Lightweight and Standalone IoT Based WiFi Sensing for Active Repositioning and Mobility

Abstract

Channel state information (CSI) provides rich insight into the physical characteristics of an environment through radio subcarrier frequencies in orthogonal frequency-division multiplexing (OFDM) systems. Many recent studies explore this rich source of data to produce quite accurate results in device-free localization, human-body pose recognition, and device-free person identification under the umbrella of WiFi Sensing. Most works thus far rely on the use of the Intel 5300 Network Interface Card (NIC), a device requiring connection to a host computer to function. Because of this requirement, the weight and form factor of CSI recording capable devices (receiver or RX) has limited the abilities of researchers to explore certain aspects of WiFi sensing such as active repositioning and mobility of RX devices. To address this, in this paper, we use the ESP32 microcontroller to develop a simple and lightweight solution for CSI collection leveraging recent additions to the Espressif IoT Development Framework which allows user developed programs to access CSI directly. The system can work standalone or attached to a smartphone for advanced online computations. Thus, it can be easily deployed, repositioned, and carried on mobile objects, which can then help improve the performance of sensing tasks. We evaluate the performance of our proposed system through several deep-learning based human activity recognition experiments and show that the repositioning and mobility of RX devices can provide increases in accuracy upwards of 29.4% and 28.2%, respectively, compared to the commonly considered static RX scenario. Finally, we produce an easy to use open source codebase for researchers to immediately begin exploring the new possibilities (e.g., massive deployment) available by the usage of the proposed system.