Spectral Matrix Decomposition-Based Motion Artifacts Removal in Multi-Channel PPG Sensor Signals

The intelligent wearable heart rate measurement requirement has attracted more and more attention, and the related applications of Internet of Things are emerging. However, under intensive physical exercises, motion artifacts are strong interference sources for wrist-type photoplethysmography (PPG) sensor signals, thus significantly affecting the accurate estimation of heart rate and other physiological parameters. Currently, how to effectively remove the motion artifacts from PPG sensor signals is becoming an active and challenging research realm. In this paper, we propose a multi-channel spectral matrix decomposition (MC-SMD) model to accurately estimate heart rate in the presence of intensive physical activities. Motivated by the observation that the PPG signal spectrum and the acceleration spectrum have almost the same spectral peak positions in the frequency domain, we first model the removal of motion artifacts as a spectral matrix decomposition optimization problem. After removing motion artifacts, we propose a new spectral peak tracking method for estimating heart rate. Experimental results on the well-known PPG data sets recorded from 12 subjects during intensive movements demonstrate that MC-SMD can efficiently remove the motion artifacts and retrieve an accurate heart rate using multi-channel PPG sensor signals.