Filter Bank Design for Minimizing Mean-Squared Estimation Error in Subband Adaptive Filtering

Abstract

This paper considers the problem of prototype filter design for subband adaptive filtering applied to system identification. The minimum mean-squared estimation error (MMSE) depends only the subband analysis filter and the response of the unknown system. We use MMSE as a design criterion to select the best analysis filter response. We show how this can be done independently of the unknown system. The connection between a minimizing MSE and a least-squares estimation problem inspires a new adaptive filter structure in which different analysis filters are used in the two branches of the subband adaptive filter