An Approach of Moment-Based Algorithm for Noisy ICA Models

Factor analysis is well known technique to uncorrelate observed signals with Gaussina noises before ICA (Independent Component Analysis) algorithms are applied. However, factor analysis is not applicable when the number of source signals are more than that of Ledermann’s bound, and when the observations are contaminated by non-Gaussian noises. In this paper, an approach is proposed based on higher-order moments of signals and noises in order to overcome those constraints.