SVR-EEMD: An Improved EEMD Method Based on Support Vector Regression Extension in PPG Signal Denoising

Abstract

Photoplethysmography (PPG) has been widely used in noninvasive blood volume and blood flow detection since its first appearance. However, its noninvasiveness also makes the PPG signals vulnerable to noise interference and thus exhibits nonlinear and nonstationary characteristics, which have brought difficulties for the denoising of PPG signals. Ensemble empirical mode decomposition known as EEMD, which has made great progress in noise processing, is a noise-assisted nonlinear and nonstationary time series analysis method based on empirical mode decomposition (EMD). The EEMD method solves the "mode mixing" problem in EMD effectively, but it can do nothing about the "end effect," another problem in the decomposition process. In response to this problem, an improved EEMD method based on support vector regression extension (SVR-EEMD) is proposed and verified by simulated data and real-world PPG data. Experiments show that the SVR-EEMD method can solve the "end effect" efficiently to get a better decomposition performance than the traditional EEMD method and bring more benefits to the noise processing of PPG signals.