摘要
介绍比例微分变步长法在盲源分离中的应用,着重阐述对于非高斯混合输出信号,高阶累积量对比例微分变步长盲源分离结果的影响.研究表明,比例微分变步长法不仅适用于高斯型混合输出信号,也适用于非高斯型混合输出信号.对于非高斯型信号,应考虑高阶累积量的贡献.将二阶、三阶和四阶累积量分别作为比例微分算法的控制量,结果表明,三者均能获得相同量级的收敛速度和分离精度,但四阶累积量的初始步长取值范围最大,为0.1~1.2;而二阶和三阶累积量的初始步长取值范围相对较小,仅为0.1~0.4.因此,基于比例微分变步长算法的盲源分离技术,通过合理选择累积量阶数,可适用于不同类型的混合源信号的分离.
This paper introduced the application of proportional differential variable step method in blind source separation. The influence of high order cumulant on the separation results of the proportional differential variable step was emphasized. The results showed that the proportional dif- ferential variable step method not only could be applies to gaussian mixed signals, but also applies to non-gaussian mixed output signals. For non-gaussian signals, the contribution of high order cumulants should be considered. Compared the second-order, third-order and fourth-order cumulants as percentage differential variable step size, respectively, the results showed that they had reached the same order of magnitude of the convergence speed and the separation accuracy, but the initial step length value range of fourth-order cumulant was maximum （ about 0. 1 - 1.2 ） , while the second order and third order cumulants of the initial step length scope were smaller （ about 0. 1 - 0. 4）. It shows that the blind source separation technology based on proportional differential variation step algorithm can be applied to the separation of mixed source signals of different types through the ap- propriate selection order number.
出处
《福建师范大学学报（自然科学版）》
CAS
CSCD
北大核心
2018年第2期23-27,共5页
Journal of Fujian Normal University：Natural Science Edition
基金
福建省教育厅重点项目(JA3341)
关键词
高阶累积量
比例微分控制
盲源分离
high order cumulant
proportional differential control
blind source separation
