"Wi-Fi Sensing for Human Activity Recognition: A Monostatic Approach via Beamforming Feedback Matrix"

"Wi-Fi\u2013based Human Activity Recognition (HAR) faces significant deployment barriers because most existing systems rely on Channel State Information (CSI), which requires multiple transceiver devices, including specific hardware, driver access, or firmware modification. Recent studies have explored Beamforming Feedback Matrix (BFM) signals as a lightweight alternative; however, these approaches typically employ bistatic configurations with physically separated transmitters and receivers, limiting practicality in residential deployments. This paper presents a firmware-agnostic, monostatic HAR framework that leverages BFM signals captured by a single commercial IEEE 802.11ac router, eliminating the need for auxiliary transceivers or hardware modification. A principled signal processing and feature engineering pipeline is introduced, incorporating Signal-to-Noise Ratio (SNR) filtering and BFM\u2013acceleration extraction to mitigate noise and nonstationarity in compressed beamforming data. The proposed system is evaluated using 150 experimental sessions, involving five subjects across three distinct environments, under a LeaveOne-Subject-Out (LOSO) cross-validation protocol. Experimental results demonstrate classification accuracy of up to 98.75% for unseen subjects within the same environment and over 93% accuracy under cross-environment testing. These results validate that monostatic BFM sensing retains sufficient spatial and temporal structure for robust activity recognition, establishing a practical pathway toward scalable, firmware-agnostic Wi-Fi sensing using commodity devices.  "

基于Wi-Fi的人体活动识别（Human Activity Recognition, HAR）存在显著的部署壁垒，因绝大多数现有系统均依赖信道状态信息（Channel State Information, CSI），而该技术需配备多台收发设备，涵盖特定硬件、驱动程序访问权限或固件修改。近期已有研究探索将波束成形反馈矩阵（Beamforming Feedback Matrix, BFM）信号作为轻量化替代方案；然而此类方法通常采用收发物理分离的双静态配置，限制了其在住宅场景中的实用性。本文提出一种与固件无关的单静态HAR框架，该框架利用单台商用IEEE 802.11ac路由器捕获的BFM信号，无需额外收发器或硬件修改。本文还引入一套系统化的信号处理与特征工程流程，包含信噪比（Signal-to-Noise Ratio, SNR）滤波与BFM加速度提取，以缓解压缩波束成形数据中的噪声与非平稳性问题。所提系统通过150组实验会话开展评估，涉及5名受试者在3种不同环境下的测试，并采用留一受试者交叉验证（LeaveOne-Subject-Out, LOSO）协议。实验结果表明，同一环境下未知受试者的分类准确率最高可达98.75%，跨环境测试下的准确率亦超过93%。上述结果证实，单静态BFM传感保留了足够的时空结构以实现鲁棒的活动识别，为使用商用设备实现可扩展、与固件无关的Wi-Fi感知开辟了实用路径。