A Research On Blind Deconvolution Based On The Spectral Similarity Of Mixed Models

Blind Source Separation (BSS) is a process of obtaining and recovering source signals from the observed signals through sensors array without the knowledge of the mixing system and source signals. Without the need of prior knowledge on the process of blind source separation, BSS has extensive applications in the field of signal process.This thesis is focused on the research of convolution mixture model. The main research works are outlined as follows:First, this thesis illustrates the background and significance of research on BSS, and expounds its research status and some important areas of applications.Secondly this thesis analyzes and researches convolution mixture model with special attention to the typical complex domain instantaneous separation algorithm, i.e., jointly approximate diagonalisation of eigenmarices(JADE). Meanwhile this thesis elaborates on the similarity of speech signals in frequency domain ,short-term analysis and sets forth a solution to optimize permutation inconsistency problem by using the genetic algorithm as well as an approach to optimize amplitude problem by adopting the steepest descent method.Finally this thesis researches and explores the problem of convolution mixture model by carrying out simulation experiments. By applying JADE algorithm into simulation experiments, we observe and study the performances for different kinds of synthetic mixing speech signals and for speech signals recorded in real world. When separating speech signals recorded in real world, we study the effects of the distance between speaker and microphone on separation results.With simulation experiments, we arrive at some conclusions and inspirations which might be useful for further research in this area.