A practical Approach Based on Gaussianization for
Stefano Squartini, Alessandro Bastari · 2006
Thisworkdeals withtheBlindSourceSeparation (BSS)probleminpresence ofmoresources thansensors and Post-Nonlinear (PNL)mixing. Theinterest onthesubject is increased bythefact thatveryfewrelated contributions have appeared intheliterature sofar. Theproposed methodismade ofthree separate steps: compensation ofnonlinearity (based on theGaussianization concept), mixing matrix recovery andfinal unknownsourceestimation. Thefirst onehasbeenalready considered fornonlinear complete BSS,butnotintheover- complete case, whereas thelatter tworepresent thetypical two- stepapproach inunderdetermined BSS.Performed computer simulations haveshowntheeffectiveness oftheidea, evenin presence ofstrong nonlinearities andsynthetic mixture ofreal worlddata(like speech signals).