基于特征值分解的循环平稳离散时间信号的盲分离

盲分离的目的是从观测到的混叠信号中恢复出各个未知的源信号,现今的很多方法都是利用了信号时域表示的某些统计特性来解决这个问题。从信号频域分析的角度提出了一种利用信号的循环平稳特性来处理离散时间信号的频域盲分离方法。该方法构造两个二阶统计矩阵的乘积,并对该乘积矩阵进行特征值分解,从而实现源信号的分离;同时,还对特征值分解的条件进行了分析。该方法在低维信号的情况下可以取得相当满意的分离效果,仿真结果表明该方法具有良好的性能。

Eigenvalue decomposition based blind separation for cyclo-stationary discrete-time singals

The goal of blind source separation methods is to restore the unknown source signals from their mixtures. Many approaches proposed exploiting some statistics associated to the temporal representation of the signals. From the point of signal frequency analysis, this paper put forward a new approach used for processing frequency-domain discrete-time signals exploiting the cyclo-stationarity of the signals. First, It constructed a product of two second-order statistics matrices, and then implemented eigenvalue decomposition of this product matrix. It also analyzed the condition of eigenvalue decomposition. The proposed method cauld derive a satisfying separation result in the case of low dimensional source signals. The simulation results using artificial sources prove the good performance of the proposed method.