基于K-均值聚类和势函数法的欠定盲分离

Underdetermined Blind Source Separation Algorithm Based on K-Means Clustering and Potential Function

K-均值聚类法能估计出观测信号聚类直线方向,利用主成分分析(PCA)提取主成分,可以提高直线估计的精准度和鲁棒性。在此思想的指导下,本文提出基于K-均值聚类的势函数法。势函数度量了聚类中心与所有观测点的距离,对势函数求导得到更新聚类中心的迭代公式,利用该公式对K-均值聚类法得到的聚类中心进行调整得到精估计。该算法计算量较小,能有效估计出混合矩阵。仿真实验验证了算法的有效性。

K-means clustering method can estimate the observed clustering signal line direction, using principal component analysis can improve the accuracy and robustness of linear estimation. Under the guidance of this thinking, method of combine K-means clustering with potential function is proposed. Potential ＂function measure the distance of clustering center with all observation points, based on the potential function find clustering center iteration formulas, using this formula to adjust clustering center. The presented algorithm is characterized by high accuracy and less computation. Simulation results illustrate the efficiency and the good performance of the algorithm.