Array Processing and Deep Learning for Non-Contact Human Vital Sign Estimation with MIMO Radar
Date Issued
May 2026
Author(s)
Moadi, Abdel-Kareem
The University of Tennessee
Advisor(s)
Aly Fathy
Additional Advisor(s)
Seddik Djouadi
Ahmedullah Aziz
Abstract
This dissertation traces the evolution of contactless vital sign monitoring from early single-input single-output (SISO) impulse-radio ultra-wideband (IR-UWB) radar to modern multiple-input multiple-output (MIMO) frequency-modulated continuous-wave (FMCW) radar, and presents a MIMO radar-based framework for monitoring respiration and cardiac activity in complex environments. The work integrates advanced array processing, radar signal processing, and deep learning to advance the state of the art in human sensing in three main contributions. First, it introduces new neural-network-based localization and tracking methods that enable accurate and consistent multi-subject tracking in real time on an embedded processor and support subject-specific spatial filtering, reducing mean absolute error for respiration and heart-rate estimation to 0.68 and 2.7 BPM, respectively. Second, it develops novel range-bin combination methods that exploit distributed target energy across adjacent bins to improve waveform reconstruction, reducing respiration waveform mean squared error to 0.07 in two- and three-subject scenarios. Finally, for the first time end-to-end deep learning methods are implemented for through-wall respiration waveform reconstruction. Experimental evaluation across diverse multi-person, obstructed, and through-wall scenarios shows that the proposed methods improve signal quality, tracking reliability, and vital sign estimation accuracy relative to conventional baselines. In particular, the proposed through-wall respiration reconstruction approach reduces mean squared error by up to 43% compared with state-of-the-art methods while maintaining robust performance in interference-prone settings. Collectively, these contributions have achieved the goal of advancing the state-of-the-art in pursuit of the optimal array weights for human vital signs through classical and deep learning-based approaches and have introduced new ways of combining range bin information to further enhance vital sign estimation accuracy.
Subjects
Degree
Doctor of Philosophy
Major
Electrical Engineering
Embargo Date
May 1, 2027