Blind source separation using higher order time-frequency representations

A novel blind separation approach using higher-order time-frequency distributions is presented. The concept of higher-order time-frequency distribution matrix is also introduced. It is devised to primarily separate sources with temporal nonstationary signal characteristics. So far, this problem has been solved using statistical information available on the source signals. In contrast to well known blind source separation approaches using second-order statistics (SOS) and/or higher-order statistics (HOS) which rely on stationarity properties of the signals, the proposed approach allows separation of the sources with nonstationarity properties. In addition, the effect of spreading the noise power while localizing of the source energy in the time-frequency domain amounts to increasing the signal to noise ratio. A computationally feasible implementation is presented based on joint diagonalization of the matrices of the principal slices of the time-multifrequency domain of support of the cumulant-based Wigner trispectrums. Numerical examples demonstrate the effectiveness of the proposed approach.