An improved natural gradient algorithm for blind source separation

Authors:
Zhao Jia;Yang Jing-shu;Gao Jun-yao
Affiliations:
702 laboratory Electronical Engineering Institute, Hefei, China;702 laboratory Electronical Engineering Institute, Hefei, China;702 laboratory Electronical Engineering Institute, Hefei, China
Venue:
CAR'10 Proceedings of the 2nd international Asia conference on Informatics in control, automation and robotics - Volume 1
Year:
2010

Citing 3
Cited 0

Adaptive online learning algorithms for blind separation: maximum entropy and minimum mutual information

Neural Computation
Flexible Independent Component Analysis

Journal of VLSI Signal Processing Systems
Equivariant adaptive source separation

IEEE Transactions on Signal Processing

Quantified Score

Hi-index	0.00

Visualization

Abstract

This paper proposes an improved natural gradient algorithm for blind source separation (BSS) based on the constrained optimization method. The improved algorithm introduces a scaling factor that restricts the training process by the balance spot, which adds little computational complexity and overcomes the conflict between the convergence rate and the steady-state accuracy. Therefore, the new algorithm exhibits fast convergence and excellent performance. Computer simulation results show that the new algorithm is effective. And compared with the conventional natural gradient algorithm and the adaptive step-size algorithm, the performance of the improved algorithm is obviously better.