A MISO acoustic echo cancellation algorithm based on a two-layer filter decomposition

The recursive least-squares (RLS) algorithm is a promising algorithm in acoustic echo cancellation (AEC) thanks to its fast convergence rate and competitive performance. However, its complexity is rather high, particularly when the system operates in acoustic environments with long acoustic impulse responses. This paper deals with the problem of AEC in a multiple-input and single-output (MISO) audio system, which consists of multiple loudspeakers and a microphone at the receiving room. A method is developed in the short-time-Fourier-transform (STFT) domain, which operates on a subband basis. In every STFT subband, the convolutive-transfer-function (CTF) model is adopted, so the echo path is modeled with a finite impulse response (FIR) filter. A two-layer decomposition (TLD) of the filter matrix is then applied and an RLS-type of algorithm is subsequently deduced to achieve channel identification and echo cancellation. This algorithm is able to achieve echo cancellation performance comparable to RLS algorithm with significantly lower complexity.