白化处理的自然梯度盲源分离统一算法

Integrated Blind Source Separation Algorithm with Natural Gradient Based on Whitening Process

分析了基于自然梯度的信号白化处理和白化后的盲信号分离,提出了一种基于白化处理的自然梯度盲源分离算法.算法统一了信号的白化和分离,而不需要单独再对信号进行白化预处理,通过采用自然梯度学习规则提高了算法的性能,并理论证明了算法的可分离性、等变化性和分离矩阵的非奇异性.仿真表明,算法能够有效地分离和重构源信号,相比信号未白化的随机梯度算法以及传统的FastICA算法,收敛速度快、分离效果好,更适合盲源分离.

Through analyzing the whitening and separation process of signal based on natural gradient,a blind source separation algorithm with natural gradient based on whitening process is proposed.The algorithm colligates the whitening and separation of signal,and need not whiten the signal individually.By the application of natural gradient,the performance of algorithm is improved.Deduction proves that the algorithm satisfies the separation,equal variety and nonsingularity of separation matrix.Simulation shows that the algorithm can separate and reconstruct source signal effectively.Compared with the random gradient algorithm without signal＇s whitening and the conventional FastICA algorithm,it exhibits faster convergence as well as better separation effect and is more favorable for blind source separation.