论文部分内容阅读
分析了盲源分离算法中互信息准则与概率密度核函数的关系,利用广义高斯模型,提出了一种基于含参数的核概率密度估计的独立分量分析算法。该算法利用观测样本求峰度,通过分段函数给出相应高斯指数值,并利用样本数据进一步修正源信号的概率密度函数,实现对分离信号评价函数的精确估计。在此评价函数基础上,采用互信息最小化准则,推导出分离矩阵的迭代更新规则。所提算法在一定程度上解决了ICA算法中信号评价函数估计的难题,且能对任意源混合信号进行有效盲分离,仿真实验验证了算法的性能。
The relationship between mutual information criterion and probability density kernel function in blind source separation algorithm is analyzed. Based on generalized Gaussian model, an independent component analysis algorithm based on kernel probability density estimation with parameters is proposed. The algorithm uses the observed samples to calculate the kurtosis, and gives the corresponding Gaussian index value by the piecewise function. The sample data is used to further correct the probability density function of the source signal to achieve an accurate estimation of the evaluation function of the separated signals. Based on this evaluation function, the minimization of mutual information criterion is used to deduce the iterative updating rule of the separation matrix. The proposed algorithm solves the problem of signal evaluation function estimation in ICA to a certain extent, and can effectively blindly separate the mixed signal of any source. The simulation results verify the performance of the algorithm.