Improving sparse echo cancellation via convex combination of two NLMS filters with different lengths

In this paper, we propose a scheme for sparse echo cancellation which uses a convex combination of two normalized least-mean-squares (NLMS) filters with different lengths. As is normally the case in acoustic echo cancellation, the first filter includes a large number of taps to guarantee that the active (i.e., non-null) coefficients of the true echo path are correctly identified. The second filter is a shorter and faster one, intended to span just the region of active coefficients. To identify this active region, we present a method based on clustering of the combined filter coefficients. We also propose two different combination strategies that simultaneously improve steady-state and convergence performance. When the echo path is very sparse, the computational cost incurred by our schemes is just slightly higher than that of a single NLMS filter. Simulation results show the superior performance of the proposed schemes when compared to other methods in the literature.