Robust Vital Signs Monitoring of Speaking Subjects Through mmWave Radar

Abstract

Vital signs measurement using radio frequency (RF) signals, particularly mmWave-based methods, has gained widespread attention. Speaking, which modifies the pattern of chest wall motion during phonation, leads to intricate couplings with vital signals in both the temporal and spectral domains. Respiratory spectrum broadening and phase noise interference are challenges in vital signs monitoring of speaking subjects. To tackle these issues, a novel method is proposed. First, a dynamic composite model of chest wall motion is developed, and the inherent challenges in vital signs monitoring of speaking subjects are analyzed in detail. Second, a recursive autocorrelation periodic enhancement (RAPE) algorithm is proposed, leveraging the periodicity of respiration and the intermittency of speech to enhance the respiratory signal. A recursive strategy is employed, which incorporates an adaptive termination mechanism based on Shannon entropy and spectral concentration. Third, an adaptive low-rank decomposition (ALRD) algorithm is proposed, exploiting the low-rank property of the Hankel matrix from the heartbeat signal to transform the denoising challenge into a matrix decomposition task. It also models phase noise as energy-bounded interference and adaptively selects the optimal parameter, achieving high-quality separation of weak heartbeat signals from phase noise. Extensive experimental results demonstrate that the proposed method facilitates accurate and reliable vital signs monitoring for speaking subjects. This study bridges a critical gap in the current body of noncontact vital signs measurement methods.