Adaptive blind separation of convolved sources based on minimization of the generalized instantaneous energy

Generalization of the energy concept is proposed resulting in a class of novel on-line algorithms for blind separation of convolved signals. Signal separation is achieved when signal energy of the appropriated order is minimal. The resulting learning rules have the similar form as those recently discussed to be optimal for blind separation of instantaneously mixed signals. Algorithms are tested on the separation of two real-world signals. It is believed that for the first time the blind signal separation (BSS) theory is applied to the light sources localization problem. With proposed algorithms better separation quality is obtained than when using adaptive decorrelation, recently proposed separation algorithm based on entropy maximization and neural network separator based on nonlinear odd activation functions.