Information Quality Aware Routing in Event-Driven Sensor Networks Hwee-Xian TAN 1, Mun Choon CHAN 1, Wendong XIAO 2, Peng-Yong KONG 2 and Chen-Khong THAM.

Information Quality Aware Routing in Event-Driven Sensor Networks

Hwee-Xian TAN1, Mun Choon CHAN1, Wendong XIAO2, Peng-Yong KONG2 and Chen-Khong THAM2

1School of Computing, National University of Singapore (NUS)2Institute for Infocomm Research (I2R), Singapore

Overview

• Introduction • Related Work and Motivation• System Model• Topology-Aware Histogram-Based Aggregation• IQAR – Information Quality Aware Routing Protocol• Performance Evaluation• Concluding Remarks

2Information Quality Aware Routing in Event-Driven Sensor Networks

Introduction

• Event-driven sensor networks– Deployed specifically for detection of Phenomenon of Interest (PoIs)– Converge-cast traffic characteristics– Sensory data is generated by multiple sensors only when PoI is

detected• Severe data implosion and redundancy


v0v1

v2

v3

v4

v5

v6

v7v8 v9

PoI

Node that does not detect PoI

Node that detects PoI

Fusion center v0

Introduction

• Data aggregation and/or fusion– Mitigates congestion and medium access contention– Suppresses data to reduce traffic load and energy consumption by

exploiting spatio-temporal correlation among sensory data– Comes at expense of loss in information quality (IQ) of data collected

at fusion center– Results in reduction of (event) detection accuracy


• Considers information content of data during aggregation/fusion and forwarding.

• Neighbor with highest information gain is selected to be next-hop.

• Fused data is transmitted to fusion center when IQ threshold is satisfied.

• Typically query-based rather than event-based.

• Alleviates medium contention, reduces transmission costs and reduces e2e delays.

• All sensory data is forwarded to fusion center, resulting in high data redundancy and energy costs.

Related Work and Motivation

Information Quality Aware Routing in Event-Driven Sensor Networks 5

Aggregation/Fusion based RoutingAggregation/Fusion based Routing IQ Aware RoutingIQ Aware Routing

IQAR considers information content, and addresses both event-detection and multi-hop networks.

System Model

• Network has n sensors and fusion center v0

• H1 denotes presence of PoI

• H0 denotes absence of PoI

• P(H1) = p, P(H0) = 1-p 0 < p < 1

• Sensor observations are assumed to be i.i.d. at each sensor as well as across sensors.


System Model


Event Detection at SensorEvent Detection at Sensor

• Independent signal yi observed by node vi is:

where wi is noise and ri is distance between vi and PoI.

• For each sampled signal yi, vi makes a per-sample binary decision bi {0,1}:

where Ti is the per-sample threshold.

Event Detection at Fusion CenterEvent Detection at Fusion Center

present) is (PoI H if

absent) is (PoI H if

)( 1

0

ii

ii wrf

wy

otherwise

if

1

0 iii

Τ yb

System Model


Event Detection at SensorEvent Detection at Sensor

• Fusion center v0 detects presence of PoI by making a global binary decision H={H0,H1} based on data received.

• Optimal fusion rule is the Likelihood Ratio Test (LRT):

where B={b1,b2,..,b|Va|} is the set of per-sample binary decisions received; and Va is set of activated nodes.

Event Detection at Fusion CenterEvent Detection at Fusion Center

p

p

HbbbP

HbbbPB

H

H

V

V

a

a

1

)|,...,,(

)|,...,,()(

0

1

0||21

1||21

• Independent signal yi observed by node vi is:

where wi is noise and ri is distance between vi and PoI.

• For each sampled signal yi, vi makes a per-sample binary decision bi {0,1}:

where Ti is the per-sample threshold.

present) is (PoI H if

absent) is (PoI H if

)( 1

0

ii

ii wrf

wy

otherwise

if

1

0 iii

Τ yb

System Model


Sequential DetectionSequential Detection

• Data acquisition can terminate at earliest subsequence of data which satisfies a pre-determined IQ threshold.

• Reduces amount of data required to make an accurate global binary decision H = {H0,H1}.

• Cumulative log-likelihood ratio at fusion center v0 is:

where Va is set of activated nodes.

• Cumulative log-likelihood at vi is:

||

10 )(log)(log

aV

iibBS

vivi

v1v1

v2v2

v3v3

},,{ 321 vvvV ui

upstream nodes of vi

u

ij Vvjii SbS )(log

IQ of node vi

Topology-Aware Histogram-based Aggregation


v0 0.40.41.01.0

0.50.5 0.30.3

0.20.2

0.80.8

0.30.3 0.10.1 1.71.7

1.21.20.40.4

0.60.6

v1 v4 v9

v2

v5

v6

v7

v8

v3

v10

v11

v12

With Global View & Topological KnowledgeWith Global View & Topological Knowledge

Required IQ IT Min-cost aggregation tree

1.0 {v3}

2.0 {v3, v7, v8}

{v2, v3, v8}

{v2, v3, v5}

{v1, v3, v8}

{v1, v4, v9}

4.5 {v2, v3, v5, v6, v8, v12}

{v1, v2, v3, v4, v5, v9}

High communication costs and overheads! High communication costs and overheads!



v0 0.40.41.01.0

0.50.5 0.30.3

0.20.2

0.80.8

0.30.3 0.10.1 1.71.7

1.21.20.40.4

0.60.6

v1 v4 v9

v2

v5

v6

v7

v8

v3

v10

v11

v12

{0.3, 2.1, 3, {2, 0, 1, 0, 0}}

IQ of v1

Max IQ using subtree rooted at v1

Max cost using subtree rooted at v1

histogram



v0 0.40.41.01.0

0.50.5 0.30.3

0.20.2

0.80.8

0.30.3 0.10.1 1.71.7

1.21.20.40.4

0.60.6

v1 v4 v9

v2

v5

v6

v7

v8

v3

v10

v11

v12

{0.4, 3.2, 6, {1, 2, 2, 1, 0}}

IQ of v2

Max IQ using subtree rooted at v2

Max cost using subtree rooted at v2

histogram

IQ-Aware Routing Protocol


InitializationInitialization Aggregation and UpdateAggregation and Update PruningPruning

v0

v1 v4 v9

v2

v5

v6

v7

v8

v3

v10

v11

v12




v0 0.40.41.01.0

0.50.5 0.30.3

0.20.2

0.80.8

0.10.1

0.40.4

v1 v4 v9

v2

v5

v6

v7

v8

v3

v10

v11

v12




• Objective is to prune off as many nodes as possible from initial distance-based aggregation tree such that:

1. IQ constraint is still satisfied. 2. Total transmission cost is minimized.

v0 0.40.41.01.0

0.50.5 0.30.3

0.20.2

0.80.8

0.10.1

0.40.4

v1 v4 v9

v2

v5

v6

v7

v8

v3

v10

v11

v12

Performance Evaluation

• Simulator: Qualnet 4.0• Fusion center near bottom left-hand corner of terrain. • Exponential sensing model.


Parameter Value

Terrain size 100 meters × 100 meters

Sensing interval 1 second

Transmission range ~ 8 meters

Target detection probability Pd 0.9

Target false alarm probability Pf 0.001



aggTreeaggTree

brute-forcebrute-force

walkwalk1.11.10.80.8

0.60.6

0.50.5

0.30.3

v0v1

v2

v3

v4

v5

v6

v7v8 v9

1.11.10.80.8

0.60.6

0.50.5

0.30.3

v0v1

v2

v3

v4

v5

v6

v7v8 v9

1.11.10.80.8

0.60.6

0.50.5

0.30.3

v0v1

v2

v3

v4

v5

v6

v7v8 v9





Concluding Remarks

• Considers individual IQ contributions of each sensory data, and collects only sufficient data for PoI to be detected reliably.

• Utilizes a compact topology-aware histogram to represent the IQ contributions of nodes in the network.

• Redundant data is suppressed for time interval to reduce traffic load and alleviate medium access contention.

• Achieves significant energy and delay savings while maintaining IQ.


IQARIQAR

Information Quality Aware Routing in Event-Driven Sensor Networks Hwee-Xian TAN 1, Mun Choon CHAN 1, Wendong XIAO 2, Peng-Yong KONG 2 and Chen-Khong THAM.

Documents

sensor fusion center

information quality

information content

poi fusion center

fusion center slide

routing iq aware routing

sensor observations

node v i