A transmultiplexer-based adaptive filtering structure using a new adaptation scheme

Abstract

A maximally decimated (MD) adaptive filtering structure (AFS) is presented based on perfect reconstruction (PR) filter banks (FBs) and their dual transmultiplexers (TMUXs). The system under identification is modeled as a parallel structure of interpolated FIR (IFIR) filters and its output is analysed with the PR-FB this model results from. The new AFS results then from this IFIR system model and the TMUX corresponding to it. The proposed structure adapts at a low rate a number of coefficients practically equal to the length of the system under identification. Hence, this method is computationally efficient for the large system lengths usually encountered in practice. A new algorithm is presented for the adaptation of the subband filter coefficients employing a new power normalisation (PN) scheme that significantly increases the convergence rate (CR) for colored input processes. Experimental results demonstrate the good performance of the proposed multi-rate (MR) structure using the new adaptive algorithm and the improvement in the CR it offers when is compared to other recently proposed approaches or the full-band nLMS filter.