Source Separation of Phase-Locked Subspaces

摘要

ICA can be interpreted as the minimization of a disorder parameter, such as the sum of the mutual informations between the estimated sources. Following this interpretation, we present a disorder parameter minimization algorithm for the separation of sources organized in subspaces when the phase synchrony within a subspace is high and the phase synchrony across subspaces is low. We demonstrate that a previously reported algorithm for this type of situation has poor performance and present a modified version, called Independent Phase Analysis (IPA), which drastically improves the quality of results. We study the performance of IPA for different numbers of sources and discuss further improvements that are necessary for its application to real data.