A novel approach to convolutive blind separation of close-to-orthogonal pulse sources using second-order statistics

We introduce a novel approach suitable for blind separation of convolutive Multiple-Input-Multiple-Output mixtures of orthogonal pulse sources. The approach is based on a joint-diagonalization of the algebraic combinations of whitened correlation matrices at non-zero lags. It supposes constant and finite system impulse responses and more measurements than sources. Preliminary tests on the synthetic signals prove 100% accuracy in detection of source pulses even if the sources are not completely orthogonal and when the measurements are noisy.