Research On Single Channel Blind Source Separation For Target Echo

When applying the active sonar into the target detection and identification in the underwater environment, due to amount of reasons such as the complication and variety of sea environment, it seriously effects the extraction and judgement of demanding information with so much noise/reverberation in the obtained echo observed signal for various situations.Blind source separation, however, with the advantage of nearly requiring any prior acknowledge of the source signal or the environment, that it could be utilized in the separation between the target echo and the noise/reverberation, realizing the rising of signal-to-noise or signal-to-reverberation ratio. Even though, facing the limited condition of providing single observed signal, through the pre-processing of decomposition in this paper, one could realize the interference restrain of observed signal when processing the single channel blind source separation.In this paper, starting from the characteristic and property of common underwater acoustic signal, respectively, three common used blind source separation are adopted here,including Second Order Nonstationary Source Separation (SONS), Time Frequency Blind Source Separation (TFBSS) and Blind Source Separation based on SNR Maxing, then their owe advantages and characteristics for underwater acoustic signal are obtained after specifically deducing the theory. In the meantime, for better comparison of discussing the separation results of each algorithm, not only common separation criteria being used as numerical value comparison, but also the WVD time-frequency figure being adopted to more direct comparison after analyzing the final results. For there, the pre-processing of single observed signal, the empirical mode decomposition (EMD) and smoothing noise based advanced ensemble empirical mode decomposition (EEMD) are invited here, then to compare the results and advantages for various environments.For the specific comparison, in this paper, amount of simulations of separation between different types of signal and noise are given. Also, to verity the property of widely used, the noise restrain for multiple-source signals is showed afterwards. For one of the important simulation results, combining method of single channel blind source separation with a common method of signal detection (energy detection) of wireless communication is discussed as an emphasis, which could improve the normal energy detection for spectrum sensing by providing the better estimation of noise power and leads to a better detection capability. At last, successfully processing the real data demonstrates the accuracy and effectiveness of the provided approaches. Through the single channel blind source separation implementation, target echoes of clearness are shown in the WVD figure as well as the correlation coefficients are raised, realizing the restrain of the noise/reverberation.