Wi-Chat: Large Language Model Powered Wi-Fi Sensing
Haopeng Zhang, Yili Ren, Haohan Yuan, Jingzhe Zhang, Yitong Shen
TL;DR
Wi-Chat demonstrates that large language models can be guided to interpret raw Wi-Fi CSI signals for human activity recognition by embedding Wi-Fi sensing physical models into prompts, enabling zero-shot inference without traditional signal processing. The approach combines four activity models (Walking, Falling, Breathing, No-event) with prompting strategies (Base, In-context Learning, Chain-of-Thought, Multi-modal) to map CSI representations to activity labels. Experiments on a self-collected CSI dataset show that vision-enabled, CoT-augmented prompting (e.g., GPT-4o-mini + CoT) can achieve high zero-shot accuracy (~0.90), while supervised methods still outperform in labeled settings, highlighting a promising direction for data-efficient sensing. The work expands the frontier of LLM applications into wireless sensing, offering a scalable, data-efficient paradigm for IoT and mobile sensing tasks, with future work on robustness and broader sensing modalities.
Abstract
Recent advancements in Large Language Models (LLMs) have demonstrated remarkable capabilities across diverse tasks. However, their potential to integrate physical model knowledge for real-world signal interpretation remains largely unexplored. In this work, we introduce Wi-Chat, the first LLM-powered Wi-Fi-based human activity recognition system. We demonstrate that LLMs can process raw Wi-Fi signals and infer human activities by incorporating Wi-Fi sensing principles into prompts. Our approach leverages physical model insights to guide LLMs in interpreting Channel State Information (CSI) data without traditional signal processing techniques. Through experiments on real-world Wi-Fi datasets, we show that LLMs exhibit strong reasoning capabilities, achieving zero-shot activity recognition. These findings highlight a new paradigm for Wi-Fi sensing, expanding LLM applications beyond conventional language tasks and enhancing the accessibility of wireless sensing for real-world deployments.