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

Ziyi Wang; Kenuo Xu; Jichu Jiang; Yumeng Yang; Zheng Chen; Xiaofei Bai; Muge Chen; Xuyang Chen

cs.LGOct 5, 2026

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

Authors: Ziyi Wang, Kenuo Xu, Jichu Jiang, Yumeng Yang, Zheng Chen, Xiaofei Bai, Muge Chen, Xuyang Chen, +3 more

Organizations: Technische Universität Berlin, Berlin, Germany · Huawei Heisenberg Research Center (Munich) · Huawei Technologies Co., Ltd., Shanghai, China · Technical University of Munich, Munich, Germany

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.

Figures & tables

Figure 1: Overview of TGT, with dashed boxes corresponding to Sections 2.1–2.4. One antenna link is shown using $P=8$ samples per patch; all links share the tokenizer.

Figure 2: Accuracy comparison of TGT and CSI encoding baselines across three datasets. Lines show means; shaded bands indicate

\pm

one sample standard deviation.

Variant	Presence	EHUNAM	Widar
w/o Helmert	$64.93\pm 3.37$	$70.69\pm 1.64$	$60.02\pm 2.01$
w/ Key Center	$90.86\pm 0.46$	$87.80\pm 5.93$	$70.33\pm 2.31$
w/o V Center	$89.62\pm 1.46$	$87.53\pm 1.15$	$70.96\pm 1.08$
TGT	$\mathbf{92.83\pm 0.35}$	$\mathbf{90.44\pm 1.02}$	$\mathbf{72.14\pm 1.41}$

Table 1: TGT ablation accuracy (%), mean

\pm

sample standard deviation.

Explore similar work

Physics-Guided Attention in a Lightweight TCN for Efficient WiFi CSI-Based Human Activity Recognition
Jun 1, 2026Chinthaka Ranasingha, Tharindu Fernando, Sridha Sridharan +2Human Activity Recognition Channel State Information
Signal Processing, Artificial Intelligence and Vision Technologies (SAIVT) Research Group, School of Electrical Engineering and Robotics, Queensland University of Technology (QUT), Brisbane, QLD 4000, Australia
TrajTok: Adaptive Spatial Tokenization for Trajectory Representation Learning
May 19, 2026Zhen Xiong, Shang-Ling Hsu, Cyrus ShahabiSpatial Tokens Traccia
University of Southern California
CSI-JEPA: Towards Foundation Representations for Ubiquitous Sensing with Minimal Supervision
May 13, 2026Xuanhao Luo, Zhizhen Li, Yuchen LiuChannel State Information Self-Supervised Representations
North Carolina State University, USA

Jun 1, 2026, cs.CVJ/K move · Enter open · S save

Physics-Guided Attention in a Lightweight TCN for Efficient WiFi CSI-Based Human Activity Recognition

Chinthaka Ranasingha, Tharindu Fernando, Sridha Sridharan, Clinton Fookes+1

Signal Processing, Artificial Intelligence and Vision Technologies (SAIVT) Research Group, School of Electrical Engineering and Robotics, Queensland University of Technology (QUT), Brisbane, QLD 4000, Australia

Human Activity Recognition Channel State Information Temporal Convolutional Networks Wifi Physics-Guided Machine Learning Inductive Bias

Open paper AskPDF arXiv

Human Action Recognition (HAR) using WiFi Channel State Information (CSI) has gained increasing attention due to its non-contact, low-cost, and privacy-preserving nature. However, existing learning-based approaches largely rely on deep, computationally intensive architectures to implicitly capture motion dynamics from CSI measurements, thereby increasing model complexity and reducing efficiency. Instead, we argue that incorporating appropriate inductive biases tailored to the physical characteristics of CSI signals enables more efficient and effective learning. In this work, we propose a compact temporal convolutional network (TCN)-based framework that explicitly incorporates motion-aware inductive biases into feature learning. Specifically, we introduce a Doppler-energy-guided temporal attention mechanism in feature space to emphasize motion-salient time segments, and a variance-driven channel attention module to weight informative subcarriers based on temporal motion statistics adaptively. By integrating these domain-specific priors, the proposed model effectively captures motion dynamics without increasing architectural depth. Extensive experiments on multiple benchmark datasets demonstrate that our approach achieves superior performance compared to deeper baselines, while significantly reducing parameter count and computational cost.

Abstract

Explore similar work

Physics-Guided Attention in a Lightweight TCN for Efficient WiFi CSI-Based Human Activity Recognition

TrajTok: Adaptive Spatial Tokenization for Trajectory Representation Learning

CSI-JEPA: Towards Foundation Representations for Ubiquitous Sensing with Minimal Supervision