Time Delay Estimation Of Blind Deconvolution Based On Correlation Analysis

Blind Source Separation (BSS) is a newly arisen research direction in modern signalprocessing. The key problem of blind source separation is how to separate the target sourcesignal from the mixed-signals without any priori information. Many of the existing researches setsome priori conditions artificially, for example, the number of sources, mixing matrix aregenerated randomly and so on. However, there is no priori information or convolution mixingmodel close to the actual environment in totally blind circumstances,which may lead to theineffective separation or even failure. In accordance with the frequency domain blinddeconvolution problem, the study in this thesis is focused on the following aspects:Firstly, This paper briefly introduces the theory of blind source separation, three MixtureModels and three classic ICA algorithm of the blind separation in the beginning, and alsoanalyzes the frequency domain blind-deconvolution method in detail.Secondly, This paper introduces the microphone array signal model and selects thegeneralized cross-correlation time delay estimation algorithm after comparing and analyzing ofthe three kinds of time delay estimation algorithm. It also proves that using the time delayestimation methods to determine the number of the blind source signal is feasible. This paperputs forward a array model in theory, and this model is used to solve the misjudgement of sourcenumber problem which results from the time delay overlap.Finally,This dissertation constructs two models: attenuation delay convolution channel anddynamic convolution channel.It also deduces and analyzes the correlation function of the signal.This dissertation focuses on the signal correlation problem based on Low-order Dynamicchannel model and High-order Dynamic channel model respectively.