Noncircular Complex ICA by Generalized Householder Reflections

We develop an efficient algorithm for noncircular complex independent component analysis (ICA) by optimizing the rows of separation matrix independently using a generalized Householder reflector and Newton method. We apply the new algorithm to the separation of complex generalized Gaussian distributed (GGD) sources, and propose a convenient way to learn the noncircularity of the sources. Simulation results are reported to study the convergence behavior, and confirm the superior performance of the new algorithm.