Blind Source Separation Based On ICA

Blind Source Separation (BSS) has attracted a great deal of attentions from the signal processing community recently. The thesis not only deals with the BSS models and realization method, but also reviews some existed typical BSS algorithms based on Independent component analysis (ICA). Ant Colony Optimization (ACO) algorithm is a new intelligent optimization algorithm with the ability of global optimization, positive feedback characteristics and robust. A new BSS algorithm based ACO is proposed, which can avoid the problem that many algorithms based on gradient descent approach get into part extremum.The problem of over-determined BSS is considered. Beginning with the Singular Value Decomposition (SVD) of the separation matrix, a cost function is presented based on Minimization of Mutual Information (MMI), and then the ACO algorithm is developed. Computer simulations verify the validity of the algorithm. The Blind Source Extraction (BSE) based on ACO is considered. A deflation algorithm based on Schmidt orthogonal is proposed, which can avoid extract signals repeatedly.Research on nonlinear Blind Source Separation (NBSS). Network structure of post nonlinear BSS is introduced. A NBSS criteria based on cross-cumulates of output signals in higher order statistics is put forward and a novel NBSS algorithm based on the criteria and ACO is proposed. The simulation result demonstrates the effectiveness of the proposed NBSS algorithm.A BSS experiment plantform based on DSP is developed. The platform is based on the DSP+RTDX+PC/Windows frame structures, Using DSP to achieve BSS algorithm, Using RTDX for host and target plate bi-directional real-time transmission and developing a good host GUI for operation. The platform is easy to use and provides a good environment for BSS experiment.