideration, beacon nodes
  should perform localization estimation, which
  leads to questions about the required resources
  and cost of the base stations to be answered.
• In order for the technique to be resilient to
  loss of information, the system should be able to
  detect beacon failure and able to handle it

16
Robust Design

• Methodology
• Decentralized location estimation protocol
• GOAL compute the mobile nodes location in a way
  that only relies upon local communication and at
  the same time to achieve low communication
  overhead.
• Distributing the reference signature database to
  beacon nodes
• GOAL ensure balanced distribution of reference
  signatures (improve robustness) while attempting
  to assign reference signatures to their closest
  beacon nodes (guarantee accuracy)
• Adaptive signature distance metric
• GOAL handle beacon failures

17
Decentralized location estimation protocol

• TRY_1 k beacon nodes send their reference
  signature slice
• mobile node acquires its signature s by listening
  to beacon nodes
• mobile node broadcasts a request for reference
  signatures and gathers the slices of the
  reference database from k nearby beacon nodes
• The mobile node then computes its location using
  the received reference signatures
• Advantage
• very accurate
• Disadvantage
• requires a great deal of communication
  overhead
• Alternative contacting nltk nearby beacon nodes
  and ask each one only send m reference signatures
  that are closest to s

18
Decentralized location estimation protocol

• TRY_2 k beacon nodes send their location
  estimate
• mobile node acquires its signature s by listening
  to beacon nodes
• mobile node broadcasts its signature s to k
  nearby beacons
• the beacon node then computes the mobile nodes
  location estimate and sends it back
• mobile node receives K estimate and compute the
  final estimate with these values (centroid of
  centroids)
• Advantage
• less communication overhead
• Disadvantage
• does not produce accurate location
  estimates

19
Decentralized location estimation protocol

• FINAL-SOLUTION Max-RSSI beacon node sends its
  location estimate
• mobile node acquires its signature s by listening
  to beacon nodes
• mobile node broadcasts its signature s to the
  beason with the strongest RSSI
• the beacon node computes the mobile nodes
  location estimate and sends it back
• Advantage
• less communication overhead
• as long as the beacon stores an appropriate slice
  of reference signature database, this should
  produce very accurate results

20
Decentralized location estimation protocol

• FINAL-SOLUTION Max-RSSI beacon node sends its
  location estimate
• mobile node acquires its signature s by listening
  to beacon nodes
• mobile node broadcasts its signature s to the
  beason with the strongest RSSI
• the beacon node computes the mobile nodes
  location estimate and sends it back
• Advantage
• less communication overhead
• as long as the beacon stores an appropriate slice
  of reference signature database, this should
  produce very accurate results

21
Distributing the reference signature database to
beacon nodes

• Greedy distribution algorithm
• maxRefSigs specifies the maximum signatures each
  beacon node will store
• For each reference signature, the beacon accepts
  and stores it if
• The current reference signature num is less than
  maxRefSigs
• The new reference signature contains a higher
  RSSI (average) value than one of the stored
  signature
• Advantage
• simplicity and no requirement for global
  knowledge or coordination between nodes
• Disvantage
• some reference signatures may be stored
  many times with some other not stored at all

22
Adaptive signature distance metric

• Greedy distribution algorithm
• Always stores the reference signature with the
  strongest RSSI to the beacon node.
• Advantage
• simplicity and no requirement for global
  knowledge or coordination between nodes
• Disvantage
• some reference signatures may be stored
  many times with some other not stored at all

23
Distributing the reference signature database to
beacon nodes

• Balanced distribution algorithm
• Variant of a stable marriage algorithm
• refer to algorithm design Jon Kleinberg
  for details
• Advantage
• ensure balanced distribution of reference
  signatures while attempting to assign reference
  signatures to their closest beacon nodes
• Disadvantage
• requires global knowledge of all reference
  signature and beacon node pairings
• individually update of beacon nodes is
  impossible
• Note both of those two algorithms are
  implemented and examined in this paper

24
Adaptive signature distance metric

• Bidirectional signature distance metric

Indicates mobile nodes signature is taken at a
different place rather than place of reference
node r
Indicates either mobile nodes signature is taken
at a different place or beacon nodes failure
Note Bidirectional signature distance metric put
a penalty on both distance and nodes failure.
is gained from experiments 0.951.0
25
Adaptive signature distance metric

• Unidirectional signature distance metric

Note unidirectional signature distance metric
only penalizes distance
Eg.
26
Adaptive signature distance metric

• Scheme dynamically switches between the
  unidirectional and bidirectional metrics based on
  the fraction of local beacon nodes failure.
• When few beacon nodes fail, bidirectional
  distance metric achieves greater accuracy
• When a lot beacon nodes fail, unidirectional
  distance metric achieves greater accuracy (only
  operational nodes are considered)
• Beacon nodes failure are determined dynamically
  by beacons periodically measure their local
  neighborhood.

27
Adaptive signature distance metric
28
Implementation

• MoteTrack is implemented on the Mica2 mote
  platform using TinyOS operating system
• 20 beacon nodes are deployed at Hard Universitys
  CS building measuring 1742 m2, with 412 m2
  hallway area and 1330 m2 in room area.
• 482 reference signatures are measured, each with
  7 power levels

29
Implementation
30
Evaluation

• Location estimation protocols

• Employed
• protocol
• Maintains
• Similar
• Accuracy
• While
• Achieve
• Very low
• Communication
• overhead

31
Evaluation

• Selection of reference signatures

32
Evaluation

• Distribution of the reference signature database

33
Evaluation

• Transmission of beacons at multiple power levels

34
Evaluation

• Density of beacon nodes

35
Evaluation

• Density of reference signatures

36
Evaluation

• Robustness to perturbed signatures

37
Evaluation

• Time of day and different motes

38
Evaluation

• Hallways, rooms, and door position

39
Evaluation

• Robustness to beacon node failure

40
Novelty

• Decentralized location estimation protocol
• Distribution of partial reference signature
  database to beacon nodes
• Dynamic adapt to nodes failure through employing
  different distance metric
• Employ multiple power levels

41
Drawbacks

• The beacons have to be installed and the database
  be set up before the scheme could be used
• Tricky Point the system actually employs more
  beacons than needed to achieve the same accuracy
  and also stores redundancy information
• However, this enables it to handle with beacon
  nodes failure and achieve robustness

42
Relationship with our Project
Our Project Proposed In the Paper
Environment basically stable Accuracy is the first consideration Environment highly volatile Robustness is the first consideration
Deployment in a small area Typically a room Only a small amount beacons will be used Deploy in a large area One whole floor 20 beacon nodes are used
Computation using centralized server Computing within beacon nodes and is decentralized
RF tag will be small with the most basic functions and merely no computation RF tag do a small amount of computation, eg. searching for the beacon nodes with the strongest RSSI
43
References

• 1 A. Smailagic, J. Small, and D. P. Siewiorek.
  Determining User Location For Context Aware
  Computing Through the Use of a Wireless LAN
  infrastructure. December 2000.
• 2N. B. Priyantha, A. Miu, H. Balakrishnan, and
  S. Teller. The Cricket Compass for Context-Aware
  Mobile Applications. In Proc. 7th ACM MobiCom,
  July 2001.
• 3 S. Ray, D. Starobinski, A. Trachtenberg, and
  R. Ungrangsi. Robust Location Detection with
  Sensor Networks. IEEE JSAC, 22(6), August 2004.
• 4 G. Slack. Smart Helmets Could Bring
  Firefighters Back Alive. FOREFRONT, 2003.
  Engineering Public Affairs Office, Berkeley.
• 5 P. Bahl and V. Padmanabhan, "RADAR An
  In-Building RF-Based User Location and Tracking
  System, Proc. IEEE Infocom 2000, IEEE CS Press

44
Appendix

• Pseudo code for greedy algorithm
• foreach (BN in allBNs)
• foreach (refSig in allRefSigs)
• if (BN.size lt maxNbrRefSigs)
• BN.assign(refSig)
• else if (refSig.RSSIValFromBN(BN) gt BN.minRSSI)
• BN.remove(BN.minRSSI)
• BN.assign(refSig)

• Pseudo code for balanced algorithm
• Invariants
• ----------
• (1) no refSig is assigned more than one
  additional
• time from any other refSig (i.e., every refSig
• has to be assigned at least once before a
• refSig can be assigned a second time)
• (2) no BN is assigned a refSig more than one
• additional time from any other BN
• Algorithm
• ---------
• L lt construct a list of all ltBN,refSiggt pairs
• and sort them by distance between BN and refSig
• while (there are more elements to assign)
• if (possible to assign the next pair
• from L such that no invariant is violated)
• make assignment
• else // resolve deadlock

45
End

• Thanks !
• Questions?