pith. sign in

arxiv: 2605.00869 · v1 · submitted 2026-04-23 · 📡 eess.SP · cs.CV· cs.LG

Robust Cross-Domain WiFi Fall Detection via Physics-Driven Attention-Enhanced Transformers

Yingzhe Wang , Cunhua Pan , Ruijing Liu , Shaokai Li , Hong Ren , Kezhi Wang , Jiangzhou Wang This is my paper

Pith reviewed 2026-05-09 21:27 UTC · model grok-4.3

classification 📡 eess.SP cs.CVcs.LG
keywords WiFi CSIfall detectioncross-domain generalizationtransformerattention mechanismphysics-driven augmentationdevice-free sensingIoT health monitoring
0
0 comments X

The pith

A physics-driven attention mechanism lets WiFi fall detection maintain 98.8 percent accuracy in completely unseen indoor environments without retraining.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper introduces a hybrid CNN-Transformer model for device-free fall detection from WiFi channel state information that remains effective when the system is moved to new rooms or buildings. Standard deep learning approaches overfit to static background signals and signal paths of the training space, causing sharp drops in performance under new layouts or blocked signals. The authors add a Dynamic Variance Gate that uses local temporal variance to mask unchanging environmental components and highlight moving human bodies, plus data augmentation that forces the network to learn motion shapes rather than room-specific noise. Cross-domain tests across four indoor settings reach 97.6 percent accuracy under non-line-of-sight conditions and 98.8 percent in rooms never encountered during training, with successful real-time operation on commercial edge hardware. This robustness addresses a key barrier to practical home monitoring systems that must function immediately in varied living spaces.

Core claim

The central claim is that a CNN-Transformer architecture equipped with a physics-driven Dynamic Variance Gate, Physics-Aware Data Augmentation, and Convolutional Block Attention Module learns environment-invariant motion features from WiFi CSI. This enables 97.6 percent fall detection accuracy in NLoS scenarios and 98.8 percent accuracy in entirely unseen environments without any target-domain fine-tuning, as shown by extensive cross-domain evaluations and live edge-deployment tests.

What carries the argument

The Dynamic Variance Gate, a soft-attention mask that computes local temporal variance on the CSI sequence to suppress static environmental DC components while amplifying dynamic human motion signatures.

If this is right

  • The detector can be installed in arbitrary new homes without collecting fresh labeled data or performing fine-tuning.
  • Performance holds in non-line-of-sight conditions where walls or objects attenuate the WiFi signals.
  • Real-time inference runs on low-cost commercial WiFi network interface cards at the edge for continuous monitoring.
  • The same pipeline reduces the data-collection burden for other WiFi-based activity monitoring tasks.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The variance-gate idea could be transferred to other CSI sensing problems such as breathing detection or gesture recognition that also suffer from environment-specific overfitting.
  • Combining the method with multiple access points might further improve robustness in larger or more cluttered spaces.
  • Long-term field trials with multiple simultaneous occupants would test whether the current single-person assumption limits real-home use.

Load-bearing premise

The Dynamic Variance Gate together with the physics-aware augmentation will reliably isolate motion features that stay invariant across any indoor layouts, materials, furniture, and occupant differences beyond the four environments tested.

What would settle it

Testing the trained model in a fifth indoor environment with markedly different geometry, wall materials, and multipath characteristics, then observing accuracy drop below 90 percent, would falsify the domain-generalization claim.

Figures

Figures reproduced from arXiv: 2605.00869 by Cunhua Pan, Hong Ren, Jiangzhou Wang, Kezhi Wang, Ruijing Liu, Shaokai Li, Yingzhe Wang.

Figure 1
Figure 1. Figure 1: Overview of the proposed physics-aware, attention-enhanced WiFi [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: The full-network architecture for FallDetection. [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: (a)Detailed view of the Dynamic Variance Gate module. (b)Detailed view of the CBAM module. (c)Detailed view of Transformer Encoder module. [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Dataset Environments four directions (front, back, left, and right) and varying impact intensities, preceded by random movements such as standing still, walking forwards, or moving backwards. The non-fall activities include bending to pick up an object, sitting down, standing up, walking, and waving arms. Furthermore, these ac￾tivities were executed on both dominant LoS paths and NLoS paths to capture real… view at source ↗
Figure 5
Figure 5. Figure 5: Classification Performance in Lecture Room D (Scenario 3). [PITH_FULL_IMAGE:figures/full_fig_p009_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Classification Performance in Living Room A (Scenario 3). [PITH_FULL_IMAGE:figures/full_fig_p009_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Classification Performance in Left Living Room C (Scenario 4). [PITH_FULL_IMAGE:figures/full_fig_p009_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Classification Performance in Right Living Room C (Scenario 4). [PITH_FULL_IMAGE:figures/full_fig_p009_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Real-time testing flowchart [PITH_FULL_IMAGE:figures/full_fig_p011_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Field Testing Environment (a) (b) (c) (d) [PITH_FULL_IMAGE:figures/full_fig_p011_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Falls including forward, lateral, and backward falls [PITH_FULL_IMAGE:figures/full_fig_p011_11.png] view at source ↗
read the original abstract

Device-free fall detection utilizing WiFi Channel State Information (CSI) has emerged as a promising, privacy-preserving solution for elderly health monitoring in the Internet of Things (IoT) era. However, existing deep learning approaches suffer from severe performance degradation when deployed in unseen environments due to static background overfitting and Non-Line-of-Sight (NLoS) signal attenuation. To address these critical bottlenecks, we propose a robust, domain-generalizable framework featuring a novel Attention-Enhanced CNN-Transformer hybrid architecture. First, we design a physics-driven \textbf{Dynamic Variance Gate (DVG)} to dynamically calculate local temporal variance, acting as a soft-attention mask that eliminates static environmental DC components while amplifying dynamic human motion. Second, we introduce a Physics-Aware Data Augmentation strategy to force the network to learn invariant morphological signatures rather than environment-specific noise. Furthermore, a Convolutional Block Attention Module (CBAM) is integrated to refine spatiotemporal features prior to Transformer-based sequence modeling. Extensive cross-domain evaluations across four distinct indoor environments demonstrate that our method achieves 97.6\% accuracy in NLoS scenarios and 98.8\% in completely unseen environments without target-domain fine-tuning. Finally, we deploy the proposed framework on an edge computing system equipped with commercial WiFi NICs. Real-world live inference field tests confirm the system's robustness against unseen environmental layouts and its capability for continuous, low-latency whole-home safety monitoring.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 2 minor

Summary. The paper proposes a CNN-Transformer hybrid for device-free WiFi CSI fall detection that incorporates a physics-driven Dynamic Variance Gate (DVG) to suppress static environmental DC components, physics-aware data augmentation to promote invariant motion features, and CBAM attention before Transformer sequence modeling. It reports 97.6% accuracy in NLoS scenarios and 98.8% accuracy in completely unseen environments (no fine-tuning) across four indoor test domains, with an edge deployment demonstration.

Significance. If the cross-domain generalization claims hold under broader validation, the work would meaningfully advance practical WiFi-based elderly monitoring by mitigating environment-specific overfitting, a persistent barrier in CSI sensing. The explicit injection of physics priors (variance gating and augmentation) is a constructive direction that could transfer to other RF sensing tasks; however, the current empirical base on four environments limits immediate impact.

major comments (3)
  1. [Abstract / Experiments] Abstract and experimental evaluation: the headline claim of 98.8% accuracy in completely unseen environments without fine-tuning rests on results from only four indoor environments whose differences in volume, wall materials, furniture density, and LoS/NLoS geometry are not quantified or statistically tested (e.g., no leave-one-environment-out analysis or diversity metrics). This directly undercuts the assertion that DVG plus physics-aware augmentation extracts universally invariant features.
  2. [Method / Experiments] Method and experiments: no ablation study isolates the contribution of the Dynamic Variance Gate versus standard attention mechanisms or quantifies residual environment-specific DC components after gating. Without these controls, it is impossible to attribute the reported cross-domain gains specifically to the physics-driven components rather than to the overall architecture or augmentation.
  3. [Abstract / Results] Experimental reporting: the abstract and results lack baseline comparisons, subject diversity counts, exact train/test splits, statistical significance tests, or confidence intervals for the 97.6% NLoS and 98.8% unseen accuracies, preventing assessment of whether the performance exceeds prior art by a meaningful margin.
minor comments (2)
  1. [Method] Notation for the DVG variance calculation and augmentation factors should be defined explicitly with equations in the method section to allow reproducibility.
  2. [Figures] Figure captions and axis labels in the experimental results should include error bars or per-environment breakdowns to clarify variability.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive comments that highlight opportunities to strengthen the experimental validation of our cross-domain claims. We address each major point below and will revise the manuscript to incorporate additional analyses, quantifications, and reporting details where appropriate.

read point-by-point responses

  1. Referee: [Abstract / Experiments] Abstract and experimental evaluation: the headline claim of 98.8% accuracy in completely unseen environments without fine-tuning rests on results from only four indoor environments whose differences in volume, wall materials, furniture density, and LoS/NLoS geometry are not quantified or statistically tested (e.g., no leave-one-environment-out analysis or diversity metrics). This directly undercuts the assertion that DVG plus physics-aware augmentation extracts universally invariant features.

    Authors: We selected the four environments to capture meaningful diversity in indoor settings, including variations in room size, construction materials, and line-of-sight conditions, as described in the experimental setup. We acknowledge that explicit quantification and statistical testing would strengthen the generalization argument. In the revision, we will add quantitative descriptors of the environments (e.g., approximate volumes, material types, and geometry classifications) and include a leave-one-environment-out analysis with diversity metrics to provide statistical support for the invariance of features learned by the DVG and physics-aware augmentation. revision: yes

  2. Referee: [Method / Experiments] Method and experiments: no ablation study isolates the contribution of the Dynamic Variance Gate versus standard attention mechanisms or quantifies residual environment-specific DC components after gating. Without these controls, it is impossible to attribute the reported cross-domain gains specifically to the physics-driven components rather than to the overall architecture or augmentation.

    Authors: We agree that dedicated ablations are needed to isolate the DVG's role. The manuscript motivates the gate via physics but does not provide the requested controls. In the revised manuscript, we will add ablation experiments that (i) compare the full model against a variant without DVG, (ii) replace DVG with standard attention mechanisms, and (iii) quantify residual DC components (via temporal variance metrics before and after gating). These results will clarify the specific contribution of the physics-driven components to the observed cross-domain performance. revision: yes

  3. Referee: [Abstract / Results] Experimental reporting: the abstract and results lack baseline comparisons, subject diversity counts, exact train/test splits, statistical significance tests, or confidence intervals for the 97.6% NLoS and 98.8% unseen accuracies, preventing assessment of whether the performance exceeds prior art by a meaningful margin.

    Authors: We will revise both the abstract and the results section to include the requested elements. Specifically, we will add comparisons to relevant prior WiFi CSI fall-detection baselines, report subject counts and diversity (age/gender distribution), specify exact train/test splits and environment assignments, include statistical significance tests (e.g., paired t-tests against baselines), and provide 95% confidence intervals for the 97.6% NLoS and 98.8% unseen accuracies. These additions will enable readers to evaluate the performance margins more precisely. revision: yes

Circularity Check

0 steps flagged

No circularity; claims rest on empirical cross-domain evaluation

full rationale

The paper introduces a CNN-Transformer architecture augmented by a Dynamic Variance Gate (DVG) and Physics-Aware Data Augmentation to promote domain-invariant features for WiFi CSI fall detection. All performance numbers (97.6% NLoS, 98.8% unseen environments) are presented as outcomes of supervised training followed by held-out cross-domain testing across four indoor settings. No derivation chain, uniqueness theorem, fitted-parameter-as-prediction, or self-citation load-bearing step is present; the architecture choices are design decisions whose effectiveness is asserted via external empirical results rather than by construction or tautology. The work is therefore self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

2 free parameters · 2 axioms · 1 invented entities

Central claim rests on domain assumptions about WiFi signal physics and the ability of the proposed modules to enforce invariance; limited information available from abstract only.

free parameters (2)
  • DVG variance calculation parameters
    Thresholds or scaling factors for local temporal variance computation are likely tuned or learned.
  • Physics-aware augmentation factors
    Parameters controlling how signals are perturbed to simulate new environments.
axioms (2)
  • domain assumption Temporal variance in WiFi CSI can separate dynamic human motion from static environmental components
    Invoked as the basis for the Dynamic Variance Gate design.
  • domain assumption Data augmentation can force learning of morphological signatures invariant to environment-specific noise
    Core premise of the Physics-Aware Data Augmentation strategy.
invented entities (1)
  • Dynamic Variance Gate (DVG) no independent evidence
    purpose: Soft-attention mask to eliminate static DC components and amplify dynamic motion
    New module introduced to address background overfitting; no independent evidence outside the paper.

pith-pipeline@v0.9.0 · 5579 in / 1291 out tokens · 47671 ms · 2026-05-09T21:27:14.068845+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

35 extracted references · 35 canonical work pages

  1. [1]

    Step safely: Strategies for preventing and managing falls across the life-course,

    World Health Organization, “Step safely: Strategies for preventing and managing falls across the life-course,” WHO, Geneva, Switzerland, Tech. Rep., 2021

    work page 2021

  2. [2]

    Wifi-based human sensing with deep learning: Recent advances, challenges, and opportunities,

    I. Ahmad, A. Ullah, and W. Choi, “Wifi-based human sensing with deep learning: Recent advances, challenges, and opportunities,”IEEE Open J. Commun. Soc., vol. 5, pp. 3595–3623, 2024

    work page 2024

  3. [3]

    Device-free Wi-Fi localization via deep learning on two-way CSI,

    J. Wanget al., “Device-free Wi-Fi localization via deep learning on two-way CSI,”IEEE Internet Things J., vol. 13, no. 2, pp. 1123–1135, 2026

    work page 2026

  4. [4]

    RT-Fall: A real-time and contactless fall detection system with commodity wifi devices,

    H. Wang, D. Zhang, Y . Wang, J. Ma, Y . Wang, and S. Li, “RT-Fall: A real-time and contactless fall detection system with commodity wifi devices,”IEEE Trans. Mob. Comput., vol. 16, no. 2, pp. 511–526, 2017

    work page 2017

  5. [5]

    Falldefi: Ubiquitous fall detection using commodity wi-fi devices,

    S. Palipana, D. Rojas, P. Agrawal, and D. Pesch, “Falldefi: Ubiquitous fall detection using commodity wi-fi devices,”Proc. ACM Interact. Mob. Wearable Ubiquitous Technol., vol. 1, no. 4, pp. 1–25, Jan 2018

    work page 2018

  6. [6]

    Wi-Fi-based fall detection using spectrogram image of channel state information,

    T. Nakamura, M. Bouazizi, K. Yamamoto, and T. Ohtsuki, “Wi-Fi-based fall detection using spectrogram image of channel state information,” IEEE Internet Things J., vol. 9, no. 18, pp. 17 220–17 234, 2022

    work page 2022

  7. [7]

    Defall: Environment-independent passive fall detection using wifi,

    Y . Hu, F. Zhang, C. Wu, B. Wang, and K. J. R. Liu, “Defall: Environment-independent passive fall detection using wifi,”IEEE In- ternet Things J., vol. 9, no. 11, pp. 8515–8530, 2022

    work page 2022

  8. [8]

    AFall: Wi- Fi-based device-free fall detection system using spatial angle of arrival,

    S. Chen, W. Yang, Y . Xu, Y . Geng, B. Xin, and L. Huang, “AFall: Wi- Fi-based device-free fall detection system using spatial angle of arrival,” IEEE Trans. Mob. Comput., vol. 22, no. 8, pp. 4471–4484, 2023

    work page 2023

  9. [9]

    Fallaware: An explainable learning approach to robust fall detection with wifi,

    S. D. Regani, B. Wang, Y . Hu, G. Zhu, and K. J. R. Liu, “Fallaware: An explainable learning approach to robust fall detection with wifi,”IEEE J. Sel. Areas Sensors, vol. 2, pp. 71–83, 2025

    work page 2025

  10. [10]

    Rethinking fall detection with Wi- Fi,

    Z. Yang, Y . Zhang, and Q. Zhang, “Rethinking fall detection with Wi- Fi,”IEEE Trans. Mob. Comput., vol. 22, no. 10, pp. 6126–6143, 2023

    work page 2023

  11. [11]

    Attention-based hybrid deep learning network for human activity recognition using wifi signals,

    P. F. Moshiri, R. Shahbazian, and S. A. Ghorashi, “Attention-based hybrid deep learning network for human activity recognition using wifi signals,”Appl. Sci., vol. 13, no. 15, p. 8884, 2023

    work page 2023

  12. [12]

    Fall-attention: An attention-based fall detection method for adjoint activities,

    Y . Xiao, J. Zhu, S. Zhang, X. Liu, and S. Guo, “Fall-attention: An attention-based fall detection method for adjoint activities,”IEEE Trans. Mob. Comput., vol. 23, no. 7, pp. 7895–7909, 2024

    work page 2024

  13. [13]

    Vision transformers for human activity recognition using wifi channel state information,

    F. Luo, S. Khan, B. Jiang, and K. Wu, “Vision transformers for human activity recognition using wifi channel state information,”IEEE Internet Things J., vol. 11, no. 18, pp. 28 111–28 122, 2024

    work page 2024

  14. [14]

    Wifi-enabled vision via spatially-variant pose estimation based on convolutional transformer network,

    H.-J. Lee and S.-J. Buu, “Wifi-enabled vision via spatially-variant pose estimation based on convolutional transformer network,”IEEE Access, vol. 13, pp. 84 855–84 869, 2025

    work page 2025

  15. [15]

    Zero- parameter attention sharing transformer for joint human activity and identity recognition,

    S. Huang, P.-Y . Chen, P. Zhou, K. Li, and J. A. McCann, “Zero- parameter attention sharing transformer for joint human activity and identity recognition,”IEEE Trans. Artif. Intell., vol. 7, no. 2, pp. 960– 972, 2025

    work page 2025

  16. [16]

    Crosssense: Towards cross-site and large-scale WiFi sensing,

    J. Zhang, Z. Tang, M. Li, D. Fang, P. Nurmi, and Z. Wang, “Crosssense: Towards cross-site and large-scale WiFi sensing,” inProc. 24th Annu. Int. Conf. Mobile Comput. Netw. (MobiCom), 2018, pp. 305–320

    work page 2018

  17. [17]

    Commercial Wi-Fi based fall detection with environment influence mitigation,

    L. Zhang, Z. Wang, and L. Yang, “Commercial Wi-Fi based fall detection with environment influence mitigation,” inProc. 16th Annu. IEEE Int. Conf. Sensing, Commun., Netw. (SECON), 2019, pp. 1–9

    work page 2019

  18. [18]

    Device-free cross location activity recognition via semi-supervised deep learning,

    R. Zhou, Z. Gong, K. Tang, B. Zhou, and Y . Cheng, “Device-free cross location activity recognition via semi-supervised deep learning,”Neural Comput. Appl., vol. 34, no. 12, pp. 10 189–10 203, 2022

    work page 2022

  19. [19]

    Cross-domain CSI-based activity recognition via iterative soft labeling and domain alignment,

    Z. Gong, R. Zhou, Y . Cheng, and Y . Luo, “Cross-domain CSI-based activity recognition via iterative soft labeling and domain alignment,” IEEE Syst. J., vol. 17, no. 1, pp. 1599–1609, 2023

    work page 2023

  20. [20]

    Metaganfi: Cross-domain unseen individual identification using wifi signals,

    J. Zhang, Z. Chen, C. Luo, B. Wei, S. S. Kanhere, and J. Li, “Metaganfi: Cross-domain unseen individual identification using wifi signals,”Proc. ACM Interact. Mob. Wearable Ubiquitous Technol., vol. 6, no. 3, pp. 1–21, 2022

    work page 2022

  21. [21]

    Multi-source domain gener- alization for csi-based human activity recognition,

    T. Fan, S. Qiu, W. Gong, and Y . Fang, “Multi-source domain gener- alization for csi-based human activity recognition,”IEEE Trans. Mob. Comput., vol. 24, no. 10, pp. 11 034–11 045, 2025

    work page 2025

  22. [22]

    Wifi csi based temporal activity detection via dual pyramid network,

    F. Zouet al., “Wifi csi based temporal activity detection via dual pyramid network,”arXiv preprint arXiv:2412.16233, 2024

    work page arXiv 2024

  23. [23]

    Wi-SensiNet: Through-wall human activity recognition based on WiFi sensing,

    X. Liet al., “Wi-SensiNet: Through-wall human activity recognition based on WiFi sensing,” inProc. IEEE Int. Conf. Sensing, Commun., Netw. (SECON), 2024

    work page 2024

  24. [24]

    Exploring los/nlos channel effects on wifi sensing performance,

    E. Bulutet al., “Exploring los/nlos channel effects on wifi sensing performance,” inProc. IEEE INFOCOM Workshops, 2025

    work page 2025

  25. [25]

    RaliSense: Extending WiFi respiratory detection range by rapid alignment of dynamic components,

    H. Liuet al., “RaliSense: Extending WiFi respiratory detection range by rapid alignment of dynamic components,”IEEE Trans. Mob. Comput., vol. 24, no. 4, pp. 3421–3435, 2025

    work page 2025

  26. [26]

    Danger-pose detection system using commodity wi-fi for bathroom monitoring,

    K. Nodaet al., “Danger-pose detection system using commodity wi-fi for bathroom monitoring,”Sensors, vol. 19, no. 4, p. 884, 2019

    work page 2019

  27. [27]

    Efficientnet: Rethinking model scaling for con- volutional neural networks,

    M. Tan and Q. Le, “Efficientnet: Rethinking model scaling for con- volutional neural networks,” inProc. Int. Conf. Mach. Learn. (ICML). PMLR, 2019, pp. 6105–6114

    work page 2019

  28. [28]

    Wifi-csi- based fall detection by spectrogram analysis with cnn,

    T. Nakamura, M. Bouazizi, K. Yamamoto, and T. Ohtsuki, “Wifi-csi- based fall detection by spectrogram analysis with cnn,” inProc. IEEE Global Commun. Conf. (GLOBECOM), 2020, pp. 1–6

    work page 2020

  29. [29]

    WiFi-based cross-domain gesture recognition using attention mechanism,

    X. Chenet al., “WiFi-based cross-domain gesture recognition using attention mechanism,”arXiv preprint arXiv:2512.04521, 2025

    work page arXiv 2025

  30. [30]

    XFall: Domain adaptive Wi-Fi-based fall detection with cross-modal supervision,

    G. Chi, G. Zhang, X. Ding, Q. Ma, Z. Yang, and Z. Du, “XFall: Domain adaptive Wi-Fi-based fall detection with cross-modal supervision,”IEEE J. Sel. Areas Commun., vol. 42, no. 9, pp. 2457–2471, 2024

    work page 2024

  31. [31]

    Wi-cbr: Salient- aware adaptive wifi sensing for cross-domain behavior recognition,

    R. Zhang, S. Tang, H. Yan, X. Zhang, and J. Guo, “Wi-cbr: Salient- aware adaptive wifi sensing for cross-domain behavior recognition,” arXiv preprint arXiv:2506.11616, 2025

    work page arXiv 2025

  32. [32]

    FewSense: Towards a scalable and cross-domain Wi- Fi sensing system using few-shot learning,

    Y . Zhanget al., “FewSense: Towards a scalable and cross-domain Wi- Fi sensing system using few-shot learning,”IEEE Trans. Mob. Comput., vol. 22, no. 12, pp. 7052–7065, 2022

    work page 2022

  33. [33]

    DapFall: Dynamic amplitude probability density profile-based Wi-Fi CSI sensing for fall detection,

    Z. Chen, Y . Zhao, G. Chen, Y . Yang, W. Huang, and C.-Z. Xu, “DapFall: Dynamic amplitude probability density profile-based Wi-Fi CSI sensing for fall detection,”IEEE Sensors J., vol. 24, no. 5, pp. 6534–6545, 2024

    work page 2024

  34. [34]

    Deep learning-based fall detection using WIFI channel state informa- tion,

    Y . Chu, K. Cumanan, S. K. Sankarpandi, S. Smith, and O. A. Dobre, “Deep learning-based fall detection using WIFI channel state informa- tion,”IEEE Access, vol. 11, pp. 83 763–83 780, 2023

    work page 2023

  35. [35]

    Attention is all you need,

    A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, Ł. Kaiser, and I. Polosukhin, “Attention is all you need,”Adv. Neural Inf. Process. Syst., vol. 30, 2017

    work page 2017