RadarNet: Noncontact ECG Signal Measurement Based on FMCW Radar

• Published in: IEEE transactions on instrumentation and measurement Vol. 73; pp. 1 - 9
• Main Authors: Li, Bin, Li, Wenlong, He, Yuchen, Zhang, Wei, Fu, Hong
• Format: Journal Article
• Language: English
• Published: New York IEEE 2024; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Biomedical monitoring; Continuous radiation; Datasets; Electrocardiogram (ECG) signal measurement; Electrocardiography; Filtering; Frequency measurement; frequency-modulated continuous wave (FMCW) radar; Heart rate; Measurement methods; micromotion signal disentanglement; Monitoring; noncontact health monitor; Physiology; Radar; Radar measurements; Root-mean-square errors; Signal measurement; Signal monitoring; Signal reconstruction; Telemedicine; Time synchronization; Time-frequency analysis; Ultra wideband radar; Waveforms
• ISSN: 0018-9456; 1557-9662
• DOI: 10.1109/TIM.2024.3476545

Continuous monitoring of electrocardiogram (ECG) signals plays an important role in the prevention and diagnosis of cardiovascular diseases. Conventional ECG signal measurement requires skin-contact electrodes, which are unfriendly to skin sensitive, burn patients and infants. In this article, we propose a noncontact ECG signal measurement method based on frequency-modulated continuous wave (FMCW) radar, facilitated by a signal amplification network named RadarNet. First, a multidomain joint modulation micromotion signal disentanglement module (MSDM) is introduced to extract cardiac mechanical motion (CMM) signals from complex radar waveforms. Then, a signal reconstruction network RadarNet based on time-frequency domain transformation is proposed to learn the nonlinear relationship between CMM and ECG signal. Finally, the first annotated dataset with synchronized FMCW radar-physiological signal RadarPhys-30 is established, which contains 30 subjects in two physiological states (sitting and lying down). The experimental results on the RadarPhys-30 dataset demonstrate that the mean average error (MAE) and the root-mean-square error (RMSE) of the heart rate (HR) estimated from the reconstructed ECG signal are less than 0.20 beats per minute (bpm) and 0.53 bpm in the sitting state, and 0.15 and 0.46 bpm in the lying down state, respectively. These results indicate that the proposed method achieves accurate noncontact ECG signal monitoring.