🤖 AI Summary
Wi-Fi sensing systems suffer from poor generalization when deployed across diverse real-world domains, while scarce labeled data hinder conventional domain adaptation/generalization approaches. This paper addresses cross-domain generalization for CSI-based human activity recognition (HAR) by proposing the Antenna Response Consistency (ARC) self-supervised alignment criterion—the first to embed physical-layer antenna response characteristics into a contrastive learning framework, jointly preserving semantic fidelity and robustness to channel noise. We further integrate physics-informed CSI modeling, time-frequency domain enhancement, and the ARC constraint to devise a lightweight, efficient pretraining method. Evaluated on multiple WiFi-HAR benchmarks, our approach achieves accuracy improvements exceeding 5%, reaching up to 94.97%, significantly outperforming standard contrastive learning baselines. Results demonstrate ARC’s effectiveness for low-resource, cross-domain generalization in Wi-Fi sensing.
📝 Abstract
Self-supervised learning (SSL) for WiFi-based human activity recognition (HAR) holds great promise due to its ability to address the challenge of insufficient labeled data. However, directly transplanting SSL algorithms, especially contrastive learning, originally designed for other domains to CSI data, often fails to achieve the expected performance. We attribute this issue to the inappropriate alignment criteria, which disrupt the semantic distance consistency between the feature space and the input space. To address this challenge, we introduce extbf{A}ntenna extbf{R}esponse extbf{C}onsistency (ARC) as a solution to define proper alignment criteria. ARC is designed to retain semantic information from the input space while introducing robustness to real-world noise. Moreover, we substantiate the effectiveness of ARC through a comprehensive set of experiments, demonstrating its capability to enhance the performance of self-supervised learning for WiFi-based HAR by achieving an increase of over 5% in accuracy in most cases and achieving a best accuracy of 94.97%.
