On the Relation between Data-Dependent Beamforming and Multichannel Linear Prediction for Dereverberation

The generalized sidelobe canceler, a data-dependent beamformer that is commonly used in noise suppression, is able to perform dereverberation if the source signal is a random white signal. A similar statement can be made for multichannel linear prediction, which may be used to blindly invert any time-invariant multichannel transmission system that is excited by a random white signal, provided that the transmission channels do not share common zeros. If the source signal is colored on the other hand, as it is the case for speech signals, both the generalized sidelobe canceler and multichannel linear prediction tend to additionally invert the source coloration, and different approaches have been proposed to tackle this problem. In this paper we give an overview on multichannel linear prediction methods and formally analyze the generalized sidelobe canceler for dereverberation, which reveals close relations between the two approaches.