Computer Engineering and Applications ›› 2014, Vol. 50 ›› Issue (6): 210-212.

Previous Articles     Next Articles

Research on improved FastICA algorithms

ZHANG Jie, LIU Hui, OU Lunwei   

  1. Institute of Physics and Information Science, Hunan Normal University, Changsha 410081, China
  • Online:2014-03-15 Published:2015-05-12


张  杰,刘  辉,欧伦伟   

  1. 湖南师范大学 物理与信息科学学院,长沙 410081

Abstract: Independent Component Analysis(ICA) is the blind source separation algorithm which is one of the most commonly used methods. And the Fast Independent Component Analysis(FastICA) with its convergence speed is widely used. But FastICA is sensitive to the choice of initial value, and in the use of Newton iterative method, each iteration step is needed to calculate a function value and a derivative value. When the function is more complex, computing its derivatives is often not convenient. This paper uses the single point string section method to iterate. Combining the steepest descent method with the single point string section method, while ensuring the separation effect, it makes FastICA iterative times reduce. At the same time it makes the calculation type more concise, and reduces the sensitivity to the initial value.

Key words: Fast Independent Component Analysis(ICA), Newton method, string section method, the steepest descent method, negative entropy

摘要: 独立分量分析是目前盲源分离算法中最常用的一种方法,其中快速独立分量分析(FastICA)以其收敛速度快而被广泛应用,但FastICA对初始值的选择比较敏感,而且在使用牛顿迭代法时,每迭代一步都需要计算一次函数值和一次导数值,当函数比较复杂时,计算它的导数值往往不方便,用单点弦截法进行迭代,将最速下降法与单点弦截法结合,在保证分离效果的同时使FastICA的迭代次数减少,同时使计算式更加简洁,而且减小了对初始值的敏感性,仿真实验验证了其有效性。

关键词: Fast独立分量分析(ICA), 牛顿法, 弦截法, 最速下降法, 负熵