pith. sign in

arxiv: 2604.23310 · v1 · submitted 2026-04-25 · 💻 cs.NI

RadTwin: Generalizable Wireless Digital Twin for Dynamic Environments

Yuru Zhang , Ming Zhao , Qiang Liu , Ahmed Alkhateeb , Abhishek K. Agrawal , Qi Qu This is my paper

Pith reviewed 2026-05-08 07:11 UTC · model grok-4.3

classification 💻 cs.NI
keywords wireless digital twinradio propagation modelingdynamic environmentspoint cloud conditioningphysics-informed attentiongeneralizable neural networksindoor wireless simulation
0
0 comments X

The pith

RadTwin models radio propagation in dynamic indoor scenes by conditioning neural networks directly on point-cloud geometry, allowing adaptation to new layouts without retraining.

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

The paper seeks to build a wireless digital twin capable of predicting radio signal behavior in environments that change over time, such as rooms with rearranged furniture. Traditional ray tracing requires detailed 3D models while neural radiance field methods tie representations to fixed static scenes and demand retraining on any change. RadTwin extracts high-level latent features from point clouds, generates physics-informed sparse attention masks via electromagnetic ray tracing to focus on signal-contributing voxels, and uses a neural decoder with masked cross-attention to learn propagation patterns. A sympathetic reader would care because this removes the need for scene-specific retraining, opening the door to practical, updatable digital twins for wireless networks. The reported gains in image quality metrics on varying indoor datasets support the claim of improved generalization and data efficiency.

Core claim

RadTwin is a generalizable wireless digital twin framework that explicitly conditions on scene geometry extracted from point clouds. It consists of a scenario representation network for latent scene features, an electromagnetic ray tracing module that produces physics-informed sparse attention masks for relevant voxels, and a neural propagation decoder that aggregates features through masked cross-attention. On a dataset of indoor scenes with varying furniture arrangements, this yields 31.6% higher SSIM and 91.96% lower LPIPS than NeRF2 while showing superior cross-scale performance, generalization, and data efficiency.

What carries the argument

The electromagnetic ray tracing module that computes physics-informed sparse attention masks identifying voxels physically contributing to signals toward each query direction, which then guide masked cross-attention in the neural propagation decoder.

If this is right

  • RadTwin can predict radio propagation for new furniture arrangements in the same room without any retraining.
  • The framework maintains higher structural similarity and lower perceptual error than NeRF2 baselines across tested indoor variations.
  • Cross-scale generalization improves because conditioning is tied to geometry rather than fixed scene representations.
  • Data efficiency rises since the model learns general propagation rules instead of memorizing specific environments.

Where Pith is reading between the lines

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

  • The method could extend to outdoor or multi-room settings if accurate point clouds from LiDAR or depth sensors are available in real time.
  • Integration with live sensing hardware might enable continuously updating digital twins for network optimization in changing spaces.
  • Similar physics-informed masking could apply to other wave phenomena such as acoustics or light transport in dynamic scenes.
  • Reducing reliance on full 3D material models might lower the cost of creating digital twins for large-scale wireless planning.

Load-bearing premise

High-level latent features from point clouds plus physics-informed sparse attention masks capture all relevant radio-propagation effects across arbitrary dynamic changes without retraining or scene-specific fine-tuning.

What would settle it

A new indoor scene where point clouds omit material properties or small objects, yet measured radio signal strength or multipath patterns deviate sharply from RadTwin predictions while matching a method that includes those details.

Figures

Figures reproduced from arXiv: 2604.23310 by Abhishek K. Agrawal, Ahmed Alkhateeb, Ming Zhao, Qiang Liu, Qi Qu, Yuru Zhang.

Figure 1
Figure 1. Figure 1: Impact of dynamic scene changes on radio propagation. view at source ↗
Figure 2
Figure 2. Figure 2: Illustration of spatial spectrum. When the RX employs a directional antenna, it can selec￾tively receive signals from a specific direction d = (θ, φ), where θ ∈ [0◦ , 360◦ ) denotes the azimuth angle and φ ∈ [0◦ , 180◦ ] denotes the elevation angle, as shown in view at source ↗
Figure 3
Figure 3. Figure 3: An overview of the RadTwin framework. RadTwin consists of three main components. The point cloud of a 3D scene is first processed by the scenario representation network, which partitions space into a voxel grid and extracts voxel-wise and global features. Given an RX query, the electromagnetic ray tracing module computes LOS voxel maps to generate sparse attention masks. Finally, the neural propagation dec… view at source ↗
Figure 4
Figure 4. Figure 4: Architecture of Scenario Representation Network. view at source ↗
Figure 5
Figure 5. Figure 5: Architecture of Electromagnetic Ray Tracing Module. view at source ↗
Figure 6
Figure 6. Figure 6: Indoor office scene for dataset generation (small size). view at source ↗
Figure 7
Figure 7. Figure 7: Path loss variation across directions for two small scenes. 1 2 3 4 5 6 7 8 9 10 11 12 Scene Index 50 75 100 125 150 175 200 Path Loss (dB) view at source ↗
Figure 9
Figure 9. Figure 9: Synthesis of spatial spectrums. Results at eight RX positions across two test scenes. RadTwin accurately captures both dominant propagation paths and fine-grained multipath structures, while NeRF2 exhibits noticeable discrepancies in detailed patterns. 0.0 0.2 0.4 0.6 0.8 SSIM 0.0 0.2 0.4 0.6 0.8 1.0 CDF RadTwin MLP NeRF 2 view at source ↗
Figure 10
Figure 10. Figure 10: CDF of SSIM values for spatial spectrum synthesis. 0.0 0.2 0.4 0.6 0.8 LPIPS 0.0 0.2 0.4 0.6 0.8 1.0 CDF RadTwin MLP NeRF 2 view at source ↗
Figure 13
Figure 13. Figure 13: shows the SNR CDF curves for each scene size. The small scene achieves the highest median SNR of 11.36 dB due to its simpler propagation environment with fewer multipath components. As scene size increases, the propagation envi￾ronment becomes more complex with longer path lengths and more potential reflectors. The performance gap also reflects the decreasing spatial density of training samples, as the sa… view at source ↗
read the original abstract

Precisely modeling radio propagation in dynamic wireless environments is fundamental to the realization of wireless digital twins. Traditional ray tracing methods rely on accurate 3D models with detailed environment parameters, while recent neural radiance field approaches learn representations tied to specific static scenes, requiring retraining when environments change. In this paper, we propose RadTwin, a generalizable wireless digital twin framework that explicitly conditions on scene geometry, enabling adaptation to dynamic environments without retraining. RadTwin comprises three key components: 1) a scenario representation network that extracts high-level latent scene features from point clouds, 2) an electromagnetic ray tracing module that computes physics-informed sparse attention masks identifying voxels that physically contribute signals toward each query direction, and 3) a neural propagation decoder that aggregates relevant scene features through masked cross-attention to learn how radio propagation behaves within the given scene geometry. We evaluate RadTwin on a customized dataset of indoor scenes with varying furniture arrangements. Experimental results show that RadTwin achieves 31.6% higher SSIM (0.846 vs. 0.643) and 91.96% lower LPIPS (0.023 vs. 0.286) compared to NeRF2. RadTwin further demonstrates superior cross-scale performance and high generalization and data efficiency, representing a significant advancement toward practical digital network twins for dynamic wireless environments.

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 / 3 minor

Summary. The paper proposes RadTwin, a generalizable wireless digital twin for radio propagation modeling in dynamic environments. It conditions on scene geometry via a scenario representation network that extracts high-level latent features from point clouds, an electromagnetic ray tracing module that generates physics-informed sparse attention masks from geometric voxel-ray intersections, and a neural propagation decoder that uses masked cross-attention to predict propagation behavior. On a customized indoor dataset of scenes with varying furniture arrangements, RadTwin reports 31.6% higher SSIM (0.846 vs. 0.643) and 91.96% lower LPIPS (0.023 vs. 0.286) than NeRF2, plus superior cross-scale performance, generalization, and data efficiency without retraining.

Significance. If the results hold under rigorous validation, the work offers a concrete advance toward practical wireless digital twins by enabling geometry-conditioned adaptation to dynamic scenes without per-scene retraining. The explicit incorporation of ray-tracing-derived attention masks to guide neural decoding is a strength that grounds the model in propagation physics while retaining flexibility; the reported perceptual metric gains suggest improved fidelity over pure neural baselines for indoor wireless modeling tasks.

major comments (3)
  1. [§3 (scenario representation network and EM ray tracing module)] The scenario representation network (described in the three-component overview and §3) takes only point-cloud geometry as input. Radio propagation depends on surface electromagnetic parameters (permittivity, conductivity, roughness) to compute reflection/transmission coefficients; these are absent, so the decoder can at best learn material-specific behaviors implicit in the training scenes. This directly undermines the central no-retraining generalization claim for arbitrary dynamic changes that introduce new materials or material combinations.
  2. [Evaluation section / abstract results paragraph] The evaluation reports concrete metric improvements but provides no information on dataset size, number of distinct scenes, training/validation splits, number of runs for statistical significance, or whether the NeRF2 baseline received identical geometry inputs. Without these details the 31.6% SSIM and 91.96% LPIPS gains cannot be assessed for robustness or fair comparison.
  3. [EM ray tracing module description] The sparse attention masks are generated solely from geometric voxel intersections with query rays. While this injects some physics, it omits material-dependent effects (absorption, diffuse scattering) that alter path loss and multipath structure; the decoder must therefore compensate implicitly, limiting reliability when furniture arrangements alter surface properties even if geometry is provided.
minor comments (3)
  1. [§3.1] Clarify the precise point-cloud format (e.g., density, coordinate system, inclusion of normals) and the dimensionality of the extracted latent features.
  2. [Experimental setup] Add explicit statements on whether all compared methods were given the same point-cloud input or whether NeRF2 operated under different conditioning.
  3. [Figures] Ensure figure captions fully describe axis scales, color mappings, and what each sub-figure visualizes (e.g., predicted vs. ground-truth power maps).

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive and detailed feedback on RadTwin. We address each major comment point by point below, providing clarifications on the model's scope and committing to revisions that strengthen the manuscript without misrepresenting its contributions.

read point-by-point responses

  1. Referee: [§3 (scenario representation network and EM ray tracing module)] The scenario representation network (described in the three-component overview and §3) takes only point-cloud geometry as input. Radio propagation depends on surface electromagnetic parameters (permittivity, conductivity, roughness) to compute reflection/transmission coefficients; these are absent, so the decoder can at best learn material-specific behaviors implicit in the training scenes. This directly undermines the central no-retraining generalization claim for arbitrary dynamic changes that introduce new materials or material combinations.

    Authors: We appreciate this observation on the model's inputs. RadTwin is explicitly designed for generalization to dynamic environments through geometry conditioning via point clouds, with electromagnetic material properties learned implicitly from the training scenes in our customized indoor dataset (where furniture arrangements vary but material characteristics remain consistent). The no-retraining claim pertains to geometric changes under these conditions, not arbitrary introduction of new materials. We will revise the manuscript in §3 and the discussion section to explicitly state this assumption and acknowledge the limitation for scenarios with varying surface properties, thereby clarifying the scope of our generalization results. revision: partial

  2. Referee: [Evaluation section / abstract results paragraph] The evaluation reports concrete metric improvements but provides no information on dataset size, number of distinct scenes, training/validation splits, number of runs for statistical significance, or whether the NeRF2 baseline received identical geometry inputs. Without these details the 31.6% SSIM and 91.96% LPIPS gains cannot be assessed for robustness or fair comparison.

    Authors: We agree that these experimental details are required for a complete assessment of robustness and fairness. In the revised manuscript, we will expand the evaluation section (and update the abstract results paragraph if space permits) to report the total number of scenes and samples, the training/validation/test splits, the number of independent runs with mean and standard deviation, and explicit confirmation that NeRF2 and other baselines received identical point-cloud geometry inputs. This addition will directly address the concern and improve the rigor of the evaluation. revision: yes

  3. Referee: [EM ray tracing module description] The sparse attention masks are generated solely from geometric voxel intersections with query rays. While this injects some physics, it omits material-dependent effects (absorption, diffuse scattering) that alter path loss and multipath structure; the decoder must therefore compensate implicitly, limiting reliability when furniture arrangements alter surface properties even if geometry is provided.

    Authors: This point is related to the first comment. The electromagnetic ray tracing module generates sparse attention masks from geometric voxel-ray intersections to inject physics-based guidance on contributing paths, while material-dependent effects such as absorption are handled implicitly by the neural propagation decoder trained on our dataset. We will revise the description of the EM ray tracing module and add a limitations paragraph to note that this design assumes fixed material properties and may not fully capture changes in surface characteristics; we will also outline potential future extensions (e.g., material parameter embeddings) to broaden applicability while retaining the current geometry-focused strengths. revision: partial

Circularity Check

0 steps flagged

No circularity: architecture and claims are independently motivated and empirically validated

full rationale

The paper defines RadTwin via three explicit components—point-cloud feature extraction, geometric ray-tracing masks, and a masked cross-attention decoder—none of which are defined in terms of the radio-propagation outputs they produce. Performance numbers (SSIM, LPIPS) are reported from direct comparison against NeRF2 on a held-out dataset of furniture rearrangements; they are not obtained by fitting a parameter to the same quantity and relabeling it a prediction. No equations, uniqueness theorems, or self-citations are invoked to force the architecture or the generalization claim. The derivation chain therefore remains self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

Based on the abstract alone, the framework rests on standard neural-network training assumptions and the domain premise that point-cloud geometry plus ray-tracing masks capture the dominant propagation physics; no explicit free parameters, new physical entities, or ad-hoc axioms are stated.

axioms (1)
  • domain assumption Point clouds of indoor scenes contain sufficient geometric information to determine radio propagation behavior when combined with physics-informed attention.
    Invoked in the description of the scenario representation network and electromagnetic ray tracing module.

pith-pipeline@v0.9.0 · 5551 in / 1399 out tokens · 47171 ms · 2026-05-08T07:11:03.935014+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

32 extracted references · 3 canonical work pages

  1. [1]

    Digital twin of wireless systems: Overview, taxonomy, challenges, and opportunities,

    L. U. Khan, Z. Han, W. Saad, E. Hossain, M. Guizani, and C. S. Hong, “Digital twin of wireless systems: Overview, taxonomy, challenges, and opportunities,”IEEE Communications Surveys & Tutorials, vol. 24, no. 4, pp. 2230–2254, 2022

    2022

  2. [2]

    Digital twin networks: A survey,

    Y . Wu, K. Zhang, and Y . Zhang, “Digital twin networks: A survey,”IEEE Internet of Things Journal, vol. 8, no. 18, pp. 13 789–13 804, 2021

    2021

  3. [3]

    Network digital twin towards networking, telecommunica- tions, and traffic engineering: A survey,

    R. Poorzare, D. N. Kanellopoulos, V . K. Sharma, P. Dalapati, and O. P. Waldhorst, “Network digital twin towards networking, telecommunica- tions, and traffic engineering: A survey,”IEEE Access, 2025

    2025

  4. [4]

    Network digital twin: Context, enabling technologies, and opportuni- ties,

    P. Almasan, M. Ferriol-Galm ´es, J. Paillisse, J. Su ´arez-Varela, D. Perino, D. L ´opez, A. A. P. Perales, P. Harvey, L. Ciavaglia, L. Wonget al., “Network digital twin: Context, enabling technologies, and opportuni- ties,”IEEE Communications Magazine, vol. 60, no. 11, pp. 22–27, 2022

    2022

  5. [5]

    Digital twin-empowered intelligent computation offloading for edge computing in the era of 5g and beyond: A state-of-the-art survey,

    H. Tran-Dang and D.-S. Kim, “Digital twin-empowered intelligent computation offloading for edge computing in the era of 5g and beyond: A state-of-the-art survey,”ICT Express, 2025

    2025

  6. [6]

    Digital twin enhanced multi- agent reinforcement learning for large-scale mobile network coverage optimization,

    H. Liu, W. Su, T. Li, W. Huang, and Y . Li, “Digital twin enhanced multi- agent reinforcement learning for large-scale mobile network coverage optimization,”ACM Transactions on Knowledge Discovery from Data, vol. 19, no. 1, pp. 1–23, 2024

    2024

  7. [7]

    Digital twin channel for 6g: Concepts, architectures and potential applications,

    H. Wang, J. Zhang, G. Nie, L. Yu, Z. Yuan, T. Li, J. Wang, and G. Liu, “Digital twin channel for 6g: Concepts, architectures and potential applications,”IEEE Communications Magazine, 2024

    2024

  8. [8]

    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, vol. 3, pp. 1089–1100, 2015

    2015

  9. [9]

    Radiotwin: A digital building material twin for wideband, cross-link, cross-band wireless channel prediction,

    Z. An, L. Shangguan, J. Kaewell, P. Pietraski, and K. Jamieson, “Radiotwin: A digital building material twin for wideband, cross-link, cross-band wireless channel prediction,” in2025 IEEE International Symposium on Dynamic Spectrum Access Networks (DySPAN). IEEE, 2025, pp. 1–10

    2025

  10. [10]

    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, vol. 65, no. 1, pp. 99–106, 2021

    2021

  11. [11]

    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, pp. 1–15

    2023

  12. [12]

    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

  13. [13]

    Differentiable monte carlo ray tracing through edge sampling,

    T.-M. Li, M. Aittala, F. Durand, and J. Lehtinen, “Differentiable monte carlo ray tracing through edge sampling,”ACM Transactions on Graph- ics (TOG), vol. 37, no. 6, pp. 1–11, 2018

    2018

  14. [14]

    Learnable wireless digital twins: Reconstructing electromagnetic field with neural representations,

    S. Jiang, Q. Qu, X. Pan, A. Agrawal, R. Newcombe, and A. Alkhateeb, “Learnable wireless digital twins: Reconstructing electromagnetic field with neural representations,”IEEE Open Journal of the Communications Society, 2025

    2025

  15. [15]

    Deepsdf: Learning continuous signed distance functions for shape rep- resentation,

    J. J. Park, P. Florence, J. Straub, R. Newcombe, and S. Lovegrove, “Deepsdf: Learning continuous signed distance functions for shape rep- resentation,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2019, pp. 165–174

    2019

  16. [16]

    Deep point cloud reconstruction,

    J. Choe, B. Joung, F. Rameau, J. Park, and I. S. Kweon, “Deep point cloud reconstruction,”arXiv preprint arXiv:2111.11704, 2021

    work page arXiv 2021

  17. [17]

    Pointpronets: Consolidation of point clouds with convolutional neural networks,

    R. Roveri, A. C. ¨Oztireli, I. Pandele, and M. Gross, “Pointpronets: Consolidation of point clouds with convolutional neural networks,” in Computer Graphics Forum, vol. 37, no. 2. Wiley Online Library, 2018, pp. 87–99

    2018

  18. [18]

    Pointcleannet: Learning to denoise and remove outliers from dense point clouds,

    M.-J. Rakotosaona, V . La Barbera, P. Guerrero, N. J. Mitra, and M. Ovsjanikov, “Pointcleannet: Learning to denoise and remove outliers from dense point clouds,” inComputer graphics forum, vol. 39, no. 1. Wiley Online Library, 2020, pp. 185–203

    2020

  19. [19]

    Score-based point cloud denoising,

    S. Luo and W. Hu, “Score-based point cloud denoising,” inProceedings of the IEEE/CVF international conference on computer vision, 2021, pp. 4583–4592

    2021

  20. [20]

    Pmp-net: Point cloud completion by learning multi-step point moving paths,

    X. Wen, P. Xiang, Z. Han, Y .-P. Cao, P. Wan, W. Zheng, and Y .-S. Liu, “Pmp-net: Point cloud completion by learning multi-step point moving paths,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2021, pp. 7443–7452

    2021

  21. [21]

    Snowflakenet: Point cloud completion by snowflake point deconvolution with skip-transformer,

    P. Xiang, X. Wen, Y .-S. Liu, Y .-P. Cao, P. Wan, W. Zheng, and Z. Han, “Snowflakenet: Point cloud completion by snowflake point deconvolution with skip-transformer,” inProceedings of the IEEE/CVF international conference on computer vision, 2021, pp. 5499–5509

    2021

  22. [22]

    V oxelnet: End-to-end learning for point cloud based 3d object detection,

    Y . Zhou and O. Tuzel, “V oxelnet: End-to-end learning for point cloud based 3d object detection,” inProceedings of the IEEE conference on computer vision and pattern recognition, 2018, pp. 4490–4499

    2018

  23. [23]

    An efficient and robust ray-box intersection algorithm,

    A. Williams, S. Barrus, R. K. Morley, and P. Shirley, “An efficient and robust ray-box intersection algorithm,” inACM SIGGRAPH 2005 Courses, 2005, pp. 9–es

    2005

  24. [24]

    A survey of various propagation models for mobile communication,

    T. K. Sarkar, Z. Ji, K. Kim, A. Medouri, and M. Salazar-Palma, “A survey of various propagation models for mobile communication,”IEEE Antennas and propagation Magazine, vol. 45, no. 3, pp. 51–82, 2003

    2003

  25. [25]

    Empirical formula for propagation loss in land mobile radio services,

    M. Hata, “Empirical formula for propagation loss in land mobile radio services,”IEEE transactions on Vehicular Technology, vol. 29, no. 3, pp. 317–325, 2013

    2013

  26. [26]

    The cost 2100 mimo channel model,

    L. Liu, C. Oestges, J. Poutanen, K. Haneda, P. Vainikainen, F. Quitin, F. Tufvesson, and P. De Doncker, “The cost 2100 mimo channel model,” IEEE Wireless Communications, vol. 19, no. 6, pp. 92–99, 2012

    2012

  27. [27]

    3gpp tr 38.901 channel model,

    Q. Zhu, C.-X. Wang, B. Hua, K. Mao, S. Jiang, and M. Yao, “3gpp tr 38.901 channel model,” inthe wiley 5G Ref: the essential 5G reference online. Wiley Press, 2021, pp. 1–35

    2021

  28. [28]

    Guidelines for evaluation of radio interface technologies for imt-advanced,

    M. Series, “Guidelines for evaluation of radio interface technologies for imt-advanced,”Report ITU, vol. 638, no. 31, 2009

    2009

  29. [29]

    Sionna rt: Differentiable ray tracing for radio propagation modeling,

    J. Hoydis, F. A ¨ıt Aoudia, S. Cammerer, M. Nimier-David, N. Binder, G. Marcus, and A. Keller, “Sionna rt: Differentiable ray tracing for radio propagation modeling,” in2023 IEEE Globecom Workshops (GC Wkshps). IEEE, 2023, pp. 317–321

    2023

  30. [30]

    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 Transactions on Wireless Communications, vol. 20, no. 6, pp. 4001–4015, 2021

    2021

  31. [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

  32. [32]

    RF-3DGS: Wireless Channel Modeling with Radio Radiance Field and 3D Gaussian Splat- ting.arXiv preprint arXiv:2411.19420, 2024

    L. Zhang, H. Sun, S. Berweger, C. Gentile, and R. Q. Hu, “Rf-3dgs: Wireless channel modeling with radio radiance field and 3d gaussian splatting,”arXiv preprint arXiv:2411.19420, 2024

    work page arXiv 2024