Filter design for decentralized state estimation in sensor networks

This paper considers decentralized state estimation with sensors exchanging information with its nearest neighbours. Under the constraint that sensors can only exchange information with its neighbours once at each discrete time instant, a lower bound on the achievable performance is derived. For networks with cycles, it is not known if the lower bound can be achieved. An approach that forms weighted combinations of local estimates is proposed, which can get very close to the lower bound when the filter gains and weights are appropriately optimized. A suboptimal way of designing the filter gains and weights, which can be computed using information available within a sensor's local neighbourhood, is also considered.