AN ITERATIVEICAMETHOD FOR SEPARATINGfMRI ACOUSTICNOISECOMPONENTS
Richard W. Briggs · 2007
Anew iterative algorithm forseparatingequalization schemes thatemploystochastic gradient descent mixtures ofmulti-channel signals isproposed. Thisalgorithm procedures tominimize non-convex costfunctions depending on extends theinstantaneous independent component analysistheequalizer output signals. TheBussgang algorithms aresimple algorithm tomulti-channel source separation algorithm.andeasytoimplement, butthey mayconverge towrongsolutions Separation isprocessed bydecomposing convolutive mixtures resulting inpoorperformance oftheequalizer (9). The toinstantaneous mixtures. Simulation results forrealfMRI deconvolution approach basedon NGA (Natural Gradient (functional Magnetic Resonance Imaging) scanner noise show Algorithm) wasdeveloped byAmari, etal., andthealgorithm thattheproposed algorithm isveryeffective inblind source usually needsfewernumberofiterations thantheordinary separation. gradient algorithms (10). However, thealgorithms cannot adaptively determine thelength ofthedeconvolutive filters. I.INTODUCTION Hence, itisdifficult tochoose thelength ofthedeconvolution uring thepastdecades, lotofattention hasbeenpaid toblind filters whenthere isnoprior knowledge about themixing process. I)source separation (BSS), because ofitssignificant potential Inthispaper, we introduce anewalgorithm forthemulti- inawidevariety offields like biomedical signal processing and channel source separation. Thisworkisbasedonasimple wireless communication systems. BSSistheprocess ofextractingdeconvolution modelwhichcantransfer theinstantaneous ICA unknownindependent source signals fromsensor measurementsalgorithm totheconvolutive source separation algorithm whichareunknowncombinations ofthesource signals (1). The directly. Therest ofthepaper isarranged asfollows. InSection II, blind caseiswhereboththeoriginal source signal andthe weformulate theBSSproblem. InSection III, theproposed multi- combination process areunknown, andonlythemeasured channel separation algorithm isdeveloped. Theperformance recordings ofthemixtures areavailable. Whenthecombination of oftheproposed algorithm isevaluated inSection IV.Finally, thesources islinear andinstantaneous, BSSissometimes referredsection V concludes thepaper. toasindependent component analysis (ICA)(2). Inthelast several years, manyalgorithms havebeendeveloped fortheICA