Differentiated Surveillance for Sensor Networks

Ting Yan, Tian He, John A. Stankovic

CS294-1Jonathan HuiNovember 20, 2003

2

Idea Exploit node

density/redundancy to maximize effective network lifetime.

Degree of coverage matters! Sensing

constraints Fault tolerance

3

Assumptions

Static placement Localization Time Synchronization Uniform expected node lifetime For simplicity of describing protocol?

Nodes on 2D plane Circular sensing radius r Communication range > 2r

4

Basic Protocol

Initialization Phase Localization, Time Sync, Determine

Working Schedule (T, Ref, Tfront, Tend)

Sensing Phase Nodes power on and off based on

working schedule

t

Sensing PhaseInit Phase

Ref

Tfront Tend

Round 0 (T)Ref

Tfront Tend

Round 1 (T)Ref

Tfront Tend

Round i (T)

5

Basic ProtocolDetermining Working Schedule

Goal: Each node determines its own working schedule such that all points within sensor coverage are covered for all time.

Approach: Represent sensor coverage with grid of points

6

Basic ProtocolDetermining Working Schedule

Reference Point Scheduling Algorithm Randomly choose Ref from [0, T) and broadcast

to all nodes within 2r. For each discrete point

Order neighboring Ref times and calculate Tfront = [Ref(i)-Ref(i-1)]/2 Tend = [Ref(i+1)-Ref(i)]/2

Final schedule = union of schedules for all points

RefA

RefA

RefBRefC

RefD RefE

Point 1

Point 2

RefA RefBRefC

RefD RefE

Node A:

7

Enhanced Protocolwith Differentiation

Working schedule for a desired coverage of degree a.

(T, Ref, Tfront, Tend, a) Working period defined as:

Power On: Power Off:

aTRefiT front aTRefiT end

Example (a = 1)

A

B

C

RefA RefB RefC

Example (a = 2)

A

B

C

RefA RefB RefCExample (a = 3)

A

B

C

RefA RefB RefC

Uh-Oh!

8

Design Issues Possible blind spots with discrete points

Choose points within sensing range conservatively Offset in time synchronization

Power on (off) slightly earlier (later) Irregular sensing regions

Okay, as long as sensing regions of neighboring nodes are known

But also requires knowledge of orientation Fault Tolerance

Awake nodes use heartbeat messages to detect failed nodes

If a node fails, wakeup all nodes within 2r and reschedule.

What if communication range < 2r?

9

Extensions and Optimizations Second Pass Optimization

After determining working schedule, broadcast schedule to all nodes within 2r.

The node which has the longest schedule: Minimize Tfront and Tend while maintaining

sensing guarantee Rebroadcasts new schedule

Done when every node has recalculated schedule or when no more can be done.

10

Extensions and Optimizations Multi-Round Extension for Energy

Balance Calculate M schedules each with

different Ref values during Init Phase. Rotate schedules during Sensing Phase.

A

B

C

RefA RefARefB RefBRefC RefCRefBRefARefC RefA RefB RefC

Schedule 1 Schedule 2 Schedule 3 Schedule 4

11

Evaluation Simulation parameters

Nodes distributed randomly with uniform distribution in 160mX160m field.

Results taken from center 140mX140m to avoid edge effects

Sensing range = 10m Communication range = 25m Ideal conditions Fault tolerance included?

Compare against sponsored approach

12

Evaluation Total energy

consumption nearly constant with changes in density.

Variation in total energy consumed decreases with greater densities.

What’s happening with the sponsored approach?

13

Evaluation Half-life increases

linearly as density increases.

Coverage provided for longer period of time.

14

Evaluation

Energy consumption increases linearly with different degrees.

Energy consumption constant with different densities.

Degree of coverage provided >= a.

a only guarantees a lower bound.

15

Comments Localized algorithm?

But still requires time synchronization and doesn’t support mobility

Inflexible mobility not supported, schedules are fixed

No notion of the “goodness” of a node Nodes that have more energy should take up a

larger portion of the working schedule Difficult to reliably broadcast Ref values to all

2r neighbors in a dense network Only have one chance to get it right! Worse in cases where communication range < 2r

(i.e. acoustic sensors)

16

Comments Working schedules determined without taking

other schedules and protocols into account How does it affect other protocols (i.e. TDMA)?

Comparison to Sponsored Coverage unfair Sponsored Coverage supports fault tolerance, limited

mobility, and is more adaptable Ability to specify degree of coverage

But current algorithm doesn’t correctly guarantee with a > 2!

Fault tolerance relies on communication range > 2r for heartbeat messages

17

Conclusion Pros

Improved performance in lifetime and workload balance

Specify a degree of coverage Cons

No upper bound on degree estimation Inflexible

Static working schedule, static nodes, time synchronization, reliable communication