Study On Blind Separation Technology And Its Application Of Underwater Acoustic Signals

In order to improve the detection and processing ability of array sonar signals, the theories and techniques of blind signal processing are studied and applied to underwater acoustic signal processing. In this dissertation, the basic theory on blind source separation (BSS) is emphasized for underwater acoustic signals, and various application methods of BSS are proposed or modified, so as to purify multi-target underwater acoustic signals and to recover signal waveforms from noise contaminated signals. The algorithm models studied and proposed are verified by using the data of computer generated, the semi-simulated data of underwater acoustic signals, the data obtained from experiments in water-tank, and the data got from experiments on the sea. The main research contents and achievements are as follows:(1) According to the current achievement on the theory and application of blind signals processing, this dissertation studies the theoritical models of underwater acoustic signals and analyzes the problems and the difficulties confronted of array underwater acoustic signal blind separation. This dissertation presents a modified blind separation algorithm for noisy ship radiated signals based on the joint diagonalization with information theory and higher-order cumulants. A blind equalization method for deconvolution is proposed by directly computing the eigenvector of special cumluents matrix, and applied to recover waveform of underwater acoustic signals. The proposed BSS approaches for underwater acoustic signals are verified with simulations and experiments.For the problem of unknown number in the mixed signals, a new principal component analysis (PCA) based method is presented and applied to estimate blindly the number of source signals. A BSS method for underwater acoustic signals is developed in beamforming domain. The number of target sources is determined with the number of mutli-beamforming for array signals. This method can not only enhance the performance of BSS under low signal-to-noise ratio conditions, but also adjust the order of blind separation outputs depending on the direction of arrival (DOA) of mutli-target signals. The method can solve the permutation problem efficiently.(2) This dissertation studied the techniques of convolutive blind source separation in frequency domain. It transforms firstly observation signals with short fast Fourier transform (FFT), and then puts forward a new frequency domain BSS approach by optimizing costfunction of multiple tap cross-correlations for non-stationary signals. It not only analyzes and verifies the influences of parameters to the performance of blind separation for both length of Fourier transform and order of separation filter, but also discusses how to choose those parameters optimally. What is more, this dissertation proposes a frequency domain blind separation method by using a few special frequency bins. The new method ensures the main components of signals be separated effectively as well as reduce the calculation complexity sharply, which is useful in the convolutive blind separation. Meanwhile, the selection rule for choosing the special frequency bins and the extracting method are suggested with a line spectrum enhancement. It proposes to a method to adjust the output order for multiple frequency bins by utilizing similarity coefficient of waveforms, so that the output order is kept coinciding with all the special frequency bins. This dissertation indicates that the more number of array elements are used, the lower SNR signals can be separated under the same condition. Computer simulation and data analysis show that the proposed frequency domain BSS algorithms are effective methods for separating signals from interferences and noises.(3) This thesis studies and analyzes the distribution characteristics of both ocean environment noise and the ship radiated underwater acoustic noise. A conclusion is obtained that underwater acoustic signals are accord with the fractional lower-order Alpha stable distribution. A new detecting model is established for underwater acoustic signals, and a new adaptive filtering method is proposed based on fractional lower order moments for underwater acoustic signals. The effect of the parameter p of fractional lower order moments is analyzed and the selecting rule is suggested. A novel algorithm of blind separation was deduced based on fractional lower order moments for underwater acoustic signals. Theoretical analysis and computer simulation show that the new blind separation method is robust under various noise conditions. This dissertation put forward a new idea for improving performance of underwater acoustic signal processing.(4) This dissertation designed a series of experiments, especially the sea experiments, in order to verify the blind separation approaches for underwater acoustic signals proposed in this dissertation. The validity and robustness of algorithms are confirmed in practice by using recorded multi-target signals and ship radiated underwater acoustic signals on ocean. The algorithms of blind separation have the enhanced ability to detect and separate weak signals by increasing the number of array elements.