- The paper introduces JOSS, a novel method using the MMV model to reduce motion artifacts in PPG signals.
- It employs spectral subtraction and robust peak tracking to enhance real-time heart rate accuracy during physical exertion.
- Experimental results show reduced average absolute error from 2.42 BPM to 1.28 BPM, validating JOSS's efficacy.
Photoplethysmography-Based Heart Rate Monitoring in Physical Activities via Joint Sparse Spectrum Reconstruction
Introduction
The paper presents a novel approach for heart rate (HR) monitoring using Photoplethysmography (PPG) in wearable devices during physical activities. It leverages a joint sparse spectrum reconstruction method, utilizing the multiple measurement vector (MMV) model in sparse signal recovery, to effectively distinguish and reduce motion artifacts (MA) that commonly contaminate PPG signals during physical exertion.
Methodology
The approach, named Joint Sparse Spectrum (JOSS), introduces the use of the MMV model to jointly estimate the spectra of PPG signals and simultaneous acceleration signals. This joint estimation exploits the common sparsity structures inherently present in these signals to enhance the accuracy and efficiency of spectral peak identification and noise reduction.
- MMV Model: The MMV model provides a more reliable reconstruction performance compared to the single measurement vector (SMV) model previously used in methodologies like TROIKA. By considering multiple measurements, it allows for a more robust identification and removal of MA in the PPG spectrum.
- Spectral Subtraction: JOSS implements a simplified spectral subtraction technique by detecting common spectral peaks across acceleration signals to identify and cleanse MA from PPG signals.
- Peak Tracking and Verification: The method simplifies spectral peak tracking with verification, reducing the need for complex signal decomposition and enhancing the real-time applicability of the algorithm.
Experimental Evaluation
The effectiveness of JOSS was validated using 12 datasets comprised of PPG and acceleration signals recorded during various physical activities. Critical metrics such as the average absolute error (Error1) and average absolute error percentage (Error2) indicated a notable improvement over the TROIKA approach:
- Performance Metrics: JOSS achieved an average absolute error of 1.28 BPM and an error percentage of 1.01%, demonstrating significant accuracy improvements over TROIKA, which recorded errors of 2.42 BPM and 1.82%, respectively.
- Low Sampling Rate Advantage: JOSS maintained its high accuracy even at lower sampling rates, which is crucial for reducing power consumption in wearable devices.
Discussion and Future Work
The use of the MMV model provides substantial benefits in collaboratively analyzing PPG and acceleration signals, greatly enhancing PPG-based HR monitoring's robustness against MA. By improving accuracy and lowering computational demands, JOSS holds promise for integration into VLSI hardware for wearable fitness and health monitoring applications.
Possible future developments could include exploring gridless compressive sensing techniques and further refinement of the adaptive noise reduction methods. Additionally, extending the application of the JOSS framework to varied skin tones and different LED wavelengths would be invaluable for generalizing its usability.
Conclusion
JOSS represents a significant step forward in the field of wearable health technology, optimizing PPG-based HR monitoring by addressing the prevalent challenge of MA during physical activities. Through innovative use of sparse signal modeling, it offers improved precision and efficiency, positioning it as a viable candidate for adoption in the next generation of fitness and health monitoring devices.