Generative Artificial Intelligence Assisted Wireless Sensing: Human Flow Detection in Practical Communication Environments

TL;DR

This paper presents G-HFD, a novel GAI-assisted wireless sensing system that accurately detects human flow and targets using channel state information and a diffusion model, demonstrating 91% accuracy in practical scenarios.

Contribution

It introduces a unified weighted conditional diffusion model (UW-CDM) for denoising and accurate detection in wireless sensing, integrating GAI with CSI-based human flow detection.

Findings

G-HFD achieves 91% accuracy in subflow size detection.

The system effectively estimates human movement parameters using CSI.

UW-CDM enhances DoA estimation accuracy in practical environments.

Abstract

Groundbreaking applications such as ChatGPT have heightened research interest in generative artificial intelligence (GAI). Essentially, GAI excels not only in content generation but also in signal processing, offering support for wireless sensing. Hence, we introduce a novel GAI-assisted human flow detection system (G-HFD). Rigorously, G-HFD first uses channel state information (CSI) to estimate the velocity and acceleration of propagation path length change of the human-induced reflection (HIR). Then, given the strong inference ability of the diffusion model, we propose a unified weighted conditional diffusion model (UW-CDM) to denoise the estimation results, enabling the detection of the number of targets. Next, we use the CSI obtained by a uniform linear array with wavelength spacing to estimate the HIR's time of flight and direction of arrival (DoA). In this process, UW-CDM solves the problem of ambiguous DoA spectrum, ensuring accurate DoA estimation. Finally, through clustering, G-HFD determines the number of subflows and the number of targets in each subflow, i.e., the subflow size. The evaluation based on practical downlink communication signals shows G-HFD's accuracy of subflow size detection can reach 91%. This validates its effectiveness and underscores the significant potential of GAI in the context of wireless sensing.