On blind separation of complex-valued sources by extended Hebbianlearning

摘要

The aim of this letter is to present a nonlinear extension tonSanger's (1989) generalized Hebbian learning algorithm forncomplex-valued data neural processing, which allows for separating mixednindependent circular source signals. The proposed generalization reliesnon an interesting interpretation of nonclassical Hebbian learningnproposed by Sudjianto and Hassoun (1994) for real-valued neural units