【摘要】 α-稳定分布可以更好地描述实际应用中所遇到的具有显著脉冲特性的随机信号和噪声。与其它统计模型不同, α稳定分布没有统一闭式的概率密度函数,其二阶及二阶以上统计量均不存在。本文先简要介绍稳定分布统计特性,再提出了适用于盲信源分离的神经网络结构与基于分数低阶统计量与子空间技术的预白化过程,并利用一种新型传递函数修正了分离算法,提出了一种基于分数阶预白化与新型传递函数的盲信源分离方法。计算机模拟和分析表明,这种算法是一种在高斯和分数低阶α稳定分布噪声条件下具有良好韧性的盲信源分离方法,是对传统的二阶统计量基础上的盲信源分离方法的改造与推广。更多还原

【Abstract】 Lower order alpha stable distribution processes can better model the impulsive random signals and noises in physical observation. After briefly introducing the statistical characteristics of stable distribution and the fractional lower order statistics, i. e. , fractional order correlation, in this paper, we propose neural network structures related to multilayer feedforward networks for performing BSS based on fractional lower order statistics and subspace technique. We modify our conventional algorithms so that their separation capabilities are greatly improved under both Gaussian and fractional lower order alpha stable distribution noise environments. The simulation experiments and analysis show that the proposed class of networks and algorithms is more robust than those of the conventional second order statistics based algorithm.更多还原