摘要
一单元参考独立成分分析是一种有效地利用先验信息抽取一个期望源信号的方法。针对基于峭度的快速算法抽取正确率较低的缺点,在两种常用近似性量度下对快速算法进行了理论分析,指出该方法抽取正确率低的原因,通过避免不等式约束失效的方法,基于新CICA提出了一种一单元ICA-R固定点算法。大量计算机模拟实验表明所提算法抽取性能和快速算法相当,但具有更快的收敛速度和更高的抽取正确率。
One-unit ICA-R is an efficient method utilizing prior information to extract an expected source signal.Aimed at flaw of the fast algorithm based on kurtosis,it performs theoretical analysis on the fast algorithm under two common closeness measurements,points out the reason of lower accurate extraction rate.It proposes a fixed-point algorithm through removing invalidation of inequality constraints based on new CICA.Computer simulations verify that the fixed-point algorithm can converge faster and extract more accurately with almost same extraction performance of the fast algorithm.
出处
《计算机工程与应用》
CSCD
北大核心
2011年第36期137-140,共4页
Computer Engineering and Applications
基金
国家自然科学基金(the National Natural Science Foundation of China under Grant No.60672049)
关键词
峭度
约束独立成分分析
固定点算法
kurtosis
Constrained Independent Component Analysis(CICA)
fixed-point algorithm