Subband adaptive algorithms based on warped cosine-modulated filter bank

Abstract

This paper investigates the properties of adaptation algorithms based on a warped cosine-modulated filter bank. The high-quality filter bank which approximates the Bark scale serves as the foundation of subband variants of three well-known adaptation algorithms. Namely, Normalized Least Mean Squares, Recursive Least Squares and Affine Projection algorithms are investigated in the following study, in terms of their better convergence behavior and the reduction of computation complexity, compared with fullband solutions.