arxiv: 2605.05844 · v1 · submitted 2026-05-07 · 📡 eess.SP · cs.IT· math.IT

Recognition: unknown

TGPP: Trajectory-Guided Plug-and-Play Priors for Sparse Radio Map Reconstruction

Jiawen Zhang , Zhiyuan Jiang , Sheng Zhou , Zhisheng Niu

Authors on Pith no claims yet

Pith reviewed 2026-05-08 07:14 UTC · model grok-4.3

classification 📡 eess.SP cs.ITmath.IT

keywords radio map reconstructiontrajectory samplingplug-and-play priorssparse reconstructionwireless networksgenerative modelsRadioMapSeer dataset

0 comments

The pith

Trajectory-guided priors can be plugged into any radio map reconstruction backbone to cut error in under-observed areas.

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

Radio map reconstruction becomes harder when measurements arrive only along mobility trajectories instead of random points, because regions far from the paths stay weakly constrained. The paper introduces TGPP as a guidance module that learns an explicit risk map in input space and an implicit feature to fuse into hidden representations of an existing model. The module attaches without changing the backbone's original loss or architecture. When tested on five trajectory sampling rates using deterministic, adversarial, graph-based, and latent generative backbones, it raises accuracy most noticeably in the distant and occluded zones. The largest reported gain is a 43.1 percent drop in normalized mean-square error relative to the same backbone without the priors.

Core claim

TGPP learns an explicit guidance map as an interpretable input-space risk prior and an implicit guide feature that is projected and fused with backbone hidden representations. The module can be attached to different reconstruction backbones without changing their original task formulation. Experiments on RadioMapSeer with trajectory-sampled observations show that such data produces spatially heterogeneous uncertainty unlike random sparse sampling, and that TGPP improves most reconstruction metrics across backbones.

What carries the argument

Trajectory-Guided Plug-and-Play Priors (TGPP), which supplies an explicit guidance map and a projected implicit guide feature to steer any backbone toward under-constrained regions.

Load-bearing premise

The learned explicit guidance map and projected implicit guide feature can be fused with arbitrary backbones to reduce uncertainty in under-constrained regions without introducing new artifacts or degrading performance near trajectories.

What would settle it

Attach TGPP to a held-out backbone and new trajectory sampling rate, then check whether normalized mean-square error rises or falls in regions farther than 20 meters from the sampled paths compared with the backbone alone.

Figures

Figures reproduced from arXiv: 2605.05844 by Jiawen Zhang, Sheng Zhou, Zhisheng Niu, Zhiyuan Jiang.

**Figure 1.** Figure 1: Comparison between random-sampled and trajectory-sampled obser view at source ↗

**Figure 2.** Figure 2: Schematic of trajectory-sampled RM reconstruction. view at source ↗

**Figure 3.** Figure 3: Overview of TGPP. The guidance map augments the model input, view at source ↗

**Figure 4.** Figure 4: Qualitative comparison of RM reconstruction under trajectory-sampled observations. TGPP-enhanced variants generally produce smoother local view at source ↗

read the original abstract

Radio map (RM) reconstruction is essential for environment-aware wireless networks, but practical measurements are often collected along mobility trajectories rather than randomly scattered over the target region. Such trajectory-sampled observations induce spatially heterogeneous uncertainty: near-trajectory regions are directly constrained, whereas distant or occluded regions remain weakly observed, leading to degraded reconstruction accuracy in under-constrained areas. To address this problem, we propose Trajectory-Guided Plug-and-Play Priors (TGPP), a general guidance module for sparse RM reconstruction. TGPP learns an explicit guidance map as an interpretable input-space risk prior, and an implicit guide feature that is projected and fused with backbone hidden representations. TGPP can be attached to different reconstruction backbones without changing their original task formulation. We further introduce RadioFlow-LDM, a latent flow-based generative backbone, and apply TGPP to deterministic, adversarial, graph-based, and latent generative reconstruction models. Experiments on RadioMapSeer with five trajectory sampling rates show that trajectory-sampled reconstruction differs substantially from random sparse interpolation. TGPP improves most reconstruction metrics across backbones, achieving up to 43.1% NMSE reduction relative to the corresponding base backbone without trajectory-guided priors.

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.

Desk Editor's Note private letter to a colleague

TGPP adds a plug-and-play guidance module that lifts reconstruction metrics on trajectory-sampled radio maps across several backbones, but the evidence for why the fusion works and where it helps most stays thin.

read the letter

The main point is that real radio map data comes from trajectories, not random points, so uncertainty is higher away from those paths. TGPP tries to handle that by learning an explicit guidance map and an implicit feature to fuse into existing models without rewriting their loss or architecture. They also add their own RadioFlow-LDM backbone and test the whole thing on RadioMapSeer at five different trajectory densities. The reported outcome is that most metrics improve, with the biggest NMSE drop reaching 43.1 percent over the plain backbones. That part is straightforward and addresses a gap between theory and how measurements are actually collected for wireless tasks. The experiments show the difference between trajectory and random sampling is real, which is worth noting. The soft spot is the lack of any breakdown on performance close to versus far from the trajectories, and no ablation on the fusion step itself. Without those, it is hard to judge whether the gains come from better handling of under-constrained regions or from something else, and whether artifacts appear near the paths. The claim that the module attaches to arbitrary backbones is plausible from the description but rests on the same limited checks. This is aimed at people working on environment-aware networks and radio map methods. It has enough of a practical hook and cross-backbone results to go to referees, even if the analysis could be tighter. I would send it for review.

Referee Report

2 major / 2 minor

Summary. The manuscript proposes Trajectory-Guided Plug-and-Play Priors (TGPP), a general module for sparse radio map reconstruction from trajectory-sampled measurements. TGPP learns an explicit guidance map (interpretable input-space risk prior) and an implicit guide feature that is projected and fused with backbone hidden representations. It is designed to attach to arbitrary backbones without altering their task formulation. The authors introduce RadioFlow-LDM, a latent flow-based generative backbone, and evaluate TGPP across deterministic, adversarial, graph-based, and generative models on the RadioMapSeer dataset at five trajectory sampling rates. Results show that trajectory sampling produces substantially different reconstruction behavior than random sparse sampling and that TGPP yields metric improvements, including up to 43.1% NMSE reduction relative to the corresponding base backbones.

Significance. If the empirical gains hold under closer scrutiny, the work addresses a realistic and under-studied sampling regime in radio-map reconstruction for environment-aware wireless systems. The plug-and-play formulation is a clear practical strength, and the introduction of RadioFlow-LDM supplies a new generative baseline. The reported gains across multiple backbones suggest the guidance mechanism can mitigate spatially heterogeneous uncertainty induced by trajectory sampling.

major comments (2)

[§3] §3 (Method description): The central claim that TGPP fuses the explicit guidance map and projected implicit guide feature with arbitrary backbones without changing their original task formulation requires explicit equations or pseudocode for the projection and fusion operations; without them it is impossible to verify that the module remains backbone-agnostic and does not introduce new artifacts near trajectories.
[§4.2] §4.2 (Experimental results): The motivation emphasizes spatially heterogeneous uncertainty (well-constrained near trajectories, under-constrained far away), yet only aggregate metrics are reported. Separate error statistics (e.g., NMSE or RMSE) computed inside versus outside a distance threshold from the trajectories are needed to confirm that gains occur primarily in under-constrained regions and that performance near trajectories is preserved or improved.

minor comments (2)

[Abstract / §4.1] The five trajectory sampling rates used in the experiments should be stated explicitly (e.g., as percentages or point densities) in the abstract or early in §4.1 for immediate reproducibility.
[Figures] Figure captions and legends should clearly annotate trajectory paths versus reconstruction error heat-maps so readers can visually correlate the reported metric gains with the spatial uncertainty pattern described in the introduction.

Simulated Author's Rebuttal

2 responses · 0 unresolved

Thank you for the constructive feedback on our manuscript. We appreciate the referee's recognition of the practical importance of addressing trajectory-sampled measurements in radio map reconstruction and the value of the plug-and-play formulation. We address each major comment below and will incorporate revisions to strengthen the paper.

read point-by-point responses

Referee: [§3] §3 (Method description): The central claim that TGPP fuses the explicit guidance map and projected implicit guide feature with arbitrary backbones without changing their original task formulation requires explicit equations or pseudocode for the projection and fusion operations; without them it is impossible to verify that the module remains backbone-agnostic and does not introduce new artifacts near trajectories.

Authors: We agree that explicit mathematical details are necessary to fully substantiate the backbone-agnostic claim. In the revised manuscript, we will expand §3 with the precise equations defining the explicit guidance map as an input-space risk prior, the projection operator that maps the implicit guide feature into the backbone's hidden representation space, and the fusion mechanism (e.g., element-wise addition or gated concatenation). We will also include pseudocode illustrating the attachment of TGPP to an arbitrary backbone while preserving its original loss and task formulation. These additions will enable verification that no new artifacts are introduced near trajectories and that the module operates without altering the backbone's core architecture or objective. revision: yes
Referee: [§4.2] §4.2 (Experimental results): The motivation emphasizes spatially heterogeneous uncertainty (well-constrained near trajectories, under-constrained far away), yet only aggregate metrics are reported. Separate error statistics (e.g., NMSE or RMSE) computed inside versus outside a distance threshold from the trajectories are needed to confirm that gains occur primarily in under-constrained regions and that performance near trajectories is preserved or improved.

Authors: We concur that disaggregated error statistics would provide stronger empirical support for the spatially heterogeneous uncertainty motivation. In the revised §4.2, we will add new results reporting NMSE and RMSE separately for regions inside and outside distance thresholds (e.g., 10 m and 20 m) from the trajectories, across all five sampling rates and all backbones. These will be presented in additional tables or figures, demonstrating that TGPP yields the largest gains in under-constrained areas while maintaining or improving accuracy near trajectories, thereby confirming the design intent. revision: yes

Circularity Check

0 steps flagged

No significant circularity; method is an independent plug-and-play module validated empirically

full rationale

The abstract and description present TGPP as an added guidance module (explicit map + implicit feature) that is attached to arbitrary backbones without altering their formulations. No equations appear that define a target quantity in terms of itself or rename a fitted parameter as a prediction. No self-citations, uniqueness theorems, or ansatzes from prior author work are invoked to justify the core construction. Experimental claims rest on external RadioMapSeer data with trajectory sampling, not on internal fitting that forces the reported metric gains. The derivation chain therefore remains self-contained and does not reduce to its inputs by construction.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 2 invented entities

Abstract-only review; the guidance map and guide feature are new constructs introduced by the paper with no stated assumptions or external validation visible.

invented entities (2)

Explicit guidance map no independent evidence
purpose: Input-space risk prior derived from trajectory sampling to highlight uncertain regions
Learned component described as interpretable prior; no independent evidence or external validation provided in abstract.
Implicit guide feature no independent evidence
purpose: Hidden representation projected and fused with backbone features
Internal fusion mechanism; no details on projection or training given.

pith-pipeline@v0.9.0 · 5523 in / 1252 out tokens · 44543 ms · 2026-05-08T07:14:01.362712+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

18 extracted references · 2 canonical work pages · 1 internal anchor

[1]

Radio environment map as enabler for practical cognitive radio networks,

H. B. Yilmaz, T. Tugcu, F. Alag ¨oz, and S. Bayhan, “Radio environment map as enabler for practical cognitive radio networks,”IEEE Commu- nications Magazine, vol. 51, no. 12, pp. 162–169, Dec. 2013

2013
[2]

On the use of radio environment maps for interference management in heterogeneous networks,

J. Perez-Romeroet al., “On the use of radio environment maps for interference management in heterogeneous networks,”IEEE Communi- cations Magazine, vol. 53, no. 8, pp. 184–191, Aug. 2015

2015
[3]

Engineering radio maps for wireless resource management,

S. Bi, J. Lyu, Z. Ding, and R. Zhang, “Engineering radio maps for wireless resource management,”IEEE Wireless Communications, vol. 26, no. 2, pp. 133–141, Feb. 2019

2019
[4]

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 Communications, vol. 28, no. 3, pp. 84–91, Mar. 2021

2021
[5]

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

Y . Zenget al., “A tutorial on environment-aware communications via channel knowledge map for 6G,”IEEE Communications Surveys & Tutorials, vol. 26, no. 3, pp. 1478–1519, Feb. 2024

2024
[6]

A Tutorial on Learning-Based Radio Map Construction: Data, Paradigms, and Physics-Awareness

X. Wang, Y . Pan, and N. Cheng, “A tutorial on learning-based ra- dio map construction: data, paradigms, and physics-awarenes,” 2026, arXiv:2603.17499. [Online]. Available: https://arxiv.org/abs/2603.17499

work page internal anchor Pith review Pith/arXiv arXiv 2026
[7]

Channel knowledge map construction based on a UA V- assisted channel measurement system,

Y . Qiuet al., “Channel knowledge map construction based on a UA V- assisted channel measurement system,”Drones, vol. 8, no. 5, Art. no. 191, May 2024

2024
[8]

3D radio map recon- struction based on generative adversarial networks under constrained aircraft trajectories,

T. Hu, Y . Huang, J. Chen, Q. Wu, and Z. Gong, “3D radio map recon- struction based on generative adversarial networks under constrained aircraft trajectories,”IEEE Transactions on V ehicular Technology, vol. 72, no. 6, pp. 8250–8255, Jan. 2023

2023
[9]

Kriging-based interference power constraint: in- tegrated design of the radio environment map and transmission power,

K. Sato and T. Fujii, “Kriging-based interference power constraint: in- tegrated design of the radio environment map and transmission power,” IEEE Transactions on Cognitive Communications and Networking, vol. 3, no. 1, pp. 13–25, Jan. 2017

2017
[10]

Tensor completion for radio map reconstruction using low rank and smoothness,

D. Sch ¨aufele, R. L. G. Cavalcante, and S. Sta ´nczak, “Tensor completion for radio map reconstruction using low rank and smoothness,” inProc. 2019 IEEE 20th Int. Workshop Signal Process. Adv. Wireless Commun. (SPA WC), Jul. 2019, pp. 1–5

2019
[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 Transactions on Wireless Communications, vol. 20, no. 6, pp. 4001–4015, Feb. 2021

2021
[12]

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, arXiv:2212.11777. [Online]. Available: https://arxiv.org/abs/2212.11777

work page arXiv 2022
[13]

Deep completion autoencoders for radio map estimation,

Y . Teganya and D. Romero, “Deep completion autoencoders for radio map estimation,”IEEE Transactions on Wireless Communications, vol. 21, no. 3, pp. 1710–1724, Aug. 2021

2021
[14]

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 of Things Journal, vol. 10, no. 20, pp. 18016–18027, May 2023

2023
[15]

RadioGAT: a joint model-based and data-driven framework for multi-band radiomap reconstruction via graph attention networks,

X. Liet al., “RadioGAT: a joint model-based and data-driven framework for multi-band radiomap reconstruction via graph attention networks,” IEEE Transactions on Wireless Communications, vol. 23, no. 11, pp. 17777–17792, Sep. 2024

2024
[16]

RadioDiff: an effective generative diffusion model for sampling-free dynamic radio map construction,

X. Wanget al., “RadioDiff: an effective generative diffusion model for sampling-free dynamic radio map construction,”IEEE Transactions on Cognitive Communications and Networking, vol. 11, no. 2, pp. 738–750, Nov. 2024

2024
[17]

RadioDiff-k 2: Helmholtz equation informed generative diffusion model for multi-path aware radio map construction,

X. Wanget al., “RadioDiff-k 2: Helmholtz equation informed generative diffusion model for multi-path aware radio map construction,”IEEE Journal on Selected Areas in Communications, vol. 44, pp. 2318–2333, Dec. 2025

2025
[18]

NeRF 2: neural radio-frequency radiance fields,

X. Zhao, Z. An, Q. Pan, and L. Yang, “NeRF 2: neural radio-frequency radiance fields,” inProceedings of the 29th Annual International Con- ference on Mobile Computing and Networking, Oct. 2023, pp. 1–15

2023