1. Energy-Efficient Signal Processing and Communication Algorithms for Scalable Distributed Fusion

2. Wireless Sensor Networks for Fusion Sensor networks for surveillance and reconnaissance target detection and tracking  environmental applications

3. Signal Processing and Communication Challenges System constraints limited energy (and bandwidth) resources per sensor need for power-efficient processing algorithms and communication protocols limitations in sensing and on-board processing equipment limitations in type and rate of processing interested in ensembles of measurements (e.g., maximum, average) need for algorithms that obtain the fusion objective, not all individual measurements large networks; changes in network topology real-time knowledge of global topology impractical  locally constructed algorithms Additional requirements scalability fault-tolerance algorithms that can compute (reduced) fusion objective over reduced topologies

4. Objectives Desired Tasks computation of statistics of the measurements over very large networks of wireless sensors maximum, average, locally-averaged (in space) signal estimates Algorithmic Objectives algorithms for data processing and relaying across the network  locally constructed, yet reliable  exploit compression benefits via distributed local fusion  designed for energy-efficient on-board processing

5. Hierarchical Networks Setting hierarchical protocol for data communication and fusion Advantages bandwidth efficient readily scalable hierarchy Disadvantages unequal distribution of resources often power usage inefficiency sensitivity to fusion node failures  robustness asymmetry

6. Ad-hoc Networks Setting two-way local communication between closely located (“connected”) sensors each sensor node receives messages send by connected nodes each sensor broadcasts messages to connected nodes Advantages robust, readily scalable space-uniform resource usage transmit power efficient Issues need for networking

7. Ad hoc Networks for Fusion Related work ad hoc networks, amorphous computing Distinct features in fusion problem interested in underlying signal in data (e.g, target location), not data info about signal “spread” over many nodes multiple destinations Remarks Advantages: data compression in fusion, inherent scalability Key problem: communication loops (contamination of information)

8. Computation of Global Maximum Objective Compute maximum among measurements Approach sequence of local maximum computations Sensor State=current maximum estimate communication step: Each node broadcasts its state fusion step: New state at its node= maximum of all received states Result: Each node state converges to the global maximum (in finite number of steps) provided network is connected

9. Computation of Weighted Averages Remarks not all local averaging rules yield global average (data contamination) Approach Locally constructed fusion rules can be designed [Scher03] which asymptotically compute weighted averages of functions of the individual measurements (e.g. average, power, variance of measurements) Advantages distributed, robust, readily scalable address non-contributing node problem in distributed fashion

10. Project Objectives Remarks Finite delays and limitations in available energy and on-board processing  finite-time approximate computations Analysis and Optimization Design energy-efficient methods for approximate computation of maxima, averages and other measurement statistics Determine trade-offs on-board processing and communication power, delays and quality of computation Non-contributing nodes  need for power-efficient data relaying