摘要
基于信息极大化和自然梯度原理,提出了一种超高斯与亚高斯混合信号的盲分离方法。该方法联合利用高斯函数与双曲正割函数平方的乘积和两个高斯函数的组合对源信号概率密度函数进行估计,放宽了约束条件,并采用峰度信息作为参数来选择概率密度模型及相应的非线性函数,对超高斯和亚高斯混合信号有较好的分离效果。并且,用小波变换与此方法相结合对含有加性噪声的混合信号进行分离。实验仿真证明了算法的有效性。
One kind of super-Gauss and the sub-Gauss composite signal blind separation method is proposed based on the information enormous and the natural gradient principle. The Gaussian function and a hyperbolic secant function square product and two Gaussian functions combination are used to estimate the source signal probability density function. And peak information is used as the parameter to choose the probability density model and the corresponding nonlinear function, This method has the better separation effect to sub-Gauss and the super-gauss composite signal. And the wavelet transformation with this method is unified to separate the composite signals including the additive noise. The experimental simulation proves that the algorithm is validity.
出处
《科学技术与工程》
2007年第22期5771-5775,共5页
Science Technology and Engineering
基金
教育部新世纪优秀人才支持计划项目(NCET-05-0897)
新疆维吾尔自治区高校科学研究计划项目(XJEDU2004E02
XJEDU2006I10)资助
关键词
盲源分离
超高斯
亚高斯
峰度
blind source separates sub-Gauss super-Gauss peak
