Blind source separation of statistically correlated sources

Standard independent component analysis is unable to separate statistically correlated machine vibration sources because the assumption of statistical independence is no more satisfied.Assuming some sub-components of correlated machine vibration sources are independent,a novel blind source separation method for correlated vibration sources seperation based on wavelet packet decompostion was proposed.In the method,subband observed signals were obtained by decomposing orignal observed signals with wavelet packet transform,the new observed signals were reconstructed by some subband observed signals with less correlation according to the mutual information creterion,the separation matrix was estimated through the reconstructed observed signals by using independent component analysis;then,the correlated machine vibration sources were separated in agreement with the estimated separation matrix.Simulations testify the validity of the proposed method.