采用EMD和奇异值分解子空间重构的信号降噪新方法

传统的奇异值降噪法对适合奇异值分解的矩阵构造及信号重构时有效秩阶次的选取缺乏具有物理意义的依据.提出一种采用EMD和奇异值分解子空间重构的信号降噪新方法,通过对EMD方法得到的各阶IMF分量构造时频矩阵进行奇异值分解,将信号的特征信息分解到各个不同的时频子空间中,根据时频子空间的特征变化,选择相应的子空间进行奇异值分解逆变换,从而实现信号降噪.对仿真合成电信号及实测机械振动信号的降噪应用,表明该方法能有效地从原始信号中提取所需的信号特征成分,具有直观的物理意义.

A New Method for Noise Reduction Based on EMD and SVD Subspaces Reconstruction

There was a lack of physical-meaning basis for traditional noise reduction based on SVD when it came to construct what kind of matrix and how to choose the rank number to reconstruct signal effectively. A novel noise reduction method based on EMD and SVD subspaces reconstruction was proposed here. A series of IMF components through EMD were obtained to form the time-frequency matrix which could be decomposed by SVD so that the feature of original signal was decomposed into different time-frequency subpaces. Then appropriate subspaces according to feature change of subspaces were selected to do SVD inverse transformation so as to achieve signal’s main feature. Simulation of electric signal and measured vibration signal results show that this approach is an effective method to extract useful component from original signal and is with intuitive physical meaning.