Robust M-estimation-based algorithm with generalized minimum error entropy criterion for active noise control

Abstract

As a prominent approach within the framework of information theoretic learning (ITL), the concept of error entropy has demonstrated significant efficacy in active noise control (ANC). However, the Gaussian kernel function used in error entropy has a fixed structure that lacks dynamic adaptability, limiting the performance of the filtered-x minimum error entropy (FxMEE) algorithm in dealing with diverse noise characteristics. To address this issue, an adaptive filtered-x generalized minimum error entropy (FxGMEE) algorithm is proposed, which employs a flexible-shape generalized Gaussian density function as its kernel. Furthermore, the filtered-x generalized minimum error entropy (MF-FxGMEE) algorithm based on the M-estimation Fair function is developed as a robust variant, aiming to mitigate the adverse effects of non-Gaussian disturbances. By incorporating M-estimation theory, the MF-FxGMEE algorithm effectively identifies and suppresses outliers via a smooth weighting mechanism, enhancing system stability in non-Gaussian noise environments. We also analyze the theoretical properties of the MF-FxGMEE algorithm, including its mean error behavior, mean square convergence, and computational complexity. Simulation studies are performed to validate the proposed algorithms, demonstrating superior convergence performance over existing methods.