Reseach On Single Channel Blind Source Separation And Its Application In Underwater Acoustic Signal Processing

Passive target detection has been a research highlight in the researches of underwater acoustics signal processing for a long time. And the Blind Source Separation (BSS) method is suitable for passive signal processing especially when received signals are mixed with too much interferential noise, because it can extract the expected signals from the observation with no or less prior knowledge of the source. The BSS methods are usually used when there are several observing channels, but in reality, sometimes only one observing channel can be used. In this extreme situation, standard BSS method based on array signal processing don't work any more. Thus it is important and challenging for us to deal with the case, that only single channel signal can be used, therefore single channel BSS method and its application on underwater acoustic signal processing have been studied in this thesis.The single channel BSS algorithm of underwater acoustic signal for instantaneous mixtures had been researched firstly. In order to overcome the limitation of single channel, three methods are presented in this part:(1) A single channel BSS method based on virtual channel. According to the spatial character of underwater acoustic signal in far field, a virtual channel is constructed from the observed channel by time delay and filtering. So one single channel is extended to multiple channels, and then the standard BSS methods can be applied. Both the simulation and experimental results are obtained with this method. From the results we can see that this method is effective for the separation of ship-radiating noise in the background of ocean ambient noise with a single observing channel, and it is also feasible to double objects.(2) A single channel BSS method based on Ensemble Empirical Mode Decomposition (EEMD). With EEMD algorithm, the observing signal is decomposed and the input matrix for BBS is obtained. This method combines the advantages of EEMD and BSS, so it works well when the frequency of different sources overlaps and adapts to non-stationary signal processing. Moreover, it is immune to the interference of impulsive noise. Experimental data were acquired from the sea-experiment at Da Lian, and the results show that ship radiated noise can be extracted from background with this method, and improves the Signal-to-Noise Ratio to enhance the performance of signal detection.(3) A single channel BSS method based on interval re-sampling. When the sampling rate is high enough, a new virtual observing data matrix can be constructed by re-sampling the observing data. By this way it overcomes the limitation of single channel. Both the simulation and experimental results have shown that this method is effective. And this method needn't adjust too many parameters, so it's easily realized.In first part of thesis, the insufficiency of channel number is emphasized and three BSS methods are developed. Since multi-path effect always exists during sound transmission, which degrades the performance of single channel BSS algorithm. Thus, single channel BSS under multipath transmission condition is investigated in the second part.In order to fully utilize the information of multipath and improve the performance of BSS, multipath focusing is introduced to the single channel BSS. Firstly, multipath time delay is estimated by autocorrelation and the signals are focused with phase compensation, therefore SNR of the observation is improved. Then the method of Signal Channel BSS can be used. Autocorrelation method is fit to broad band signal processing, and has a good ability to reduce noise. Simulation and experimental results show that the method suppresses the influence of multipath greatly and improves the single channel BSS.When the observation preserves strong line spectrum components, we can not estimate the multipath time delay any longer with autocorrelation. Fortunately the cepstrum method is feasible to treat this problem, and an improved cepstrum method is presented which degrades the noise. The method can also focus the multipath, extract the line spectrum signal efficiently, and it will work better when the ship-radiating noise contains low frequency line spectrum components.To sum up, the single channel blind source separation is presented in this thesis. The feasibility of the method is verified with experimental data and good performance has been shown on the processing of underwater passive signal. Moreover, these methods can simplify the experimental hardware system. And they have an explicit application prospect under the condition of limitation of channels. At the same time, they also have advantages in ocean dynamic monitoring and passive sonar signal processing.