Wireless vital sign sensing (VSS) of respiration, capturing signal variations caused by chest movements, is sensitive to signal propagation paths, with current research primarily focusing on coarse-grained changes in environments. However, the frequently changing relative positions between the chest and devices result in varying signal propagation degradation, contributing to adverse monitoring states such as non-line-of-sight, long distance, blind spots, and suboptimal orientations, making the sensing performance highly susceptible. In this paper, we propose a robust VSS scheme with WiFi under varying signal degradation (RoSe). The core mechanism lies in adaptively recovering signals by learning from expert respiratory features under direct-path propagation. For broader respiratory patterns, we employ a diffusion model to expand the manually collected expert database under ideal conditions. To amplify the respiration features, we establish a signal model with multi-dimensional information, including climbing speed, evolution amplitude, peak position, and duration, providing effective guidance for subsequent recovery. The expert characteristics serve as a reference, from which imitation learning identifies strategies to adaptively recover respiration features disrupted under diverse adverse states. According to our proposed robustness metric, accounting for performance variance, balance, and loss, experimental results on 20,000 samples demonstrate that RoSe achieves superior robustness across varying signal degradation.
Zhao et al. (Tue,) studied this question.