Radio Frequency Ray Tracing with Neural Object Representation

TL;DR

RFScape introduces a modular RF simulation framework that decouples geometry and material properties into per-object neural primitives and integrates them with a differentiable ray-tracing engine. It employs neural SDFs to capture detailed geometry and a neural material model to encode directional attenuation, enabling end-to-end optimization with limited RF measurements and differentiable scene edits. The approach yields substantial improvements over conventional ray tracing and competitive gains over state-of-the-art neural baselines, especially at mmWave frequencies, and supports dynamic scene updates with minimal retraining. This work offers a scalable, editable, and efficient path toward high-fidelity RF propagation modeling for design, planning, and inverse RF reconstruction tasks.

Abstract

Radio frequency (RF) propagation modeling poses unique electromagnetic simulation challenges. While recent neural representations have shown success in visible spectrum rendering, the fundamentally different scales and physics of RF signals require novel modeling paradigms. In this paper, we introduce RFScape, a novel framework that bridges the gap between neural scene representation and RF propagation modeling. Our key insight is that complex RF-object interactions can be captured through object-centric neural representations while preserving the composability of traditional ray tracing. Unlike previous approaches that either rely on crude geometric approximations or require dense spatial sampling of entire scenes, RFScape learns per-object electromagnetic properties and enables flexible scene composition. Through extensive evaluation on real-world RF testbeds, we demonstrate that our approach achieves 13 dB improvement over conventional ray tracing and 5 dB over state-of-the-art neural baselines in modeling accuracy while requiring only sparse training samples.

Figures (9)

• Figure 1: An overview of the RFScape framework.
• Figure 2: RF Ray Tracing with RFScape's Neural Object Representations. While electromagnetic waves physically propagate from Tx (router) to Rx (device), the ray tracing computation traces paths backward from Rx to Tx for efficiency. Objects along these paths are modeled by our learned neural representations.
• Figure 3: Room-level Wireless Channel Distribution Collection Setup.
• Figure 4: Object-level Wireless Channel Response Collection Setup.
• Figure 5: RF Characterization Prediction Results.