The reviewed record of science sign in
Pith

arxiv: 2606.03531 · v1 · pith:GGN4LUAN · submitted 2026-06-02 · eess.SP

Voxel-CKM: Voxelized Radio Frequency Radiance Fields for Fast and Few-Shot CKM Construction

Hanlei Li , Guangyi Zhang , Kequan Zhou , Yunlong Cai , Guanding Yu This is my paper

Reviewed by Pith2026-06-28 09:10 UTCgrok-4.3pith:GGN4LUANopen to challenge →

classification eess.SP
keywords channel knowledge mapsradio frequency radiance fieldsvoxel gridsfew-shot learningwireless channel predictionvector-matrix decompositiontotal variation regularization
0
0 comments X

The pith

Voxel grids with vector-matrix decomposition build channel knowledge maps faster from sparse measurements.

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

The paper introduces Voxel-CKM to cut the hours-to-days training time and dense data needs of prior channel knowledge map methods. It swaps implicit neural representations for explicit voxel grids that directly encode spatial channel variation. These grids are parameterized via a compact vector-matrix decomposition that speeds convergence, while a transmitter prior supplies inductive bias for learning from few samples and total-variation regularization curbs overfitting. If correct, the method lowers the measurement and compute cost of obtaining reliable CSI predictions from user locations.

Core claim

Voxel-CKM constructs channel knowledge maps by replacing implicit neural radiance fields with explicit voxel grids that are parameterized through vector-matrix decomposition, guided by a transmitter prior for sparse-data learning and stabilized by total-variation regularization.

What carries the argument

Explicit voxel grids parameterized by vector-matrix decomposition that capture spatial variation of wireless channels, augmented by a transmitter prior as inductive bias.

If this is right

  • CKM training converges substantially faster than implicit neural methods.
  • Prediction accuracy improves when only a small fraction of possible measurements is available.
  • Overfitting on limited data is reduced by the added regularization term.
  • Overall deployment cost for CKM-based CSI acquisition drops because fewer measurements and shorter training suffice.

Where Pith is reading between the lines

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

  • The explicit grid representation could allow direct fusion with geometric ray-tracing outputs or other physics-based priors not tested in the paper.
  • Because the parameterization is compact, the same grids might support incremental updates when the environment changes, enabling online CKM adaptation.
  • The few-shot regime gains might transfer to related location-based prediction tasks such as coverage mapping or interference estimation.

Load-bearing premise

Explicit voxel grids with vector-matrix decomposition can efficiently capture the spatial variation of wireless channels, and incorporating a transmitter prior as inductive bias enables effective learning from sparse measurements.

What would settle it

A controlled experiment that measures wall-clock training time and prediction error on identical sparse measurement sets, checking whether Voxel-CKM reaches target accuracy in minutes instead of hours while outperforming implicit baselines.

Figures

Figures reproduced from arXiv: 2606.03531 by Guanding Yu, Guangyi Zhang, Hanlei Li, Kequan Zhou, Yunlong Cai.

Figure 1
Figure 1. Figure 1: Overview of the voxelized RF radiance field. The wireless environ [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Pipeline of the Voxel-CKM framework. The framework builds on a voxelized RF radiance field and integrates a VM decomposition, a transmitter-aware [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Visualization of the learned voxel density field. A slice plane is [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Architecture of the proposed lightweight decoder. [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: The geometries of three indoor 3D environment models. [PITH_FULL_IMAGE:figures/full_fig_p007_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Comparison of training convergence curves between Voxel-CKM and NeWRF under different sampling budgets: (a) [PITH_FULL_IMAGE:figures/full_fig_p009_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Visualization of learned CKMs from different models. The first column presents the top-down floor plans of two indoor environments. The second [PITH_FULL_IMAGE:figures/full_fig_p012_7.png] view at source ↗
read the original abstract

Channel knowledge maps (CKMs) are designed to predict channel state information (CSI) from user locations, thereby enabling low-overhead CSI acquisition. However, existing CKM construction methods often require hours-to-days of training time and dense measurements, resulting in substantial deployment cost. In this paper, we propose Voxel-CKM, a novel voxelized radio frequency (RF) radiance field framework for fast and few-shot CKM construction. The core idea is to replace implicit neural representations with explicit voxel grids to efficiently capture the spatial variation of wireless channels. Building upon this, we further introduce a compact vector-matrix (VM) decomposition to parameterize these voxel grids using a small set of matrices and vectors, which significantly accelerates convergence and facilitates fast CKM construction. To enable few-shot learning, we incorporate a transmitter prior as an inductive bias to guide the learning process under sparse measurements. Additionally, a total-variation (TV) regularization loss is proposed to mitigate overfitting and stabilize optimization. Experiments show that Voxel-CKM substantially accelerates training convergence and improves performance in the few-shot regime.

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 / 1 minor

Summary. The paper proposes Voxel-CKM, a voxelized RF radiance field framework that replaces implicit neural representations with explicit voxel grids parameterized via vector-matrix (VM) decomposition. It adds a transmitter location prior as inductive bias and a total-variation (TV) regularization loss to support fast convergence and learning from sparse measurements for channel knowledge map (CKM) construction.

Significance. If the empirical claims hold, the method could meaningfully reduce the training time and measurement density required for CKMs, lowering deployment costs relative to existing implicit approaches. The explicit representation plus targeted inductive biases directly target known pain points in radiance-field-style CKM work.

major comments (1)
  1. [Abstract] Abstract: the central claim that 'Experiments show that Voxel-CKM substantially accelerates training convergence and improves performance in the few-shot regime' is unsupported by any quantitative results, baselines, error metrics, dataset sizes, or experimental protocol. Without these details the primary contribution cannot be evaluated.
minor comments (1)
  1. The manuscript should define 'few-shot' and 'fast' with concrete numbers (e.g., number of measurements or training iterations) rather than qualitative statements.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the positive evaluation of Voxel-CKM's potential impact and for the constructive comment on the abstract. We address the point below.

read point-by-point responses

  1. Referee: Abstract: the central claim that 'Experiments show that Voxel-CKM substantially accelerates training convergence and improves performance in the few-shot regime' is unsupported by any quantitative results, baselines, error metrics, dataset sizes, or experimental protocol. Without these details the primary contribution cannot be evaluated.

    Authors: We agree that the abstract would be strengthened by including concrete quantitative support for the claims. The full manuscript (Sections IV and V) reports the supporting experiments, including NMSE and training-time comparisons against implicit NeRF baselines, specific dataset sizes (e.g., number of measurement locations), few-shot regimes (e.g., 5-20% sampling density), and the exact evaluation protocol. To address the referee's concern directly, we will revise the abstract to incorporate key quantitative highlights such as training-time reduction factors and NMSE improvements under the reported few-shot conditions. This change will be made in the next version. revision: yes

Circularity Check

0 steps flagged

No significant circularity detected

full rationale

The abstract and high-level description introduce voxel grids with VM decomposition, a transmitter prior as inductive bias, and TV regularization as design choices targeting efficiency and few-shot learning. No equations, derivations, or load-bearing steps are presented that reduce by construction to fitted inputs, self-definitions, or self-citation chains. The central claims are framed as empirical outcomes from experiments rather than theoretical guarantees derived from the method itself. The approach builds on external concepts (neural radiance fields) without internal circular reductions.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Only the abstract is available, which does not detail any free parameters, axioms, or invented entities.

pith-pipeline@v0.9.1-grok · 5736 in / 1118 out tokens · 35667 ms · 2026-06-28T09:10:13.842789+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

43 extracted references · 1 canonical work pages

  1. [1]

    Communicating with extremely large-scale ar- ray/surface: Unified modeling and performance analysis,

    H. Lu and Y . Zeng, “Communicating with extremely large-scale ar- ray/surface: Unified modeling and performance analysis,”IEEE Trans. Wireless Commun., vol. 21, no. 6, pp. 4039–4053, Jun. 2022

    2022

  2. [2]

    Deep learning for efficient CSI feedback in massive MIMO: Adapting to new environments and small datasets,

    Z. Liu, L. Wang, L. Xu, and Z. Ding, “Deep learning for efficient CSI feedback in massive MIMO: Adapting to new environments and small datasets,”IEEE Trans. Wireless Commun., vol. 23, no. 9, pp. 12 297– 12 312, Sep. 2024

    2024

  3. [3]

    Feature allocation for semantic communication with space-time impor- tance awareness,

    K. Zhou, G. Zhang, Y . Cai, Q. Hu, G. Yu, and A. Lee Swindlehurst, “Feature allocation for semantic communication with space-time impor- tance awareness,”IEEE Trans. Wireless Commun., vol. 24, no. 10, pp. 8847–8862, Oct. 2025

    2025

  4. [4]

    FDD massive MIMO: How to optimally combine UL pilot and limited DL CSI feedback?

    J. Kim, J. Choi, J. Park, A. Alkhateeb, and N. Lee, “FDD massive MIMO: How to optimally combine UL pilot and limited DL CSI feedback?”IEEE Trans. Wireless Commun., vol. 24, no. 2, pp. 926– 939, Feb. 2025

    2025

  5. [5]

    Channel estimation method and phase shift design for reconfigurable intelligent surface assisted MIMO networks,

    J. Mirza and B. Ali, “Channel estimation method and phase shift design for reconfigurable intelligent surface assisted MIMO networks,”IEEE Trans. Cognit. Commun. Netw., vol. 7, no. 2, pp. 441–451, Jun. 2021

    2021

  6. [6]

    Channel estimation approach for RIS assisted MIMO systems,

    E. Shtaiwi, H. Zhang, S. Vishwanath, M. Youssef, A. Abdelhadi, and Z. Han, “Channel estimation approach for RIS assisted MIMO systems,” IEEE Trans. Cognit. Commun. Netw., vol. 7, no. 2, pp. 452–465, Jun. 2021

    2021

  7. [7]

    A structure-shared dual-polarization endfire MIMO antenna based on uniform circular array for positioning EMIS,

    P. Wang, B. He, B. Yin, H. Zeng, and W. Wang, “A structure-shared dual-polarization endfire MIMO antenna based on uniform circular array for positioning EMIS,”IEEE Trans. Veh. Technol., vol. 73, no. 4, pp. 5925–5929, Apr. 2024

    2024

  8. [8]

    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 Tuts., vol. 26, no. 3, pp. 1478–1519, 3rd Quart. 2024

    2024

  9. [9]

    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, pp. 84–91, Jun. 2021

    2021

  10. [10]

    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., early access, Mar. 12, 2026, DOI: 10.1109/MWC.2026.3668205

    work page doi:10.1109/mwc.2026.3668205 2026

  11. [11]

    Channel gain map tracking via distributed Kriging,

    E. Dall’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. 13

    2011

  12. [12]

    Space-frequency- interpolated radio map,

    K. Sato, K. Suto, K. Inage, K. Adachi, and T. Fujii, “Space-frequency- interpolated radio map,”IEEE Trans. Veh. Technol., vol. 70, no. 1, pp. 714–725, Jan. 2021

    2021

  13. [13]

    Spatially consistent street-by-street path loss model for 28-GHz channels in micro cell urban environments,

    A. Karttunen, A. F. Molisch, S. Hur, J. Park, and C. J. Zhang, “Spatially consistent street-by-street path loss model for 28-GHz channels in micro cell urban environments,”IEEE Trans. Wireless Commun., vol. 16, no. 11, pp. 7538–7550, Nov. 2017

    2017

  14. [14]

    Calibration of NYURay, a 3D mmWave and Sub-THz ray tracer using indoor, outdoor, and factory channel measurements,

    O. Kanhere and T. S. Rappaport, “Calibration of NYURay, a 3D mmWave and Sub-THz ray tracer using indoor, outdoor, and factory channel measurements,” inProc. IEEE Int. Conf. Commun. (ICC), Rome, Italy, May 2023, pp. 5054–5059

    2023

  15. [15]

    Calibration of ray- tracing with diffuse scattering against 28-GHz directional urban channel measurements,

    R. Charbonnier, C. Lai, T. Tenoux, D. Caudill, G. Gougeon, J. Senic, C. Gentile, Y . Corre, J. Chuang, and N. Golmie, “Calibration of ray- tracing with diffuse scattering against 28-GHz directional urban channel measurements,”IEEE Trans. Veh. Technol., vol. 69, no. 12, pp. 14 264– 14 276, Dec. 2020

    2020

  16. [16]

    Channel knowledge map for environment-aware communications: EM algorithm for map construc- tion,

    K. Li, P. Li, Y . Zeng, and J. Xu, “Channel knowledge map for environment-aware communications: EM algorithm for map construc- tion,” inProc. IEEE Wireless Commun. Netw. Conf. (WCNC), Austin, TX, USA, Apr. 2022, pp. 1659–1664

    2022

  17. [17]

    Ray tracing accelera- tion using total variation norm minimization for radio map simulation,

    N. Suga, R. Sasaki, M. Osawa, and T. Furukawa, “Ray tracing accelera- tion using total variation norm minimization for radio map simulation,” IEEE Wireless Commun. Lett., vol. 10, no. 3, pp. 522–526, Mar. 2021

    2021

  18. [18]

    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

  19. [19]

    A scalable and generalizable pathloss map prediction,

    J.-H. Lee and A. F. Molisch, “A scalable and generalizable pathloss map prediction,”IEEE Trans. Wireless Commun., vol. 23, no. 11, pp. 17 793–17 806, Nov. 2024

    2024

  20. [20]

    An I2I inpainting approach for efficient channel knowledge map construction,

    Z. Jin, L. You, J. Wang, X.-G. Xia, and X. Gao, “An I2I inpainting approach for efficient channel knowledge map construction,”IEEE Trans. Wireless Commun., vol. 24, no. 2, pp. 1415–1429, Feb. 2025

    2025

  21. [21]

    Channel gain map construction based on subregional learning and prediction,

    J. Chen, R. Gao, J. Wang, S. Sun, and Y . Wu, “Channel gain map construction based on subregional learning and prediction,”IEEE Trans. Veh. Technol., vol. 74, no. 6, pp. 9852–9857, Jun. 2025

    2025

  22. [22]

    WiFi-Diffusion: Achieving fine-grained WiFi radio map estimation with ultra-low sam- pling rate by diffusion models,

    Z. Liu, S. Zhang, Q. Liu, H. Zhang, and L. Song, “WiFi-Diffusion: Achieving fine-grained WiFi radio map estimation with ultra-low sam- pling rate by diffusion models,”IEEE J. Sel. Areas Commun., vol. 43, no. 11, pp. 3796–3812, Nov. 2025

    2025

  23. [23]

    Generative CKM construction using partially observed data with diffusion model,

    S. Fu, Z. Wu, D. Wu, and Y . Zeng, “Generative CKM construction using partially observed data with diffusion model,” inProc. IEEE Veh. Technol. Conf. (VTC), Oslo, Norway, Jun. 2025, pp. 1–5

    2025

  24. [24]

    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 Trans. Cognit. Commun. Netw., vol. 11, no. 2, pp. 738–750, Apr. 2025

    2025

  25. [25]

    RME-GAN: A learning framework for radio map estimation based on conditional generative adversarial network,

    S. Zhang, A. Wijesinghe, and Z. Ding, “RME-GAN: A learning framework for radio map estimation based on conditional generative adversarial network,”IEEE Internet Things J., vol. 10, no. 20, pp. 18 016–18 027, Oct. 2023

    2023

  26. [26]

    RadioPix2pix: A GAN-based style transfer model for 2D radio map estimation,

    Z. Liu and S. Shao, “RadioPix2pix: A GAN-based style transfer model for 2D radio map estimation,” inProc. IEEE Veh. Technol. Conf. (VTC), Chengdu, China, Oct. 2025, pp. 1–5

    2025

  27. [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,”Commun. ACM, vol. 65, no. 1, pp. 99–106, Dec. 2021

    2021

  28. [28]

    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 Trans. Graph., vol. 42, no. 4, pp. 1–14, Jul. 2023

    2023

  29. [29]

    NeRF2: Neural radio-frequency radiance fields,

    X. Zhao, Z. An, Q. Pan, and L. Yang, “NeRF2: Neural radio-frequency radiance fields,” inProc. ACM Int. Conf. Mob. Comput. Netw. (Mobi- Com), Madrid, Spain, Oct. 2023, pp. 1–15

    2023

  30. [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,” inProc. Int. Conf. Mach. Learn. (ICML), Vienna, Austria, Jul. 2024, pp. 33 147–33 159

    2024

  31. [31]

    F4-CKM: Learning channel knowledge map with radio frequency radiance field rendering,

    K. Zhou, G. Zhang, H. Li, Y . Cai, S. Liu, and G. Yu, “F4-CKM: Learning channel knowledge map with radio frequency radiance field rendering,” IEEE Trans. Commun., vol. 74, pp. 5684–5700, 2026

    2026

  32. [32]

    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 Trans. Wireless Commun., vol. 25, pp. 7490–7504, 2026

    2026

  33. [33]

    GSRF: Complex- valued 3D Gaussian splatting for efficient radio-frequency data synthe- sis,

    K. Yang, G. Dong, S. Ji, W. Du, and M. Srivastava, “GSRF: Complex- valued 3D Gaussian splatting for efficient radio-frequency data synthe- sis,” inAdv. Neural Inf. Process. Syst. (NeurIPS), San Diego, CA, USA, Dec. 2025

    2025

  34. [34]

    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, 2026

    2026

  35. [35]

    Total variation based image restoration with free local constraints,

    L. Rudin and S. Osher, “Total variation based image restoration with free local constraints,” inProc. Int. Conf. Image Process. (ICIP), Austin, TX, USA, Nov. 1994, pp. 31–35

    1994

  36. [36]

    Born and E

    M. Born and E. Wolf,Principles of Optics, 7th ed. Cambridge, U.K.: Cambridge Univ. Press, 1999

    1999

  37. [37]

    Tse and P

    D. Tse and P. Viswanath,Fundamentals of Wireless Communication. Cambridge, U.K.: Cambridge Univ. Press, 2005

    2005

  38. [38]

    TensoRF: Tensorial radiance fields,

    A. Chen, Z. Xu, A. Geiger, J. Yu, and H. Su, “TensoRF: Tensorial radiance fields,” inProc. Eur. Conf. Comput. Vis. (ECCV), Tel Aviv, Israel, Oct. 2022, pp. 333–350

    2022

  39. [39]

    Sigmoid-weighted linear units for neural network function approximation in reinforcement learning,

    S. Elfwing, E. Uchibe, and K. Doya, “Sigmoid-weighted linear units for neural network function approximation in reinforcement learning,” Neural Netw., vol. 107, pp. 3–11, Nov. 2018

    2018

  40. [40]

    A shooting and bouncing ray (SBR) modeling framework involving dielectrics and perfect conductors,

    R. Brem and T. F. Eibert, “A shooting and bouncing ray (SBR) modeling framework involving dielectrics and perfect conductors,”IEEE Trans. Antennas Propag., vol. 63, no. 8, pp. 3599–3609, Aug. 2015

    2015

  41. [41]

    Massive MIMO channel measurement data set for localization and communication,

    A. Colpaert, S. De Bast, R. Beerten, A. P. Guevara, Z. Cui, and S. Pollin, “Massive MIMO channel measurement data set for localization and communication,”IEEE Commun. Mag., vol. 61, no. 9, pp. 114–120, Sep. 2023

    2023

  42. [42]

    Understanding the difficulty of training deep feedforward neural networks,

    X. Glorot and Y . Bengio, “Understanding the difficulty of training deep feedforward neural networks,” inProc. Int. Conf. Artif. Intell. Statist. (AISTATS), Sardinia, Italy, May 2010, pp. 249–256

    2010

  43. [43]

    k*-nearest neighbors: From global to local,

    O. Anava and K. Levy, “k*-nearest neighbors: From global to local,” in Adv. Neural Inf. Process. Syst. (NeurIPS), Barcelona, Spain, Dec. 2016, pp. 4916–4924

    2016