ELECTION: Energyefficient and LowlatEncy sCheduling Technique for wIreless sensOr Networks

1
ELECTION Energy-efficient and Low-latEncy
sCheduling Technique for wIreless sensOr Networks

S. Begum, S. Wang, B. Krishnamachari, A.
Helmy Electrical Engineering-Systems University
of Southern California
2
Motivation
R
r
BS

• Sensor network of homogenous active sensors
• Monitor some phenomenon to detect abnormalities
• Application chemical monitoring, machine fault
  detection
• Exhibits spatio-temporal correlation
• Phases of operation
• Phase1 (normal operation) Energy efficiency
• Phase2 (event detection) Latency and
  responsiveness

3
Motivation

• LEACH Heinzleman et. al., HICSS 2000
• Data driven, passive sensor
• Achieves energy efficiency
• Periodic clustering
• Rotation of cluster head
• High latency
• TEEN Manjeshwar et. al., IPDPS 2001
• Event driven, passive sensor
• Periodic cluster and rotation of cluster head
• Sleeps with fixed sleep cycle
• Achieves low latency
• Sense continuously
• Stay awake when the event is detected (threshold
  reached)
• ELECTION
• Event driven, active sensor
• Takes advantage of the spatio-temporal
  correlation to adaptively adjust sleep cycle
• Achieve energy efficiency in phase 1 turn radios
  off
• Ensures low latency and high responsiveness in
  phase2

4
Assumptions

• Active/smart sensors
• Able to sense the environment in a responsive and
  timely manner
• Schedules sensors and communication radios
  independently
• The underlying phenomenon exhibits
  spatio-temporal correlation

5
Outline

• Motivation
• Description of Algorithms
• Performance Analysis
• Conclusion

6
System Parameters

• Initial sleep cycles Sin
• Data threshold Dth
• Gradient threshold Gth
• Gradient rate of change of the phenomenon
• Sleep reduction function Fsr

7
Basic Algorithms
Timing Diagram
Phase0Synchronization
CH formation
TDMA aggregation
Phase2 Report (sense communication)
Phase1Monitor (sense only with
phenomenon dependant scheduling)
State Transition Diagram
g(t) lt Gth ? s(t1) s(t)
CH Selection
CH
CH
d(t) gt Dth
Synch
Sleep
Active
Init
d(t) lt Dth
D(t) lt Dth, g(t) gt Gth ? s(t1) Fsr(s(t), g(t))
CM
CH Advertisement
CM
Phase 1 Radio off
Phase 2
Point at which threshold crosses
8
Adapting Sleep Cycles
s(t1) Fsr(s(t), g(t))

• Adjust sleep cycle based previous sleep cycle and
  gradient
• Temporal correlation ? a node wakeup at the event
  of threshold crossing
• Spatial correlation ? All sensors measuring same
  phenomenon wake up at the same time

System Parameters Sin 250 sec, Dth 95 degrees
9
Performance Metrices

• Energy
• Total energy dissipation
• Sensing energy
• Communication Cluster formation Reporting
• Latency
• Delay between report generation and actual time
  of threshold being reached
• Responsiveness
• Difference between reported data value and
  threshold (e.g. degree of temperature)

10
Energy Analysis
ELECTION ?AEsT1/?s ?AEsT2/Tr ?A/?Ec ?A/?
Er T2/Tr LEACH ?AEsT/Tr ?A/?EcT/Tc
?A/?ErT/Tr TEEN ?AEsT ?A/?EcT/Tc
?A/?ErT2/Tr
Ec gtgt Es ? Savings in cluster formation
Es gt Ec ? Savings in sensing (w.r.t. TEEN)
Es Energy dissipation of a single sensing
operation Ec Energy dissipation in a single
cluster formation Er energy dissipation in a
single report T Network life T1, T2 duration of
phase 1, phase 2 Tr Reporting interval
Tc Cluster formation interval (Le, Te) ? Node
density ? Average node degree A Total area of
the network ? Percentage of node CH (Le, Te) ?s
Expected sleep duration (El)
11
Latency and Responsiveness
Gmax Max gradient threshold it responds to
(El) Sin Initial sleep duration (El) S Fixed
sleep cycle (Le, Te)
12
Simulation Setup

• High level simulation
• ELECTION
• TEEN
• Hybrid
• Fixed sleep cycle (like TEEN)
• On demand cluster formation (like ELECTION)
• Network simulated
• 36 uniformly distributed sensors
• Network divided into 4 quadrant
• Each quadrant is assigned a sensing pattern
• Phenomenon simulated
• Phenomenon 1 changes 100 times during entire
  simulation
• Phenomenon 2 changes 20 times

13
Simulation Parameters

• Simulation time 600K seconds
• ELECTION
• Geared sleep reduction function
• Initial sleep cycle (Sin) 256 secs
• TEEN
• Cluster formation interval (Tc) 6K secs
• Fixed sleep cycle 50 secs
• Hybrid
• Cluster formation on demand
• Fixed sleep cycle 50 secs

14
Remaining Energy Analysis
Average Remaining Energy (in unit) Phenomenon 1
(changes 100 times) Es/Etx 10
Phenomenon 1 (changes 100 times) Es/Etx 1
Phenomenon 2 (changes 20 times) Es/Etx 10
15
Delay and Responsiveness
Delay (in seconds)
Responsiveness (in degrees)
16
Limitations

• Dependency on the underlying phenomenon
• A priori information of the environment may not
  be available
• Not suitable for phenomenon that does not exhibit
  spatio-temporal correlation (e.g. seismic
  monitoring)
• Synchronization problem in phase 1

17
Conclusion

• New sleep scheduling scheme for wireless active
  sensor networks
• Exploit spatio-temporal correlation of physical
  phenomenon
• Adaptively adjust sleep cycle
• Outperforms LEACH and TEEN with respect to
  energy, latency and responsiveness