:ant: Awesome Wi-Fi Sensing Generalization

Fei Wang, Tingting Zhang, Wei Xi, Han Ding, Ge Wang, Di Zhang, Yuanhao Cui, Fan Liu, Jinsong Han, Jie Xu, Tony Xiao Han, A Survey on Wi-Fi Sensing Generalizability: Taxonomy, Techniques, Datasets, and Future Research Prospects, IEEE Communications Surveys & Tutorials, 2026. IEEE, arXiv

Sensing Dataset Platform: https://sdp8.org/

@article{wang2026wifisurvey,
  author    = {Fei Wang and Tingting Zhang and Wei Xi and Han Ding and Ge Wang and Di Zhang and Yuanhao Cui and Fan Liu and Jinsong Han and Jie Xu and Tony Xiao Han},
  title     = {A Survey on Wi-Fi Sensing Generalizability: Taxonomy, Techniques, Datasets, and Future Research Prospects},
  journal   = {IEEE Communications Surveys \& Tutorials},
  year      = {2026}
}

Fig.3: Growth of research in Wi-Fi sensing generalizability: from a handful of studies between year of 2015 and 2018 to a surge of publications since 2019. We employed a linear regression to fit the growth trend. In the figure, the $R^2$ score indicates the correlation between the estimated results and the ground truth, where a value closer to 1 denotes a higher degree of model fitting. (We selected one representative paper per year based on the highest citation count, ensuring each choice employs a technical approach distinct from those featured in previous years.)

Highlight

While this survey focuses on Wi-Fi sensing, it is worth noting that many of the taxonomy and generalization challenges discussed herein are also applicable to millimeter-wave radar–based human sensing, which has gained increasing attention with the availability of commodity radars. Although mmWave radar relies on different signal representations, such as range–Doppler or range–angle maps rather than Wi-Fi CSI, it faces similar fundamental challenges related to device heterogeneity, human body diversity, and environment diversity.

For example, variations in radar hardware configurations and antenna arrays resemble device heterogeneity in Wi-Fi sensing; differences in human body shape, motion patterns, and orientation affect radar reflections in a manner analogous to CSI-based sensing; and environmental factors such as clutter, multipath, and deployment geometry similarly impact sensing robustness.

As a result, learning paradigms and deployment strategies aimed at improving generalization, such as multi-view sensing, large-scale in-the-wild datasets, domain adaptation, and continual learning, can be conceptually extended to millimeter-wave radar sensing, despite differences in signal modality.

:turtle: Tools

:dog: Datasets

Table VII: Wi-Fi Datasets. The main results column mainly denotes the in-domain results from the original publications.

🦙 Latest Dataset Update

You are welcome to contribute datasets via issues or pull requests.

1. Yuan, D., Zhang, X., Hou, W., Lyu, S., Yu, Y., Yu, L. J. T., ... & Wu, C. OctoNet: A Large-Scale Multi-Modal Dataset for Human Activity Understanding Grounded in Motion-Captured 3D Pose Labels. In The Thirty-ninth Annual Conference on Neural Information Processing Systems Datasets and Benchmarks Track. 2025.

2. Cheng, X., Huang, Z., Yu, Y., Bai, L., Sun, M., Han, Z., ... & Li, S. SynthSoM: A synthetic intelligent multi-modal sensing-communication dataset for Synesthesia of Machines (SoM). Scientific Data, 12(1), 819. 2025

🐼 Papers

Stage 1: 🐘 Experimental Setup

Table II: Comparison of Experimental Setup Strategies for Enhanced Generalization.

Distributed Antennas

[63] Y . Qin, S. Sigg, S. Pan, and Z. Li, "Direction-agnostic gesture recog- nition system using commercial wifi devices,"Computer Communications, vol. 216, pp. 34--44, 2024. paper

[64] Y . Qin, S. Pan, and Z. Li, "Cross-domain extendable gesture recognition system using wifi signals,"Electronics Letters, vol. 59, no. 16, p. e12931, 2023. paper

[65] F. Wang, W. Gong, and J. Liu, "On spatial diversity in wifi-based human activity recognition: A deep learning-based approach,"IoT-J, vol. 6, no. 2, pp. 2035--2047, 2018. paper

Distributed Devices

[66] N. Bahadori, J. Ashdown, and F. Restuccia, "Rewis: Reliable wi-fi sensing through few-shot multi-antenna multi-receiver csi learning," in WoWMoM. IEEE, 2022, pp. 50--59. paper

[67] Y . Zheng, Y . Zhang, K. Qian, G. Zhang, Y . Liu, C. Wu, and Z. Yang, "Zero-effort cross-domain gesture recognition with wi-fi," inMobiSys, 2019, pp. 313--325. paper

[68] R. Xiao, J. Liu, J. Han, and K. Ren, "Onefi: One-shot recognition for unseen gesture via cots wifi," in SenSys, 2021, pp. 206--219. paper

[69] D. Wu, Y . Zeng, R. Gao, S. Li, Y . Li, R. C. Shah, H. Lu, and D. Zhang, "Witraj: Robust indoor motion tracking with wifi signals,"IEEE TMC, vol. 22, no. 5, pp. 3062--3078, 2021. paper

Scaling Up Dataset

[70] F. Wang, Y . Lv, M. Zhu, H. Ding, and J. Han, "Xrf55: A radio frequency dataset for human indoor action analysis,"IMWUT, vol. 8, no. 1, pp. 1--34, 2024. paper

[67] Y . Zheng, Y . Zhang, K. Qian, G. Zhang, Y . Liu, C. Wu, and Z. Yang, "Zero-effort cross-domain gesture recognition with wi-fi," in MobiSys, 2019, pp. 313--325. paper

[71] Y . Zhang, Y . Yin, Y . Wang, J. Ai, and D. Wu, "Csi-based location- independent human activity recognition with parallel convolutional networks,"Computer Communications, vol. 197, pp. 87--95, 2023. paper

[72] J. Yang, H. Huang, Y . Zhou, X. Chen, Y . Xu, S. Yuan, H. Zou, C. X. Lu, and L. Xie, "Mm-fi: Multi-modal non-intrusive 4d human dataset for versatile wireless sensing,"NeurIPS, vol. 36, 2024. paper

[73] G. Zhu, Y . Hu, W. Gao, W.-H. Wang, B. Wang, and K. Liu, "Csi- bench: A large-scale in-the-wild dataset for multi-task wifi sensing," arXiv preprint arXiv:2505.21866, 2025. paper

Key Takeaways and Lessons Learned: Experimental Setup

• Structural Invariance through Spatial Diversity:
  Distributed antenna topologies (e.g., staggered or circular layouts) and multi-device configurations (e.g., 1-TX with 4–6 RXs) provide a physical-layer solution to orientation sensitivity. This “structural invariance” ensures multi-view signal observations that effectively mitigate body-shadowing and occlusion.

• Placement as a Robustness Multiplier:
  Optimizing the geometric placement of transmitters and receivers is critical for capturing spatially heterogeneous channel responses. Strategic layout design not only maximizes the sensing range but also significantly reduces the impact of environmental multipath interference.

• Natural Invariance through Data Scaling:
  A key lesson from large-scale datasets (e.g., XRF55, CSI-Bench) is the existence of a quantitative threshold where models begin to exhibit “natural” invariance. Massive data exposure allows models to inherently decouple human motion from specific propagation conditions, facilitating easier adaptation to novel domains.

• Exploiting Diversity to Counter Heterogeneity:
  To overcome device heterogeneity, moving from single-link to multi-link coordinated sensing is essential. Distributed setups exploit spatial and hardware diversity to average out the unique radio artifacts and hardware imperfections of individual commercial Wi-Fi chipsets.

Stage 2: :snail: Signal Preprocessing

Table III: Comparison of Signal Preprocessing Techniques for Wi-Fi Sensing Generalization.

Signal Analysis Methods

[74] Y . Lu, S. Lv, and X. Wang, "Towards location independent gesture recognition with commodity wifi devices,"Electronics, vol. 8, no. 10, p. 1069, 2019. paper

[75] R. Zhou, Z. Gong, K. Tang, B. Zhou, and Y . Cheng, "Device-free cross location activity recognition via semi-supervised deep learning,"Neural Computing and Applications, vol. 34, no. 12, pp. 10 189--10 203, 2022. paper

[76] J. Yang, Y . Liu, Y . Wu, P. Yang, and Z. Liu, "Gait-enhance: Robust gait recognition of complex walking patterns based on wifi csi," inSmart World Congress. IEEE, 2023, pp. 1--9. paper

[77] H. Zhuo, X. Wu, Q. Zhong, and H. Zhang, "Position-free breath detection during sleep via commodity wifi,"IEEE Sensors Journal, 2023. paper

[78] F. Abuhoureyah, S. K. Swee, and W. Y . Chiew, "Multi-user human activity recognition through adaptive location-independent wifi signal characteristics,"IEEE Access, 2024. paper

Advanced Handcrafted Indicators

[79] W. Meng, X. Chen, W. Cui, and J. Guo, "Wihgr: A robust wifi-based human gesture recognition system via sparse recovery and modified attention-based bgru,"IoT-J, vol. 9, no. 12, pp. 10 272--10 282, 2021. paper

[80] J. Huang, B. Liu, C. Miao, Y . Lu, Q. Zheng, Y . Wu, J. Liu, L. Su, and C. W. Chen, "Phaseanti: An anti-interference wifi-based activity recognition system using interference-independent phase component," IEEE TMC, vol. 22, no. 5, pp. 2938--2954, 2021. paper

[81] R. Gao, W. Li, Y . Xie, E. Yi, L. Wang, D. Wu, and D. Zhang, "Towards robust gesture recognition by characterizing the sensing quality of wifi signals,"IMWUT, vol. 6, no. 1, pp. 1--26, 2022. paper

[82] J. Zhang, Y . Li, H. Xiong, D. Dou, C. Miao, and D. Zhang, "Handgest: Hierarchical sensing for robust-in-the-air handwriting recognition with commodity wifi devices,"IoT-J, vol. 9, no. 19, pp. 19 529--19 544, 2022. paper

[83] X. Yu, T. Jiang, X. Ding, Z. Yao, X. Zhou, and Y . Zhong, "Towards position-independent gesture recognition based on wifi by subcarrier selection and gesture code," inWCNC. IEEE, 2023, pp. 1--6. paper

[84] Y . Zhang, Q. Liu, Y . Wang, and G. Yu, "Csi-based location-independent human activity recognition using feature fusion,"IEEE TIM, vol. 71, pp. 1--12, 2022. paper

[85] C. Peng, L. Gui, B. Sheng, Z. Guo, and F. Xiao, "Rosefi: A robust sedentary behavior monitoring system with commodity wifi devices," IEEE TMC, 2023. paper

[86] X. Wang, A. Yu, K. Niu, W. Shi, J. Wang, Z. Yao, R. C. Shah, H. Lu, and D. Zhang, "Understanding the diffraction model in static multipath- rich environments for wifi sensing system design,"IEEE TMC, 2024. paper

[87] X. Ding, X. Yu, Y . Zhong, W. Xie, B. Cai, M. You, and T. Jiang, "Robust gesture recognition method toward intelligent environment using wi-fi signals,"Measurement, vol. 231, p. 114525, 2024. paper

[88] M. Peng, X. Fu, H. Zhao, Y . Wang, and C. Kai, "Likey: Location- independent keystroke recognition on numeric keypads using wifi signal,"Computer Networks, vol. 245, p. 110354, 2024. paper

[89] Q. Bu, G. Yang, J. Feng, and X. Ming, "Wi-fi based gesture recognition using deep transfer learning," inSmartWorld, Ubiquitous Intelligence & Computing, Advanced & Trusted Computing, Scalable Computing & Communications, Cloud & Big Data Computing, Internet of People and Smart City Innovation. IEEE, 2018, pp. 590--595. paper

[90] C. Zhang, W. Jiao, and W. Du, "Enhancing human activity recognition performance in small-sample wi-fi sensing using data augmentation methods," inICCT. IEEE, 2023, pp. 473--478. paper

Motion Indicators

[91] F. Meneghello, D. Garlisi, N. Dal Fabbro, I. Tinnirello, and M. Rossi, "Sharp: Environment and person independent activity recognition with commodity ieee 802.11 access points,"IEEE TMC, vol. 22, no. 10, pp. 6160--6175, 2022. paper

[92] L. Zhao, R. Xiao, J. Liu, and J. Han, "One is enough: Enabling one-shot device-free gesture recognition with cots wifi," inINFOCOM. IEEE, 2024, pp. 1231--1240. paper

[93] Y . Yin, Z. Zhang, X. Yang, F. Yan, and Q. Niu, "Towards fully domain- independent gesture recognition using cots wifi device,"Electronics Letters, vol. 57, no. 5, pp. 232--234, 2021. paper

[94] C. Feng, N. Wang, Y . Jiang, X. Zheng, K. Li, Z. Wang, and X. Chen, "Wi-learner: Towards one-shot learning for cross-domain wi-fi based gesture recognition,"IMWUT, vol. 6, no. 3, pp. 1--27, 2022. paper

[95] K. Niu, F. Zhang, X. Wang, Q. Lv, H. Luo, and D. Zhang, "Under- standing wifi signal frequency features for position-independent gesture sensing,"IEEE TMC, vol. 21, no. 11, pp. 4156--4171, 2021. paper

[96] J. Chen, S. Bi, X.-H. Lin, and Z. Quan, "Lager: Label-free domain- adaptive wireless gesture recognition via latent feature alignment and augmentation,"IoT-J, vol. 11, no. 23, pp. 37 928--37 941, 2024. paper

[97] S. Li, Z. Liu, Q. Lv, Y . Zou, Y . Zhang, and D. Zhang, "Wilife: Long-term daily status monitoring and habit mining of the elderly leveraging ubiquitous wi-fi signals,"ACM Transactions on Computing for Healthcare. paper

[98] B. van Berlo, C. Oerlemans, F. L. Marogna, T. Ozcelebi, and N. Merat- nia, "Mini-batch alignment: A deep-learning model for domain factor- independent feature extraction for wi-fi--csi data,"Sensors, vol. 23, no. 23, p. 9534, 2023. paper

[99] B. van Berlo, T. Ozcelebi, and N. Meratnia, "Insights on mini- batch alignment for wifi-csi data domain factor independent feature extraction," inInternational Conference on Pervasive Computing and Communications Workshops and other Affiliated Event. IEEE, 2022, pp. 527--532. paper

[69] D. Wu, Y . Zeng, R. Gao, S. Li, Y . Li, R. C. Shah, H. Lu, and D. Zhang, "Witraj: Robust indoor motion tracking with wifi signals,"IEEE TMC, vol. 22, no. 5, pp. 3062--3078, 2021. paper

[67] Y . Zheng, Y . Zhang, K. Qian, G. Zhang, Y . Liu, C. Wu, and Z. Yang, "Zero-effort cross-domain gesture recognition with wi-fi," inMobiSys, 2019, pp. 313--325. paper

[100] B.-B. Zhang, D. Zhang, Y . Hu, and Y . Chen, "Unsupervised domain adaptation for wifi gesture recognition," inWCNC. IEEE, 2023, pp. 1--6. paper

[101] M. Yang, H. Zhu, R. Zhu, F. Wu, L. Yin, and Y . Yang, "Witransformer: A novel robust gesture recognition sensing model with wifi,"Sensors, vol. 23, no. 5, p. 2612, 2023. paper

[102] W. Shi, M. Duan, H. He, L. Lin, C. Yang, C. Li, and J. Zhao, "Location adaptive motion recognition based on wi-fi feature enhancement," Applied Sciences, vol. 13, no. 3, p. 1320, 2023. paper

[103] I. Bulugu, "Gesture recognition system based on cross-domain csi extracted from wi-fi devices combined with the 3d cnn,"Signal, Image and Video Processing, vol. 17, no. 6, pp. 3201--3209, 2023. paper

[30] Y . Zhang, Y . Zheng, G. Zhang, K. Qian, C. Qian, and Z. Yang, "Gaitid: Robust wi-fi based gait recognition," inWASA. Springer, 2020, pp. 730--742. paper

[68] R. Xiao, J. Liu, J. Han, and K. Ren, "Onefi: One-shot recognition for unseen gesture via cots wifi," inSenSys, 2021, pp. 206--219. paper

[104] G. Chi, G. Zhang, X. Ding, Q. Ma, Z. Yang, Z. Du, H. Xiao, and Z. Liu, "Xfall: Domain adaptive wi-fi-based fall detection with cross- modal supervision,"IEEE JASAC, 2024. paper

[105] R. Gao, M. Zhang, J. Zhang, Y . Li, E. Yi, D. Wu, L. Wang, and D. Zhang, "Towards position-independent sensing for gesture recogni- tion with wi-fi,"IMWUT, vol. 5, no. 2, pp. 1--28, 2021. paper

[106] C. Shi, J. Liu, N. Borodinov, B. Leao, and Y . Chen, "Towards environment-independent behavior-based user authentication using wifi," inMASS. IEEE, 2020, pp. 666--674. paper

[107] Y . Wang, Y . Tian, and R. Peng, "Position and orientation independent wireless gesture recognition," inWCSP. IEEE, 2022, pp. 466--471. paper

[108] D. Wu, D. Zhang, C. Xu, Y . Wang, and H. Wang, "Widir: Walking direction estimation using wireless signals," inUBICOMP, 2016, pp. 351--362. paper

[109] H. Wang, D. Zhang, J. Ma, Y . Wang, Y . Wang, D. Wu, T. Gu, and B. Xie, "Human respiration detection with commodity wifi devices: Do user location and body orientation matter?" inUBICOMP, 2016, pp. 25--36. paper

[63] Y . Qin, S. Sigg, S. Pan, and Z. Li, "Direction-agnostic gesture recog- nition system using commercial wifi devices,"Computer Communica- tions, vol. 216, pp. 34--44, 2024. paper

[110] L. Zhang, C. Wang, M. Ma, and D. Zhang, "Widigr: Direction- independent gait recognition system using commercial wi-fi devices," IoT-J, vol. 7, no. 2, pp. 1178--1191, 2019. paper

[111] L. Zhang, C. Wang, and D. Zhang, "Wi-pigr: Path independent gait recognition with commodity wi-fi,"IEEE TMC, vol. 21, no. 9, pp. 3414--3427, 2021. paper

[112] F. Zhang, C. Wu, B. Wang, H.-Q. Lai, Y . Han, and K. R. Liu, "Widetect: Robust motion detection with a statistical electromagnetic model," IMWUT, vol. 3, no. 3, pp. 1--24, 2019. paper

[113] G. Zhu, B. Wang, W. Gao, Y . Hu, C. Wu, and K. R. Liu, "Wifi-based robust human and non-human motion recognition with deep learning," inPerCom Workshops. IEEE, 2024, pp. 769--774. paper

Angle Indicators

[114] Y . Li, T. Jiang, X. Ding, and Y . Wang, "Location-free csi based activity recognition with angle difference of arrival," inWCNC. IEEE, 2020, pp. 1--6. paper

[115] Z. Han, Z. Lu, Z. Hu, Y . Chen, and X. Wen, "On position-independency passive gesture tracking with commodity wi-fi,"IEEE Sensors Journal, vol. 23, no. 14, pp. 16 264--16 275, 2023. paper

[116] Y . Ren and J. Yang, "Robust person identification: A wifi vision-based approach,"arXiv preprint arXiv:2210.00127, 2022. paper

Key Takeaways and Lessons Learned: Signal Preprocessing

• Signal Analysis for Denoising:
  Algorithms like LRSD, SSA, and PCA are fundamental for extracting action-related components from background noise. These methods enhance resilience to location variations and enable robust monitoring (e.g., breath detection) even in non-ideal positions.

• Handcrafted IQ & Phase Features:
  Advanced indicators derived from phase differences or the IQ complex plane (e.g., SPD, rotational features) capture fine-grained patterns. Notably, features like NLPEV provide independence from CCI, which is crucial for real-world robustness.

• Physical Invariance via Motion Indicators:
  Transitioning to speed/velocity profiles, particularly the BVP and the Fresnel Zone Model, represents a major shift toward domain invariance. By mapping signals to speed or body-relative coordinates, systems decouple sensing from the user's orientation and environment.

• Geometric Consistency with Angle Indicators:
  Angle-based measurements, such as AoA and ADoA, provide spatial geometric paths that remain consistent across domains. These allow for real-time trajectory reconstruction and 3D body representation, enabling complex tasks like person re-identification.

• Signal-to-Visual Feature Enrichment:
  Converting 1D CSI time-series into 2D representations (e.g., Recurrence Plots, Grayscale images) bridges the gap between signal processing and computer vision, allowing models to benefit from powerful pre-trained visual architectures and data augmentation.

Stage 3: 🦊 Feature Learning

Table IV: Comparison of Pros, Cons, and Best-fit Scenarios of Feature Learning Techniques in Wi-Fi Sensing.

Domain Alignment

Domain Alignment with Domain Discriminator

Fig 7: Domain alignment with domain discriminator or domain classifier. Both source and target data are mapped into a shared feature space. Through adversarial competition between an action classifier and a domain discriminator, the network learns to produce domain-agnostic representations that enhance cross-domain performance.

[117] Z. Wang, S. Chen, W. Yang, and Y . Xu, "Environment-independent wi-fi human activity recognition with adversarial network," inICASSP. IEEE, 2021, pp. 3330--3334. paper

[118] L. Sheng, Y . Chen, S. Ning, S. Wang, B. Lian, and Z. Wei, "Da- har: Dual adversarial network for environment-independent wifi human activity recognition,"Pervasive and Mobile Computing, vol. 96, p. 101850, 2023. paper

[119] L. Zhang, W. Cui, B. Li, Z. Chen, M. Wu, and T. S. Gee, "Privacy- preserving cross-environment human activity recognition,"IEEE Trans- actions on Cybernetics, vol. 53, no. 3, pp. 1765--1775, 2021. paper

[120] Z. Zhou, F. Wang, and W. Gong, "i-sample: Augment domain adver- sarial adaptation models for wifi-based har,"ACM ToSN, vol. 20, no. 2, pp. 1--20, 2024. paper

[121] H. Zou, J. Yang, Y . Zhou, and C. J. Spanos, "Joint adversarial domain adaptation for resilient wifi-enabled device-free gesture recognition," inICMLA. IEEE, 2018, pp. 202--207. paper

[122] X. Li, L. Chang, F. Song, J. Wang, X. Chen, Z. Tang, and Z. Wang, "Crossgr: Accurate and low-cost cross-target gesture recognition using wi-fi,"IMWUT, vol. 5, no. 1, pp. 1--23, 2021. paper

[123] H. Zhang, Z. Zhou, and W. Gong, "Wi-adaptor: Fine-grained domain adaptation in wifi-based activity recognition," inGLOBECOM. IEEE, 2021, pp. 1--6. paper

Domain Alignment with Domain Classifier

[124] W. Jiang, C. Miao, F. Ma, S. Yao, Y . Wang, Y . Yuan, H. Xue, C. Song, X. Ma, D. Koutsonikolas, W. Xu, and L. Su, "Towards environment independent device free human activity recognition," inMobiCom, 2018, pp. 289--304. paper

[125] A. Khattak and A. Khan, "Cross-location activity recognition using adversarial learning," inSOICT, 2022, pp. 59--65. paper

[126] F. Wang, J. Liu, and W. Gong, "Wicar: Wifi-based in-car activity recog- nition with multi-adversarial domain adaptation," inIWQoS, 2019, pp. 1--10. paper

[127] X. Zhang, Y . Feng, J. Huang, H. Yan, P. Zhao, J. Liu, T. Liu, M. Li, Z. Liu, and B. Liu, "Objective gesture recognition based on wifi," Authorea Preprints, 2024. paper

[128] A. Zinys, B. van Berlo, and N. Meratnia, "A domain-independent generative adversarial network for activity recognition using wifi csi data,"Sensors, vol. 21, no. 23, p. 7852, 2021. paper

[129] J. Strohmayer, R. Sterzinger, M. W ¨odlinger, and M. Kampel, "Datta: Domain-adversarial test-time adaptation for cross-domain wifi-based human activity recognition,"arXiv preprint arXiv:2411.13284, 2024. paper

[130] S. Liu, Z. Chen, M. Wu, C. Liu, and L. Chen, "Wisr: Wireless domain generalization based on style randomization,"IEEE TMC, 2023. paper

[93] Y . Yin, Z. Zhang, X. Yang, F. Yan, and Q. Niu, "Towards fully domain- independent gesture recognition using cots wifi device,"Electronics Letters, vol. 57, no. 5, pp. 232--234, 2021. paper

[131] W. Yang, X. Jiawei, W. Ao, X. Huijuan, Z. Chuanxin, and J. Yimu, "Domain-generalization human activity recognition model based on csi instance normalization."Journal on Communication/Tongxin Xuebao, vol. 45, no. 6, 2024. paper

[64] Y . Qin, S. Pan, and Z. Li, "Cross-domain extendable gesture recognition system using wifi signals,"Electronics Letters, vol. 59, no. 16, p. e12931, 2023. paper

[132] B. v. Berlo, R. Verhoeven, and N. Meratnia, "Use of domain labels dur- ing pre-training for domain-independent wifi-csi gesture recognition," Sensors, vol. 23, no. 22, p. 9233, 2023. paper

[133] S. Zhou, L. Guo, Z. Lu, X. Wen, W. Zheng, and Y . Wang, "Subject- independent human pose image construction with commodity wi-fi," inIEEE ICC, 2021, pp. 1--6. paper

[134] D. Yan, P. Yang, F. Shang, F. Han, Y . Yan, and X.-Y . Li, "freegait: Liberalizing wireless-based gait recognition to mitigate non-gait human behaviors," inMobiHoc, 2024, pp. 241--250. paper

[135] H. Li, X. Chen, J. Wang, D. Wu, and X. Liu, "Dafi: Wifi-based device- free indoor localization via domain adaptation,"IMWUT, vol. 5, no. 4, pp. 1--21, 2021. paper

[106] C. Shi, J. Liu, N. Borodinov, B. Leao, and Y . Chen, "Towards environment-independent behavior-based user authentication using wifi," inMASS. IEEE, 2020, pp. 666--674. paper

[136] X. Chen, H. Li, C. Zhou, X. Liu, D. Wu, and G. Dudek, "Fidora: Robust wifi-based indoor localization via unsupervised domain adaptation," IoT-J, vol. 9, no. 12, pp. 9872--9888, 2022. paper

Domain Alignment with Similarity Computing

Fig 8: Maximum Mean Discrepancy (MMD) quantifies the similarity between two probability distributions by projecting data into a Reproducing Kernel Hilbert Space (RKHS) and computing the squared distance between their mean embeddings, encouraging the model to learn domain-invariant representations.

[138] Y . Zhou, J. Yang, H. Huang, and L. Xie, "Adapose: Towards cross-site device-free human pose estimation with commodity wifi,"IoT-J, 2024. paper

[139] X. Rao, L. Qin, Y . Yi, J. Liu, G. Lei, and Y . Cao, "A novel adaptive device-free passive indoor fingerprinting localization under dynamic environment,"IEEE TNSM, 2024. paper

[140] H. Kang, Q. Zhang, and Q. Huang, "Context-aware wireless-based cross-domain gesture recognition,"IoT-J, vol. 8, no. 17, pp. 13 503-- 13 515, 2021. paper

[141] Y . Liang, W. Wu, H. Li, F. Han, Z. Liu, P. Xu, X. Lian, and X. Chen, "Wiai-id: Wi-fi-based domain adaptation for appearance-independent passive person identification,"IoT-J, vol. 11, no. 1, pp. 1012--1027, 2023. paper

[142] S. Zhang, T. Jiang, X. Ding, X. Zhou, and Y . Zhong, "Device-free cross-environment human action recognition using wi-fi signals," in International Conference on Artificial Intelligence in China. Springer, 2023, pp. 141--151. paper

[143] S. Mehryar, "A domain adaptation framework for human activity monitoring using passive wi-fi sensing," inICCCMLA. IEEE, 2023, pp. 263--268. paper

[144] R. Zhou, H. Hou, Z. Gong, Z. Chen, K. Tang, and B. Zhou, "Adaptive device-free localization in dynamic environments through adaptive neural networks,"IEEE Sensors Journal, vol. 21, no. 1, pp. 548--559, 2020. paper

[145] Z. Xiao, S. Zhou, X. Wen, S. Ling, and X. Yang, "Pattern-independent human gait identification with commodity wifi," inWCNC. IEEE, 2024, pp. 1--6. paper

[147] W. Cui, K. Wu, M. Wu, X. Li, and Z. Chen, "Wicau: Comprehen- sive partial adaptation with uncertainty-aware for wifi-based cross- environment activity recognition,"IEEE TIM, vol. 73, pp. 1--10, 2024. paper

[148] W. Jiao, C. Zhang, W. Du, and S. Ma, "Wisda: Subdomain adaptation human activity recognition method using wi-fi signals,"IEEE TMC, 2024. paper

[149] K. Gong, Y . Gao, and W. Dong, "Privacy-preserving and cross- domain human sensing by federated domain adaptation with semantic knowledge correction,"IMWUT, vol. 8, no. 1, pp. 1--26, 2024. paper

[150] B. Sheng, R. Han, H. Cai, F. Xiao, L. Gui, and Z. Guo, "Cdfi: Cross- domain action recognition using wifi signals,"IEEE TMC, 2024. paper

[96] J. Chen, S. Bi, X.-H. Lin, and Z. Quan, "Lager: Label-free domain- adaptive wireless gesture recognition via latent feature alignment and augmentation,"IoT-J, vol. 11, no. 23, pp. 37 928--37 941, 2024. paper

[151] X. Zhan and Z. Wu, "Indoor positioning based on channel state information and deep learning domain adaptation," inICCAID, vol. 13105. SPIE, 2024, pp. 918--930. paper

[152] D. Wang, J. Yang, W. Cui, L. Xie, and S. Sun, "Airfi: empowering wifi-based passive human gesture recognition to unseen environment via domain generalization,"IEEE TMC, vol. 23, no. 2, pp. 1156--1168, 2022. paper

[153] X. Ding, T. Jiang, Y . Li, W. Xue, and Y . Zhong, "Device-free location- independent human activity recognition using transfer learning based on cnn," inICC Workshops. IEEE, 2020, pp. 1--6. paper

Domain Alignment with Generative Adversarial Networks

Fig 9: Generative Adversarial Networks. Generator (G): The network responsible for creating synthetic data. Its primary objective is to learn the underlying distribution of the real data and produce samples that are indistinguishable from the ground truth to fool the discriminator. Discriminator (D): The network acting as a binary classifier. Its role is to evaluate both real data (x) and synthetic data (G(z)), outputting a probability that a given sample is real (from the actual dataset) rather than fake (produced by the generator). It provides the gradient signal that guides the generator to improve its realism.

[154] J. Zhang, Z. Chen, C. Luo, B. Wei, S. S. Kanhere, and J. Li, "Metaganfi: Cross-domain unseen individual identification using wifi signals,"IMWUT, vol. 6, no. 3, pp. 1--21, 2022. paper

[155] H. Zhang, X. Chen, and S. Chen, "Cross-domain wi-fi sign language recognition with gans," inICCBN, 2022, pp. 60--65. paper

[156] Y . Mao, Z. Guo, B. Sheng, L. Gui, and F. Xiao, "Wi-cro: Wifi-based cross domain activity recognition via modified gan,"IEEE TVT, 2024. paper

[157] C. Xiao, Y . Lei, C. Liu, and J. Wu, "Mean teacher-based cross-domain activity recognition using wifi signals,"IoT-J, vol. 10, no. 14, pp. 12 787--12 797, 2023. paper

Domain Alignment with Multi-task Learning

[158] J. Zhang, Z. Tang, M. Li, D. Fang, P. Nurmi, and Z. Wang, "Crosssense: Towards cross-site and large-scale wifi sensing," inMobiCom, 2018, pp. 305--320. paper

[159] Y . Sugimoto, H. Rizk, A. Uchiyama, and H. Yamaguchi, "Towards environment-independent activity recognition using wi-fi csi with an encoder-decoder network," inBodySys, 2023, pp. 13--18. paper

[160] Y . Zhang, A. Cheng, B. Chen, Y . Wang, and L. Jia, "A location- independent human activity recognition method based on csi: System, architecture, implementation,"IEEE TMC, 2023. paper

Domain Alignment with Cross-modal Embedding

[161] M. T. Islam and S. Nirjon, "Wi-fringe: Leveraging text semantics in wifi csi-based device-free named gesture recognition," inDCOSS. IEEE, 2020, pp. 35--42. paper

[70] F. Wang, Y . Lv, M. Zhu, H. Ding, and J. Han, "Xrf55: A radio frequency dataset for human indoor action analysis,"IMWUT, vol. 8, no. 1, pp. 1--34, 2024. paper

[162] H. Zhang, Y . Ren, H. Yuan, J. Zhang, and Y . Shen, "Wi-chat: Large lan- guage model powered wi-fi sensing,"arXiv preprint arXiv:2502.12421, 2025. paper

[163] I. Nirmal, A. Khamis, M. Hassan, W. Hu, R. Li, and A. Kalyanaraman, "Wifi2radar: Orientation-independent single-receiver wifi sensing via wifi to radar translation,"IoT-J, vol. 11, no. 9, pp. 15 750--15 766, 2024. paper

Component Disentangle

[164] I. Elujide, C. Feng, A. Shiran, J. Li, and Y . Liu, "Location indepen- dent gesture recognition using channel state information," inAnnual Consumer Communications & Networking Conference. IEEE, 2022, pp. 841--846. paper

[165] Y .-J. Chen, W. Chen, S. Q. Zhang, H.-Y . Huang, and H. Kung, "A task- oriented deep learning approach for human localization,"IEEE TCDS, 2024. paper

[45] F. Wang, S. Zhou, S. Panev, J. Han, and D. Huang, "Person-in-wifi: Fine-grained person perception using wifi," inICCV, 2019, pp. 5452-- 5461. paper

[166] Z. Hao, J. Niu, X.-c. Dang, and Z. Qiao, "Wi-piga: A personnel- independentmethod for actions recognition based on wi-fi," inBigCom. IEEE, 2021, pp. 52--59. paper

[167] B. Wu, T. Jiang, J. Yu, X. Ding, S. Wu, and Y . Zhong, "Device-free human activity recognition with identity-based transfer mechanism," in WCNC. IEEE, 2021, pp. 1--6. paper

Metric Learning for Zero/Few-shot Learning

Metric Learning with Triplet Loss

Fig 10: Triplet loss aims to structure the feature space such that samples from the same class are mapped close together ($x_1^+, x_2^+$), while samples from different classes are pushed farther apart ($x^+, x^-$).

[167] B. Wu, T. Jiang, J. Yu, X. Ding, S. Wu, and Y . Zhong, "Device-free human activity recognition with identity-based transfer mechanism," in WCNC. IEEE, 2021, pp. 1--6. paper

[168] Q. Zhou, Q. Yang, and J. Xing, "Enabling efficient wifi-based occupant behavior recognition using insufficient samples,"Building and Environ- ment, vol. 212, p. 108806, 2022. paper

[169] Z. Wei, W. Chen, W. Tao, S. Ning, B. Lian, X. Sun, and J. Zhao, "Catfsid: A few-shot human identification system based on cross- domain adversarial training,"Computer Communications, vol. 224, pp. 275--284, 2024. paper

[170] Q. Bu, X. Ming, J. Hu, T. Zhang, J. Feng, and J. Zhang, "Transfersense: towards environment independent and one-shot wifi sensing,"Personal and Ubiquitous Computing, pp. 1--19, 2022. paper

[171] Y . Liu, A. Yu, L. Wang, B. Guo, Y . Li, E. Yi, and D. Zhang, "Unifi: A unified framework for generalizable gesture recognition with wi-fi signals using consistency-guided multi-view networks,"IMWUT, vol. 7, no. 4, pp. 1--29, 2024. paper

[172] X. Zhang, J. Huang, H. Yan, P. Zhao, G. Zhuang, Z. Liu, and B. Liu, "Wiopen: A robust wi-fi-based open-set gesture recognition framework,"arXiv preprint arXiv:2402.00822, 2024. paper

[174] C. Xiao, S. Chen, F. Zhou, and J. Wu, "Self-supervised few-shot time-series segmentation for activity recognition,"IEEE TMC, vol. 22, no. 11, pp. 6770--6783, 2022. paper

Learning with Contrastive Learning

[175] K. Xu, J. Wang, L. Zhang, H. Zhu, and D. Zheng, "Dual-stream contrastive learning for channel state information based human activity recognition,"JBHI, vol. 27, no. 1, pp. 329--338, 2022. paper

[176] Y . Wang, G. Yu, Y . Zhang, D. Liu, and Y . Zhang, "Csi-based location- independent human activity recognition by contrast between dual stream fusion features,"IEEE Sensors Journal, 2024. paper

[177] C. Xiao, Y . Han, W. Yang, Y . Hou, F. Shi, and K. Chetty, "Diffusion model-based contrastive learning for human activity recognition,"IoT- J, 2024. paper

Metric Learning with Siamese Neural Networks

Fig 11: Siamese Neural Networks. (1) The network takes a pair of inputs (e.g., two CSI segments) and feeds them into two identical sub-networks. Both branches share the same parameters (weights), mapping the inputs into a common low-dimensional feature space. (2) The model computes a similarity metric (usually Euclidean distance or Cosine similarity) between the two generated feature vectors. (3) Based on the distance, the system determines whether the two inputs belong to the same class (small distance) or different classes (large distance), often using a contrastive loss function.

[116] Y . Ren and J. Yang, "Robust person identification: A wifi vision-based approach,"arXiv preprint arXiv:2210.00127, 2022. paper

[178] Y . Liang, H. Li, W. Wu, and P. Xu, "Map-sgan: Multi-anchor point siamese gan for wi-fi csi-based cross-domain gait recognition,"ESWA, vol. 251, p. 124083, 2024. paper

[179] Z. Zhao, T. Chen, Z. Cai, X. Li, H. Li, Q. Chen, and G. Zhu, "Crossfi: A cross domain wi-fi sensing framework based on siamese network," arXiv preprint arXiv:2408.10919, 2024. paper

Metric Learning with Relation Network

Fig 12: Relation Network. (1) Both the support set and the query sample are passed through an embedding backbone to extract their respective feature maps. (2)The query feature map is paired and concatenated with each feature map in the support set, creating combined feature representations. (3) These concatenated pairs are fed into a Relation Module (usually a non-linear CNN or MLP), which acts as a learnable metric to calculate a relation score (between 0 and 1) for each pair. (4) The query is assigned to the class with the highest relation score, representing the strongest similarity determined by the deep neural network rather than a fixed distance metric.

[180] J. Wang, Q. Gao, X. Ma, Y . Zhao, and Y . Fang, "Learning to sense: Deep learning for wireless sensing with less training efforts,"IEEE Wireless Communications, vol. 27, no. 3, pp. 156--162, 2020. paper

[181] P. Hu, C. Tang, K. Yin, and X. Zhang, "Wigr: a practical wi-fi- based gesture recognition system with a lightweight few-shot network," Applied Sciences, vol. 11, no. 8, p. 3329, 2021. paper

[182] L. Zhang, S. Wu, T. Zhang, and Q. Zhang, "Learning to locate: Adaptive fingerprint-based localization with few-shot relation learning in dynamic indoor environments,"IEEE TWC, vol. 22, no. 8, pp. 5253-- 5264, 2023. paper

[183] R. Zhang, C. Jiang, S. Wu, Q. Zhou, X. Jing, and J. Mu, "Wi-fi sensing for joint gesture recognition and human identification from few samples in human-computer interaction,"IEEE JASAC, vol. 40, no. 7, pp. 2193-- 2205, 2022. paper

Metric Learning with Matching Network

Fig 13: Matching Network. (1) Both the small support set and the query sample are mapped into a shared low-dimensional feature space using a neural network. (2) The model computes the cosine similarity (or distance metric) between the query embedding and each embedding in the support set. (3) These similarity scores are processed through a softmax function to generate attention weights, indicating which support examples most closely resemble the query. (4) The final label is predicted by a weighted sum of the support set labels, allowing the model to recognize new classes without any parameter updates.

[184] Z. Shi, Q. Cheng, J. A. Zhang, and R. Y . Da Xu, "Environment-robust wifi-based human activity recognition using enhanced csi and deep learning,"IoT-J, vol. 9, no. 24, pp. 24 643--24 654, 2022. paper

[185] Z. Shi, J. A. Zhang, R. Xu, Q. Cheng, and A. Pearce, "Towards environment-independent human activity recognition using deep learn- ing and enhanced csi," inIEEE GLOBECOM. IEEE, 2020, pp. 1--6. paper

[186] Z. Shi, J. A. Zhang, R. Y . Xu, and Q. Cheng, "Environment-robust device-free human activity recognition with channel-state-information enhancement and one-shot learning,"IEEE TMC, vol. 21, no. 2, pp. 540--554, 2020. paper

[188] X. Ding, T. Jiang, Y . Zhong, Y . Huang, and Z. Li, "Wi-fi-based location-independent human activity recognition via meta learning," Sensors, vol. 21, no. 8, p. 2654, 2021. paper

Metric Learning with Prototypical Network

Fig 14: Prototypical Network. (1) For each class in the support set, the model computes a prototype (mean vector) by averaging the feature embeddings of all samples belonging to that class. (2) The query sample is mapped into the same feature space using the shared embedding backbone. (3) The model calculates the Euclidean distance between the query embedding and each class prototype. (4) The query is classified based on a softmax distribution over the negative distances; it is assigned to the class whose prototype is nearest in the embedding space.

[66] N. Bahadori, J. Ashdown, and F. Restuccia, "Rewis: Reliable wi-fi sensing through few-shot multi-antenna multi-receiver csi learning," in WoWMoM. IEEE, 2022, pp. 50--59. paper

[189] X. Ding, T. Jiang, Y . Zhong, J. Yang, Y . Huang, and Z. Li, "Device-free location-independent human activity recognition via few-shot learning," inIEEE/CIC ICC in China, 2021, pp. 106--111. paper

[190] Y . Gu, H. Yan, M. Dong, M. Wang, X. Zhang, Z. Liu, and F. Ren, "Wione: one-shot learning for environment-robust device-free user authentication via commodity wi-fi in man--machine system,"IEEE Transactions on Computational Social Systems, vol. 8, no. 3, pp. 630-- 642, 2021. paper

[191] D. Wang, J. Yang, W. Cui, L. Xie, and S. Sun, "Caution: A robust wifi- based human authentication system via few-shot open-set recognition," IoT-J, vol. 9, no. 18, pp. 17 323--17 333, 2022. paper

[192] B. Yang, H. Wang, L. Hu, H. Zhu, C.-T. Lam, and K. Fang, "Few-shot cross-domain based wifi sensing system for online learning in iot," IEEE Sensors Journal, 2023. paper

[187] T. Zhao, N. Wang, G. Cao, S. Mao, and X. Wang, "Functional data analysis assisted cross-domain wi-fi sensing using few-shot learning," inIEEE ICC. IEEE, 2024, pp. 4780--4785. paper

[193] X. Zhang, C. Tang, K. Yin, and Q. Ni, "Wifi-based cross-domain gesture recognition via modified prototypical networks,"IoT-J, vol. 9, no. 11, pp. 8584--8596, 2021. paper

Meta Learning for Few-shot Learning

Fig 15: Model-Agnostic Meta-Learning (MAML) includes a meta-training stage and a meta-testing stage. The meta-training stage is designed to find a good initialization of model parameters such that the model can quickly adapt to new tasks with only a few gradient steps and a small amount of labeled data in the meta-testing stage.

[195] S. Huang, Y . Chen, D. Wu, G. Yu, and Y . Zhang, "Few-shot learning for human activity recognition based on csi," inCACML. IEEE, 2022, pp. 403--409. paper

[196] Z. Zhou, F. Wang, J. Yu, J. Ren, Z. Wang, and W. Gong, "Target- oriented semi-supervised domain adaptation for wifi-based har," in IEEE INFOCOM. IEEE, 2022, pp. 420--429. paper

[197] J. Gao, D. Wu, F. Yin, Q. Kong, L. Xu, and S. Cui, "Metaloc: Learning to learn wireless localization,"IEEE JASAC, 2023. paper

[92] L. Zhao, R. Xiao, J. Liu, and J. Han, "One is enough: Enabling one-shot device-free gesture recognition with cots wifi," inINFOCOM. IEEE, 2024, pp. 1231--1240. paper

[94] C. Feng, N. Wang, Y . Jiang, X. Zheng, K. Li, Z. Wang, and X. Chen, "Wi-learner: Towards one-shot learning for cross-domain wi-fi based gesture recognition,"IMWUT, vol. 6, no. 3, pp. 1--27, 2022. paper

[198] B. Sheng, R. Han, F. Xiao, Z. Guo, and L. Gui, "Metaformer: Domain- adaptive wifi sensing with only one labelled target sample,"IMWUT, vol. 8, no. 1, pp. 1--27, 2024. paper

[199] S. Ding, Z. Chen, T. Zheng, and J. Luo, "Rf-net: A unified meta- learning framework for rf-enabled one-shot human activity recogni- tion," inSenSys, 2020, pp. 517--530. paper

[154] J. Zhang, Z. Chen, C. Luo, B. Wei, S. S. Kanhere, and J. Li, "Metaganfi: Cross-domain unseen individual identification using wifi signals,"IMWUT, vol. 6, no. 3, pp. 1--21, 2022. paper

[200] Z. Gao, J. Xue, J. Zhang, and W. Xiao, "Ml-wigr: A meta-learning- based approach for cross-domain device-free gesture recognition,"Soft Computing, vol. 26, no. 13, pp. 6145--6155, 2022. paper

Data Augmentationand Data Synthesize

[122] X. Li, L. Chang, F. Song, J. Wang, X. Chen, Z. Tang, and Z. Wang, "Crossgr: Accurate and low-cost cross-target gesture recognition using wi-fi,"IMWUT, vol. 5, no. 1, pp. 1--23, 2021. paper

[180] J. Wang, Q. Gao, X. Ma, Y . Zhao, and Y . Fang, "Learning to sense: Deep learning for wireless sensing with less training efforts,"IEEE Wireless Communications, vol. 27, no. 3, pp. 156--162, 2020. paper

[201] Y . Zhang, X. Wang, Y . Wang, and H. Chen, "Human activity recogni- tion across scenes and categories based on csi,"IEEE TMC, 2020. paper

[202] J. Zhang, Y . Dai, J. Chen, C. Luo, B. Wei, V . C. Leung, and J. Li, "Sida: Self-supervised imbalanced domain adaptation for sound enhancement and cross-domain wifi sensing,"IMWUT, vol. 7, no. 3, pp. 1--24, 2023. paper

[203] T. Huang, S. Wang, and S. Li, "Low-cost and user independent action recognition using wifi signals,"Available at SSRN 4160593. paper

[182] L. Zhang, S. Wu, T. Zhang, and Q. Zhang, "Learning to locate: Adaptive fingerprint-based localization with few-shot relation learning in dynamic indoor environments,"IEEE TWC, vol. 22, no. 8, pp. 5253-- 5264, 2023. paper

[204] Y . Wang, L. Yao, Y . Wang, and Y . Zhang, "Robust csi-based human activity recognition with augment few shot learning,"IEEE Sensors Journal, vol. 21, no. 21, pp. 24 297--24 308, 2021. paper

[205] L. Zhang, Y . Jiang, Y . Ma, S. Mao, W. Huang, Z. Yu, X. Zheng, L. Shu, X. Fan, and G. Xu, "Toward robust and effective behavior based user authentication with off-the-shelf wi-fi,"IEEE TIFS, 2024. paper

[136] X. Chen, H. Li, C. Zhou, X. Liu, D. Wu, and G. Dudek, "Fidora: Robust wifi-based indoor localization via unsupervised domain adaptation," IoT-J, vol. 9, no. 12, pp. 9872--9888, 2022. paper

[206] C. Xiao, D. Han, Y . Ma, and Z. Qin, "Csigan: Robust channel state information-based activity recognition with gans,"IoT-J, vol. 6, no. 6, pp. 10 191--10 204, 2019. paper

[207] X. Chen, H. Li, C. Zhou, X. Liu, D. Wu, and G. Dudek, "Fido: Ubiquitous fine-grained wifi-based localization for unlabelled users via domain adaptation," inProceedings of The Web Conference, 2020, pp. 23--33. paper

[134] D. Yan, P. Yang, F. Shang, F. Han, Y . Yan, and X.-Y . Li, "freegait: Liberalizing wireless-based gait recognition to mitigate non-gait human behaviors," inMobiHoc, 2024, pp. 241--250. paper

[177] C. Xiao, Y . Han, W. Yang, Y . Hou, F. Shi, and K. Chetty, "Diffusion model-based contrastive learning for human activity recognition,"IoT- J, 2024. paper

[208] D. Yan, F. Shang, P. Yang, F. Han, Y . Yan, and X.-Y . Li, "freeloc: Wireless-based cross-domain device-free fingerprints localization to free user's motions,"IoT-J, vol. 11, no. 14, pp. 25 099--25 110, 2024. paper

[209] B. Zhou, R. Zhou, Y . Luo, and Y . Cheng, "Towards cross domain csi action recognition through one-shot bimodal domain adaptation," in MobiQuitous. Springer, 2022, pp. 290--309. paper

[210] F. Wang, J. Han, S. Zhang, X. He, and D. Huang, "Csi-net: Unified human body characterization and pose recognition,"arXiv preprint arXiv:1810.03064, 2018. paper

[211] B. Lan, F. Wang, L. Xia, F. Nai, S. Nie, H. Ding, and J. Han, "Bullydetect: Detecting school physical bullying with wi-fi and deep wavelet transformer,"IoT-J, 2024. paper

[212] H. He, X. Huan, J. Wang, Y . Luo, H. Hu, and J. An, "P3id: A privacy- preserving person identification framework towards multi-environments based on transfer learning,"IEEE TMC, pp. 1--16, 2024. paper

[213] H. Liu, L. Xi, W. Wang, F. Zhang, and Z. J. Haas, "Openfi: Open-set wifi human sensing via virtual embedding confidence-aware,"IEEE TNSE, pp. 1--12, 2024. paper

[152] D. Wang, J. Yang, W. Cui, L. Xie, and S. Sun, "Airfi: empowering wifi-based passive human gesture recognition to unseen environment via domain generalization,"IEEE TMC, vol. 23, no. 2, pp. 1156--1168, 2022. paper

[215] W. Hou and C. Wu, "Rfboost: Understanding and boosting deep wifi sensing via physical data augmentation,"IMWUT, vol. 8, no. 2, pp. 1--26, 2024. paper

[216] S. Liu, Z. Chen, M. Wu, H. Wang, B. Xing, and L. Chen, "Generalizing wireless cross-multiple-factor gesture recognition to unseen domains," IEEE TMC, 2023. paper

[156] Y . Mao, Z. Guo, B. Sheng, L. Gui, and F. Xiao, "Wi-cro: Wifi-based cross domain activity recognition via modified gan,"IEEE TVT, 2024. paper

[217] A. Virmani and M. Shahzad, "Position and orientation agnostic gesture recognition using wifi," inMobiSys, 2017, pp. 252--264. paper

[174] C. Xiao, S. Chen, F. Zhou, and J. Wu, "Self-supervised few-shot time-series segmentation for activity recognition,"IEEE TMC, vol. 22, no. 11, pp. 6770--6783, 2022. paper

[68] R. Xiao, J. Liu, J. Han, and K. Ren, "Onefi: One-shot recognition for unseen gesture via cots wifi," inSenSys, 2021, pp. 206--219. paper

[92] L. Zhao, R. Xiao, J. Liu, and J. Han, "One is enough: Enabling one-shot device-free gesture recognition with cots wifi," inINFOCOM. IEEE, 2024, pp. 1231--1240. paper

[218] S. Wang, L. Wang, and W. Liu, "Feature decoupling and regeneration towards wifi-based human activity recognition,"Pattern Recognition, vol. 153, p. 110480, 2024. paper

Pseudo Labeling

[75] R. Zhou, Z. Gong, K. Tang, B. Zhou, and Y . Cheng, "Device-free cross location activity recognition via semi-supervised deep learning,"Neural Computing and Applications, vol. 34, no. 12, pp. 10 189--10 203, 2022. paper

[96] J. Chen, S. Bi, X.-H. Lin, and Z. Quan, "Lager: Label-free domain- adaptive wireless gesture recognition via latent feature alignment and augmentation,"IoT-J, vol. 11, no. 23, pp. 37 928--37 941, 2024. paper

[219] B.-B. Zhang, D. Zhang, Y . Li, Y . Hu, and Y . Chen, "Unsupervised domain adaptation for rf-based gesture recognition,"IoT-J, vol. 10, no. 23, pp. 21 026--21 038, 2023. paper

[118] L. Sheng, Y . Chen, S. Ning, S. Wang, B. Lian, and Z. Wei, "Da- har: Dual adversarial network for environment-independent wifi human activity recognition,"Pervasive and Mobile Computing, vol. 96, p. 101850, 2023. paper

[100] B.-B. Zhang, D. Zhang, Y . Hu, and Y . Chen, "Unsupervised domain adaptation for wifi gesture recognition," inWCNC. IEEE, 2023, pp. 1--6. paper

Key Takeaways and Lessons Learned: Feature Learning

• Domain Alignment for Distribution Matching:
  Minimizing domain divergence via adversarial discriminators or statistical metrics (e.g., MMD, EMD) is fundamental. A key lesson is using cross-modal embeddings (e.g., BERT, GPT-4o) as “semantic anchors” to bridge the gap between raw CSI and high-level human activities.

• Component Disentanglement for Interpretability:
  Explicitly separating environment, identity, and motion components using GANs or specialized encoders ensures that the model ignores environmental noise. This disentangled latent space is critical for handling multi-factor variations in real-world settings.

• Metric Learning for Few-shot Scalability:
  Structuring the feature space using triplet loss, contrastive learning, or prototypical networks allows for clustering similar actions. The major takeaway is the high scalability of similarity-based matching, which enables zero/few-shot recognition of novel classes without retraining the backbone.

• Meta-Learning for Rapid Adaptation:
  Paradigms like MAML and Reptile focus on “learning how to learn” by optimizing for a superior parameter initialization. This ensures that the sensing system can converge to a site-specific optimum with minimal gradient steps (e.g., one-shot or few-shot).

• Data Synthesis as a Robustness Plug-in:
  Beyond simple augmentations, generative models (GANs, VAEs, and Diffusion) are essential for simulating target-domain conditions. These modules mitigate the Sim-to-Real gap by expanding training diversity with physically plausible synthetic signals.

• Pseudo Labeling for Iterative Refinement:
  Leveraging abundant unlabeled target data through virtual labels reduces annotation costs. The iterative refinement of these labels (e.g., via confidence thresholds or centroid-based reassignment) is vital for long-term performance improvement in unfamiliar domains.

Stage 4: :monkey: Model Deployment

Table V: Comparison of Model Deployment Strategies in Wi-Fi Sensing.

Transfer Learning

[70] F. Wang, Y . Lv, M. Zhu, H. Ding, and J. Han, "Xrf55: A radio frequency dataset for human indoor action analysis,"IMWUT, vol. 8, no. 1, pp. 1--34, 2024. paper

[210] F. Wang, J. Han, S. Zhang, X. He, and D. Huang, "Csi-net: Unified human body characterization and pose recognition,"arXiv preprint arXiv:1810.03064, 2018. paper

[153] X. Ding, T. Jiang, Y . Li, W. Xue, and Y . Zhong, "Device-free location- independent human activity recognition using transfer learning based on cnn," inICC Workshops. IEEE, 2020, pp. 1--6. paper

[220] Y .-S. Chen, Y .-C. Chang, and C.-Y . Li, "A semi-supervised transfer learning with dynamic associate domain adaptation for human activity recognition using wifi signals,"Sensors, vol. 21, no. 24, p. 8475, 2021. paper

[221] J. K. Brinke and N. Meratnia, "Scaling activity recognition using channel state information through convolutional neural networks and transfer learning," inProceedings of the International Workshop on Challenges in Artificial Intelligence and Machine Learning for Internet of Things, 2019, pp. 56--62. paper

[222] Y . Fang, B. Sheng, H. Wang, and F. Xiao, "Witransfer: A cross-scene transfer activity recognition system using wifi," inTURC, 2020, pp. 59--63. paper

[223] G. Yin, J. Zhang, G. Shen, and Y . Chen, "Fewsense, towards a scalable and cross-domain wi-fi sensing system using few-shot learning,"IEEE TMC, vol. 23, no. 1, pp. 453--468, 2022. paper

[224] L. Zheng, S. Bi, S. Wang, Z. Quan, X. Li, X. Lin, and H. Wang, "Resmon: Domain-adaptive wireless respiration state monitoring via few-shot bayesian deep learning,"IoT-J, vol. 10, no. 23, pp. 20 914-- 20 927, 2023. paper

[225] S. Bi, X. Chen, L. Zheng, H. Hou, and X. Lin, "Roger: Few-shot learning based robust gesture recognition with multi-modal wi-fi csi measurements," inICMLCN. IEEE, 2024, pp. 517--522. paper

[226] H. Hou, S. Bi, L. Zheng, X. Lin, and Z. Quan, "Sample-efficient cross- domain wifi indoor crowd counting via few-shot learning," inWOCC. IEEE, 2022, pp. 132--137. paper

[227] H. Hou, S. Bi, L. Zheng, X. Lin, Y . Wu, and Z. Quan, "Dasecount: Domain-agnostic sample-efficient wireless indoor crowd counting via few-shot learning,"IoT-J, vol. 10, no. 8, pp. 7038--7050, 2022. paper

Federated Learning

[228] S. M. Hernandez and E. Bulut, "Wifederated: Scalable wifi sensing using edge-based federated learning,"IoT-J, vol. 9, no. 14, pp. 12 628-- 12 640, 2021. paper

[229] H. Geng, D. Deng, W. Zhang, P. Ji, and X. Wu, "Personalized federated learning based on bidirectional knowledge distillation for wifi gesture recognition,"Electronics, vol. 12, no. 24, p. 5016, 2023. paper

[230] X. Li, F. Song, M. Luo, K. Li, L. Chang, X. Chen, and Z. Wang, "Towards collaborative and cross-domain wi-fi sensing: A case study for human activity recognition,"IEEE TMC, vol. 23, no. 2, pp. 1674-- 1688, 2023. paper

[231] W. Qi, R. Zhang, J. Zhou, H. Zhang, Y . Xie, and X. Jing, "A resource-efficient cross-domain sensing method for device-free gesture recognition with federated transfer learning,"IEEE TGCN, vol. 7, no. 1, pp. 393--400, 2023. paper

[232] S. Zhang, T. Jiang, X. Ding, Y . Zhong, and H. Jia, "A cloud-edge col- laborative framework for cross-environment human action recognition based on wi-fi," inIEEE/CIC ICC in China, 2023, pp. 1--6. paper

[119] L. Zhang, W. Cui, B. Li, Z. Chen, M. Wu, and T. S. Gee, "Privacy- preserving cross-environment human activity recognition,"IEEE Trans- actions on Cybernetics, vol. 53, no. 3, pp. 1765--1775, 2021. paper

[233] N. Zheng, Y . Li, S. Jiang, Y . Li, R. Yao, C. Dong, T. Chen, Y . Yang, Z. Yin, and Y . Liu, "Adawifi, collaborative wifi sensing for cross- environment adaptation,"IEEE TMC, pp. 1--15, 2024. paper

[234] J. Zhang, Y . Li, Q. Li, and W. Xiao, "Variance-constrained local--global modeling for device-free localization under uncertainties,"IEEE TII, 2023. paper

Continual Learning

[192] B. Yang, H. Wang, L. Hu, H. Zhu, C.-T. Lam, and K. Fang, "Few-shot cross-domain based wifi sensing system for online learning in iot," IEEE Sensors Journal, 2023. paper

[235] Y . Bai, Z. Wang, K. Zheng, X. Wang, and J. Wang, "Widrive: Adaptive wifi-based recognition of driver activity for real-time and safe takeover," inICDCS. IEEE, 2019, pp. 901--911. paper

[236] E. Soltanaghaei, R. A. Sharma, Z. Wang, A. Chittilappilly, A. Luong, E. Giler, K. Hall, S. Elias, and A. Rowe, "Robust and practical wifi hu- man sensing using on-device learning with a domain adaptive model," inProceedings of the ACM International Conference on Systems for Energy-Efficient Buildings, Cities, and Transportation, 2020, pp. 150-- 159. paper

[237] S. Zhai, Z. Tang, P. Nurmi, D. Fang, X. Chen, and Z. Wang, "Rise: Robust wireless sensing using probabilistic and statistical assessments," inMobiCom, 2021, pp. 309--322. paper

[238] Q. Fu, F. Wang, M. Zhu, H. Ding, J. Han, and T. X. Han, "Ccs: Con- tinuous learning for customized incremental wireless sensing services," arXiv preprint arXiv:2412.04821, 2024. paper

[239] T. Zhang, Q. Fu, H. Ding, G. Wang, and F. Wang, "Carec: Continual wireless action recognition with expansion-compression coordination," Sensors, 2025. paper

[240] Y . Zhang, F. He, Y . Wang, D. Wu, and G. Yu, "Csi-based cross- scene human activity recognition with incremental learning,"Neural Computing and Applications, vol. 35, no. 17, pp. 12 415--12 432, 2023. paper

Key Takeaways and Lessons Learned: Model Deployment

• Hierarchical Efficiency in Transfer Learning:
  While fine-tuning the entire network yields superior accuracy, freezing the backbone and updating only the task head is most viable for resource-constrained edge devices. This strategy balances global feature representation with site-specific adaptation.

• Privacy-Preserving Collaborative Generalization:
  Federated Learning is a powerful engine for building robust global models without moving raw CSI data. A major lesson is managing communication overhead during parameter exchange and addressing the performance drops caused by shifted data distributions across heterogeneous clients.

• Evolution through Continual Learning:
  To handle long-term environmental drifts (e.g., furniture moving), models must incrementally acquire knowledge. The core challenge is mitigating catastrophic forgetting; strategies like data replay and dynamic model expansion are essential to preserve performance on original tasks.

• Minimizing User Burden in Online Adaptation:
  Incorporating “User-in-the-loop” mechanisms allows for continuous refinement. The key takeaway is using anomaly detection and high-confidence sampling to minimize the labeling effort required from users during the adaptation phase.

• Systems-level Latency and Resource Trade-offs:
  Beyond accuracy, real-world deployment on COTS chipsets requires balancing model update frequency with inference latency. For high-responsiveness applications, local updates must be optimized to prevent system bottlenecks.

🐱&🦩 Challenges and Future Directions

🐬 Data, Data, Data

The scaling law, which involves training large-scale models with large-scale datasets, has proven its success since the advent of AlexNet [Krizhevsky et al., 2012], achieving remarkable breakthroughs in fields such as computer vision (CV), natural language processing (NLP), robotics, and science. Large-scale datasets like ImageNet [Deng et al., 2009], COCO [Lin et al., 2014], and Kinetics [Kay et al., 2017] have significantly accelerated advancements in these areas. In contrast, the field of Wi-Fi human sensing has not progressed as rapidly, despite the widespread adoption of deep learning methods. A key reason lies in the limited scale and diversity of available Wi-Fi sensing datasets.

Recent initiatives are actively narrowing the data gap; for instance, CSI-Bench [Zhu et al., 2025] achieves a milestone in realism with 461 hours of recordings, XRF55 [Wang et al., 2024] significantly expands the sample count to 42,900, and OctoNet [Yuan et al., 2025] comprises 12 heterogeneous modalities. Despite these advancements in data duration and environmental complexity, Wi-Fi sensing datasets still lack the vast scale of CV benchmarks like ImageNet or Kinetics. The gap remains evident not only in raw sample counts but also in the breadth of action categories and participant demographics, emphasizing the need for continued expansion in data diversity.

Collecting and annotating Wi-Fi sensing data is inherently time-consuming and labor-intensive, creating a data-scarcity bottleneck for deep learning. To address this, synthetic data generation has emerged as a promising solution. By leveraging high-fidelity 3D modeling and computational electromagnetics, researchers can simulate diverse human-environment interactions. For example, by placing dynamic 3D human models into various virtual environments and employing ray-tracing algorithms to emulate signal propagation, realistic Wi-Fi CSI data can be synthesized. This approach provides the massive datasets required to train large-scale foundation models, potentially revolutionizing the scalability of Wi-Fi sensing. However, several open challenges, such as the domain gap between simulated and physical signals, remain to be addressed.

Open Challenges in Synthetic Data Generation

1. Modeling signal propagation: How to model the propagation of signals through people and environments? Can methods like ray-tracing and Fresnel zone models effectively simulate Wi-Fi signals' interactions with objects and people? How to enable modern approaches, such as diffusion models [Ho et al., 2020], Variational Autoencoders (VAEs) [Kingma & Welling, 2013], Neural Radiance Fields (NeRF) [Mildenhall et al., 2021], or 3D Gaussian Splatting [Kerbl et al., 2023] to learn probabilistic mappings of signal propagation patterns from observed data?

2. Constructing diverse environments: Should we rely on manual modeling of environments, even though it is resource-intensive? Or should we leverage open-source tools like Unity, Unreal Engine, or Infinigen Indoors [Raistrick et al., 2024], which provide scalable solutions for generating dynamic, realistic environments tailored to Wi-Fi sensing scenarios?

3. Synthesizing diverse human body models: Is it more effective to use game engines like Unity or Unreal Engine to create diverse human models? Alternatively, could tools such as SMPL [Loper et al., 2015] or MANO [Romero et al., 2022] be utilized to extract human body/hand models from large-scale video datasets for synthetic data generation?

4. Bridging synthetic and real data: Could Generative Adversarial Networks [Goodfellow et al., 2014] help enhance the realism of synthetic features or adapt synthetic data to better match real-world characteristics? What role could domain adaptation and domain-invariant feature learning play in aligning synthetic and real data to improve model performance?

Roadmap for Synthetic Data Generation

• Short-term Goals (1–2 years): Focus on enhancing signal realism using advanced generative models (Diffusion Models, GANs). Establish a precise "signal-to-physical-entity" mapping to ensure synthesized CSI samples are indistinguishable from real-world data in both frequency and time domains.

• Long-term Goals (3–5 years): Realize large-scale, multi-environment closed-loop simulation systems. Develop simulation engines capable of automatically generating complex 3D indoor scenes and dynamic human-environment interactions. The core challenge is overcoming the "Sim-to-Real" domain gap, ensuring models trained entirely on simulated data can be deployed in physical hardware environments.

🐺 Wi-Fi Sensing Foundation Model Pre-Training

Foundation models in vision and language, such as CLIP [Radford et al., 2021], GPT [Brown et al., 2020], and BERT [Devlin et al., 2019], have demonstrated remarkable success in adapting to diverse downstream tasks. Similarly, Wi-Fi sensing encompasses a wide range of applications, including human activity recognition, indoor localization, presence detection, and pose estimation. These varied tasks make the development of a Wi-Fi sensing foundation model both promising and necessary. However, training such a model raises several critical open questions:

Open Questions for Foundation Model Design

1. Model architectures: Should we utilize traditional architectures like CNNs and RNNs, or opt for Transformers [Vaswani et al., 2017]? Emerging architectures like Mamba [Gu et al., 2023], known for smaller parameters, reduced memory usage, and faster inference, may also be viable. How to tailor architecture design for different hardware constraints?

2. Leveraging unlabeled real-world data: How can large-scale unlabeled real-world Wi-Fi data be incorporated effectively? Should data be transmitted to a central server, or processed at the edge? How can privacy and computational constraints be maintained?

3. Self-supervised learning strategies: What proxy tasks are most suitable? Contrastive learning, masking-then-reconstruction (as in BERT or MAE [He et al., 2022]), or domain-specific tasks for capturing spatiotemporal Wi-Fi signal characteristics?

4. Incorporating environmental and human factors: Should environmental parameters, human body factors, and device configurations be embedded explicitly, or inferred implicitly by the model?

Roadmap for Wi-Fi Foundation Models

• Short-term Goals (1–2 years): Construct task-agnostic universal feature extractors using self-supervised learning (e.g., Masked Signal Modeling) on medium-scale datasets. Ensure robustness against device heterogeneity.

• Long-term Goals (3–5 years): Train large-scale perception foundation models with billions of parameters. Integrate ultra-large-scale unlabeled real-world data with synthetic datasets for zero-shot generalization. Address computational costs and globally distributed device coordination while preserving privacy.

🐳 When Wi-Fi Sensing Meets Large Multimodal Models

Another promising direction is leveraging existing large multimodal models (LMMs) by fine-tuning them to process Wi-Fi data. LMMs trained on diverse images, text, and audio demonstrate impressive adaptability. Integrating Wi-Fi raises several challenges:

Open Questions

1. Fine-tuning strategy: LoRA [Hu et al., 2022], adapters [Rebuffi et al., 2017], or prompt-based fine-tuning [Liu et al., 2021]? How to balance computational efficiency and task performance?

2. Aligning Wi-Fi with existing modalities: Should Wi-Fi CSI data be converted to spectrograms/images [Zhang et al., 2023; Jiao et al., 2024] or processed in native time-series format? How to bridge the semantic gap via cross-modal contrastive learning or other alignment techniques?

3. Incorporating domain knowledge: Use Wi-Fi propagation principles, IEEE 802.11bf protocols, or Retrieval-Augmented Generation (RAG) [Lewis et al., 2020] for interpretability and efficiency.

4. Multimodal fusion: Wi-Fi combined with cameras or IMUs can enhance performance. Techniques like ALBEF [Li et al., 2021], Cross-Attention [Chen et al., 2024], or Mixture of Experts [Yun et al., 2024] may be applied. Challenges include modality-specific noise, temporal synchronization, and computational overhead.

Roadmap for LMM Integration

• Short-term Goals (1–2 years): Establish efficient semantic alignment between Wi-Fi signals and visual/text modalities. Use cross-modal translation (e.g., WiFi2Radar [Zhang et al., 2020]) to convert Wi-Fi into LMM-compatible tokens.

• Long-term Goals (3–5 years): Develop native sensing agents with inherent wireless perception. Bridge the semantic gap between low-level Wi-Fi signals and high-level reasoning in LMMs.

🐍 System Deployment

Wi-Fi sensing systems can follow a hierarchical "Cloud-to-Edge" deployment pipeline:

Open Challenges

1. Efficient inference: Deploying large models on COTS chips (e.g., ESP32) requires handling frequency-domain sparsity. Use structure-aware pruning [Han et al., 2015], quantization [Jacob et al., 2018], and knowledge distillation [Hinton et al., 2015] to retain subcarriers sensitive to human motion.

2. On-device training and fine-tuning: Environmental shifts require edge adaptation with on-device training [Lin et al., 2022] to calibrate models to local environments.

3. Sensing-communication coexistence: Optimize CSI streaming and resource allocation during congestion via downsampling or temporal-spectral compression. Consider ISAC-aware scheduling.

4. Adversarial resilience: Protect against physical adversarial attacks [Liu et al., 2022] and explore Generative AI solutions for physical-layer security (e.g., DFSS [Wang et al., 2025]).

5. Data collection in consumer scenarios: Minimize manual labeling via implicit feedback loops and cross-modal supervision (e.g., smartphone IMUs). Challenges include noisy labels, privacy restrictions, and concept drift.

Roadmap for Deployment

• Short-term Goals (1–2 years): Develop lightweight adaptive deployment, low-power calibration-free algorithms, and standardized sensing interfaces (e.g., IEEE 802.11bf).

• Long-term Goals (3–5 years): Realize large-scale, privacy-preserving "Sensing-as-a-Service" integrated into smart city infrastructure.

@inproceedings{a-survey-on-wi-fi-sensing-generalizability-taxonomy-techniques-datasets-and-future-research-prospects,
  title = {A Survey on Wi-Fi Sensing Generalizability: Taxonomy, Techniques, Datasets, and Future Research Prospects},
  author = {Fei Wang and Tingting Zhang and Wei Xi and Han Ding and Ge Wang and Di Zhang and Yuanhao Cui and Fan Liu and Jinsong Han and Jie Xu and Tony Xiao Han},
  booktitle = {IEEE Communications Surveys & Tutorials, 2026},
  year = {2026}
}