A New Algorithm for Complex-Valued Independent Component Analysis with Unitary Constraint

In this paper, we present a novel one-step learning algorithm based on geodesic approximation for the problem of complex-valued independent component analysis (CICA) under unitary constraint. With the special intrinsic geometrical structure of the problem, we can perform the optimization directly on the curved space without any additional projecting operation. Compared with the classical algorithms, the algorithm proposed in this paper possesses a more compact and simpler form. Finally, experimental results indicate its validity and excellent performance.