A Novel Cluster-Based Approach for PPG Signal Preprocessing: Motion Artifact Filtering and Ambient Light Correction

Photoplethysmography (PPG) is widely used in wearable devices, but wrist- and arm-worn recordings are sensitive to motion artifacts, ambient light, and changes in sensor-skin contact. This study presents an adaptive preprocessing pipeline that combines accelerometer-derived motion-state clustering, an ambient-light-based Contact Quality Index (CQI), and cluster-based Symlet 4 wavelet filtering. Motion states are obtained from less posture-dependent features of 10 s accelerometer windows, whereas the CQI is a heuristic indicator of ambient-light-related optical reliability, scaled from 0 to 1, that flags low-reliability segments without changing their motion label. The wavelet stage removes baseline drift, soft-thresholds the high-frequency detail bands with thresholds selected on training data, and preserves the bands that overlap the cardiac frequency range. The model was trained on 8 participants and evaluated on 27 held-out participants against a simultaneously recorded ECG. Compared with a band-pass filter and an accelerometer-referenced adaptive filter, the proposed method detected more beats (median sensitivity 1.00 vs. 0.91 and 0.92) and reduced the inter-beat-interval (IBI) error when missed beats were included (128.0 vs. 181.2 and 156.8 ms), particularly during movement, at the cost of a lower positive predictive value (0.92 vs. 1.00), whereas the reference filters yielded lower IBI error on detected beats (24.6 and 26.6 vs. 38.7 ms). The method also outperformed a fixed Daubechies 4 wavelet filter on all timing metrics. In a band-energy analysis, it reduced out-of-band energy by a median of 27.7 dB but also reduced useful-band energy by 41.0%. These results show that a lightly thresholded wavelet filter recovers more beats than fixed filters during daily-life movement, whereas the benefit of cluster-specific thresholds remains to be established in other cohorts and during higher-intensity activities.