Independent Components Analysis (ICA) has recently received considerable attention because of the potentialities showed when applied to problems such as Blind Sources Separation (BSS). ICA can be used whenever you desire to recover signals generated from several statistical independent sources and only some linear combinations of them are available.
In my thesis work I have compared the basic criterions used to impose the independence constraint on the recovered components and the most important ICA's algorithms recently proposed in literature. For each algorithm the theoretical limits for the asymptotic estimation error, the performances in terms of convergence rate and asymptotic SIR (Signal-to-Interference Ratio) and the computational complexity have been evaluated. Moreover an extension of the ICA techniques to the Multichannel Blind Deconvolution problem, based on the use of the space-time independence assumption among the source signals, has been proposed. At last some typical applications of Blind Source Separation have been considered.
In the following pages you will find a preview of the principal arguments treated in my thesis work. For further information don't esitate to contact me.
BSS problem formulation.
ICA as minimization of contrast functions.
Stochastic gradient algorithms.
Next | Back to Polimi
If you need to contact the author: firstname.lastname@example.org