mmWave Sensing for Vital Sign Monitoring and User Identification: Enhancing Accuracy and Robustness via Deep Learning

Wang, Yingqi

mmWave Sensing for Vital Sign Monitoring and User Identification: Enhancing Accuracy and Robustness via Deep Learning

Wang, Yingqi

Permalink

Publication Type:: Thesis
Issue Date:: 2025

Open Access

Copyright Clearance Process

Recently Added
In Progress
Open Access

This item is open access.

Adobe PDF

Download thesisAdobe PDF (5.34 MB)

View statistics

Full metadata record

Field	Value	Language
dc.contributor.advisor	au.edu.uts.lib/cph
dc.contributor.author	Wang, Yingqi
dc.date.accessioned	2026-03-25T02:56:02Z
dc.date.available	2026-03-25T02:56:02Z
dc.date.issued	2025
dc.identifier.uri	http://hdl.handle.net/10453/194445
dc.description	University of Technology Sydney. Faculty of Engineering and Information Technology.	en_US.UTF-8
dc.description.abstract	Millimeter-wave (mmWave) sensing has emerged as a powerful sensing modality for non-contact, privacy-preserving human monitoring applications. However, such technology could still lose the efficiency in real-world sensing scenarios. This thesis explores the integration of mmWave sensing for vital sign monitoring and user identification, leveraging deep learning and advanced signal processing to enhance accuracy and robustness in dynamic environments. The first focus of this research is on vital sign monitoring, where mmWave radar is employed to detect respiratory rate, heart rate, arterial pulse, and blood pressure (BP). A multi-modal sensing technology and a novel multi-channel variational mode decomposition (VMD) method are investigated to separate vital sign components from interference caused by body movements and environmental noise. Further, a robust heartbeat and wrist localization algorithm and a physics-driven deep learning approach are developed, incorporating Neural Ordinary Differential Equations (Neural ODEs) and Temporal Convolutional Networks (TCNs) to reconstruct high-level physiological signals from radar Doppler shifts. The system achieves high-accuracy estimation of physiological parameters, demonstrating its feasibility for continuous, non-invasive health monitoring. The second focus is user identification, aiming to associate the user’s ID with the previous vital signs sensing. A novel Inverse Synthetic Aperture Radar (ISAR) sensing is investigated to reconstruct body profiles of individuals walking past the radar. A ResNet-based deep learning model with Additive Angular Margin Loss is developed to enhance feature discrimination, enabling high-precision user identification in long-term. Unlike traditional biometric methods, this approach identifies individuals based on their radio imaging profiles, making it robust to occlusions and appearance variations. The contributions of this work include: (1) a novel multi-modal mmWave localization and signal decomposition is proposed and address the challenges of low accuracy in practical vital sign monitoring, (2) an adaptive mmWave heart and wrist detection, and the physics-embedded learning scheme is proposed to improve flexibility, robustness and explainability in ECG (Electrocardiogram), pulse and BP monitoring, and (3) an ISAR-based user identification method optimized with deep learning for real-world scenarios, significantly improved the performance in long-term mmWave user Re-ID. This research paves the way for future applications in smart healthcare, security, and human-computer interaction.	en_US.UTF-8
dc.format	Thesis (PhD)
dc.language.iso	en_US	en_US.UTF-8
dc.relation	https://opus.lib.uts.edu.au/bitstream/10453/194445/1/thesis.pdf
dc.rights	info:eu-repo/semantics/openAccess
dc.rights	The author owns the copyright in this thesis including all reproduction and reuse rights for the work. The work may not be altered without the permission of the copyright owner. Attribution is essential when quoting or paraphrasing from this thesis.
dc.rights	© 2025 Yingqi Wang
dc.rights	au.edu.uts.lib/cph
dc.title	mmWave Sensing for Vital Sign Monitoring and User Identification: Enhancing Accuracy and Robustness via Deep Learning	en_US.UTF-8
dc.type	Thesis
utslib.copyright.status	open_access	*

Abstract:

Millimeter-wave (mmWave) sensing has emerged as a powerful sensing modality for non-contact, privacy-preserving human monitoring applications. However, such technology could still lose the efficiency in real-world sensing scenarios. This thesis explores the integration of mmWave sensing for vital sign monitoring and user identification, leveraging deep learning and advanced signal processing to enhance accuracy and robustness in dynamic environments. The first focus of this research is on vital sign monitoring, where mmWave radar is employed to detect respiratory rate, heart rate, arterial pulse, and blood pressure (BP). A multi-modal sensing technology and a novel multi-channel variational mode decomposition (VMD) method are investigated to separate vital sign components from interference caused by body movements and environmental noise. Further, a robust heartbeat and wrist localization algorithm and a physics-driven deep learning approach are developed, incorporating Neural Ordinary Differential Equations (Neural ODEs) and Temporal Convolutional Networks (TCNs) to reconstruct high-level physiological signals from radar Doppler shifts. The system achieves high-accuracy estimation of physiological parameters, demonstrating its feasibility for continuous, non-invasive health monitoring. The second focus is user identification, aiming to associate the user’s ID with the previous vital signs sensing. A novel Inverse Synthetic Aperture Radar (ISAR) sensing is investigated to reconstruct body profiles of individuals walking past the radar. A ResNet-based deep learning model with Additive Angular Margin Loss is developed to enhance feature discrimination, enabling high-precision user identification in long-term. Unlike traditional biometric methods, this approach identifies individuals based on their radio imaging profiles, making it robust to occlusions and appearance variations. The contributions of this work include: (1) a novel multi-modal mmWave localization and signal decomposition is proposed and address the challenges of low accuracy in practical vital sign monitoring, (2) an adaptive mmWave heart and wrist detection, and the physics-embedded learning scheme is proposed to improve flexibility, robustness and explainability in ECG (Electrocardiogram), pulse and BP monitoring, and (3) an ISAR-based user identification method optimized with deep learning for real-world scenarios, significantly improved the performance in long-term mmWave user Re-ID. This research paves the way for future applications in smart healthcare, security, and human-computer interaction.

Please use this identifier to cite or link to this item:

http://hdl.handle.net/10453/194445