Constrained independent component analysis techniques

摘要

Independent component analysis (ICA) is a promising statistical signal processing technique. To overcome the inherent drawbacks encountered in the conventional ICA method, a general framework of constrained ICA is introduced. The prior knowledge of reference is incorporated into a negentropy based objective function so as to construct a constrained ICA problem. Subsequently, a flexible constrained ICA algorithm is derived for extraction of one or a few desired source signals. The utility of the proposed algorithm is demonstrated by computer simulations on real ECG data.