【作者】 肖瑞；

【导师】 刘丁；

【作者基本信息】 西安理工大学 ， 控制理论与控制工程， 2007， 硕士

【摘要】 盲源分离问题是目前信号处理领域中最热门的新兴技术之一。本文讨论了盲源分离的数学模型、实现方法等，并回顾和总结了多种典型的基于独立分量分析的盲源分离方法。蚁群算法作为新兴的智能优化算法，具有全局寻优能力、正反馈特性、强鲁棒性等特点，利用蚁群算法对代价函数进行全局寻优，可以很好的克服梯度算法容易陷入局部最小的缺陷，因此将它应用于盲源分离问题的求解。针对超定盲源分离问题，本文从分离矩阵的奇异值分解出发，采用基于最小互信息的超定盲源分离代价函数，并用蚁群算法对代价函数进行优化，最终解决了一类超定的盲源分离问题；对于同步盲信号提取问题的研究，采用Schmidt消源去相关的方法避免了蚁群算法对已提取信号的重复提取，提高了算法的计算效率。本文对非线性盲源分离问题也进行了阐述，针对后非线性混合模型，从基于联合累积量的非线性统计独立性判据出发，提出了基于该准则与蚁群算法相结合的非线性盲源分离算法。仿真表明该算法对后非线性混合盲源分离问题的可行性和有效性。最后，开发了基于DSP的盲源分离实验平台，该平台是基于DSP+RTDX+PC／Windows构架搭建的，借助DSP实现盲源分离核心算法的实时运算，利用RTDX技术实现PC主机和DSP的双向实时数据传输，开发主机端界面来进行界面操作并显示实验结果。该实验平台的开发为借助DSP进行盲源分离实验提供了便利。更多还原

【Abstract】 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.更多还原