摘要
本文提出了一种以多层神经网络来估计概率密度函数的后非线性盲源分离算法.该算法将PReLU函数作为激活函数,并对概率密度函数进行自适应逼近,以最小互信息作为基本准则来构建目标函数测试独立性.最后用改进后的自然梯度算法推导出分离矩阵和迭代公式,以此来更新目标函数.仿真实验证明所提算法可以有效分离非线性混合信号.
In this paper,a post-nonlinear blind signal separation algorithm based on multilayer neural network is proposed to estimate the probability density function,which uses PReLU function as activation function to perform adaptive approximation to the probability density function,and contrasting independence based on cost function by taking the minimum mutual information as basic criterion.At last,the separation matrix and iteration formula are derived by using the improved algorithm,can effectively separation nonlinear mixed signals.
作者
陈曦
李炜
张亚丽
薛青松
CHEN Xi;LI Wei;ZHANG Yali;XUE Qingsong(Anhui Key Laboratory of Detection Technology and Energy Saving Devices,Anhui Polytechnic University,Wuhu,Anhui 241000;School of Electrical Engineering,Anhui Polytechnic University,Wuhu,Anhui 241000)
出处
《绵阳师范学院学报》
2020年第2期25-30,46,共7页
Journal of Mianyang Teachers' College
基金
安徽省高校自然科学研究项目重点项目(KJ2016A795)
安徽工程大学检测技术与节能装置安徽省重点实验室开放研究基金资助项目(2017070503B026-A04)
安徽工程大学国家自然科学基金预研项目(2017yyzr01)
安徽工程大学大学生科研项目(2017DZ15)
关键词
盲源分离
后非线性
多层神经网络
最小互信息
自然梯度
blind source separation
Post-nonlinear BSS
multilayer neural network
minimum mutual information
natural gradient