4 - Estimateurs du maximum de vraisemblance étendus à la séparation de sources non gaussiennes

A combination of signais emitted by several sources is received on an array ofsensors. In this situation frequently realized in application, we want to separate the sources in obtaining a transformation of the observed signais giving components related for each of them uniquely to a specific source . This separation cannot be successfull using the spectral matrix and then people use a priori information . We show here that this separation can be donc using Higher Order Statistics of the observed signais and we give the maximum likelihood estimator of the source models that defines the separation .