Distributed decorrelation in sensor networks with application to distributed particle filtering

Most distributed statistical signal processing methods assume conditionally uncorrelated sensor measurements although this assumption is often not satisfied. Here, we propose a distributed algorithm for decorrelating the sensor measurements in a wireless sensor network. The algorithm employs a matrix-valued Chebyshev approximation to achieve an approximate decorrelation using only local computations and communication between neighboring sensors. We apply the algorithm to consensus-based distributed particle filtering in a target tracking problem with correlated measurement noises. Simulations show that the decorrelation yields a substantial accuracy improvement while causing only a small communication overhead.