Distributed filtering for multi-sensor networked systems with stochastic communication protocol and correlated noises

Abstract

In multi-sensor networked systems, to solve data conflicts caused by limited bandwidths during measurement data exchange among sensors, a stochastic communication protocol scheduling is employed to ensure that each sensor can only receive one component of measurement vector from neighbor nodes at each instant. The measurement components selected for communication at each instant are indicated by a set of random variables. When each sensor knows which component of the measurement vector from each of its neighbor nodes is passed at each instant, an optimal distributed Kalman filter (ODKF) where both estimator and gain rely on values of random variables is designed in the linear minimum variance sense. To reduce the online computational burden, a suboptimal DKF (SDKF) where the filtering gain relies on probabilities of random variables is designed and its steady-state property is analyzed. When each sensor doesn't know which component of the measurement vector is passed at each instant, a probability-dependent DKF (PDKF) where both estimator and gain rely on probabilities of random variables is designed and its steady-state property is analyzed. Finally, the performances of the proposed three DKF algorithms are compared. Two simulation examples verify the effectiveness of algorithms.