摘要
Simulation experiments are carried out to test the blind separation ability of the most popular independent component analysis (ICA) algorithm, FastICA. Experimental results show that the performance of FastICA is highly affected by the selection of non-linear activation functions which depends on the distributions of source signals, the separation result is unsatisfied. To solve this problem, a new ICA algorithm is proposed, where mutual entropy is used as a measure of statistical independence, genetic algorithm is applied to find the separation matrix. Computation of object function which is the key issue in this process is solved. Simulation results show that the proposed ICA algorithm possesses better separation ability than FastICA does. The proposed ICA algorithm is free from selection of activation function and especially suits for engineering practice.
| 源语言 | 英语 |
|---|---|
| 页(从-至) | 1867-1874 |
| 页数 | 8 |
| 期刊 | Journal of Computational Information Systems |
| 卷 | 9 |
| 期 | 5 |
| 出版状态 | 已出版 - 1 3月 2013 |
学术指纹
探究 'A novel genetic algorithm based independent component analysis algorithm' 的科研主题。它们共同构成独一无二的学术指纹。引用此
- APA
- Author
- BIBTEX
- Harvard
- Standard
- RIS
- Vancouver