pith. sign in

arxiv: 2605.29538 · v1 · pith:3QVHWLNJnew · submitted 2026-05-28 · 💻 cs.CV

RadioFormer3D: Weakly Supervised 3D Radio Map Estimation in Low-Altitude Airspace via Generative Modeling

Zheng Fang , Junjie Liu , Kangjun Liu , Jianguo Zhang , Yaowei Wang , Ke Chen This is my paper

Pith reviewed 2026-06-29 08:41 UTC · model grok-4.3

classification 💻 cs.CV
keywords 3D radio map estimationweakly supervised learningvolumetric reconstructiongenerative modelingJoint Spectrum Integrity Losslow-altitude airspaceFourier sampling encoder
0
0 comments X

The pith

RadioFormer3D reconstructs 3D radio maps from sparse horizontal measurements by enforcing spectrum integrity across volumes, maps, and pixels.

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

The paper presents RadioFormer3D to estimate signal strength throughout a full 3D volume of low-altitude airspace when only sparse measurements at a few fixed heights are available. It extends a prior dual-stream architecture with a Fourier-based sampling encoder and volumetric decoder, then trains the whole system under a Joint Spectrum Integrity Loss that supplies missing vertical information through volume-level pseudo-labels, map-level geometry-aware rendering, and pixel-level constraints. If the approach holds, networks could characterize coverage at every height without dense vertical sampling campaigns. A reader would care because drone corridors and 3D wireless systems need accurate height-dependent propagation maps to avoid blind spots. Experiments report better quality at unseen altitudes than prior methods while preserving a practical speed-accuracy balance.

Core claim

RadioFormer3D, built on the dual-stream multi-granularity fusion of RadioFormer, adds a Fourier-based sampling encoder and volumetric decoder to process sparse 3D measurements; its Joint Spectrum Integrity Loss unifies volume-level pseudo-label supervision, map-level geometry-aware radio rendering, and pixel-level localized constraints so the model can recover complex vertical structural relationships from limited horizontal data alone, yielding superior overall performance and improved reconstruction at unlabeled altitudes.

What carries the argument

The Joint Spectrum Integrity Loss, which integrates volume-level pseudo-label supervision, map-level geometry-aware radio rendering, and pixel-level localized constraints to infer vertical structure from sparse horizontal inputs.

If this is right

  • Superior reconstruction quality at unlabeled altitudes compared with representative existing methods.
  • Favorable accuracy versus inference-efficiency trade-off on multiple radio map datasets.
  • Support for future 3D environment-aware wireless networks that require volumetric spectrum awareness.
  • Effective use of weak supervision to handle increased spatial sparsity when extending from 2D to 3D radio mapping.

Where Pith is reading between the lines

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

  • The loss formulation could transfer to other anisotropic volumetric tasks where dense sampling exists in only one or two axes.
  • Deployment would need checks against real flight-collected data whose altitude fading statistics differ from the pseudo-label generation process.
  • Larger spatial extents could reveal whether the Fourier encoder continues to scale without accuracy loss.
  • The same architecture might support incremental updating when new horizontal slices become available over time.

Load-bearing premise

The assumption that the combined pseudo-label, rendering, and constraint terms can reliably recover vertical propagation patterns when only horizontal slices are directly observed.

What would settle it

A set of real continuous-altitude radio measurements in a test volume where the model's predicted signal values at the unlabeled heights deviate substantially from the measured ground truth.

Figures

Figures reproduced from arXiv: 2605.29538 by Jianguo Zhang, Junjie Liu, Kangjun Liu, Ke Chen, Yaowei Wang, Zheng Fang.

Figure 1
Figure 1. Figure 1: Illustration of the 3D electromagnetic spectrum situation in urban [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Architectural overview of the proposed RadioFormer3D. The framework adopts a dual-stream design to process heterogeneous inputs for 3D RME. The DSA module independently encodes multi-granularity features: the building-height map M is processed by ViT blocks to capture environmental context, while the sparse sample information S is encoded via a Point Encoder integrated with Fourier and FiLM processes. Thes… view at source ↗
Figure 3
Figure 3. Figure 3: Overview of the RRM. The module casts vertical rays through [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Comparison of reconstruction error (RMSE) across different altitude [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Visualization of 3D radio map reconstruction on the UrbanRadio3D dataset. The rows are categorized into Labeled Heights (1m and 19m), used during [PITH_FULL_IMAGE:figures/full_fig_p008_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Impact of supervision density on reconstruction performance. The [PITH_FULL_IMAGE:figures/full_fig_p010_6.png] view at source ↗
read the original abstract

With the emergence of wireless applications in three-dimensional environments, such as the low-altitude airspace and 3D heterogeneous networks, radio map estimation is increasingly required to characterize signal propagation across both horizontal and vertical dimensions. However, extending radio map estimation from 2D to 3D remains challenging due to increased spatial sparsity and limited supervision across continuous altitudes. In this paper, we propose \textbf{\textit{RadioFormer3D}}, a specialized model for volumetric spectrum reconstruction under weak supervision. Building on the dual-stream, multi-granularity fusion architecture of \textit{RadioFormer}, \textit{RadioFormer3D} introduces a Fourier-based sampling encoder and a volumetric decoder to efficiently process sparse measurements in 3D space. To alleviate the lack of vertical supervision, we propose the \textbf{\textit{Joint Spectrum Integrity Loss}}, which integrates volume-level pseudo-label supervision, map-level geometry-aware radio rendering, and pixel-level localized constraints within a unified optimization scheme. This design enables the model to capture complex vertical structural relationships more effectively under sparse supervision. Extensive experiments across several radio map datasets show that \textit{RadioFormer3D} achieves superior overall performance compared to representative existing methods. In particular, it demonstrates improved reconstruction quality at unlabeled altitudes while maintaining a favorable trade-off between accuracy and inference efficiency, positioning it as a highly promising solution for future 3D environment-aware wireless networks.

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

1 major / 0 minor

Summary. The paper proposes RadioFormer3D, extending the dual-stream RadioFormer architecture with a Fourier-based sampling encoder and volumetric decoder for weakly supervised 3D radio map estimation in low-altitude airspace. It introduces the Joint Spectrum Integrity Loss, which combines volume-level pseudo-label supervision, map-level geometry-aware radio rendering, and pixel-level localized constraints to address limited vertical supervision from sparse horizontal measurements. Experiments on multiple radio map datasets claim superior overall performance versus existing methods, with particular gains in reconstruction quality at unlabeled altitudes and a favorable accuracy-inference efficiency trade-off.

Significance. If the central claims hold, the work would advance 3D spectrum reconstruction for environment-aware wireless networks by demonstrating that a multi-term loss can recover vertical propagation structure under weak supervision. The extension of generative modeling to volumetric radio maps with explicit handling of altitude sparsity addresses a practical gap in low-altitude airspace applications. No machine-checked proofs or parameter-free derivations are present, but the emphasis on inference efficiency is a positive attribute if validated.

major comments (1)
  1. [Abstract] Abstract (Joint Spectrum Integrity Loss): the headline claim of improved reconstruction quality at unlabeled altitudes rests on the assertion that the three-term loss recovers complex vertical structural relationships from sparse horizontal data alone. No information is supplied on pseudo-label generation (e.g., whether derived solely from the same horizontal measurements), the explicit formulation of the geometry-aware rendering term, or any ablation isolating each component's contribution to vertical consistency. This absence makes it impossible to determine whether the loss supplies new vertical information or merely regularizes toward plausible but unverified solutions, directly undermining assessment of the central empirical result.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for their constructive feedback on the manuscript. We address the major comment point by point below.

read point-by-point responses

  1. Referee: [Abstract] Abstract (Joint Spectrum Integrity Loss): the headline claim of improved reconstruction quality at unlabeled altitudes rests on the assertion that the three-term loss recovers complex vertical structural relationships from sparse horizontal data alone. No information is supplied on pseudo-label generation (e.g., whether derived solely from the same horizontal measurements), the explicit formulation of the geometry-aware rendering term, or any ablation isolating each component's contribution to vertical consistency. This absence makes it impossible to determine whether the loss supplies new vertical information or merely regularizes toward plausible but unverified solutions, directly undermining assessment of the central empirical result.

    Authors: We agree that the abstract is too concise and omits key implementation details, which hinders evaluation of the central claim. The full manuscript details pseudo-label generation in Section 3.2 (derived solely from horizontal measurements via physics-informed interpolation), the geometry-aware rendering term in Equation (6) (a differentiable ray-integration approximation), and component ablations in Section 4.3/Table 4 (showing each term's isolated contribution to vertical consistency). We will revise the abstract to briefly reference these elements. The experiments demonstrate that removing any loss term degrades performance at unlabeled altitudes, supporting that the formulation recovers vertical structure rather than mere regularization. revision: yes

Circularity Check

0 steps flagged

No circularity: derivation is self-contained architectural and loss design.

full rationale

The abstract and description introduce RadioFormer3D as an extension of RadioFormer with a new Joint Spectrum Integrity Loss that combines three explicitly described terms (volume-level pseudo-label supervision, map-level geometry-aware radio rendering, pixel-level constraints). No equations, fitting procedures, or self-citations are shown that reduce any claimed prediction or result to the inputs by construction. The central claim of improved vertical reconstruction rests on the independent design of the loss terms rather than on any self-referential definition or fitted parameter renamed as output. This is the normal non-circular case for a methods paper proposing a new model and objective.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review supplies no explicit free parameters, axioms, or invented entities; the model and loss are presented as engineering choices without stated mathematical assumptions beyond standard deep-learning practice.

pith-pipeline@v0.9.1-grok · 5804 in / 1207 out tokens · 17577 ms · 2026-06-29T08:41:34.707074+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

39 extracted references · 1 canonical work pages

  1. [1]

    Blockage- resilient integrated sensing and communication in mmwave networks: Multi-view collaboration and efficient task allocation,

    Y . Cui, H. Ding, Y . Ma, X. Li, H. Zhang, and Y . Fang, “Blockage- resilient integrated sensing and communication in mmwave networks: Multi-view collaboration and efficient task allocation,”IEEE Trans. Mob. Comput., 2025

    2025

  2. [2]

    Cram ´er-rao bound analysis and beam- forming design for integrated sensing and communication with extended targets,

    Y . Wang, M. Tao, and S. Sun, “Cram ´er-rao bound analysis and beam- forming design for integrated sensing and communication with extended targets,”IEEE Transactions on Wireless Communications, 2024

    2024

  3. [3]

    A spatiotemporal approach for secure crowd- sourced radio environment map construction,

    Y . Hu and R. Zhang, “A spatiotemporal approach for secure crowd- sourced radio environment map construction,”IEEE/ACM Transactions on Networking, 2020. JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2021 11

    2020

  4. [4]

    Wireless communications with reconfigurable intelligent surface: Path loss modeling and experimental measurement,

    W. Tang, M. Z. Chen, X. Chen, J. Y . Dai, Y . Han, M. Di Renzo, Y . Zeng, S. Jin, Q. Cheng, and T. J. Cui, “Wireless communications with reconfigurable intelligent surface: Path loss modeling and experimental measurement,”IEEE Transactions on Wireless Communications, 2021

    2021

  5. [5]

    Radiounet: Fast radio map estimation with convolutional neural networks,

    R. Levie, C ¸ . Yapar, G. Kutyniok, and G. Caire, “Radiounet: Fast radio map estimation with convolutional neural networks,”IEEE Trans. Wirel. Commun., 2021

    2021

  6. [7]

    Dataset of pathloss and toa radio maps with localization application,

    C. Yapar, R. Levie, G. Kutyniok, and G. Caire, “Dataset of pathloss and toa radio maps with localization application,” 2022

    2022

  7. [8]

    Joint trajectory and communication design for multi-uav enabled wireless networks,

    Q. Wu, Y . Zeng, and R. Zhang, “Joint trajectory and communication design for multi-uav enabled wireless networks,”IEEE Transactions on Wireless Communications, 2018

    2018

  8. [9]

    A mobility-resilient spectrum sharing framework for operating wireless uavs in the 6 ghz band,

    J. Hu, S. K. Moorthy, A. Harindranath, Z. Zhang, Z. Zhao, N. Mas- tronarde, E. S. Bentley, S. Pudlewski, and Z. Guan, “A mobility-resilient spectrum sharing framework for operating wireless uavs in the 6 ghz band,”IEEE/ACM Transactions on Networking, 2023

    2023

  9. [10]

    Sparse bayesian learning-based hierarchical construction for 3d radio environment maps incorporating channel shadowing,

    J. Wang, Q. Zhu, Z. Lin, J. Chen, G. Ding, Q. Wu, G. Gu, and Q. Gao, “Sparse bayesian learning-based hierarchical construction for 3d radio environment maps incorporating channel shadowing,”IEEE Transactions on Wireless Communications, 2024

    2024

  10. [11]

    Radio map assisted multi-uav target searching,

    C. He, Y . Dong, and Z. J. Wang, “Radio map assisted multi-uav target searching,”IEEE Transactions on Wireless Communications, 2023

    2023

  11. [12]

    Radioformer: A multiple-granularity radio map estimation transformer with 1‱spatial sampling,

    Z. Fang, K. Liu, K. Chen, Q. Liu, J. Zhang, L. Song, and Y . Wang, “Radioformer: A multiple-granularity radio map estimation transformer with 1‱spatial sampling,”CoRR, 2025

    2025

  12. [13]

    Bayesian active learning for sample efficient 5g radio map reconstruction,

    K. D. Polyzos, A. Sadeghi, W. Ye, S. Sleder, K. Houssou, J. Calder, Z.-L. Zhang, and G. B. Giannakis, “Bayesian active learning for sample efficient 5g radio map reconstruction,”IEEE Transactions on Wireless Communications, 2024

    2024

  13. [14]

    Deep completion autoencoders for radio map estimation,

    Y . Teganya and D. Romero, “Deep completion autoencoders for radio map estimation,”IEEE Transactions on Wireless Communications, 2022

    2022

  14. [15]

    A self-supervised learning-based channel estimation for irs-aided communication without ground truth,

    Z. Zhang, T. Ji, H. Shi, C. Li, Y . Huang, and L. Yang, “A self-supervised learning-based channel estimation for irs-aided communication without ground truth,”IEEE Transactions on Wireless Communications, 2023

    2023

  15. [16]

    Wide- band millimeter-wave propagation measurements and channel models for future wireless communication system design,

    T. S. Rappaport, G. R. MacCartney, M. K. Samimi, and S. Sun, “Wide- band millimeter-wave propagation measurements and channel models for future wireless communication system design,”IEEE Transactions on Communications, 2015

    2015

  16. [17]

    Ray tracing for radio propagation modeling: Principles and applications,

    Z. Yun and M. F. Iskander, “Ray tracing for radio propagation modeling: Principles and applications,”IEEE Access, 2015

    2015

  17. [18]

    A statistical basis for lognormal shadowing effects in multipath fading channels,

    A. Coulson, A. Williamson, and R. Vaughan, “A statistical basis for lognormal shadowing effects in multipath fading channels,”IEEE Trans- actions on Communications, 1998

    1998

  18. [19]

    Radio map estimation: A data-driven approach to spectrum cartography,

    D. Romero and S.-J. Kim, “Radio map estimation: A data-driven approach to spectrum cartography,”IEEE Signal Processing Magazine, 2022

    2022

  19. [20]

    U-net: Convolutional networks for biomedical image segmentation,

    O. Ronneberger, P. Fischer, and T. Brox, “U-net: Convolutional networks for biomedical image segmentation,” inMedical Image Computing and Computer-Assisted Intervention - MICCAI 2015, N. Navab, J. Horneg- ger, W. M. W. III, and A. F. Frangi, Eds., 2015

    2015

  20. [21]

    Paying deformable attention to sparse spatial observations for deep radio map estimation,

    K. Liu, C. Qiu, K. Chen, Q. Zheng, L. Song, and Y . Wang, “Paying deformable attention to sparse spatial observations for deep radio map estimation,”IEEE Transactions on Cognitive Communications and Networking, 2026

    2026

  21. [22]

    Deformable convolutional networks,

    J. Dai, H. Qi, Y . Xiong, Y . Li, G. Zhang, H. Hu, and Y . Wei, “Deformable convolutional networks,” in2017 IEEE International Conference on Computer Vision (ICCV), 2017

    2017

  22. [23]

    Deformable convnets v2: More deformable, better results,

    X. Zhu, H. Hu, S. Lin, and J. Dai, “Deformable convnets v2: More deformable, better results,” in2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2019

    2019

  23. [24]

    Radiodun: A physics-inspired deep unfolding network for radio map estimation,

    T. Chen, Z. Zhou, Z. Fang, W. Zou, K. Liu, K. Chen, Y . Zhang, and Y . Wang, “Radiodun: A physics-inspired deep unfolding network for radio map estimation,” 2025

    2025

  24. [25]

    Pmnet: Robust pathloss map prediction via supervised learning,

    J. Lee, O. G. Serbetci, D. P. Selvam, and A. F. Molisch, “Pmnet: Robust pathloss map prediction via supervised learning,” inIEEE Global Communications Conference, GLOBECOM 2023,, 2023, pp. 4601–4606

    2023

  25. [26]

    Attention is all you need,

    A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, L. Kaiser, and I. Polosukhin, “Attention is all you need,” inAdvances in Neural Information Processing Systems 30: Annual Conference on Neural Information Processing Systems, NIPS 2017, I. Guyon, U. von Luxburg, S. Bengio, H. M. Wallach, R. Fergus, S. V . N. Vishwanathan, and R. Garne...

    2017

  26. [27]

    Nerf: Representing scenes as neural radiance fields for view synthesis,

    B. Mildenhall, P. P. Srinivasan, M. Tancik, J. T. Barron, R. Ramamoorthi, and R. Ng, “Nerf: Representing scenes as neural radiance fields for view synthesis,”Communications of the ACM, 2021

    2021

  27. [28]

    3d gaussian splatting for real-time radiance field rendering

    B. Kerbl, G. Kopanas, T. Leimk ¨uhler, G. Drettakiset al., “3d gaussian splatting for real-time radiance field rendering.”ACM Trans. Graph., vol. 42, no. 4, pp. 139–1, 2023

    2023

  28. [29]

    Nerf2: Neural radio-frequency radiance fields,

    X. Zhao, Z. An, Q. Pan, and L. Yang, “Nerf2: Neural radio-frequency radiance fields,” inProceedings of the 29th Annual International Con- ference on Mobile Computing and Networking, 2023

    2023

  29. [30]

    Newrf: A deep learning framework for wireless radiation field reconstruction and channel prediction,

    H. Lu, C. Vattheuer, B. Mirzasoleiman, and O. Abari, “Newrf: A deep learning framework for wireless radiation field reconstruction and channel prediction,”arXiv preprint arXiv:2403.03241, 2024

    work page arXiv 2024

  30. [31]

    Neural representation for wireless radiation field reconstruction: A 3d gaussian splatting approach,

    C. Wen, J. Tong, Y . Hu, Z. Lin, and J. Zhang, “Neural representation for wireless radiation field reconstruction: A 3d gaussian splatting approach,”IEEE Transactions on Wireless Communications, 2025

    2025

  31. [32]

    An image is worth 16x16 words: Trans- formers for image recognition at scale,

    A. Dosovitskiy, L. Beyer, A. Kolesnikov, D. Weissenborn, X. Zhai, T. Unterthiner, M. Dehghani, M. Minderer, G. Heigold, S. Gelly, J. Uszkoreit, and N. Houlsby, “An image is worth 16x16 words: Trans- formers for image recognition at scale,” in9th International Conference on Learning Representations, ICLR 2021, 2021

    2021

  32. [33]

    Bracewell,The Fourier Transform and its Applications, Tokyo

    R. Bracewell,The Fourier Transform and its Applications, Tokyo

  33. [34]

    Learning represen- tations by back-propagating errors,

    D. E. Rumelhart, G. E. Hinton, and R. J. Williams, “Learning represen- tations by back-propagating errors,”Nature, 1986

    1986

  34. [35]

    P. J. Huber,Robust Estimation of a Location Parameter, 1992

    1992

  35. [36]

    Radiodiff-3d: A 3d× 3d radio map dataset and generative diffusion based benchmark for 6g environment-aware communication,

    X. Wang, Q. Zhang, N. Cheng, J. Chen, Z. Zhang, Z. Li, S. Cui, and X. Shen, “Radiodiff-3d: A 3d× 3d radio map dataset and generative diffusion based benchmark for 6g environment-aware communication,” IEEE Transactions on Network Science and Engineering, 2026

    2026

  36. [37]

    Generative ai on spectrumnet: An open benchmark of multiband 3- d radio maps,

    S. Zhang, S. Jiang, W. Lin, Z. Fang, K. Liu, H. Zhang, and K. Chen, “Generative ai on spectrumnet: An open benchmark of multiband 3- d radio maps,”IEEE Transactions on Cognitive Communications and Networking, 2025

    2025

  37. [38]

    Image quality assessment: from error visibility to structural similarity,

    Z. Wang, A. C. Bovik, H. R. Sheikh, and E. P. Simoncelli, “Image quality assessment: from error visibility to structural similarity,”IEEE Trans. Image Process., 2004

    2004

  38. [39]

    Radiodiff: An effective generative diffusion model for sampling-free dynamic radio map construction,

    X. Wang, K. Tao, N. Cheng, Z. Yin, Z. Li, Y . Zhang, and X. Shen, “Radiodiff: An effective generative diffusion model for sampling-free dynamic radio map construction,”IEEE Transactions on Cognitive Communications and Networking, 2025

    2025

  39. [40]

    Physics-informed diffusion model for radio environment map reconstruction from sparse measurements,

    Z. Ye, Y . Shao, T. Fan, C. Zhang, F. Wu, F. Liu, W. Fan, B. Tang, and Y . Liu, “Physics-informed diffusion model for radio environment map reconstruction from sparse measurements,” in2025 30th Asia-Pacific Conference on Communications (APCC), 2025

    2025