Extended Hebbian learning for blind separation of complex-valued sources

The aim of this work is to present a nonlinear extension to Sanger´s generalized Hebbian learning algorithm for complex-valued signal processing by neural networks. A possible choice of the involved nonlinearity is discussed by recalling the Sudjianto-Hassoun interpretation of nonlinear Hebbian learning. An extension of this interpretation to the complex-valued case leads to a Rayleigh nonlinearity, that allows for separating mixed independent complex-valued circular source signals.