AUTHOR=Yu Guang , Yang Chenxi , Li Haobo , Wang Chaochao , Zhang Xianchao , Li Jianqing , Liu Chengyu 
  
TITLE=Non-contact seismocardiogram measurement and HRV analysis using cardiac beamforming with FMCW radar
  
JOURNAL=Frontiers in Physiology
  
VOLUME=Volume 16 - 2025
  
YEAR=2026
  
URL=https://www.frontiersin.org/journals/physiology/articles/10.3389/fphys.2025.1733573
  
DOI=10.3389/fphys.2025.1733573
  
ISSN=1664-042X
  
ABSTRACT=IntroductionHeart rate variability (HRV) is a vital metric for assessing cardiovascular health, psychological stress, and sleep quality. Non-contact HRV monitoring offers advantages in safety, comfort, and hygiene, making it an increasingly attractive solution.MethodsIn this study, we propose a high-precision, non-contact HRV analysis method using a 77 GHz multiple-input multiple-output (MIMO) frequency-modulated continuous wave (FMCW) radar system. The proposed method first employs an optimized Capon beamforming algorithm to accurately localize the heart and enhance intermediate frequency (IF) signals from the heart’s direction. A modified differentiate and cross-multiply (MDACM) algorithm is then used to demodulate the phase sequence, yielding a raw vital sign signal that includes both respiratory and cardiac components. This signal is further processed using a six-level wavelet packet transform (WPT), from which specific wavelet coefficients (6th to 12th bands at level six) are selected to reconstruct the seismocardiogram (SCG) signal. To extract precise inter-beat interval (IBI) sequences, a robust aortic valve opening (AO) point detection algorithm is developed. Time-domain HRV indices—including the standard deviation of normal-to-normal intervals (SDNN), the root mean square of successive differences (RMSSD), and the percentage of successive normal-to-normal intervals differing by more than 50 milliseconds (ms) (pNN50)—are then computed from the IBI sequence. To validate the approach, we developed a synchronized data acquisition system combining radar and electrocardiogram (ECG) sensors and collected data from 13 participants—each person collected data for 10 min.ResultsExperimental results demonstrate the effectiveness of our method, achieving average errors of 4.11 ms in SDNN, 8.05 ms in RMSSD, and 2.15% in pNN50 compared to ECG-derived ground truth.DiscussionThese results outperform existing non-contact HRV monitoring techniques and highlight the method’s potential for practical, continuous, and unobtrusive cardiovascular monitoring.