Data Aggregation in Wireless Sensor Networks

1
Data Aggregation in Wireless Sensor Networks

• Rong Zheng, Richard Barton
• University of Houston

2
Wireless Sensor Networks (WSNs)

• A WSN typically consists of one or more sinks and
  many sensors acting as information sources
• Data dissemination (sink ? sensors)
• Data aggregation (sensors ? sink)
• Key characteristics
• Limited energy budget
• Collective utility
• Asymmetry

3
Synopsis

• Question
• What is the asymptotically achievable data
  aggregation rate at the sink in a wireless sensor
  network of n nodes in a fading environment?

4
Synopsis

• Answer
• Information theoretical tight bounds in low (2 lt
  ? lt 4) and high (? gt 4) attenuation regime are
  ?(log(n)) and ?(1), respectively
• Order optimal throughput can be achieved using
  cooperative time-reversal communication (TRC) and
  a novel hierarchical network protocol
• Data aggregation using multihop relay is
  suboptimal except for ? gt 4
• TRC can improve network life time by an order of
  magnitude for low-duty cycle operations

5
Outline

• Physical layer models
• Clustering and routing
• Results
• Extensions

6
Physical Layer Models

• Nodes follow Poisson point with unit density on
  , Pn P n B(n)
• Node are maximum power constrained Pmax
• Two modes of communication
• Naive multihop relay
• Cooperative time reversal communication

7
Naive Multihop Relay

• Maximum rate for point-to-point communication
  from node i to j
• Maximum rate for point-to-point communication
  from node i to receivers R
• where

8
Background on TRCForward-Channel Impulse Response
9
Background on TRCTime-Reversed Channel Response
10
Background on TRCMultiple Receivers
11
Background on TRCCooperative Time-Reversal
12
Capacity of TRC Link
R
d

• If R ltlt d, Pi P

13
Information-theoretical Bounds

• The capacity of m-user Gaussian multiple access
  channel with CSIR is characterized by

14
A Tighter Bound for High Attenuation Regime

• Let P be a Poisson point process of unit density
  over R2. For any rate ? gt 0, the fraction of
  nodes that can receive data from all other nodes
  at that rate is at most w.h.p., where
• and I is defined as
• Theorem The total information-theoretic data
  aggregation rate ? of an extended network of unit
  density in B(n) is characterized as
• For the path loss exponent a 4, ? O(logn),
  and
• For the path loss exponent a gt 4, ? O(1).

15
Data Aggregation Using Native Multihop Relay

• Proof (sketch)
• Upper bound in extended network, nearest
  neighbor to the sink is at distance ?(1)
• Lower bound Using argument of percolation
  theory, if the sink is part of largest connected
  component then ?(1) is achievable otherwise

16
Sink tree Routing
17
Data Aggregation Using TRC
18
Data Aggregation Using TRC
19
Questions to be Addressed

• Which set of nodes should be clustered together
  in a TRC link?
• How should routing be done?
• How to schedule transmissions from different
  clusters?

20
Network Organization

• Area I, nodes are too far from sink to use TRC
  directly
• Area III, nodes are too close to benefit from TRC
• Area I, III both use naïve multihop routing
• Area II use TRC to sink directly

Sink
Area I
Area II
Area III
21
Routing on Random Networks (1)

• Partition the network into cells Vi of side
  length
• Designate a cell head hi
• Data from node u to sink O are first forward to
  cell head and then routed among cell heads whose
  cells intersect with line LuO

22
Routing on Random Networks (2)
23
Routing on Random Networks (3)
24
Achievable Rates

• Area I, III
• Leaf nodes
• Non-leaf nodes
• Area II, using TDMA among clusters
• where m and M are size and number of clusters
  respectively

25
Finally

• When ? lt 4
• we have
• When ? 4, let ? ? 0
• we have

26
Conclusion

• TRC provides an efficient way to communicate over
  long distance
• To unleash the power of TRC in a network setting,
  cross-layer design of routing, scheduling and
  communication protocols are required
  cooperation, cooperation, cooperation!

27
Questions Comments?

• Visit us at http//coco.cs.uh.edu

28
Backup Slides