Robust Cardiac Feature Monitoring Based on Millimeter-Wave Radar

Zhang, Yuanyuan (2025) Robust Cardiac Feature Monitoring Based on Millimeter-Wave Radar PhD thesis, University of Liverpool.

Text 201377323_Aug2025.pdf - Author Accepted Manuscript Download (8MB) | Preview

Abstract

Wireless sensing empowers numerous emerging industries such as autonomous driving and device-free monitoring. By introducing the contactless sensors such as millimeter-wave (mmWave) radar, a promising and challenging research area is emerging to realize robust and remote vital sign monitoring, enabling contactless monitoring for future in-cabin monitoring, elderly people caregiving and even clinical diagnosis. In recent years, frequency-modulated continuous wave (FMCW) radar with high operating frequency is becoming mainstream in radar front-end design, encouraging the related research to extract fine-grained ECG signals as the golden standard in clinical diagnosis and realize robust monitoring in the presence of real-world noises. In this thesis, radar-based ECG recovery is thoroughly investigated to provide a contactless ECG measurement that gets rid of cumbersome wired connections and adhesive electrode patches. However, several challenges need to be solved: (a) an efficient signal processing algorithm is required to ensure a high signal-to-noise ratio (SNR) for the collected radar signal, providing enough cardiac features for the later ECG recovery; (b) the domain transformation for the cardiac activities within single-cardiac cycle from mechanical to electrical domain needs to be modeled to realize a robust ECG recovery against noises; (c) the long-term ECG recovery should be modeled and realized instead of relying on the purely data-driven methods that can hardly resist noise disturbance; (d) the developed deep learning model normally relies on the large-scale radar/ECG dataset for training, and alleviating data scarcity is an important issue especially for the deployment in the new scenarios with limited data. Based on the challenges above, multiple novel algorithms and deep learning frameworks are proposed: (a) a cardio-focusing and -tracking (CFT) algorithm is proposed to iteratively approach the point with a high-SNR radar signal extracted; (b) a deep learning model called radarODE is designed to realize robust single-cycle ECG recovery; (c) a multi-task learning framework called radarODE-MTL is proposed to realize robust long-term ECG recovery; (d) a data augmentation method Horcrux and a transfer learning framework RFcardi are proposed to jointly decrease the demand for data acquisition. Extensive experiments are performed based on both public and private datasets to show the effectiveness of the proposed algorithms and frameworks. It is believed that this thesis contributes to the general development of the wireless sensing community and brings the future applications of wireless wellness monitoring closer to our daily lives.

Item Type:	Thesis (PhD)
Uncontrolled Keywords:	Contactless Vital Sign Monitoring, Radar-Based Sensing, Transfer Learning, Derivative-Free Optimization, Multi-task Learning, Augmentation
Divisions:	Faculty of Science & Engineering Faculty of Science & Engineering > School of Electrical Engineering, Electronics and Computer Science
Depositing User:	Symplectic Admin
Date Deposited:	10 Feb 2026 11:29
Last Modified:	01 Jul 2026 01:30
DOI:	10.17638/03194297
Supervisors:	Yang, Rui Guan, Sheng-Uei Yue, Yutao Huang, Yi
URI:	https://livrepository.liverpool.ac.uk/id/eprint/3194297
Disclaimer:	The University of Liverpool is not responsible for content contained on other websites from links within repository metadata. Please contact us if you notice anything that appears incorrect or inappropriate.