二阶统计量的盲源分离研究

Research of Blind Source Separation Based on Second-order Statistics

介绍一种基于二阶统计量的盲源分离方法。该方法不必估计源信号的高斯特性,利用样本数据的二阶统计量和源信号的时序结构来实现信号的盲分离,因此其计算量小,适合于工程数据的分析。仿真结果表明,在有一定噪声的情况下,此种方法仍能较好地实现信号的分离。通过对电机振动信号分析发现,基于二阶统计量的盲源分离方法能有效地实现特征信号的提取和分离,这为今后将盲源分离技术进一步应用于机械振动信号的分析提供可借鉴的方法。

A method of blind source separation based on second-order statistics is introduced. Separation of the mixed signals can be obtained by this method, which can process second-order statistics of sample data and temporal structure of the source signals and avoid to estimate Gaussian property. The method reduces the computation and is suitable for analyzing engineering data. Simulation results show that the method can separate the signals mixed with noise. It was proved that blind source separation based on second-order statistics can effectively extract and separate characteristic signals through analyzing the electromotor＇s vibration signals. All of these work may be an good reference for applying blind source separation to analyzing the machine vibration signals.