An Algorithm Research On Underdetermined Blind Source Separation With Linear Mixture

Blind source separation (BSS) aims to estimate the source from theobserved mixtures without any prior knowledge. It develops rapidly sincethe1980s. As a combination of neural network and statistics, it became acentral issue in the signal processing field. BSS play a major role in awide range of important applications, including image understanding,wireless communications, and bioinformatics.The thesis mainly focuses on the separation of instantaneousmixtures in underdetermined conditions when the number of orginalsources is more than the number of receiving sensors. The main researchwork includes:(1) Introduced the significance and current status of BSS.Summarize four characteristics of the original sources, includinghigher-order statistics properties, second-order statistical properties,non-stationary, sparsity. Then the blind source separation algorithms ofthe time domain and frequency domain are researched.(2) To solve the problem of the underdetermined blind sourceseparation, this thesis proposes an algorithm for blind identification of underdetermined mixtures based on Parafac decomposition ofcovariance matrix of the observed signal and weight enhanced alternatingleast squares, do not need the sources are quite sparse. Because theparallel factor decomposition still satisfy unique identifiability inunderdetermined situation, the proposed algorithm can solve theunderdetermined blind source separation problem successfully.Simulation results demonstrate the performance and effectiveness of theproposed algorithm is very better for underdetermined mixed. Thealgorithm is relatively simple, which can satisfy the demand ofengineering application.