With Noise Fanaticism, Extraction And Separation Technology Research

In recent years, Blind Signal Processing (BSP) algorithms arouse wide concern ofresearchers in various fields at home and abroad. Blind source separation (BSS) refersto estimate the source signal waveform by observation signal without the knowledge ofsource signal and the transmission channel characteristics. At present, it has become ahotspot in the signal processing field, and has been widely applied in biology andmedicine, noise elimination, image processing, and wireless communications etc. Then,a branch of blind signal separation is named as Blind Signal Extraction (BSE), itextracts the target signals one by one with their own features according to certain order,for example, the source signal is extracted according to kurtosis, non-gaussianmeasurement and sparsity measurement. This is different from the parallel extraction ofblind source signals.In the present various researches of BSS and BSE, most of them use the hypothesisof model without noise. Some of them suppose that the effects of noise on BSS andBSE can be ignored. However, in practice the signals will be disturbed by noise andinterference inevitably, this will even make the algorithm invalid. For example,communication signals, speech signals and biomedical signals are interfered with noise.As a result, the research of noisy Blind Signal Processing (BSP) is important and hasthe theoretical and applicable significance. For the moment, quite a few effective blindsource separation algorithms considering noise have been proposed, for example, usingthe optimized nonlinear function in non-generalized gaussian noise model, using theExpectation Maximization (EM) for processing defect data, using the method of solvingthe problem with gaussian distribution signal to extract the target signals by assumingthat noise is one of source signals, and using existing pre-filter device to filter out thenoise signal. These methods reduce the effect of noise on BSS and BSE algorithms, andimprove the accuracy of the separation results.Because the noisy blind separation is complex and the characteristics of sourcesignals and channels are unable to obtain, the current research achievement of noisyblind separation is insufficient. This dissertation investigates blind signal extraction or separation algorithms considering the effect of measurement noise, environment noiseand interference, and proposes several new signal extraction and separation algorithmsunder blind and semi-blind conditions. The Main work of this dissertation includesnoisy blind separation algorithm based on independence, time correlation, andDenoising Source Separation (DSS) framework combining with noise pre-processing.The simulation results demonstrate the validity of the proposed algorithms. Finally,future work for BSS and BSE are put forward.