RMPrior: Bridging Propagation Priors and Diffusion Refinement for Efficient Radio Map Construction

2026-06-01

AI Summary

Diffusion models face practical limitations in constructing high-fidelity radio maps due to their computationally expensive iterative sampling and neglect of propagation priors. This work proposes a midpoint-start sampling strategy that injects a matched propagation model prior into an intermediate timestep of the diffusion process, performing only subsequent reverse steps to focus on multipath-aware fine-grained reconstruction. The approach achieves the first integration of propagation priors with diffusion models and provides theoretical guarantees, including an upper bound on initialization error, sufficient conditions under which truncating early timesteps improves reconstruction quality, and an analysis of prior sensitivity. Evaluated on the IRT4HighRes dataset with a starting timestep of 0.5, the method accelerates inference by 2.01× while outperforming the full-sampling baseline across all four metrics: NMSE, RMSE, SSIM, and PSNR.

Abstract

Diffusion models achieve high-fidelity radio map construction through iterative denoising, yet their sampling cost limits practicality in dynamic wireless systems where radio maps must be refreshed repeatedly. Meanwhile, classical propagation models encode valuable scene-level knowledge that standard diffusion inference discards entirely by initializing from pure Gaussian noise. This paper bridges propagation priors and diffusion refinement through a mid-start sampling strategy. A matched propagation prior is perturbed to an intermediate diffusion timestep, and the pretrained diffusion backbone executes only the remaining reverse steps, focusing computation on multipath-aware refinement rather than full reconstruction from noise. We provide theoretical analysis establishing an upper bound on the initialization gap, a sufficient condition under which truncation improves reconstruction fidelity, and a formal characterization of prior-quality sensitivity under aggressive truncation. Experiments on IRT4HighRes show that, at $P_{\text{start}}=0.5$, the proposed method achieves a $2.01\times$ speedup while simultaneously improving NMSE, RMSE, SSIM, and PSNR over the full-step baseline. A prior-quality ablation across three propagation models of different fidelity confirms that reconstruction quality tracks prior quality, with the sensitivity amplified under shorter reverse trajectories, consistent with the theoretical predictions. These results also suggest that mid-start reconstruction quality can serve as a proxy for ranking the scene-level fidelity of different propagation models.

Problem

• radio map construction
• diffusion models
• propagation priors
• sampling efficiency
• dynamic wireless systems

Innovation

• diffusion models
• propagation priors
• mid-start sampling
• radio map construction
• efficient inference