Sparsity-aware adaptive filtering based on a Douglas-Rachford splitting

摘要

In this paper, we propose a novel online scheme for the sparse adaptive filtering problem. It is based on a formulation of the adaptive filtering problem as a minimization of the sum of (possibly nonsmooth) convex functions. Our proposed scheme is a time-varying extension of the so-called Douglas-Rachford splitting method. It covers many existing adaptive filtering algorithms as special cases. We show several examples of special choices of the cost functions that reproduce those existing algorithms. Our scheme achieves a monotone decrease of an upper bound of the distance to the solution set of the minimization under certain conditions. We applied a simple algorithm that falls under our scheme to a sparse echo cancellation problem where it shows excellent convergence performance.