Fetal ECG separation using non-parametric ICA algorithm

Abstract

Performances of former independent component analysis (ICA) algorithms depend on the true probability density function of each source, but in practice these densities are unknown. To handle this problem Non-parametric and parametric ICA algorithms have been developed. In this paper it is studied the separation of maternal and fetal heart beats from electrocardiogram (ECG) recordings based on FastICA and Non-parametric ICA algorithms and differences of algorithms are investigated on ECG signal. The most important two properties of Non-parametric algorithm are it's performance is not dependent upon prior assumptions about the source probability distribution and it is also capable of accurately and efficiently estimating unmixing matrix, and which doesn't require the selection of any tuning parameters. The simulations demonstrate that non-parametric ICA algorithm and FastICA algorithm can accurately separate fetal and maternal ECG signals.