Spectral Domain B-spline Identification in Acoustic Echo Cancellation

Abstract

Spectral domain B-spline identification is proposed for acoustic echo cancellation. Two approaches are considered. The first is based on solution of normal equations; we describe an efficient technique for such a solution, which benefits from the sparseness of the system matrix due to B-splines. The second approach is based on using local splines, enabling further simplification. We also show how the proposed techniques can be used for efficient double-talk detection. The echo cancellation performance and complexity of the proposed techniques are compared with that of a low-complexity cross-spectral technique and the affine projection (AP) algorithm possessing high cancellation performance. The Bspline identification allows cancellation performance comparable with that of the AP algorithm and complexity close to that of the cross-spectral algorithm.