AI 中文总结
本文利用商用BLE RSSI，通过Nordic nRF5340开发套件设计轻量级阈值分类器，实现低成本无设备的物联网系统材料与障碍物识别，验证了粗粒度材料识别的可行性。
AI 中文摘要
配备蓝牙低功耗（BLE）无线电的物联网（IoT）设备会提供用于链路管理的接收信号强度指示（RSSI）测量值；然而，这些现成的测量值也能捕捉到因障碍物材料阻挡传播路径而产生的信号变化。本文使用两个商用Nordic nRF5340开发套件开展材料识别研究，无需额外射频（RF）感知前端。首先，通过实验表征了收发器的距离相关RSSI基线，证明了距离感知校准的必要性。接着，针对五种知名障碍物材料采集RSSI轨迹：即木材、陶瓷、空塑料瓶、玻璃和人体。对这些轨迹的均值、标准差及相对于局部基线的瞬态偏差进行表征，进而估计路径损耗指数。利用这些估计值设计了一种轻量级基于阈值的分类器，在报告的未知试验中，该分类器可正确识别人体、玻璃和木材，而纸张、黄铜等材料则被映射到最相似的训练类别，即塑料瓶和陶瓷。结果表明，利用商用BLE RSSI进行粗粒度材料识别具有可行性。
英文摘要
Internet of Things (IoT) devices equipped with Bluetooth Low Energy (BLE) radios provide received signal strength indicator (RSSI) measurements for link management; however, these readily available measurements can also capture signal variations caused by materials obstructing the propagation path. This paper investigates material identification using two commodity Nordic nRF5340 development kits without an additional RF sensing front end. First, the distance-dependent RSSI baseline is characterized for the transmitter-receiver through experiments and demonstrates the need for distance-aware calibration. Next, RSSI traces are collected for five well-known materials as obstructions: namely, wood, ceramic, an empty plastic bottle, glass, and a human body. The traces are characterized using their mean, standard deviation, and transient deviation from a local baseline, and then the path-loss exponent is estimated. A lightweight threshold-based classifier is designed using these estimates, and thus the designed classifier is shown to correctly identify the human body, glass, and wood in the reported unknown trials, while materials like paper and brass are mapped to the most similar trained classes, namely a plastic bottle and ceramic. The results demonstrate the feasibility of coarse material identification using commodity BLE RSSI.
Comments6 pages and 5 figures. Submitted to a conference, under review
