31 A BLINDSIGNALSEPARATION TECHNIQUEUSINGCOMBINATIONOF SECOND-ORDERAND HIGHER-ORDER APPROACHES
Fakhry Mahmmoud · 2006
Inthis paper, anovel technique toblind signal separation based oncombination ofthesecond order andthehigher order approach isintroduced. We transfer theproblem of blind signal separation tothewavelet domain. Themain idea behind this approach isthat themixing signal canbe decomposed into sumofuncorrelated and/or independent subbands using thewavelet transform. Inthebeginning, the observed signal ispre-whitened inthetimedomain then, theinitial separation matrix will beestimated fromsecond order statistics decorrelation method inthewavelet domain. Theestimating matrix will beusedasinitial separating matrix inthehigher order statistics method toestimate the final separation matrix. Thealgorithm istested using different examples different examples (chirp signals, sound signal andnatural images). Ourextensive experiments haveconfirmed thattheuseoftheproposed procedure