Algebraic differential decorrelation for nonstationary source separation

Abstract

A differential correlation is introduced which is able to capture the time-varying statistics of nonstationary signals and it is shown that minimisation of differential cross-correlations between observation signals can achieve nonstationary source separation. For implementation. an algebraic method is employed, the joint approximate diagonalisation. therefore the resulting method is referred to as Algebraic Differential DEcorrelation (ADDE). The useful behaviour of the method is confirmed by computer simulations.