An Efficient Distributed Kalman Filter Over Sensor Networks With Maximum Correntropy Criterion
Chen Hu, Badong Chen
Abstract
We consider the distributed Kalman filtering (DKF) with non-Gaussian noises problem, where each sensor exchanges information between its neighbors with limited communication. Inspired by the ability to capture higher-order statistics of maximum correntropy criterion (MCC) to deal with non-Gaussian noises, we utilizes a matrix weight instead of a scalar obtained by MCC to improve the estimation performance comparing with existing MCC based DKFs. We approximate the centralized estimate by the covariance intersection method, and propose a new MCC based distributed Kalman filter, named CI-DMCKF. The proposed algorithm only needs to communicate once with neighbors in a sampling period, which is more efficient for low bandwidth communication than existing MCC based DKFs. Under the condition of global observability, we show that the consistency, stability and asymptotic unbiasedness properties of proposed CI-DMCKF algorithm. Finally, we experimentally demonstrate the effectiveness of the proposed algorithm on a cooperating target tracking task.