CoRF: Cross-Scene RF Synthesis by Learning Propagation and Preserving Array Physics

2026-09-28

AI Summary

This study addresses the high deployment cost of existing radio frequency neural fields, which require per-scene training. We propose a cross-scene, training-free spatial spectrum synthesis framework that decouples scene-dependent propagation characteristics from receiver array observation models. Specifically, a reference-conditioned anchor field generates directions of arrival and signal powers, which are directly synthesized into spatial spectra via an analytical mapping, enabling single-pass inference at arbitrary query positions. Furthermore, we introduce a local reference capacity bound and error decomposition theory to facilitate rapid instantiation with frozen pretrained models. Evaluated across 35 simulated scenarios, our method outperforms the strongest baseline by 7.09 dB and 6.97 dB in PSNR on unseen variants and entirely unseen categories, respectively.

Abstract

Existing radio-frequency (RF) neural fields fit each scene separately, making new-scene deployment measurement- and optimization-intensive. This work studies amortized cross-scene spatial spectrum synthesis, where a shared model learns propagation across scenes and instantiates an unseen scene from sparse target-scene measurements without scene-specific training. To achieve this, CoRF separates learned scene-dependent propagation from the known receiver-array observation model. An unordered set of spectrum-only references conditions a canonical anchor field, producing arrival directions and query-dependent component powers for each query. Analytic array physics maps these components to the receiver covariance and then to the spatial spectrum. This factorization keeps the pretrained propagation model frozen, enabling synthesis at arbitrary query locations after a single reference-conditioning pass. A necessary local reference-capacity bound and an error decomposition further characterize the formulation. Across 35 simulated scenes spanning seven categories, CoRF outperforms the strongest baseline by 7.09 dB PSNR on unseen variants of represented scene categories and by 6.97 dB on entirely unseen scene categories.

Problem

• cross-scene RF synthesis
• spatial spectrum synthesis
• amortized inference
• neural fields
• sparse measurements

Innovation

• Cross-Scene RF Synthesis
• Amortized Inference
• Array Physics
• Neural Fields
• Spatial Spectrum