V. Duong and A.R. Stubberud (USA)
Independent Component Analysis, Space time mixture, Joint Approximative Diagonalization of Eigenmatrices, higher order cumulant.
Independent Component Analysis (ICA) hasr ecently attracted a great deal of attention in the signal analysis and processing fields. ICA can recover from an observed signal a set of unobserved and underlying independent source signals that were combined to make up the observed signal. The recovered source signals are as statistically independent as the ICA algorithm allows.I n linear ICA, the observed has been generated by a set of linear combinations and/or convolutions of unknown independent signal sources. The way in which the independent signals have been mixed need not be knownf or ICA to work. In this paper, we extend the linear ICAt o separate mixtures in both space and time in which theo bserved signals are linear combinations and/or convolutions of the independent sources; and we show that a space-time ICA can be separated into twoi ndependent processes: space ICA and time ICA. Thus, the complex problem of space-time ICA is reduced to two simpler problems, one of time ICA and one of space ICA,w hich are very similar. In simulations, the proposedm ethod has succeeded in recovering original signals from mixtures, which have been mixed in both time and space.
Important Links:
Go Back
