Multichannel PPG signal denoising and heart rate estimation

Abstract

Many cardiovascular diseases only manifest during physical activity, making ac curate real-time heart rate monitoring and estimation during exercise critically important. This not only enables precise localization and classification of dis eases but also contributes to disease prediction to a certain extent. Photoplethys mography (PPG) has been widely used in wearable devices for real-time heart rate monitoring due to its low cost and ease of data acquisition. However, the presence of motion artifacts (MAs) poses significant challenges in accurately es timating heart rate from PPG signals. In this study, we propose a method based on the multichannel joint estimation of PPG signals and synchronous accelerometer signals using a multiple measure ment vector (MMV) model from sparse signal recovery theory. By leveraging the sparsity constraint of spectral coefficients, the proposed method effectively identifies and removes spectral peaks caused by motion artifacts in the PPG spectrum. Furthermore, two distinct frequency estimation approaches are utilized during the frequency estimation stage: a dual-channel PPG frequency estimation method based on a Kalman filter, and a single-channel multi-stage spectral peak track ing method. Finally, the accuracy of the proposed approach is validated using publicly available motion-contaminated PPG datasets. Relative to conventional methods—exemplified by TROIKA—the proposed algo rithm achieves a statistically significant reduction in estimation error and demon strates markedly enhanced robustness. The average absolute error was 1.53 beat per minute and the standard deviation was 0.23 beat per minute.