Removal of Motion Artifacts in PPG Signals Based on the CEEMDAN-MPE and VS_LMS Adaptive Filter

The performance of a wearable oximeter based on processing photoplethysmographic (PPG) signals is largely affected by the motion artifacts (MAs) during motion episodes. To effectively improve the estimation accuracy of arterial oxygen saturation (SpO2), the quality of a reference noise signal before it is applied to the adaptive filtering is vital. In this study, a novel method for reducing MAs in corrupted PPG signals is proposed, based on a synthetic reference noise signal generated from the corrupted PPG signal itself and an efficient adapter filter for MA reduction. Specifically, three methods are first used to individually generate the reference noise signals, and the most suitable one is selected using correlation analysis with the MA corrupted PPG signal. Thus, the selected synthetic reference noise signal is applied to four different adaptive filters to assess their effectiveness in reducing MA noise. Experimental results showed that the estimated SpO2 values, derived from the PPG signals acquired using self-designed equipment and from a publicly online dataset, had a Pearson correlation coefficient of 0.9775 when compared to the reference SpO2 values. The mean absolute error (MAE) and mean squared error (mse) during walking were 0.5721 and 0.4615, respectively, and during jogging, they were 0.7231 and 0.5202, respectively, which demonstrated the efficacy of the proposed scheme for reducing MA and accurately estimating SpO2 values. The study shows that the proposed scheme can improve the estimation accuracy of SpO2 and may be suited for modern pulse oximeters that utilize PPG signals during motion.