Single Channel Blind Source Separation Using Dual Extended Kalman Filter

Single channel Blind Source Separation (SCBSS) is an important source separation technique gaining prominence in many emerging applications. It is a special case of the well-defined Blind Source Separation (BSS) where only a single mixed signal is recorded to estimate the unknown sources. In this paper, we propose a simultaneous state-parameter estimation methodology for SCBSS using Dual Extended Kalman Filter (D-EKF). The proposed methodology eliminates the inherent frequency disjoint and statistical independence limitations of the state-of-the-art SCBSS approaches such as single channel Independent Component Analysis (SCICA). A frame- based Kalman processing technique has been proposed to ensure faster convergence of the proposed methodology. Simulation results have been presented for mixed sources with overlapping spectra and compared with SCICA and other BSS algorithms. The results demonstrate the superior performance of the proposed methodology with improved Signal-to-Interference Ratio (SIR) and Signal-to-Distortion Ratio (SDR) for real-world practical applications.