A Study On Blind Separation Of Underwater Acoustic Signals

To improve the detection performance of passive sonar, present dissertation addresses the blind separation of underwater acoustic signals. Present existing blind source separation (BSS) algorithms of real domain data are reviewed. A variety of complex algorithms for complex domain data are derived and demonstrated, and verified by tests both in pool and lake.The main contributions are as follows:(1) Both the received signal model and blind signal separation model for narrow band underwater acoustic arrays are analyzed. Computer simulation methods of blind underwater acoustic signal separation algorithms and evaluating criterion of the BSS algorithms are proposed. Based on real ship radiated noise, two methods of BSS are presented. One is the principal component analysis (PCA) for Gaussian signals, while another is independent component analysis (ICA) for non-Gaussian signals.(2) General approach for noise-free BSS and classical BSS algorithms of real domain data are discussed. A class of BSS complex algorithms is introduced. Computer simulation demonstrated the performance. The algorithms for underwater acoustic signals within isotropic uniform noise field are studied. Some encouraging results are obtained.(3) Considering that the environment noise of ocean is Gaussian distribution, a one-source BSE algorithm based on the logarithm-kurtosis is proposed. Meanwhile two deflation complex algorithms are derived. Using the one-source BSE and deflation algorithm in turn, Multiple sources can be separated while the number of sources needs not be known in advance. One-source BSE algorithms are directly usable to blind beam forming.(4) Incorporating spatial information into BSS algorithms, a class of time-spatial algorithms is derived, which restrict the search range in spatial and improved the convergence rate. The BSS complex algorithms can be applied to calibrate the manifold of acoustic array and estimate DOA of source.(5) Some BSS algorithms are exploringly studied in the case of more sources than sensors. An algorithm for blind estimation of complex mixed matrix with arbitrary number of sources is proposed. Computer simulation has demonstrated the property.