分段经验模态分解的直流漂移消除方法

在低频模拟信号采集及处理电路中，常常存在着噪声与直流漂移的问题影响信号的测量。为了从原始信号中准确去除直流漂移分量及存在的噪声从而获得有用信号，提出了一种基于分段经验模态分解直流漂移消除的方法，并通过合适且有效的去噪方法处理使得采样信号更加真实准确。首先，对信号进行经验模态分解，求取本征模态函数分量的局部极值点进行区间分段，之后分别对每一段信号再次进行经验模态分解，选取每一段信号的低频分量重构出该段信号的直流漂移分量。最后，利用自相关函数筛选出噪声占主要成分的本征模态函数分量进行能量分析，将所有分段整合，得到去除直流漂移且降噪之后的信号。本研究通过仿真对此方法进行了说明，比较了该方法与多项式拟合、小波分析、高通滤波等方法的效果。并对微创外科手术机器人力传感器的应变信号进行了处理。实验结果表明：按照已有数据计算可将信噪比提升到6.39 dB以上，均方根误差有明显减少；该方法能够有效消除应变信号中的直流漂移，并且也能达到降噪的目的。

Direct drift elimination method based on segmented empirical mode decomposition

In low frequency analog signal acquisition and processing circuit, signal measurement is often affected by noise and direct current (DC) drift. To remove DC drift and the existing noise from original signal accurately and obtain useful signal, a method of DC drift elimination based on segmented Empirical Mode Decomposition (EMD) is proposed. First, the signal is decomposed by EMD, and the local extremum points of the intrinsic mode function (IMF) components are identified for interval segmentation. Then, each segment of signal is decomposed by EMD, and the low-frequency components of each signal segment are selected to reconstruct the DC drift signal. In the end, the IMF components with noise as the main ingredients are screened by autocorrelation function for energy analysis, with all segments integrated and the signal after removing DC drift and noise reduction obtained. The simulation in the study shows the obvious effectiveness of the proposed method, compared with polynomial fitting, wavelet analysis, high-pass filtering and other methods. The strain signal of the robot force sensor in minimally invasive surgery is processed. The experimental results demonstrate that the Signal-Noise Ratio (SNR) is improved, specifically more than 6.39 dB, with the root mean square error (RMSE) significantly reduced. This method proves to be effective to eliminate the DC drift in the strain signal and achieve the purpose of noise reduction.