Semi-blind source parameter separation

Abstract

Independent Component Analysis (ICA) is a useful extension of standard Principal Component Analysis (PCA). The ICA model is utilized mainly in blind separation of unknown source signals from their linear mixtures. In some applications, the mixture coefficients are totally unknown, while some knowledge about temporal model exists. In this paper, we propose a learning system for semi-blind binary signal separation. Only second order statistics are used, and therefore the network structure is quite simple. In the experiments, the networks are succesfully applied to the CDMA (Code Division Multiple Access) mobile phone parameter estimation.