Overview: Chapter 4 (cont)

Fidelity and Yield in a Volcano Monitoring Sensor

Network
Geoff Werner-Allen, Konrad Lorincz, Jeff Johnson,
Jonathan Lees and Matt Welsh

VigilNet: An Integrated Sensor Network System for

Energy-Efficient Surveillance
TIAN HE , SUDHA KRISHNAMURTHY, LIQIAN LUO,
TING YAN, LIN GU, RADU STOLERU, GANG ZHOU,
QING CAO, PASCAL VICAIRE, JOHN A. STANKOVIC,
TAREK F. ABDELZAHER, JONATHAN HUI and BRUCE
KROGH

07/30/16

CSE 4/60484: Networked Sensor Systems

page 1

Volcano monitoring
Traditional approach: Bulky and sensitive sensors
collecting info. into flash drives. Field researchers
periodically harvest data
Volcano sensing required high fidelity and high
data rates (compared to other environment
monitoring such as temperature).
100 Hz with 24 bit accuracy
Design decision to either transmit or capture - assuming
back-to-back events will be rare

07/30/16

CSE 4/60484: Networked Sensor Systems

page 2

Evaluation
Overall network uptime
Higher for sensors than for the base station. Frequent
trips to replace batteries on the sensors. Base stations
lost wall-power frequently

Individual node uptime

Safe for durations when they had a bug - fairly good

Event detector accuracy was extremely poor ~1%

Hardware limitation

Data yield and fetch latency affected by design

decisions (simultaneous capture and transmit)
Time synchronization failure - some software bugs
not seen in lab
Time rectification to recover some timing data
07/30/16

CSE 4/60484: Networked Sensor Systems

page 3

Lessons learnt
Ground truth and self-validation are important

Coping with infrastructure and protocol failures

Cross-domain collaboration: science researchers

would want the data immediately. Waiting for timer
rectication etc. might strain relationship

07/30/16

CSE 4/60484: Networked Sensor Systems

page 4

Vigilnet Applications

Wildlife Monitoring

Alarm System
07/30/16

Flock Protection

Border Surveillance

Power Consumption: An
Important Issue in
Surveillance Systems

No power management 4 days lifetime!

Power management 10 months lifetime!

07/30/16

State of the Art: Hardware

Power efficient hardware

MICA2, MICAz, XSM, etc

Energy scavenging hardware

Vibrations. Roundy et al., A Study of

Sun light. Perpetually powered Telos.

Low Level
Vibrations as a Power Source for Wireless Sensor
Network, Computer Communications, 2003.

07/30/16

State of the Art: Software

Synchronization and coordination: Nodes turn on

only for specific tasks of which the execution
time is known in advance.
Data aggregation and compression: Nodes
reduce amount of transferred data to decrease
energy costs.
Coverage control: Nodes providing redundant
sensing coverage are turned off.
Duty cycle scheduling: Nodes alternate between
on and off states at a fast rate, which still allow
them to detect slow paced targets.
07/30/16

Power Management in
VigilNet

Combination of three schemes in real system.

Tripwire
Service

Sentry
Service

Duty Cycle
Scheduling
07/30/16

Network
Level

Group
Level

Node
Level
9

Putting Nodes Into a Sleep

State

Putting nodes to sleep as often and as long as possible.

Sleeping mode: node wakeup 1% of the time.
Toggle Period (1s)

(10ms)
Sleep

1% Wakeup

Sleep

Wakeup operation: send message with long preamble.

Preamble length = TogglePeriod * BitRate SYNC Bytes

07/30/16

DATA

CRC

10

Group Level: Sentry

Selection

Redundant sensing coverage!

07/30/16

11

Group Level: Sentry

Selection

Sentry selection and rotation.

07/30/16

Asleep
Awake
Sentry

12

Group Level: Sentry

Selection

How are the sentries selected?

07/30/16

13

Group Level: Sentry

Selection

1. Neighbors exchange hello message

(ID + position + nb of neighbors + energy).

07/30/16

14

Group Level: Sentry

Selection

2. Each node selects a delay according to its

energy resources and coverage.

Delay = Function (Energy + Coverage)

07/30/16

15

Group Level: Sentry

Selection

3. Once the delay is elapsed, a node announces

itself as a sentry.

07/30/16

16

Group Level: Sentry

Selection

Tradeoff: detection probability versus density.

Target
Detection
Probability
Sensing Radius=20m
Sensing Radius=8m
Sensing Radius=2m
10

07/30/16

100

Number of Sentries in
Area

1,000

17

Node Level: Duty Cycle

Scheduling

Target takes time to go through the network.

07/30/16

18

Node Level: Duty Cycle

Scheduling

Target takes time to go through the network

duty cycle scheduling.

Awake
(200ms)

Asleep
(800ms)

Toggle Period
07/30/16(1s)

19

Node Level: Duty Cycle

Scheduling

Putting it all together.

07/30/16

Asleep
Awake
Sentry

20

Node Level: Duty Cycle

Scheduling

Tradeoff: detection probability versus duty cycle.

Target Detection
Probability

100%

1000 Nodes, V=10m/s

1000 Nodes, V=30m/s
40%
0%

20%

Proportion of the Time

Sentries are Awake
07/30/16

21

Network Level: Tripwire

Scheduling

Exploiting knowledge about the target.

07/30/16

22

Network Level: Tripwire

Asleep
Scheduling

Putting it all together.

07/30/16

Awake

Sentry
Tripwire

23

Network Level: Tripwire

Scheduling
Each base station defines a tripwire; a node pertains
to the tripwire associated with the closest base.

Tripwire
Node
Base Station
07/30/16

24

Network Level: Tripwire

Scheduling

There are as many tripwires as base stations.

Tripwire
Node
Base Station
07/30/16

25

Network Level: Tripwire

Scheduling

Tripwire schedule specification:

Tripwire
1

Tripwire
2

Tripwire
3

Tripwire
4

Tripwire
5

Tripwire
6

07/30/16

26

Network Level: Tripwire

Scheduling

Tripwire schedule specification:

Tripwire
1

Tripwire
2

Tripwire
3

Tripwire
4

Tripwire
5

Tripwire
6

07/30/16

27

Network Level: Tripwire

Scheduling

Tripwire schedule specification:

Tripwire
1

Tripwire
2

Tripwire
3

Tripwire
4

Tripwire
5

Tripwire
6

07/30/16

28

Network Level: Tripwire

Scheduling

Tripwire schedule specification:

Tripwire
1

Tripwire
2

Tripwire
3

Tripwire
4

Tripwire
5

Tripwire
6

07/30/16

29

Network Level: Tripwire

Scheduling

Tripwire schedule specification:

Tripwire
1

Tripwire
2

Tripwire
3

Tripwire
4

Tripwire
5

Tripwire
6

07/30/16

30

Network Level: Tripwire

Scheduling

Tripwire schedule specification:

Tripwire
1

Tripwire
2

Tripwire
3

Tripwire
4

Tripwire
5

Tripwire
6

07/30/16

31

Network Level: Tripwire

Scheduling

Tripwire schedule specification:

Tripwire
1

Tripwire
2

Tripwire
3

Tripwire
4

Tripwire
5

Tripwire
6

07/30/16

32

Network Level: Tripwire

Scheduling

Tripwire schedule specification:

Tripwire
1

Tripwire
2

Tripwire
3

Tripwire
4

Tripwire
5

Tripwire
6

07/30/16

33

On-Demand Wakeup

Detection

Asleep
Awake
Base Station
07/30/16

Wakeup
Wakeup
Path Surrounding
Nodes
for Better
To Base
Station
Target Reporting
Tracking,
For Prompt
Target Classification,
& Velocity Estimation

34

Evaluation Methodology

Tripwire 1

XSMs

Tripwire 2

Tripwire 3

Field

07/30/16

35

Evaluation Methodology

Energy (mWs)

70

Sentry
Non-Sentry

60
50

Surveillance

40
30
20
10

Initia liza tion Dura tion = 5 m inute s

Initialization

Without syste m rota tion:

NonSe ntry Life Tim e : 250 da ys
Se ntry Life Tim e : 7 da ys

07/30/16

Time (s)

36

Evaluation Results: Lifetime

350
300
Lifetime (days)

250
200
150
100
50
0
No PM

07/30/16

+ Sentry

+ Duty
Cycle

+ Tripwire

37

Evaluation Results: Target

Detection and Classification

Average detection delay: 2.42 seconds.

Average classification delay: 3.56 seconds.
Classification of humans, humans with
weapons, and vehicles.
Average delay to get velocity estimation:
3.75 seconds (average error: 6%).

07/30/16

38

Summary

Successfully integrate 3 power management

strategies into real surveillance system having
a 10 months lifetime (source code available
online).
Analytical model to predict system performance
under various system configurations.
Simulation results exposing tradeoffs between
detection performance and lifetime of the
network.
07/30/16

39