Investigation into wireless sensing using channel state information

Abstract

Wireless sensing has become popular for human activity recognition (HAR). It primarily exploits channel state information (CSI), which is affected by environmental changes. Thus, it is possible to determine the corresponding human activity by measuring CSI over time. This led to research on tools that can extract CSI. However, a limitation of some of these tools is the cost and size. Thus, the ESP32 CSI Tool, which uses the ESP32 microcontroller, could potentially be a low-cost solution for CSI sensing.

In this project, the ESP32 CSI Tool was used to perform CSI-based wireless sensing in two different areas: respiration sensing and keystroke inference. A phone was used as a receiver for respiration sensing, and testing was done in two different scenarios. The first was direct contact, while the second was the phone (receiver) placed 15 cm away from the chest. On the other hand, a 1D CNN model was developed for keystroke inference that takes in CSI amplitude and predicts the corresponding keystroke.

Although the goal of building a keystroke inference system was not achieved, promising results were obtained using the ESP32 CSI Tool for respiration sensing. The respiration process could be seen in direct contact and when the receiver was 15 cm away, albeit with a potentially lowered accuracy. An application was also developed to parse, process and plot CSI data collected using the ESP32 CSI Tool.