A Study of the epsilon-NLMS Algorithm With Application to Stereophonic Acoustic Echo Cancellation

Tetsuya Hoya, Jonathon A. Chambers, Patrick Naylor

Abstract: In Stereophonic Acoustic Echo Cancellation (SAEC), one of the fundamental problems lies in the misalignment in the filter coefficients due to the two strongly correlated channel-inputs. In this paper, we study the effect of the normalisation factor, ffl, upon the convergence properties of the two-channel Normalised Least Mean Square (ffl-NLMS) algorithm, through analysis and simulation studies with real speech datasets. We then show that the optimal choice for ffl may be close to the variance of the channel-input data. Finally, a subband stereo echo canceller structure which uses a combination of the two-channel ffl-NLMS and the Fast Least Squares (FLS) algorithms is proposed as a practical and promising solution to SAEC.