Time-space-sequential distributed particle filtering with low-rate communications

We present a distributed particle filtering scheme for time-space-sequential Bayesian state estimation in wireless sensor networks. Low-rate inter-sensor communications between neighboring sensors are achieved by transmitting Gaussian mixture (GM) representations instead of particles. The GM representations are calculated using a clustering algorithm. We also propose a ?look-ahead? technique for designing the proposal density used for importance sampling. Simulation results for a target tracking application demonstrate the performance of our distributed particle filter and, specifically, the advantage of the look-ahead proposal design over a conventional design.