Blind separation of signal and reverberation by minimizing the circular variance of the phase

A new canonical correlation-like algorithm is developed for blindly separating signal and interfering reverberation components using a small vertical array in a synthetic aperture sonar application when no signal-free interference data is available for training. The new approach exploits differences in the ping-to-ping coherence properties of the signal and reverberation and is based on estimating the filter weights through minimization of the output ping-to-ping phase circular variance. As a result, the performance is close to the Cramer-Rao lower bounds (CRLB) and is competitive with parametric high resolution bearing estimation and in addition does not need a precisely calibrated array.