Blind separation of sources applied to convolutive mixtures in shallow water

Abstract

In underwater acoustics, the signal received by sensors is a mixture of different elementary sources, filtered by the environment. In blind separation of sources, we can isolate each source from different mixtures of sources without any a priori information, except for assuming statistical independence of the different sources. Jutten and Herault (1991) proposed a neuromimetic solution to the problem. In our work, we use this solution to separate convolutive mixtures of simulated complex underwater signals in a shallow water environment. To allow multipath identification a whitening step has to be introduced. We propose a local whitening procedure that does not impact the separated signal output and preserves the signal characteristics. This promising technique can be improved using non causal whitening filters more adapted to the target environment.