SLIM-SENSE: a resource efficient WiFi sensing framework towards integrated sensing and communication

Abstract

WiFi signals have become pivotal in sensing applications by leveraging Channel State Informa tion (CSI). WiImg, a novel system, utilizes Generative Adversarial Networks (GAN) for image inpainting to perform WiFi sensing under low transmission rates. This project fully implements WiImg while adapting it to four different datasets: Exposing CSI, SimWiSense, SharpAX, and HeadGest. The implementation follows the outlined methodology in the original paper but is applied to these datasets, which leads to variation in results due to their specific charac teristics. The focus is on transforming raw CSI data into meaningful insights while maintaining low packet rates and ensuring high sensing accuracy. The report provides a step-by-step guide to implementing WiImg, discusses challenges encountered, and presents a comparative analysis of performance across datasets.