摘要
相比标准梯度而言,自然梯度算法以其更快的收敛速度和更好的分离性能在盲源分离中占据着重要地位。由于常用的自然梯度算法是基于固定步长的,因此无法真正解决收敛速度和稳态误差之间的矛盾。通过建立步长因子与分离矩阵相互差异之间的非线性关系,提出了一种新的自然梯度算法。由于该算法采用的步长是时变的,加快了收敛速度,减小了稳态误差,从而很好地解决了固定步长的内在矛盾。计算机仿真结果证实了理论分析,并说明了该算法明显优于通常的自然梯度算法。
Compared with the standard gradient, natural gradient algorithm has faster convergence rate and better separation performance, so it occupies importance position in blind source separation. Because all of usual natural gradient algorithms adopt fix-step-size, they can not resolve the contradiction between convergence speed and the error in steady state. By building a nonlinear function relationship between the step size factor and the difference among the separating matrixes,the paper proposes a new natural gradient algorithm. Due to the algorithm's step-size is time-variable, the algorithm can improve convergence speed and reduce the steady state error,thus solves the internal contradiction of fix-step-size. Computer simulation result confirms the theoretical analysis and shows that the algorithms performance is superior to the usual natural gradient algorithm.
出处
《舰船电子对抗》
2007年第4期65-68,共4页
Shipboard Electronic Countermeasure
关键词
盲源分离
变步长
自然梯度算法
blind source separation
variable step size
natural gradient algorithm