用噪声残差似然估计改进经验模态分解基信号去噪方法

Improved empirical mode decomposition based signal de-noising approach using likelihood estimation of residual noise

现有的基于经验模态分解(EMD)的信号去噪算法,用于幅值取阈以及信号重建的本征模函数(IMF)基本都是靠经验筛选,影响了算法的去噪性能。为解决这一问题,引入噪声残差似然估计(LE-RN)测度以建立IMF的优化筛选准则。由消噪后的重建信号与原始带噪信号,可以形成一个噪声残差(RN)分量。在不同去噪参数组合下,通过最大化RN分量与标准正态分布的对数似然,实现IMF分量的优化筛选。最终,形成一种改进的EMD基信号去噪方法。借助仿真消噪试验,通过与现有的EMD基去噪方法进行对比,验证了所提出的去噪参数优选准则及其改进EMD基去噪方法的有效性。

In existing signal de-noising approaches based on empirical mode decomposition （EMD）, the intrinsic mode functions （IMFs） used for amplitude thresholding and signal reconstruction are almost selected empirically, which affects the de-noising performance of the algorithm. In order to solve this problem, a measure named likelihood estimation of residual noise （LE-RN） was introduced for setting up an optimal criterion for IMFs selection. A residual noise （RN） component can be produced from the original noisy signal and reconstructed denoised signal. Under different combinations of denoising coefficients, the optimal IMFs sets were fixed by maximizing the logarithm likelihood of residual noise （LE-RN） and the standard Gaussian distribution. Thus, an improved EMD based approach for signal de-noising was achieved. In simulation denoising test, through comparing with the existing EMD based signal de-noising technique, the effectiveness of the de-noising parameter optimal criterion and the resultant improved approach is verified.