A New Step-Adaptive Natural Gradient Algorithm for Blind Source Separation
The main differences between the mixed signals and origin signals: Gaussian probability density function, statistical independence and temporal predictability. The proposed BSS algorithms mainly derived from the Gaussian probability density function and statistical independence. A new adaptive method is proposed in the paper. The method uses the temporal predictability as cost function which is not studied as much as other generic differences between the properties of signals and their mixtures. Step-adaptive nature gradient algorithm is proposed to separate signals, which is more robust and effective. Compared to fixed step natural gradient algorithm, Simulations show a good performance of the algorithm.
KeywordsGradient Algorithm Blind Source Separation Statistical Independence Correlation Coefficiency Fixed Step
Unable to display preview. Download preview PDF.
- 1.James, V. Stone.: Blind Source Separation Using Temporal Predictability, Neural computation (in press) (2001) 1196–1199Google Scholar
- 3.Amari, S I.:Natural Gradient Works Efficiently in Learning. Neural Computation(1998) 251–276Google Scholar
- 4.Amari, S. I., Chen, T. P., Cichocki, A.: Stability Analysis of Adaptive Blind Source Separation, Neural Networks(1997) 1345–1351.Google Scholar
- 5.Sergio, A., Cruces-Alvarez, Andrzej Cichocki, Shun-Ichi Amari.: On A New Blind Signal Extraction Algorithm: Different Criteria and Stability Analysis., IEEE SIGNAL PROCESSING LETTERS, VOL.9, NO.8, AUGUST (2002)Google Scholar
- 6.Yan, Li, Peng Wen, David Powers.: Methods for The Blind Signal Separation Problem. IEEE Int. Conf. Neural Networks&Signal Processing. December (2002)Google Scholar