Abstract: This paper utilizes radial basis function to construct the kernel function and forms a sparse representation of the probability density of the mixed signal based on SVM regression method of Neural network theory，so the expressions of the PDF of the output signals is gotten and a new method of estimating the active function is put forward.A novel BSS algorithm is presented in this paper by combining the PDF estimation with the fixed-point algorithm.The simulations have verified that this algorithm can successfully separate the mixed sub-Gaussian and super-Gaussian source signals.

Key words: support vector machine, probability density function, active function, fixed-point algorithm

摘要： 基于神经网络理论中的支持向量机回归方法，利用径向基函数构造核函数，给出类高斯函数的混合信号概率密度稀疏表达，进而得到输出信号的概率密度的显式表达；提出一种估计激活函数的新方法，与盲信号抽取定点算法相结合，形成一种新的盲分离算法。通过仿真实验，验证了该方法能成功地分离超、亚高斯混合信号。

关键词: 支持向量机, 概率密度, 激活函数, 定点抽取算法