BLIND SIGNAL SEPARATION .2. LINEAR, CONVOLUTIVE COMBINATIONS

Blind signal separation is the process of extracting unknown independe nt source signals from sensor measurements which are unknown combinati ons of the source signals. The term ''blind'' is used as the source si gnals and the method of combination are unknown, and hence the problem is related to the problems of blind deconvolution and blind equalizat ion. Blind signal separation is sometimes referred to as independent c omponent analysis (InCA), as it generalizes principal component analys is to produce independent signals rather than simply uncorrelated sign als. The problem of blind signal separation has been investigated in d etail during the past ten years. The work has been driven by a wide va riety of interests and areas of application, such as array beam-formin g, higher-order statistics, neural networks and artificial learning, n oise cancellation, and speech enhancement. However, no review of the a vailable literature has been published. This paper is one of two paper s seeking to redress this point. Whereas Part I focused on the separat ion of sources that have combined in a linear, instantaneous fashion, this paper considers a more complicated separation problem in which th e combination of the sources is linear and convolutive. In addition, a variety of issues of importance to blind signal separation problems i n general are also discussed. (C) 1996 Academic Press, Inc.