Adaptive Data Collection Strategies for Lifetime-Constrained Wireless Sensor Networks
Xueyan Tang Jianliang Xu Sch. of Comput. Eng., Nanyang
Technol. Univ., Singapore;
Parallel and Distributed Systems, IEEE Transactions onJune 2008
Outline Introduction Problem Formulation
Single-hop networks Optimal Data Update Solution (Off-line) Adaptive Data Update Strategy (On-line) Adaptive Aggregate Data Update
Multi-hop networks Performance Evaluation Conclusion
Data Report Problem (1/3)-Single-hop Networks
Consider 10 solar radiation readings 369, 330, 264, 266, 274, 279, 260, 233, 225
Assume the total energy budget of a sensor is three updates (i.e., send only three updates)
Periodically update strategy Sends the 1-th, 4-th, and 7-th readings 369, skip, skip, 266, skip, skip, 260, skip, skip
Approximate readings 369, 369, 369, 266, 266, 266, 260, 260, 260
Reconstructed data
Data Report Problem (2/3)- Single-hop Networks
Data Error (Deviation) Exact readings:
369, 330, 264, 266, 274, 279, 260, 233, 225
Approximate readings: 369, 369, 369, 266, 266, 266, 260, 260, 260
Error = 0+39+105+0+8+13+0+27+35 = 227.
error
Data Report Problem (3/3)- Single-hop Networks
Better Update Strategy sends the 1-th, 4-th, and 8-th
readings369, skip, skip, 266, skip, skip, skip,
233, skip approximate readings
369, 369, 369, 266, 266, 266, 266, 233, 233
Error = 0+39+0+2+10+15+4?+0+ 8 = 78error
Problem Formulation (1/3)-Single-hop Networks
Problem: Exact readings:
369, 330, 264, 266, 274, 279, 260, 233, 225…………
Find M updates such that root-mean-square of collected data error is minimized.
Problem Formulation (2/3)- Single-hop Networks
Assume Exact readings (T: given network lifetime):
d1, d2, …, dT Energy budget (at most): M updates Data updates at times: v1=1, v2, v3,…, vM
Ex: v1=1 1-th reading (first update) v2=3 3-th reading (second update)
Approximate readings: MMM vvvvvvvvv ddddddddd ,....,,...,,,,...,,,
222111
Problem Formulation (3/3)-Single-hop Networks
Find v1=1, v2, v3,…, vM such that is minimized. where
Optimal Data Update Solution (Off-line Version)
Assume that all sensor readings are known a priori Exact readings d1, d2, …, dT are known
Solve by a dynamic programming algorithm.
Dynamic Programming (1/4) Let be an optimal
solution to the (t, m)-optimization problem.
Claim: must be an optimal
solution to the (t -1, m -1)-optimization problem.
Dynamic Programming (2/4)
Proof Assume there exists a better solution
Dynamic Programming (3/4)
Dynamic Programming (4/4) Let A(t, m) be the minimal achievable total square
error to the (t, m)-optimization problem. Let B(t, m) be the time of the last data update in th
e optimal solution.
Adaptive Data Update Strategy (On-line Version)
Idea Let the sensor node update a new
reading with the base station only when the new reading substantially differs from the last update.
i.e., update only ifWdd elast updatnew ||
Example: W = 40
369, 330, 264, 266, 274, 310, 260, 233, 225
Adaptive Data Update Strategy (On-line Version)
Issues The number of updates are decided by W How to dynamically adjust W
Assume that the energy budgets: 3 updates
Expected data update period : Once every 3 time units
369, 330, 264, 266, 274, 279, 260, 233, 225
Adaptive Data Update Strategy (On-line Version)
Measure the data update period every time a new reading is updated. Estimate of data update period
Compare with the expected data update period IE :
oldupdatelastc ITTI )1() (
)1( if )1(
)1( if )1(
E
E
IIWW
IIWW
Adaptive Data Update Strategy (Algorithm)
Initialization
Adaptive Data Update Strategy (Algorithm)
Adaptive Aggregate Data Update-Multi-hop networks
Problem in multi-hop networks
bottleneck
Node A : receive 6 updates
sends 3 updates
Adaptive Aggregate Data Update-Multi-hop networks
Node A : receive 6 updates
sends 8 updates
Node A : receive 6 updates
sends 3 updates
Allocating Number of Updates
The number of updates that node can send is
bottleneck
)( vCs
e
i
i
send
receive
Total energy
Allocating Number of Updates-Idea
2022
24
Round t
Round t+1
22tAVG
Assume thresholds WA = 3, WB=2, WC=2
2119
22
|22-19| > WB
22 20
22
19
|21-20| < WC
3.20tAVG
|22-20.3| < WA
6
3
3 3
6
3 3 3 3 3
6 6
Goal The objective is to let the sensor nodes se
nd as many updates as possible subject to the energy constraints
Update Allocation Algorithm-An Example ui : unused energy budget
xi: min(xi , xpi)
ci: allocated number of updates Assume that s = 1 units (send) and v = 1 units (receive)
ui/xi/ci
A:
ui = 12 (initial)
xi = 12/(2+1) = 4
ci = min(4, ∞)=4
Round 1
Update Allocation Algorithm-An Example ui : unused energy budget
xi: min(xi , xpi)
ci: allocated number of updates Assume that s = 1 units (send) and v = 1 units (receive)
ui/xi/ci
B:
ui = 12 (initial)
xi = 12/(3+1) = 3
ci = min(4, 3) = 3
Round 1
Update Allocation Algorithm-An Example ui : unused energy budget
xi: min(xi , xpi)
ci: allocated number of updates
Round 2
A:
ui = 12-4-6 = 2
xi = 2/(0+1) = 2
ci = min(2, ∞)+4=6
Update Allocation Algorithm
Performance Evaluation
Experimental Setup
Performance Evaluation-Single-hop (without aggregation)
Performance for Parameter Settings
Performance Evaluation-Multi-hop (MAX Aggregation)
Performance Evaluation-Multi-hop (Average Aggregation)
Conclusion This paper developed adaptive strategies
for both individual and aggregate data collections to make full use of the energy budgets of sensor nodes.
Experimental results show that, compared to the periodic strategy, adaptive strategies significantly improve the accuracy of collected data.