Two novel architectures for the self adaptive separation of signals

It is known that self-adaptive separation of a linear mixture of non-Gaussian independent signals can be achieved with a feedback structure that is adapted by the Jutten-Herault algorithm. Two novel architectures are presented. The first one is direct (no feedback). The second one is a mixed architecture, where a feedback structure is cascaded after a direct structure. Equlibrium points and their stability conditions are investigated for two mixed signals. The direct architecture is more general than the feedback one in that it involves no implementation constraints. It converges faster when there is an even number of minus signs in the mixture matrix. The mixed architecture's transient period is hardly dependent on the mixture matrix. Computer simulations corroborate the theory.<<ETX>>