Research On Multi-target Identification And Orientation Based On Blind Source Separation

During the past two decades, Blind Source Separation (BSS) has been a focus in the research area of signal processing. As a signal pre-processing technique, there are many potential exciting applications in science and technology, especially in multi-user communication signal processing, speech processing, image processing, biomedical engineering, array signal processing and seism signal processing.To improve the detection performance of passive sonar, present thesis addresses the identification and Direction-of-Arrival (DOA) of underwater acoustic signals using BSS technique.The main contributions are as follows:(1) This paper studies the basic principle and method of BSS, in detail which are conception, theorem, structure, several representative algorithm, function for evaluating the BSS algorithm's performance, and some application and so no.(2) Natural Gradient (NG) algorithm is derived. To be aimed at the problem of separating blindly underwater target-radiated acoustic signals, the method of pre-processing data using eigenvalue decompose is proposed, and a new evaluation function is build up in frequency domain which improves the performance of evaluating target identification.(3) A modified NG (MNG) algorithm, whose non-linear function decided by probability, is introduced into BSS for underwater target-radiated acoustic signals. The results show that the MNG algorithm can separate the mixture signals of sub- or/and super-signals, while the NG algorithm can not.(4) This paper tries to study BSS with noise. The MNG algorithm's separation performance is analyzed in some different SNR. The simulation and experimental results show that the MNG algorithm's anti-noise interference capability is excellent.(5) A new DOA method which based BSS using complex MNG (CMNG) algorithm is introduced. Building up the received signal model of underwater target-radiated acoustic signals, the simulation results verify the validity the CMNG algorithm. The experiment of DOA is carried out using real target-radiated acoustic signals, and their inaccuracies are less than 5%.