Human-to-human interaction behaviors sensing based on complex-valued neural network using Wi-Fi channel state information

Next generation wireless communications will not only be able to transmit information faster but also can sense the physical world, i.e., integrated sensing and communications (ISAC). How to use or share part of the equipment to realize simultaneous communication and sensing is becoming a hot topic in both industry and academia. In this paper, we propose a human-to-human interaction behaviors sensing method based on the complex-valued neural network (CVNN), which can sense human-to-human interaction behaviors through the channel state information (CSI) of Wi-Fi signals. Specifically, according to the propagation characteristics of wireless signals in indoor scenarios and the Fresnel zone model, we analyze the changing relationship between the position or posture of the human body and the wireless signal. Then, by analyzing Wi-Fi signals captured in an indoor scenario, we explore the differences between different channels with the same human-to-human behavior and between different subcarriers of the same channel, and investigate the impacts of different human interaction behaviors on Wi-Fi signal propagation characteristics. In addition, to reduce the computation complexity and improve the sensing speed, we combine the existing channel models and remove the static signal components. Finally, we design a sensing model based on CVNN, and realize the sensing of human-to-human interaction behaviors through the CSI. The proposed method can sense 12 kinds of human-to-human interaction behaviors, and the overall sensing accuracy is 82%. This work contributes to the theoretical exploration and system design of Wi-Fi sensing.