Efficient Multi-Channel Acoustic Echo Cancellation Using Constrained Sparse Filter Updates in the Subband Domain

In this paper we present constrained sparse tap-selection schemes for updating Multi-Channel Acoustic Echo Cancellation (MAEC) filters in the subband domain. At first, M-Max tap-selection on the complete multi-channel reference spectra is performed and the effect of the subsequent sparse filter update on the speed of convergence of the MAEC filters is investigated. We consider a measure to quantify signal sparsity, and use it to investigate the spectral and inter-channel sparsity present in real-world surround sound signals. A heuristic tap-selection scheme is proposed which exploits the signal sparsity properties, and performs tap-selection with much lower computational effort as compared to the M-Max and the full-update tapselection schemes, while giving similar echo cancellation performance.