Trajectory-Guided Tokenization of Complex CSI for Wi-Fi Sensing

2026-10-05

AI Summary

This study addresses the loss of critical dynamic information during the compression of high-dimensional complex-valued channel state information (CSI) time-series data by proposing a trajectory-guided tokenization method for Wi-Fi sensing. The approach introduces an orthogonal Helmert transform to perform complex trajectory decomposition on temporal patches, effectively decoupling local centroids from trajectory coordinates. Furthermore, an asymmetric attention mechanism is designed to aggregate subcarrier features, which, combined with a TokenMLP architecture, constructs compact and continuous token representations. Experimental results demonstrate that the proposed method achieves an average accuracy of 92.83% on a self-constructed dataset. Its superior generalization capability is further validated across cross-domain presence detection and gesture recognition tasks.

Abstract

Wi-Fi channel state information (CSI) enables contactless presence detection and gesture recognition. Its high-dimensional complex-valued time series require input representations that preserve informative temporal variations during compression. We propose Trajectory-Guided Tokenization (TGT), which combines complex trajectory decomposition with asymmetric attention to construct compact continuous tokens. For each antenna link and subcarrier, an orthonormal Helmert transform decomposes short, ordered temporal patches into local-center and centered-trajectory coordinates. Keys are learned from the centered-trajectory coordinates, while values retain both components. Learnable queries aggregate subcarriers into frequency slots, which are fused into temporal tokens. Trained jointly from scratch, TGT with TokenMLP achieves the highest mean accuracy of 92.83% among all evaluated frontend-backend combinations on the self-collected dataset. Experiments on EHUNAM and Widar further support the applicability of TGT to cross-domain presence detection and gesture recognition.

Problem

• Wi-Fi sensing
• channel state information
• complex-valued time series
• tokenization
• gesture recognition

Innovation

• Trajectory-Guided Tokenization
• Complex CSI
• Helmert Transform
• Asymmetric Attention
• Wi-Fi Sensing