摘要
稀疏成分分析是信号处理中解决欠定盲源分离问题的新方法,本文研究了稀疏成分分析中的混合矩阵估计问题,提出了无需预知源个数利用一种相似性函数估计混合矩阵的方法.首先,估计相似性函数中的核参数,使得算法适应不同的稀疏信号.然后,给出了估计混合矩阵的不动点算法.最后,实验结果表明提出的算法通过适当地选取参数,能够准确有效地估计出具有不同源个数的混合矩阵,对不太稀疏的源也有令人满意的结果.
Sparse component analysis is a new approach to solve the problem of underde- termined blind source separation in signal processing area. In this paper, a clustering method based on similarity function is developed to estimate the mixing matrix in the case of unknowing the number of source signals. Firstly, the kernel parameter of similarity function is estimated so that the proposed algorithm adapts to different kinds of sparse signals. Then, a fixed point algorithm is formulated to estimate the mixing matrix. Finally, the simulations show that through choosing parameters appropriately, our algorithm effectively and accurately estimates the mixing matrix with different numbers of sources. The results for insufficient sparsity sources is also satisfactory.
出处
《工程数学学报》
CSCD
北大核心
2013年第5期642-654,共13页
Chinese Journal of Engineering Mathematics
基金
国家自然科学基金(70871015
71171030)
新世纪优秀人才支持计划(NCET-11-0050)~~
关键词
稀疏成分分析
欠定盲源分离
相似性函数
不动点算法
凝聚层次聚类
sparse component analysis
underdetermined blind source separation
similarity function
fixed point algorithm
agglomerative hierarchical clustering
