Distributed Maximum Correntropy Kalman Filter with Consensus Strategies

In wireless sensor networks, finding a consensus in which the system works collectively is a fundamental problem of distributed estimation. Meanwhile, non-Gaussian problems arise in many real scenarios, such as target tracking and digital communications. The recently proposed distributed maximum correntropy Kalman filter (D-MCKF) has performed better than the conventional distributed Kalman filter (DKF) in non-Gaussian environments. In the DKF and D-MCKF, the nodes with more neighbors can achieve better performance than those with fewer neighbors. The differences among different nodes may weaken the network. In this paper, the consensus strategies are incorporated into non-Gaussian KFs to mitigate state differences among different nodes. We firstly incorporate the average consensus into the D-MCKF, and secondly propose a new weighted consensus strategy based on the correntropy cost function, where the estimated state and observations of neighbors are involved. Simulations show that the proposed algorithm mitigates state differences among different nodes.