Steady-state solutions of the extended LMS algorithm for stereophonic acoustic echo cancellation with leakage or signal conditioning

Tetsuya Hoya, Jonathon A. Chambers, Neil Forsyth, Patrick Naylor

Abstract: Acoustic echo cancellers are widely employed in teleconferencing systems to reduce the undesired echos resulting from coupling between loudspeaker and microphone. The single-channel case has been widely studied. Of particular recent interest is the stereophonic case which is not difficult to solve due to the strongly correlated two-channel inputs. In this paper we compare the steady-state solutions of the Leaky extended LMS (XLMS) algorithm with XLMS having inputs conditioned by additional zero-memory non-linearities. Modification of the correlation matrix of the two channel-inputs is analysed. We also describe a new configuration for the zero-memory non-linearities which does not impact upon sound quality whilst maintaining improved algorithm convergence properties. Simulation results to support the analyses of these two different de-correlation methods are also included, which suggest that for deterministic parameter settings the performance of the Leaky XLMS algorithm is superior to the case where a half-wave rectifying non-linearity is used. Moreover, simulation results where time variations in the transmission room are considered in order to represent more realistic situations are given, which, on the other hand, suggest that the performance of the XLMS with a signal conditioning method using an Half-Wave Rectifier (HWR) is superior to that of the Leaky XLMS algorithm.