Research On Blind Source Separation Of Underwater Acoustic Signal With Time-frequency Analysis

Sonars play a significant role in underwater communication, which are main electronic devices during the process of underwater signal processing. Mostly, working frequencies of active sonars converge from about 3KHz to 97 KHz, and they reach the range from 3Hz to 97 KHz when happened to passive sonars. Underwater acoustic channels can be considered as channels which show time-varying and space-variant characteristics, meanwhile, some nonlinear components of hydrophones receiving devices lead to situations that observed signals are commonly mixtures of multiple non-stationary signals. Numerous external disturbances and noises exist in the underwater acoustic environment, which interfere with later performances of object detection severely. Time-frequency transformation is a way to analyze non-stationary signals, unknown hybrid signals can be separated through blind source separation. Combining time-frequency transformation with blind source separation is the most direct way to analyze non-stationary signals in the underwater acoustic channels.This paper is based on applications of blind source separation in the underwater acoustic field, and it focuses on the non-stationary characteristic of underwater acoustic signals to search for more practical methods of blind source separation. Reason and the researching significance for analyzing underwater acoustic channels based on algorithms time-frequency and blind source separation were given, and the status of researches at home or abroad was elaborated in the paper. It simply introduces the models for blind source separation, and evaluates performances of the algorithm through signal-to-interference ratio index. Time-frequency transformation is a strong means to analyze non-stationary signals, and there are three kinds of time-frequency transformations: linear time-frequency transformation, nonlinear time-frequency transformation and Hilbert-Huang transformation. The simulation combines these three kinds of time-frequency transformations with blind source separation respectively for blind source separation of underwater acoustic signals. Results of simulation can clearly embody algorithm performances of the blind source separation for underwater acoustic signals, they will provide some references for future researches.In terms of blind source separation for signals based on nonlinear time-frequency transformation, In this paper firstly introduces theoretical basis of common bilinear time-frequency transformation; then applies methods of blind source separation suitable for non-stationary signals, and proves them by computer simulation next. It uses linear time-frequency to overcome the drawbacks of cross terms existed in bilinear time-frequency analysis, and searches for single source domain of observed signals by time-frequency ratio to estimate hybrid matrix and source signals, by which to achieve signals’ blind source separation. Aimed at Empirical Mode Decomposition’s application of decomposing multi-component signals into single-component signals and hybrid signals’ common characteristic of multi-component signals consisting of single-component signals, calculates IMF and observed underwater acoustic signals’ correlation coefficients, eliminates false IMF and achieves blind source separation for multi-component underwater acoustic signals by Hilbert-Huang transformation. Non-stationary low-frequency underwater acoustic signals simulated prove feasibility and validity of Hilbert-Huang transformation algorithm. Finally, it discusses applications of several algorithms to signal processing of single vector hydrophones, which provide probability for future applications.