Pith. sign in

REVIEW 4 major objections 5 minor 31 references

3D Gaussian splatting can build and update radio channel gain maps with sub-decibel accuracy, even from sparse new measurements.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · deepseek-v4-flash

2026-08-01 08:27 UTC pith:DMLV5564

load-bearing objection Solid 3DGS-based CGM construction and a genuinely useful incremental update for new blockers, but the dynamic-update claim runs ahead of the one-sided residual mechanism. the 4 major comments →

arxiv 2607.21099 v1 pith:DMLV5564 submitted 2026-07-23 cs.IT math.IT

Construction and Dynamic Update of Channel Gain Maps via 3D Gaussian Splatting

Yilong Chen , Yuan Guo , Juncong Zhou , Jie Xu , Rui Zhang This is my paper
classification cs.IT math.IT MSC 94A12
keywords channel knowledge mapschannel gain maps3D Gaussian splattingradio propagation modelingincremental learningenvironment-aware communicationswireless channel modelingray tracing
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

Channel knowledge maps let wireless networks predict coverage without costly pilot training, but constructing and refreshing them in dynamic environments remains hard. This paper argues that a grid-averaged channel gain can be decomposed into an incoherent sum of separable path powers, and exploits that structure to render maps with physics-informed 3D Gaussian splatting. It introduces the GS-CG model, where Gaussian primitives act as scatterers and attenuators, and an incremental update that freezes the static representation while adding a compact active set initialized from residuals. Simulations show sub-dB construction error and roughly 1-2.3 dB update error from only 4% new measurements, outperforming MLP-based, virtual-scatterer, and Kriging baselines.

Core claim

On its own terms, the paper claims that the propagation environment can be faithfully represented by a set of 3D Gaussian primitives whose geometry, opacity, and directional features map directly to the physical factors of a grid-level channel gain: distance-dependent attenuation, path transmittance, and effective scattering response. The central result is that grid-averaged channel gain equals the incoherent sum of direct and one-bounce scattering path powers, which turns CGM construction into a differentiable rendering problem. The same representation supports dynamic updates by freezing the reference Gaussians, introducing a small active set placed via residual-guided back-projection, and

What carries the argument

The GS-CG model is the central object: a set of learnable Gaussian primitives (center, scale, rotation, opacity, spherical-harmonic features) that render grid-averaged channel gain through explicit transmittance products along each propagation segment and one-bounce scattering gains using opacity, geometric support, and directional SH factors. The incremental mechanism freezes the static Gaussians and adds an active Gaussian set initialized along transmitter-receiver rays where the static model overestimates the measured gain, turning dynamic updating into a localized residual-learning problem.

Load-bearing premise

The whole additive path-power model rests on the assumption that distinct propagation paths arrive at a receiver grid from sufficiently separated directions, so their cross-terms average to zero; if a scene has closely spaced or unresolvable paths, the grid averaging no longer cancels interference and the map model is structurally invalid.

What would settle it

Simulate a scene where two propagation paths reach a receiver grid with nearly identical directions of arrival (small angular separation relative to the grid size), compute the exact phase-aware channel gain averaged over the grid, and compare it to the incoherent sum of the two path powers. If the difference exceeds the reported sub-dB accuracy, the model's foundation fails. The same test can be run indoors or in dense urban canyons with many overlapping paths.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

Share X Bluesky LinkedIn Reddit HN

If this is right

  • CKM construction becomes a fast, differentiable optimization that can start from coarse geometry and adaptively refine only where propagation structure is complex.
  • Dynamic CGM refresh from sparse samples (4% of grids) becomes practical, enabling real-time coverage awareness when vehicles or temporary obstacles alter the scene.
  • The learned Gaussian primitives localize to physical blockage and scattering regions, giving operators a degree of interpretability in why a map looks the way it does.
  • The MLP-GS variant trades a little accuracy for a roughly five-fold construction speed-up, so the approach can be tuned for resource-limited deployments.
  • The rendering structure extends naturally to different CGM granularities because the Gaussians are continuous and only the supervision grid changes.

Where Pith is reading between the lines

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

  • The same physics-informed splatting could be reused for other CKM families (beam index maps, angle maps) since the primitives capture geometry-dependent propagation structure rather than just scalar gain.
  • The residual-guided initialization targets attenuating blockers; using signed residuals could extend the method to handle newly appearing reflective scatterers that increase local gain.
  • The incoherent-path assumption predicts that accuracy will degrade in rich-scattering environments; incorporating path clustering or a residual interference term could be a testable extension.
  • The grid size relative to wavelength and the angular separation of paths give a principled criterion for when grid-based CGM is valid, which could be used to choose grid resolution across scenes.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

4 major / 5 minor

Summary. The paper proposes a physics-informed 3D Gaussian splatting (3DGS) framework for constructing grid-based channel gain maps (CGMs) and updating them in dynamic environments. The authors first derive a grid-averaged channel gain model that decomposes the gain into distance-dependent attenuation, path transmittance, and effective scattering contributions. They then represent the propagation environment with Gaussian primitives whose geometry, opacity, and directional SH features are mapped to these physical factors, yielding a differentiable rendering model (GS-CG). A low-complexity MLP variant (MLP-GS) is also introduced. For dynamic scenarios, the reference Gaussians are frozen and a compact set of active Gaussians is initialized by a residual-guided back-projection scheme and optimized on sparse new measurements. Simulations on three Sionna-RT scenes (campus, road, low-altitude) report sub-dB MAE for static construction (0.54--0.60 dB) and 1.07--2.26 dB MAE for dynamic updating using 4% new samples, outperforming MLP-RF, virtual scatterer, and Kriging baselines. The paper also provides an ablation study of the incremental design and a grid-resolution analysis.

Significance. If validated, this work offers a novel explicit and differentiable scene representation for CGM construction and updating, combining physical interpretability with the efficient optimization of 3DGS. The grid-level averaging formulation is practically motivated for large-scale deployments, and the incremental update mechanism is a meaningful step toward maintaining CKMs in changing environments. The paper is clearly written and the central static-construction derivation (Proposition 1) is standard. The authors provide a useful baseline comparison and an ablation that supports the importance of prior reuse, parameter freezing, and residual-guided initialization. However, the dynamic update mechanism is structurally one-sided and the experimental evidence is limited to single-run, blocker-addition scenarios with unspecified hyperparameters, which narrows the validity of the claimed scope.

major comments (4)
  1. [Section IV.B and Eq. (25)-(26), Algorithm 2] The incremental update mechanism is structurally one-sided. The residual e(p_r) in (25) is nonzero only when the static model overestimates the measured gain, and the active Gaussian rendering in (12) and (21) can only attenuate existing paths or add nonnegative scattering power; there is no mechanism to increase transmittance or reduce the opacity of reference Gaussians. Consequently, gain-increasing events such as the removal of a blocker or the appearance of a new reflector cannot be represented. All reported dynamic experiments (campus, road, low-altitude) are blocker-addition scenarios, consistent with Footnote 3's admission that the initialization mainly targets blockages along direct paths, but not with the abstract's and Section IV's promise of adaptation to moving 'blockers/scatterers' in general. The authors should either restrict the claims to attenuation-only changes or exten
  2. [Section VI.A and Section IV.B] The residual-guided active initialization depends on several threshold hyperparameters whose values are not specified: tau_theta, tau_min_e, tau_max_e, and jitter scale sigma in Eqs. (27)-(30). The simulation setup also omits the ADC frequency, pruning thresholds, and the convergence criterion for the active set. Without these values, the reported dynamic-updating results are not reproducible, and the reader cannot assess the sensitivity of the method to these choices. Please provide the exact settings or an ablation showing that the results are stable across a reasonable range.
  3. [Table I and Section VI.B] All performance numbers in Table I are single-run results without error bars, despite the presence of randomness in training (e.g., Gaussian initialization, jitter, Adam, adaptive density control). The claims of 'favorable accuracy-complexity tradeoff' and the moderate gaps between GS-CG and MLP-GS in some cases would be more convincing with repeated trials and standard deviations. The authors should report mean and standard deviation over multiple random seeds, or at least specify a fixed seed and justify the single-run comparison.
  4. [Proposition 1 and Appendix A, Eq. (39)] The incoherent sum decomposition in Eq. (6) relies on the condition that distinct paths have sufficiently separated DoAs so that the grid-averaged cross-correlation in (39) is negligible. The simulations are conducted in Sionna-RT scenes that are likely sparse in resolvable paths. The general claim that GS-CG 'accurately reconstructs grid-based CGMs' may not transfer to dense multipath environments (e.g., indoor or rich-scattering scenarios) where the cross terms survive. The authors should add a discussion of this validity range or include a synthetic test with closely spaced paths to quantify the breakdown of the approximation.
minor comments (5)
  1. [Algorithm 1] The 'Return Theta_s' statement appears inside the 'foreach' loop; it should be placed after loop termination.
  2. [Eq. (26)] If all residuals e(p_r) are zero, the normalized weight rho(p_r) is undefined. The algorithm should specify a fallback, e.g., uniform initialization or no active Gaussians.
  3. [Fig. 10] The ablation states that the variants yield a 'much larger MAE' but does not report numerical values. Please add a table or numbers in the caption for quantitative comparison.
  4. [Eq. (4)] The notation d^in_n and d^out_n is used without explicit definition in the text; clarify that these are the transmitter-to-scatterer and scatterer-to-receiver distances.
  5. [Table II] The wall-clock training times compare GPU-trained differentiable models (GS-CG, MLP-GS, MLP-RF) with CPU or non-differentiable baselines (VS, Kriging). Please specify the hardware and implementation details for the baselines, and note that the comparison is not entirely apples-to-apples.

Circularity Check

0 steps flagged

No significant circularity: the GS-CG rendering is trained on held-out grids and the physical decomposition is derived with stated assumptions; self-citations are not load-bearing.

full rationale

The paper's central derivation (Proposition 1, Eq. (6)) is an approximation of grid-averaged channel gain grounded in the field-response model and the assumption of sufficiently separated directions of arrival. The proof is provided in Appendix A, extending the 2D derivation of [26] to 3D, so it is not merely a restatement of a fitted parameter. The GS-CG model (Eqs. (11)-(12)) uses learnable Gaussian surrogates for the physical quantities in (7), and is trained on only 25% of grid measurements with evaluation over all grids, breaking the fitting loop. Dynamic updating (Algorithm 2) trains on 4% new measurements and is also evaluated on all grids; the residual-guided initialization (Eq. (25)) is a heuristic for placing active Gaussians, not a prediction forced by construction. The one-sided residual in Eq. (25) is a scope limitation (it cannot represent gain-increasing changes such as blocker removal), but that is a correctness/claim-width concern, not circularity. Self-citations to [4], [18], [26], and [27] are contextual references to prior work by overlapping authors, and none of them is invoked as a uniqueness theorem or as the sole justification of a central claim; the relevant mathematical derivation is presented in the paper itself. The model is benchmarked against MLP-RF, virtual scatterer, and Kriging baselines in Table I, providing external comparison independent of the fitted values. Overall, no step in the derivation reduces to its own inputs by definition, and no 'prediction' is statistically forced from a fitted parameter.

Axiom & Free-Parameter Ledger

4 free parameters · 5 axioms · 1 invented entities

The central claim rests on a small set of physical assumptions inherited from prior scatterer-based channel models, plus an ad hoc representation assumption that Gaussian primitives can surrogate arbitrary radio interactions. The paper contributes the rendering/update machinery, while the physical decomposition and simulator ground truth are taken from prior work.

free parameters (4)
  • Initial Gaussian count N (static construction) = 5000 (Section VI-A)
    Chosen by hand; controls model capacity for representing propagation structures. No sensitivity analysis is reported.
  • Active Gaussian count N_act (dynamic update) = 1000 (Section VI-A)
    Chosen by hand; controls the granularity of residual compensation for newly introduced blockers/scatterers.
  • Residual-guided update thresholds tau_theta, tau_min_e, tau_max_e, and jitter scale sigma = not reported
    Section IV-B depends on several hand-set thresholds and a jitter scale. Their values are not given, blocking exact reproduction.
  • SH degree L_SH and MLP latent feature dimension = L_SH=3; latent dim=16 (Section VI-A)
    Architecture capacity choices made without ablation or sensitivity study.
axioms (5)
  • domain assumption Distinct propagation paths have sufficiently separated directions of arrival so cross terms vanish in grid averaging.
    Used in Proposition 1 and Appendix A to reduce the squared channel magnitude to an incoherent sum of path powers. The paper footnotes that closely spaced DoAs should be merged.
  • domain assumption Grid size is much larger than the carrier wavelength and much smaller than propagation distances, so large-scale path parameters are invariant and phase varies linearly within the grid.
    Section II-A; required for the local plane-wave approximation and for defining grid-level averaged channel gain.
  • domain assumption Effective one-bounce scattering with free-space spherical spreading (path-loss exponent 2) per segment captures multi-bounce propagation, diffraction, and other effects.
    Section II-A and footnote 2; multi-bounce effects are absorbed into effective scatterers rather than explicitly modeled.
  • ad hoc to paper Gaussian primitives with opacity, spherical-harmonic features, and segment-overlap transmittance can serve as differentiable surrogates for arbitrary scatterers and blockers.
    Section III-A/C; no capacity theorem is given, and the surrogate is validated only on three simulated scenes.
  • domain assumption Sionna-RT ray-tracing output faithfully represents real wireless propagation for validation.
    Section VI-A; all ground-truth CGMs are generated by the Sionna simulator, not by field measurements.
invented entities (1)
  • GS-CG Gaussian primitives (latent radio-interaction structures) no independent evidence
    purpose: Represent effective scatterers and volumetric attenuators; map geometry, opacity, and SH features to scattering responses and transmittances.
    Introduced as a fitted surrogate for the unknown propagation environment. There is no falsifiable handle outside the fitted channel gains; the visualization in Fig. 7 is post-training interpretation.

pith-pipeline@v1.3.0-alltime-deepseek · 16172 in / 12220 out tokens · 118520 ms · 2026-08-01T08:27:35.581892+00:00 · methodology

0 comments
Cite this review

Pith. "Pith review of Construction and Dynamic Update of Channel Gain Maps via 3D Gaussian Splatting." pith.science (2026). https://pith.science/paper/DMLV5564

@misc{pith2026260721099,
  author       = {Pith},
  title        = {Pith review of: Construction and Dynamic Update of Channel Gain Maps via 3D Gaussian Splatting},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/DMLV5564}},
  note         = {Machine review of arXiv:2607.21099}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Channel knowledge maps (CKMs) have emerged as a promising technique for providing scene-specific and location-dependent propagation knowledge to enable environment-aware wireless network design. This paper investigates the construction and dynamic updating of a particular type of CKM, namely grid-based channel gain maps (CGMs), for large-scale networks using three-dimensional Gaussian splatting (3DGS). First, we formulate a grid-based channel gain model, where each map entry is defined as the locally averaged channel gain over a receiver grid, thereby suppressing phase-sensitive small-scale fluctuations. The resulting channel gain is decomposed into distance-dependent attenuation, path transmittance, and effective scattering contributions. Based on this decomposition, we develop a physics-informed Gaussian-splatting-based channel gain (GS-CG) model, which represents the propagation environment as a set of Gaussian primitives. The proposed model maps Gaussian geometry, opacity, and directional features to propagation-related factors and renders grid-level channel gains through a differentiable process. To accommodate real-time environmental changes, we further propose an incremental learning mechanism that updates a static reference GS-CG representation into a dynamic CGM. Specifically, the reference Gaussian primitives are frozen, while a compact set of tunable Gaussians is introduced to capture newly induced local channel-gain variations from sparse measurements.Numerical results demonstrate that the proposed GS-CG methods accurately reconstruct grid-based CGMs, efficiently adapt to dynamic environmental changes, and achieve a favorable accuracy-complexity tradeoff for fast CGM refinement.

Figures

Figures reproduced from arXiv: 2607.21099 by Jie Xu, Juncong Zhou, Rui Zhang, Yilong Chen, Yuan Guo.

Figure 1
Figure 1. Figure 1: Considered system and grid-based channel gain modeling. [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: GS-CG pipeline for static CGM construction. [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Incremental GS-CG pipeline for dynamic CGM updating. [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Test scenes for CGM construction and dynamic updating. [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: CGM construction results in the campus scene. [PITH_FULL_IMAGE:figures/full_fig_p010_5.png] view at source ↗
Figure 7
Figure 7. Figure 7: Learned Gaussian representation for the campus scene. [PITH_FULL_IMAGE:figures/full_fig_p010_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: CGM prediction results by GS-CG in the road scene. [PITH_FULL_IMAGE:figures/full_fig_p011_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: CGM prediction results by GS-CG in the low-altitude scene. [PITH_FULL_IMAGE:figures/full_fig_p011_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Ablation study on incremental GS-CG for CGM updating. [PITH_FULL_IMAGE:figures/full_fig_p012_10.png] view at source ↗

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

31 extracted references · 5 linked inside Pith

  1. [1]

    Toward environment-aware 6G communications via channel knowledge map,

    Y . Zeng and X. Xu, “Toward environment-aware 6G communications via channel knowledge map,”IEEE Wireless Commun., vol. 28, no. 3, Jun. 2021

    2021

  2. [2]

    A tutorial on environment-aware communications via channel knowledge map for 6G,

    Y . Zeng, J. Chen, J. Xu, D. Wu, X. Xu, S. Jin, X. Gao, D. Gesbert, S. Cui, and R. Zhang, “A tutorial on environment-aware communications via channel knowledge map for 6G,”IEEE Commun. Surveys Tut., vol. 26, no. 3, pp. 1478–1519, thirdquarter 2024

    2024

  3. [3]

    Channel gain map tracking via distributed kriging,

    E. D. Anese, S.-J. Kim, and G. B. Giannakis, “Channel gain map tracking via distributed kriging,”IEEE Trans. Veh. Technol., vol. 60, no. 3, pp. 1205–1211, Mar. 2011

    2011

  4. [4]

    Channel knowledge map construction: Recent advances and open challenges,

    Z. Ren, J. Zhou, J. Xu, L. Qiu, Y . Zeng, H. Hu, J. Zhang, and R. Zhang, “Channel knowledge map construction: Recent advances and open challenges,”IEEE Wireless Commun., pp. 1–7, Mar. 2026

    2026

  5. [5]

    Channel knowledge maps for 6G wireless networks: construction, applications, and future challenges,

    X. Liu, S. Sun, M. Tao, A. Kaushik, and H. Yan, “Channel knowledge maps for 6G wireless networks: construction, applications, and future challenges,”arXiv preprint arXiv:2505.24151, 2025

    Pith/arXiv arXiv 2025

  6. [6]

    An efficient radio map learning scheme based on kernel density function,

    Z. Xu, B. Huang, and B. Jia, “An efficient radio map learning scheme based on kernel density function,”IEEE Trans. Veh. Technol., vol. 70, no. 12, pp. 13 315–13 324, Dec. 2021

    2021

  7. [7]

    Propagation map reconstruction via interpolation assisted matrix completion,

    H. Sun and J. Chen, “Propagation map reconstruction via interpolation assisted matrix completion,”IEEE Trans. Signal Process., vol. 70, pp. 6154–6169, Dec. 2022

    2022

  8. [8]

    Site-specific propagation prediction for wireless in-building personal communication system design,

    S. Seidel and T. Rappaport, “Site-specific propagation prediction for wireless in-building personal communication system design,”IEEE Trans. Veh. Technol., vol. 43, no. 4, pp. 879–891, Nov. 1994. 6We assume that distinct paths have sufficiently separated local directions. If two paths have very similar DoAs, their cross-correlation may not vanish after sp...

    1994

  9. [9]

    COST 231 Hata adaptation model for urban conditions in LTE networks,

    R. V . Akhpashev and A. V . Andreev, “COST 231 Hata adaptation model for urban conditions in LTE networks,” inProc. 2016 17th International Conference of Young Specialists on Micro/Nanotechnologies and Elec- tron Devices (EDM), Jun. 2016, pp. 64–66

    2016

  10. [10]

    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, Jul. 2015

    2015

  11. [11]

    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. Wireless Commun., vol. 20, no. 6, pp. 4001–4015, Jun. 2021

    2021

  12. [12]

    RMTransformer: Accurate radio map construction and coverage prediction,

    Y . Li, C. Zhang, W. Wang, and Y . Huang, “RMTransformer: Accurate radio map construction and coverage prediction,” inProc. 2025 IEEE 101st Vehicular Technology Conference (VTC2025-Spring), Jun. 2025, pp. 1–5

    2025

  13. [13]

    CKMDiff: A generative diffusion model for CKM construction via inverse problems with learned priors,

    S. Fu, Y . Zeng, Z. Wu, D. Wu, S. Jin, C.-X. Wang, and X. Gao, “CKMDiff: A generative diffusion model for CKM construction via inverse problems with learned priors,”arXiv preprint arXiv:2504.17323, 2025

    Pith/arXiv arXiv 2025

  14. [14]

    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, Dec. 2021

    2021

  15. [15]

    3D Gaussian splatting for real-time radiance field rendering,

    B. Kerbl, G. Kopanas, T. Leimk ¨uhler, and G. Drettakis, “3D Gaussian splatting for real-time radiance field rendering,”ACM Transactions on Graphics, vol. 42, no. 4, Jul. 2023. [Online]. Available: https://repo-sam.inria.fr/fungraph/3d-gaussian-splatting/

    2023

  16. [16]

    NeRF2: Neural radio-frequency radiance fields,

    X. Zhao, Z. An, Q. Pan, and L. Yang, “NeRF2: Neural radio-frequency radiance fields,” inProc. the 29th Annual International Conference on Mobile Computing and Networking (ACM MobiCom ’23), Oct. 2023

    2023

  17. [17]

    WRF-GS: Wireless radiation field reconstruction with 3D Gaussian splatting,

    C. Wen, J. Tong, Y . Hu, Z. Lin, and J. Zhang, “WRF-GS: Wireless radiation field reconstruction with 3D Gaussian splatting,” inProc. IEEE INFOCOM 2025 - IEEE Conference on Computer Communications, May 2025, pp. 1–10

    2025

  18. [18]

    6D channel knowledge map construction via bidirectional wireless Gaussian splatting,

    J. Zhou, C. Hu, G. Wu, Z. Ren, H. Hu, J. Zhang, R. Zhang, and J. Xu, “6D channel knowledge map construction via bidirectional wireless Gaussian splatting,”arXiv preprint arXiv:2510.26166, 2025

    arXiv 2025

  19. [19]

    RF-3DGS: Wireless channel modeling with radio radiance field and 3D Gaussian splatting,

    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,”IEEE Trans. Wireless Commun., vol. 25, pp. 10 419–10 433, Jan. 2026

    2026

  20. [20]

    RF-PGS: Fully-structured spatial wireless channel representation with planar Gaussian splatting,

    L. Zhang, Z. Li, and H. Sun, “RF-PGS: Fully-structured spatial wireless channel representation with planar Gaussian splatting,”arXiv preprint arXiv:2508.16849, 2025

    Pith/arXiv arXiv 2025

  21. [21]

    Deformable 2D Gaussian splatting for efficient wireless radiance field rendering,

    M. Liu, C. Zhang, Q. Yang, Y . Cao, Y . Xu, Y . Xu, S. Sun, M. Dai, and Y . Guan, “Deformable 2D Gaussian splatting for efficient wireless radiance field rendering,”IEEE Trans. Vis. Comput. Graph., pp. 1–17, Apr. 2026

    2026

  22. [22]

    GSpaRC: Gaussian splatting for real-time reconstruc- tion of RF channels,

    B. S. Nukapotula, R. Tripathi, S. Pregler, D. Kalathil, S. Shakkottai, and T. S. Rappaport, “GSpaRC: Gaussian splatting for real-time reconstruc- tion of RF channels,”arXiv preprint arXiv:2511.22793, 2025

    Pith/arXiv arXiv 2025

  23. [23]

    A Gaussian splatting approach to continuous radio map construction,

    A. Chen, S. Mao, Z. Li, H. Zhang, D. Niyato, and Z. Han, “A Gaussian splatting approach to continuous radio map construction,” inProc. IEEE INFOCOM 2025 - IEEE Conference on Computer Communications Workshops (INFOCOM WKSHPS), May 2025, pp. 1–6

    2025

  24. [24]

    RadSplatter: Extend- ing 3D Gaussian splatting to radio frequencies for wireless radiomap extrapolation,

    Y . Wang, Y . Xue, S. Zhang, and T.-H. Chang, “RadSplatter: Extend- ing 3D Gaussian splatting to radio frequencies for wireless radiomap extrapolation,”arXiv preprint arXiv:2502.12686, 2025

    Pith/arXiv arXiv 2025

  25. [25]

    Wideband RF radiance field modeling using frequency- embedded 3D Gaussian splatting,

    Z. Li, L. Yang, Y . Bian, H. Pan, Y . Fu, Y . Wang, Z. Chen, Y .-C. Chen, and G. Xue, “Wideband RF radiance field modeling using frequency- embedded 3D Gaussian splatting,”arXiv preprint arXiv:2505.20714, 2025

    arXiv 2025

  26. [26]

    Channel gain map estimation for wireless networks based on scatterer model,

    H. Sun, L. Zhu, and R. Zhang, “Channel gain map estimation for wireless networks based on scatterer model,”IEEE Trans. Wireless Commun., vol. 24, no. 8, pp. 7012–7028, Aug. 2025

    2025

  27. [27]

    Channel gain map estimation based on 3-D virtual scatterer model,

    H. Sun, L. Zhu, J. Xu, and R. Zhang, “Channel gain map estimation based on 3-D virtual scatterer model,”IEEE Trans. Wireless Commun., vol. 25, pp. 15 741–15 757, Apr. 2026

    2026

  28. [28]

    Goldsmith,Wireless Communications

    A. Goldsmith,Wireless Communications. Cambridge, U.K.: Cambridge University Press, 2005

    2005

  29. [29]

    Planet dump retrieved from https://planet.osm.org,

    OpenStreetMap contributors, “Planet dump retrieved from https://planet.osm.org,” https://www.openstreetmap.org, 2017

    2017

  30. [30]

    Hoydis, S

    J. Hoydis, S. Cammerer, F. Ait Aoudia, M. Nimier-David, L. Maggi, G. Marcus, A. Vem, and A. Keller, “Sionna,” https://nvlabs.github.io/ sionna/, 2022

    2022

  31. [31]

    Kriging: a method of interpolation for ge- ographical information systems,

    M. A. Oliver and R. Webster, “Kriging: a method of interpolation for ge- ographical information systems,”International Journal of Geographical Information Systems, vol. 4, no. 3, pp. 313–332, 1990

    1990