Separation Performance of ICA Algorithms Dependent on the Source Number

Abstract

In order to apply ICA (Independent Component Analysis) to EEG (electroencephalogram) analysis, this paper clarifies that the separation performance of the ICA algorithms depend on the number of unknown source signals. From several experimental results, when the number of the source signals is equal to or less than that of the mixture signals, it is found that the original source signals can be recovered by using ICA algorithm and a scaling adjuster. It is also found that the ICA algorithms works well to extract feature of the sources even if the sources are larger in number than the mixtures.