Spectral coherence as a method for evaluating adaptive filtering techniques in photoplethysmography

In this study, we introduce a method that utilizes spectral coherence as a metric for assessing the restoration of the photoplethysmographic (PPG) pulse signal morphology when filtering myokinetic noise using adaptive filtering techniques. Unlike other approaches that simply focus on detecting peaks and valleys after filtering, our research concentrates on recovering most of the components of the PPG pulse curve, as these reveal highly important information about cardiovascular health status, and which is essential for a proper diagnosis. To record PPG signals, a photoplethysmography system was placed on both hands of the subjects. On one hand, a three-axis accelerometer was also attached to capture the myokinetic motion, while the other hand remained stationary, allowing the PPG system to record a motion-free signal. After capturing the PPG signals with and without motion, we conducted an analysis by calculating the spectral coherence of the contaminated and recovered signals after adaptive filtering. We compared noisy PPG signals with the reference signal (myokinetic motion-free PPG signal). Based on the spectral coherence criterion, the affine projection algorithm and the variable step-size affine projection algorithm were found to be the most effective for filtering a PPG signal contaminated by myokinetic noise, achieving similarity scores of 94.25% and 94.67%, respectively, between the filtered signal and the motion-free pulse signal.