Generalized Combined Nonlinear Adaptive Filters for Nonlinear Acoustic Echo Cancellation

Abstract

This paper proposes a new and generalized algorithm of combination of nonlinear adaptive filters (CNAF) for nonlinear acoustic echo cancellation (NAEC). In contrast to combining filters in a conventional parallel manner, the candidate filters are organized to form in a network structure with two subnetworks. The nodes in each subnetwork serve as linear and nonlinear adaptive filters respectively. A generalized CNAF (GCNAF) is then obtained by linking the nodes in the network and using the diffusion adaptation strategy. The proposed GCNAF algorithm allows information exchange and sharing among the nodes, so as to maximally optimize the performance of the combined filters. Simulations with noise and speech signals demonstrate the effectiveness of the proposed GCNAF for the NAEC problem.