Distributed entropy filtering subject to DoS attacks in non-Gauss environments

This paper is concerned with distributed entropy filtering for a class of time-varying systems subject to non-Gaussian noises and denial-of-service attacks. A distributed Kalman-type filter is constructed by fusing the information from neighboring sensors in the one-step prediction estimation. The maximum correlation entropy criterion with weighted Gaussian kernels instead of the traditional least squares and minimum variance indices is then employed to evaluate the filtering performance under the more general case of non-Gaussian noises. For the considered scenario, a new algorithm of distributed maximum correntropy criterion Kalman filters is developed by utilizing traditional fixed-point iterative rules, and the desired filter gains are dependent on both the one-step prediction cross-variance and the weighted Gaussian kernel function. In light of a similar analysis process and an introduced auxiliary cost, its suboptimal version is proposed to avoid the calculation of cross-variance and thereby satisfying the requirement of real time. Furthermore, a degenerated result is also derived under the traditional correntropy criterion (ie, with unit weighted matrices). Finally, the simulation results show the merit of proposed distributed maximum correntropy criterion Kalman filters in the presence of denial-of-service attacks and non-Gaussian noises.