Multi-level Diffusion Kalman Filter Algorithm for Adaptive Networks

We study the problem of distributed state estimation over adaptive networks, where agents collaborate to estimate a common state parameter vector. If the sensing target area is too large or we want to improve the convergence speed of a large adaptive network, single-level diffusion algorithms do not have better performance, so we study the multi-level diffusion Kalman filter algorithm where a network running a diffusion strategy is enhanced by defining some special nodes called supernodes. In order to improve the estimation accuracy, we also study the weight-normalized diffusion Kalman filter algorithm. Simulation results show that multi-level diffusion Kalman filter algorithm has better accuracy and convergence performance than single-level diffusion Kalman filter algorithm. Furthermore, in order to further improve the algorithm's performance, we studied weight-normalized methods which are better than average weight methods.