A novel variable tap-length learning algorithm for low complexity, fast converging stereophonic acoustic echo cancellation

Abstract

An advanced low complexity, fast converging variable tap-length (VT) learning algorithm for stereophonic acoustic echo cancellation (SAEC) based on multiple sub-filters (MSF) approach is designed and analysed in this paper. The proposed algorithm retains advantage of both VT different error algorithm (DEA) and common error algorithm (CEA). A systematic procedure is presented to set key variables that affect the structure adaptation to provide the best performance when compared to the existing stereophonic echo cancellers. The convergence performance of the MSF-based parallel structure is studied, for the designed dynamic structure and improvements are addressed over the VT-single long filter, VT-CEA, VT-DEA. A VT-selective coefficient update method to reduce the computational cost of adaptive design is discussed in context with the proposed design. It leads to the closest possible performance to the full update algorithm. Simulated results are shown to make a comparison of the proposed design with existing SAEC algorithms.