Toward Continuous and Contactless Cardiac Monitoring: A State-Space Approach With FMCW Radar

Cardiac disease remains a leading global cause of morbidity and mortality, necessitating continuous and unobtrusive monitoring of cardiac activities. Traditional contact-based ECG monitoring is limited by patient discomfort, skin irritation, and restricted long-term usability, driving the need for non-contact alternatives. This study introduces a system for reconstructing electrocardiogram (ECG) signals from radar-based chest displacement data, treating the human body as a signal generator and radar as a state observer. Each heartbeat is modeled as an independent event, and system parameters are iteratively optimized using reinforcement learning to refine state-space matrices. This integration enables accurate reconstruction of stable physiological signals. Experimental results from an overnight sleep dataset demonstrate high fidelity in cardiac motion reconstruction, achieving a Pearson correlation coefficient approximately 0.9 of the ECG waveform and a mean absolute error of 16.45 ms in heart rate variability (HRV) compared to polysomnography (PSG) reference, approaching clinical-grade accuracy. The proposed system effectively addresses common generalization challenges in data-driven methods, achieving high accuracy with minimal training data and maintaining consistent performance across subjects, ensuring robust and scalable non-contact cardiac monitoring.