Robust Volterra Filter for Nonlinear Censored Regression and Its Applications to Nonlinear Acoustic Echo Cancellation
Guobing Qian, Luping Shen, Yan Wang, Ying‐Ren Chien, Junhui Qian, Shiyuan Wang
Abstract
Adaptive filtering algorithms play a crucial role in consumer electronic field. However, when faced with nonlinear scenarios involving censored measurement data, traditional algorithms often encounter limitations, leading to estimation biases. To tackle this challenge, this study first applies the censored regression (CR) model to the nonlinear Volterra system. Focusing on the censored Volterra model and the Gaussian mixture noise (GMN) environment, we propose a novel robust adaptive filtering algorithm. First, we design a data recovery method for censored data, effectively compensating for computational biases caused by data truncation. Then, inspired by Modified Blake-Zisserman and mixture correntorpy, we develop a robust adaptive filtering algorithm named CR Modified Blake-Zisserman with mixture correntropy Volterra (CR-MBZMC-V) algorithm. Afterward, a performance analysis of the proposed algorithm is conducted, and theoretical steady-state mean square derivation (MSD) is derived. Finally, the correctness of the theoretical analysis and the superiorities of the proposed method are validated through simulations on nonlinear system identification and nonlinear acoustic echo cancellation (NAEC) with censored measurement data.