Distributed Incremental Saturation Kernel-Based Correntropy Filter Subject to Unknown Non-Gaussian Noise and Measurement Bias

In this paper, a distributed incremental saturation correntropy filter is proposed for multi-sensor network under unknown non-Gaussian noise and measurement bias. First, to suppress non-Gaussian noise, a saturation kernel function with nonlinear compression characteristics is developed, and its properties are investigated. It combines the benefits of both Gaussian and Cauchy kernel functions. Second, an incremental measurement equation is constructed by the difference method to eliminate the unknown measurement bias. Concurrently, an adaptive noise variance estimator is designed based on the incremental measurement innovation, enabling real-time estimation of unknown incremental noise variance. Subsequently, the incremental saturation correntropy filter is derived by the presented saturation correntropy criterion, with a dynamic parameter adjustment strategy introduced to mitigate parameter selection uncertainty. The improved covariance intersection technique is then employed to achieve the consensus estimation of the whole sensor network. In addition, the unbiasedness, boundedness and convergence of the proposed algorithms are rigorously established under appropriate conditions. Finally, the simulation results of three scenarios demonstrate the effectiveness and superiority of the proposed algorithms.