Convolutive Blind Source Separation Applied To The Digital Communication Signals

Blind Source Separation(BSS) is an important branch of communication signal processing. In the early stages, since the effects of delayed and filtered when signal is transmitted was not considered, lots of classical methods for linear instantaneous mixture was proposed. However, these algorithms are disabled when they are used to the real signals which are often convolutive mixture in time domain. For this reason, this thesis focused on the theorems, algorithms and applications of determined convolutive blind source separation. For noisy signals, regardless of mixed signals and separated signals, an automatic method of wavelet denoising processing is proposed, which can improve the performance of blind source separation system, and this is confirmed in the experiment. Based on the fact that there is a good amplitude correlation between neighbour bins of a communication signal, we presented a new blind source separation method named Domino Effect relevance ranking which can eliminate the permutation indeterminacy. it is simple, robust and time saving. The algorithm does not need pre-processing of mixed data. Furthermore, results confirmed its validity in blind source separation of communication signals with same carrier frequencies. At the same time, aim at the signals that with same carrier frequencies, a novel Z-domain algorithm for convolutive blind source separation is proposed. There is only one parameter need to set up, that is, the length of sliding window. Thus reduced the computational complexity and improved separation performance. After studying the time-frequency characteristics of convolutive mixed LFM signals, a solution to LFM detection under noisy circumstances is proposed, which consider the results of time-frequency analysis as a picture. In order to achieve the effect of blind source separation, signal feature is extracted by image processing method named hough transform.