Differentiated Surveillance for
Sensor Networks
Ting Yan, Tian He, John A. Stankovic
CS294-1
Jonathan Hui
November 20, 2003
Idea


Exploit node
density/redundancy
to maximize
effective network
lifetime.
Degree of coverage
matters!


Sensing constraints
Fault tolerance
2
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
3
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
Round 0 (T)
Ref
Tfront
Init Phase
Tend
Round 1 (T)
Ref
Tfront
Tend
Sensing Phase
Round i (T)
Ref
Tfront
Tend
t
4
Basic Protocol
Determining 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
5
Basic Protocol
Determining 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
RefC RefA
RefB
Point 1
RefC RefA
Node A:
RefD
Point 2
RefA
RefE
RefD
RefB
RefE
6
Enhanced Protocol
with Differentiation

Working schedule for a desired
coverage of degree a.


(T, Ref, Tfront, Tend, a)
Working period defined as:


Power On:
Power Off:
Example (a = 2)
RefA
T  i  Ref  T front  a
T  i  Ref  Tend  a
RefB
RefC
Example (a = 1)
RefA
RefB
RefC
RefB
RefC
A
B
C
Example (a = 3)
RefA
A
A
B
B
C
C
Uh-Oh!
7
Design Issues

Possible blind spots with discrete points


Offset in time synchronization


Power on (off) slightly earlier (later)
Irregular sensing regions



Choose points within sensing range conservatively
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?
8
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.
9
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.
RefC RefA RefB
RefB
RefC RefA RefA RefB
RefCRefA
RefC
RefB
A
B
C
Schedule 1
Schedule 2
Schedule 3
Schedule 4
10
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
11
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?
12
Evaluation

Half-life increases
linearly as density
increases.

Coverage
provided for
longer period of
time.
13
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.
14
Comments

Localized algorithm?


Inflexible


mobility not supported, schedules are fixed
No notion of the “goodness” of a node


But still requires time synchronization and doesn’t support
mobility
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)
15
Comments

Working schedules determined without taking other
schedules and protocols into account


Comparison to Sponsored Coverage unfair


Sponsored Coverage supports fault tolerance, limited
mobility, and is more adaptable
Ability to specify degree of coverage


How does it affect other protocols (i.e. TDMA)?
But current algorithm doesn’t correctly guarantee with a >
2!
Fault tolerance relies on communication range > 2r
for heartbeat messages
16
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
17