@Article{cmes.2023.030144,
AUTHOR = {Liufeng Du, Shaoru Shang, Linghua Zhang, Chong Li, Jianing Yang, Xiyan Tian},
TITLE = {Multidomain Correlation-Based Multidimensional CSI Tensor Generation for Device-Free Wi-Fi Sensing},
JOURNAL = {Computer Modeling in Engineering \& Sciences},
VOLUME = {138},
YEAR = {2024},
NUMBER = {2},
PAGES = {1749--1767},
URL = {http://www.techscience.com/CMES/v138n2/54611},
ISSN = {1526-1506},
ABSTRACT = {Due to the fine-grained communication scenarios characterization and stability, Wi-Fi channel state information (CSI) has been increasingly applied to indoor sensing tasks recently. Although spatial variations are explicitly reflected in CSI measurements, the representation differences caused by small contextual changes are easily submerged in the fluctuations of multipath effects, especially in device-free Wi-Fi sensing. Most existing data solutions cannot fully exploit the temporal, spatial, and frequency information carried by CSI, which results in insufficient sensing resolution for indoor scenario changes. As a result, the well-liked machine learning (ML)-based CSI sensing models still struggling with stable performance. This paper formulates a time-frequency matrix on the premise of demonstrating that the CSI has low-rank potential and then proposes a distributed factorization algorithm to effectively separate the stable structured information and context fluctuations in the CSI matrix. Finally, a multidimensional tensor is generated by combining the time-frequency gradients of CSI, which contains rich and fine-grained real-time contextual information. Extensive evaluations and case studies highlight the superiority of the proposal.},
DOI = {10.32604/cmes.2023.030144}
}