Localized FaultTolerant Event Boundary Detection in Sensor Networks

1
Localized Fault-TolerantEvent Boundary
Detectionin Sensor Networks
(IEEE INFOCOM 2005)

• Ki Sung Lee
• 17 Nov. 05
• DB Lab.

2
Overview

• Introduction
• Background
• Motivation
• Assumptions
• Localized faulty sensor detection
• Localized event boundary detection
• Performance evaluation
• Simulation results
• Summary

3
Background

• Faulty sensor readings
• Hardware crash
• Security attack
• Environment disturbance
• Event boundary detection
• Can be more important than the entire event
  region detection
• Example) detection of the transportation front
  line of a contamination

4
Motivation

• To save energy
• Filter out faulty readings
• Transmit only the boundary information to the
  base station
• Problem of a centralized fashion
• Strict resource limitation
• ? Need to make

Localized faulty sensor detection algorithm
Localized event boundary detection algorithm
5
Assumptions

• Network model
• N sensors uniformly deployed in a b?b field in
  the plane R2
• faulty if a sensor reading deviates
  significantly from other readings of neighboring
  sensors
• Notations
• S set of all the sensors in the field
• R radio range of the sensors
• xi actual reading of the sensor Si
• E event, subset of R2
• B(E) boundary of the event E

6
Localized faulty sensor detection

• Compare the reading at Si with readings of its
  neighbors
• N(Si)
• Represents a neighborhood of the sensor Si
• Contains Si and additional k sensors(Si1, Si2, ,
  Sik)
• Example) a closed disk centered at Si with the
  radius R
• Comparison
• medi median of readings of k sensors
• Cannot use sample mean

di xi medi eq.(1)
7
Localized faulty sensor detection (contd)

• N(Si)
• Also represents a neighborhood of the sensor Si
• Contains Si and additional n-1 sensors(S1, S2,
  , Sn-1)
• Example) N(Si) N(Si)
• D d1, , dn-1, di by eq.(1)

DECISION If yi ?, gt treat Si as a faulty
sensor (where ?(gt1) is a predetermined number)
C1 set of sensors with yi ?
8
Localized faulty sensor detection (contd)

• Algorithm 1
• Construct N and N
• For each sensor Si
• Compute di by using N(Si) and eq.(1)
• Compute yi by using N(Si) and eq.(2)
• If yi ?,
• Assign Si to C1

d2
di
d1
9
Localized event boundary detection

• Limitations of C1
• may contain some normal sensors close to the
  event boundary
• but, usually does not effectively detect sensors
  close to the event boundary
• ? Should modify Algorithm 1
• NN(Si)
• a special neighborhood of Si

10
Localized event boundary detection (contd)

• Random bisection
• Place a closed disk centered at Si from S-C1
• Randomly draw a line through Si, dividing the
  disk into 2 halves
• Calculate di in each half
• Select NN(Si) half disk yielding the largest
  di
• Update di from algorithm1 to the new di from
  NN(Si)
• Calculate eq.(2) and make a decision on Si

NN(Si)
C2 set of all sensors with yi ?
11
Localized event boundary detection (contd)

• Random trisection
• Place a closed disk centered at Si from S-C1
• Randomly divide the disk into 3 sectors with an
  equal size
• Form a union using any 2 sectors and calculate di
  in each union
• Select NN(Si) union yielding the largest di
• Update di from algorithm1 to the new di from
  NN(Si)
• Calculate eq.(2) and make a decision on Si

NN(Si)
C2 set of all sensors with yi ?
12
Localized event boundary detection (contd)

• Limitations of C2
• only contains non-C1 sensors
• but C1 contains some boundary sensors
• contains some sensors that are not close to the
  boundary
• C3
• combine C1 and C2
• only include sensors that have at least one C2
  sensor nearby

13
Localized event boundary detection (contd)

• Algorithm 2
• Construct N and N
• Apply Algorithm1 to produce the set C1
• For each sensor Si ? S-C1,
• Obtain NN(Si)
• Update di from algorithm1 to the new di from
  NN(Si)
• Use eq.(2) to recompute yi
• If yi ?, assign Si to set C2
• Obtain C3

14
Performance evaluation

• Evaluation C1
• a(C1) detection accuracy
• e(C1) false alarm rate
• Evaluation C3
• a(C3, r) degree of fitting
• e(C3, R) false detection rate

15
Simulation results

• Detection accuracy false alarm rate

16
Simulation results (contd)

• Degree of fitting false detection rate

17
Summary

• Localized faulty sensor detection
• Localized event boundary detection