Robust Quadrature Information Filter Based on Minimum Error Entropy for Non-Gaussian Measurement Noises

Under the framework of minimum error entropy (MEE), a robust quadrature information filter (MEE-QIF) is proposed to address the problems of non-Gaussian measurement noises in this article. More specifically, based on the classical Gauss-Hermite Quadrature (GHQ) rule, a statistical linear re-gression model involved in Bayesian estimation is developed. Furthermore, through specialized linear combinations of some lower-level tensor products, the sparse-grid quadrature (SGQ) method is applied to approximate the multidimensional integral. Unlike the widely used maximum correntropy criterion (MCC) and the minimum mean square error (MMSE) rule, an error entropy based optimal cost function is given, and a fixed-point iterative algorithm for the filtering update steps is employed. Moreover, the sufficient conditions for convergence of our pro-posed algorithm are thoroughly analyzed. Under non-Gaussian complex noises, two sets of experiments prove the effectiveness of the proposed MEE-QIF algorithm. Compared with classical MCC and MMSE-based algorithms, there are considerable im-provements in the robustness and estimation accuracy.