基于源数估计的无约束欠定盲源分离算法  

Unconstrained underdetermined blind source separation algorithm based on the source number estimation

作　　者：付永庆[1] 郭慧[1,2] 苏东林[2] 刘焱[2] 

机构地区：[1]哈尔滨工程大学信息与通信工程学院,黑龙江哈尔滨150001 [2]北京航空航天大学电磁兼容技术研究所,北京100191

出　　处：《哈尔滨工程大学学报》2015年第5期730-735,共6页Journal of Harbin Engineering University

基　　金：国家自然科学基金资助项目(61172038;60831001)

摘　　要：信源数目的估计是欠定盲源分离的前提条件,为了提高混合信号分离的准确性,提出一种Hough加窗法。利用Hough变换的思想将观测信号转变为角度变量,对变换域中的角度直方图进行加窗获得变换量的聚类区域,其峰值数即为信号源的数目。在此基础上,通过寻找变换量与混合矩阵列向量的关系可得到混合矩阵的估计值。提出一种无约束分离算法,由内点法从散点图分布中选取合适的初始迭代值,通过梯度下降法实现信号的分离。仿真实验结果表明,Hough加窗法具有较高的估计精度、较强的抗噪声性以及较低的稀疏敏感性,无约束分离算法具有较好的分离效果。The source number estimation is the prerequisite for underdetermined blind separation. In order to im-prove the accuracy of the mixed signal separation, a new Hough-windowed algorithm is proposed in this study. First, the algorithm transforms the observed signals into angular variables based on Hough transformation. The clus-ter area is obtained by windowing the histogram of angular variables in the transform domain. The peak value is the number of sources. Next, the mixture matrix is obtained through analyzing the relationship between the maxima in each cluster area and the column vector of the mixture matrix. Finally, an unconstrained separation algorithm is presented. The interior point method enables the acquisition of the initial value and the gradient descent method separates the signals. The simulation results showed that the Hough windowing method demonstrates higher estima-tion accuracy, stronger noise resistance ability, and lower sensitivity to sparse as well. In conclusion, the uncon-strained separation algorithm has better separation effect.

关 键 词：欠定盲源分离 信源数目估计 稀疏信号 Hough加窗法 无约束分离 梯度下降法 

分 类 号：TP911.7[自动化与计算机技术]