Consensus-based distributed mean square state estimation

This paper addresses the design of a distributed steady state filter for a sensor network. In the considered scenario, an estimate of the full system state is determined at each network node. The filter structure includes both measurement update terms from neighboring nodes, and a consensus term on the state estimates. The computation of the filter gains is recast as a convex optimization problem. Convergence of the estimation error variance is ensured at each network node and a guaranteed performance in the mean square sense is achieved. Extension of the method to linear parameter-varying systems yields a gain scheduled distributed filter.