Distributed consensus-based Kalman filtering considering subspace decomposition

Abstract

The aim of this paper is to provide a new observer structure able to deal with the distributed estimation of a discrete-time linear system from a network of agents. The main result is an innovative consensus-based structure that decompose the state in the observable and unobservable subspace of the agent using the observability staircase form. The paper proposes a design in which Kalman-like gains are synthetized to minimize the variance of the error on both subspaces. Finally some simulations are shown to compare the proposed estimator with centralized Kalman filter and other distributed schemes found in literture.