New structures for adaptive filtering based on the generalised structural subband decomposition

Abstract

The system identification (SI) problem is addressed from the viewpoint of the generalised structural subband decomposition (GSSD). A system identification structure (SIS) that provides significant computational savings and a substantial increase in the convergence rate (CR) in coloured input environments is presented for the identification of the generalised polyphase components (GPC) of the unknown system. Sparsity constraints are imposed on the input for the identification of polyphase components to be feasible. The proposed structure is then modified so as to relax the imposed constraints on its input and render it appropriate for applications such as the acoustic echo cancellation. The result is an efficient-with respect its computational complexity-adaptive filtering structure that provides an attractive solution in situations where the concern is the reduction in the complexity.