Variable-regularized low complexity RLS algorithm for stereophonic acoustic echo cancellation

Abstract

The classical stereophonic acoustic echo cancellation (SAEC) scheme requires the identification of four echo paths using the same number of adaptive algorithms. The SAEC setup can be reinterpreted by employing the widely linear (WL) model, which recasts the two-input/two-output system with real random variables as a single-input/single-output system with complex random variables. The WL model improves the handling and reduces the number of adaptive algorithms to only one, requiring the same computational workload. In this paper, we employ RLS-type adaptive algorithms for the SAEC configuration and we focus on the regularization problem, which is important in the presence of additive noise. Furthermore, we analyze a low complexity version of the RLS, by combining it with the dichotomous coordinate descent (DCD) iterations. We determine a regularization parameter based on the signal-to-noise ratio (SNR). Moreover, by using a suitable estimation of the SNR, we propose a low complexity variable regularized RLS algorithm. Simulations performed in the context of SAEC demonstrate the robustness of the proposed regularization method in noisy conditions, such as the double-talk scenarios.