Design and pilot validation of a multispectral photoplethysmography platform for transparent physiological signal acquisition

This is the authors' abstract. We don't add key points for this paper.

Photoplethysmography (PPG) is a non-invasive optical sensing technique widely used for heart-rate and oxygen-saturation monitoring, but many commercial devices expose only final values and hide the raw optical data and processing steps. This work presents the design and pilot validation of a custom multispectral PPG platform intended as a transparent research prototype. The system combines a reflective optical board based on the TCS3448 multispectral sensor and broadband white light-emitting diode (LED) illumination, an ESP32 acquisition node, a desktop graphical interface for live visualization and timestamped reference entry, and an offline signal-processing and validation pipeline. The platform was evaluated on a pilot dataset of 25 subjects and 66 recordings covering controlled breathing, post-exercise recovery and post-cold recovery. Quantitative evaluation used a matched subset of 23 subjects, 59 runs and 922 timestamped reference snapshots, with subject-held-out validation where calibration was applied. The best-validated output was heart rate, reaching 2.31 bpm mean absolute error on the high-confidence subset. An exploratory peripheral oxygen saturation (SpO 2 )-style output reached 0.78 percentage points mean absolute error but did not outperform a constant train-median baseline. Target-aware respiratory-rate tracking during controlled breathing reached 0.52 breaths/min mean absolute error. The results show that the platform is useful as a transparent multispectral PPG research tool, while remaining a non-clinical prototype.