Channel characterization and adaptive learning solutions for WiFi-assisted sensing in indoor environments

由于其在智能家居、远程健康监测和辅助日常生活方面的潜在应用，室内环境中的运动传感是众多人的强烈愿望。与基于图像的传感技术相反，无线信号（例如Wi-Fi）可以在不影响日常生活的情况下以保护隐私的方式用于运动传感。考虑到无处不在的Wi-Fi信号和运动传感的重要性，我们的合作伙伴组织（PO）“认知系统公司”开发了Wi-Fi运动传感（WMS）系统。然而，现有的WMS系统有一些限制，例如(i)只能预测一个特定的运动，在分辨率上受限于系统中运行的独立Wi-Fi设备。（ii）有限的多用户分辨率，由于缺乏单用户和多用户通道干扰之间的区分特征，（iii）有限的监督机器学习（ML）解决方案的训练标签，以及（iv）只能在少数特定的室内环境中部署。在本研究项目中，我们的目标是通过提出新颖的机器学习解决方案来解决这些限制，并对WMS系统的可重构智能表面（RISs）等破坏性通信技术进行可行性分析。该项目的主要目标是(A)在多用户环境中对室内无线信道模型进行数学表征，并为各种ML解决方案生成全面的数据集。这些合成数据集将能够使我们的PO目前拥有的标记数据集多样化，从而有可能提高基于ML的WMS系统的定位精度，(B)推导新的定制损失函数并开发各种监督、自监督和半监督的ML解决方案，以及(C)通过计算机模拟分析RISs在提高WMS系统精度方面的意义。PO将充分利用这项研究项目的进展和发现，从而能够继续在全球范围内引领运动传感领域。

Motion sensing in an indoor environment is a strong desire of a multitude of people due to its potential applications in smart homes, remote health monitoring and assisted daily living. Contrary to image-based sensing technologies, wireless signals (e.g., Wi-Fi) can be leveraged for motion sensing in a privacy-preserved manner without affecting daily routines. Considering the ubiquitous presence of Wi-Fi signals and significance of motion sensing, our partner organization (PO) 'Cognitive Systems Corp' developed a Wi-Fi Motion Sensing (WMS) system. However, the existing WMS system has few limitations such as (i) predicting only a specific motion, limited in resolution to an independent Wi-Fi device operating in the system. (ii) limited multi-user resolution, due to the absence of distinguishing features between single and multi-user channel disturbances, (iii) limited training labels for supervised machine learning (ML) solutions, and (iv) can only be deployed in a few particular indoor environments. In this research project, we aim to address these limitations through proposing novel ML solutions and perform the feasibility analysis of disruptive communication technologies such as reconfigurable intelligent surfaces (RISs) for the WMS system. The primary goals of this project are (A) Characterize indoor wireless channel models mathematically in multi-user environments and generate a comprehensive dataset for various ML solutions. These synthetic datasets will be able to diversify the labeled datasets that are currently owned by our PO and thus can potentially enhance the localization accuracy of the ML-empowered WMS system, (B) Derive novel customized loss functions and develop various supervised, self-supervised, and semi-supervised ML solutions, and (C) Analyze the significance of RISs in increasing the accuracy of the WMS system through computer simulations. The PO will make full use of the advancements and findings of this research project, hence will be able to continue to lead in the field of motion sensing at a global scale.