eess.SPOct 21, 2025

Deep Learning-Enhanced Real-Time Wi-Fi Sensing Through Single Transceiver Pair

Authors: Yuxuan Liu, Chiya Zhang, Yifeng Yuan, Chunlong He, Weizheng Zhang, Gaojie Chen

Organizations: School of Electronic and Information Engineering, Harbin Institute of Technology, Shenzhen 518055, China. · School of Electronic and Information Engineering, Harbin Institute of Technology, Shenzhen 518055, China, and also with the Peng Cheng Laboratory (PCL), Shenzhen 518055, China. · Guangdong Key Laboratory of Intelligent Information Processing, Shenzhen University, Shenzhen 518060, China. · School of Flexible Electronics (SoFE), State Key Laboratory of Optoelectronic Materials and Technologies (OEMT), Sun Yat-sen University, Shenzhen, Guangdong 518107, China.

Read PDF
Ask AI
Cite

Copy citation

DOINot available for this paper

Abstract

The advancement of next-generation Wi-Fi technology heavily relies on sensing capabilities, which play a pivotal role in enabling sophisticated applications. In response to the growing demand for large-scale deployments, contemporary Wi-Fi sensing systems strive to achieve high-precision perception while maintaining minimal bandwidth consumption and antenna count requirements. Remarkably, various deep learning-driven perception technologies have demonstrated the ability to surpass conventional resolution limits. However, the theoretical underpinnings of this phenomenon have not been thoroughly investigated in existing research. We find that under hardware-constrained conditions, the performance gains of deep learning in Wi-Fi sensing primarily originate from two aspects: prior information and temporal correlation, which act as specific forms of side information that reduce the estimation error bound. We construct a deep learning-based Wi-Fi sensing system using only a single transceiver pair and design experiments to validate these gains. The system achieves an average human pose estimation error of 0.2189 m and an average localization error of 0.6124 m, while operating in real time at 42 fps on commodity hardware.

Figures & tables

Fig. 1: Angle-of-arrival estimation in a typical Wi-Fi sensing system. The angle of arrival is determined from the time difference between the arrivals of the same signal at different antennas.
Fig. 2: A flowchart of an estimation process, where g(Z)g(Z) is an estimator that relies only on observations, and g∗(Z∣Y)g^{\ast}(Z|Y) is an estimator when side information YY is known.
Fig. 3: An example of the collected CSI. The area within the red box exhibits clear sinusoidal characteristics.
Fig. 4: Schematic diagram of the deep learning-based real-time Wi-Fi sensing and visualization system.
Fig. 5: Illustration of the network structure, where we incorporate a hyperparameter in the stacked ConvGRU layers to control the maximum processable sequence length. The depth head and 2D pose estimation head work jointly to achieve 3D human pose estimation.
Fig. 6: Illustration of the three classrooms used for data collection, showing the positions of the transmitter and receivers. During data collection, different propagation environments were created within the same classroom by varying the number and spatial distribution of scatterers.
Fig. 7: (a) CDF of the localization error of the proposed system and of the MUSIC baseline under the same single-transceiver-pair setting. The MUSIC curve is plotted after excluding estimates that fall outside the room, which are spurious peaks caused by the antenna spacing of our testbed. (b) CDF of the human pose estimation error of the proposed system.
JointThoraxRight ShoulderRight ElbowRight WristLeft ShoulderLeft ElbowLeft Wrist
MAE (m)0.08860.15650.36780.47630.17990.35170.4856
TABLE I: The mean absolute error (MAE) of HPE.
Fig. 8: Training loss curves for CSI‑based human pose estimation and localization with and without image pre‑training.
Fig. 9: Average estimation error and feasible error lower bound range over time. (a) Pose estimation: average errors of HPE_I and HPE_T, and the lower bound of HPE_T. (b) Localization: average errors of HPE_I and HPE_T, and the lower bound of HPE_T.
Fig. 10: The improvement of HPE_T over HPE_I on different timestamps.
Fig. 11: Example of temporal correlation gain, where the blue skeletons represent the ground truth, while the red skeletons indicate the estimation results. It can be observed that the pose estimation results become progressively more accurate over time.
Fig. 12: Real-time sensing and visualization system in operation.
Figure 14
Figure 15
Figure 16
Figure 17
Figure 18
Figure 19
More in this topic

Wi-Fi Sensing

Wi-Fi sensing uses wireless signal changes to infer human motion, location, presence, and activity without dedicated wearable sensors.

See all 48 papers

Explore similar work

CardsList

  1. RF-LEGO: Modularized Signal Processing-Deep Learning Co-Design for RF Sensing via Deep Unrolling

    Apr 11, 2026Luca Jiang-Tao Yu, Chenshu WuDeep UnfoldingModular Neural Networks

    The University of Hong Kong

  2. DoRF++: Spherical Representation Learning over Doppler Radiance Fields for Robust Wi-Fi Sensing

    Aug 9, 2026Navid Hasanzadeh, Shahrokh ValaeeHuman Activity RecognitionGeometric Representation Learning

    Department of Electrical & Computer Engineering, University of Toronto, Toronto, ON, Canada.

  3. Seeing Through WiFi: Lightweight Human Pose Estimation with Dynamic Kernel Attention

    Jul 3, 2026Toan D. Gian, Van-Dinh Nguyen, Vo Phi Son +5Dynamic ConvolutionChannel Attention

    Smart Green Transformation Center (GREEN-X), VinUniversity, Vietnam · Trinity College Dublin, Ireland · University of Oulu, Finland +3