Separation of Synchronous Sources

Abstract

This thesis studies the Separation of Synchronous Sources (SSS) problem, which deals with the separation of signals resulting from a linear mixing of sources whose phases are synchronous. While this study is made in a form independent of the application, a motivation from a neuroscience perspective is presented. Traditional methods for Blind Source Separation, such as Independent Component Analysis (ICA), cannot address this problem because synchronous sources are highly dependent. We provide sufficient conditions for SSS to be an identifiable problem, and quantify the effect of prewhitening on the difficulty of SSS. We also present two algorithms to solve SSS. Extensive studies on simulated data illustrate that these algorithms yield substantially better results when compared with ICA methods. We conclude that these algorithms can successfully perform SSS in varying configurations (number of sources, number of sensors, level of additive noise, phase lag between sources, among others). Theoretical properties of one of these algorithms are also presented. Future work is discussed extensively, showing that this area of study is far from resolved and still presents interesting challenges.