Study on blind separation of underwater acoustic signals

It is difficult to detect and recognize underwater targets of interest due to oceanic ambient noise and signal distortion resulting from the multipath effect. An ideal method to overcome this problem is to accurately estimate original source signals from this kind of complex signal. The number of unknown sources is estimated by principal component analysis using information theoretic criterion. The presented idea shows that the source signals of interest could be reconstructed by a blind source separation technique from the simulated mixed complex underwater signals. The multipath effect in underwater acoustic signal transmission is approximately considered as a convolutive mixture. Experiments based on simulated data achieved satisfactory results.