A one step time-frequency blind identification

Blind source separation using time-frequency distributions is considered in this contribution. We propose a time-frequency source separation technique based on the joint diagonalization procedure introduced in D. T. Pham and J. F. Cardoso (1997). The approach exploits the auto-terms of the time-frequency distributions of the sources. It is based on the joint-diagonalization of the spatial time-frequency distributions of the received data. The joint-diagonalization is optimized without any orthogonality constraint, which bypasses any prior whitening of the observations. Numerical simulations are provided to demonstrate the effectiveness of our proposed method.