On blind separation of complex-valued sources by extended Hebbian learning

The aim of this letter is to present a nonlinear extension to Sanger's gene ralized Hebbian learning algorithm for complex-valued data neural processin g, which allows for separating mixed independent circular source signals, T he proposed generalization relies on an interesting interpretation of noncl assical Hebbian learning proposed by Sudjianto and Hassoun for real-valued neural units.