Acoustic Echo Cancellation During Doubletalk Using Convolutive Blind Source Separation of Signals Having Temporal Dependence

This paper describes a new algorithm for acoustic echo cancellation during doubletalk or, more precisely, acoustic echo separation, based on blind source separation (BSS) of convolutively mixed signals. The signal model assumes independence between sources, but temporal dependence between time samples, specifically that the vector signals have first-order Markov dependence. The source separation is done using a maximum likelihood approach. The source separation does not always provide separation, because of too many degrees of freedom on the separation. However, when applied to the acoustic echo cancellation problem, the constraints of the echo system neatly solve this problem. An example shows that acoustic echoes can be cleanly separated during doubletalk.