RayProNet: A Neural Point Field Framework for Radio Propagation Modeling in 3D Environments

TL;DR

RayProNet introduces a neural point-field framework for efficient, geometry-aware wireless channel modeling in complex 3D environments. By integrating point clouds for geometry, light probes to bake propagation physics, and spherical-harmonics decoding, the method predicts power maps and received signals with fast runtimes. Extensive experiments across indoor, outdoor, and large-city scenes show RayProNet closely matches ray-tracing ground truth, outperforms 2D surrogates, and achieves substantial speedups (roughly 80–200×) over GPU-based ray tracing. The work enables rapid network planning and deployment optimization, while acknowledging limitations in geometry-change scenarios and pointing to future work on flexible geometry updates without retraining.

Abstract

The radio wave propagation channel is central to the performance of wireless communication systems. In this paper, we introduce a novel machine learning-empowered methodology for wireless channel modeling. The key ingredients include a point-cloud-based neural network and a Spherical Harmonics encoder with light probes. Our approach offers several significant advantages, including the flexibility to adjust antenna radiation patterns and transmitter/receiver locations, the capability to predict radio power maps, and the scalability of large-scale wireless scenes. As a result, it lays the groundwork for an end-to-end pipeline for network planning and deployment optimization. The proposed work is validated in various outdoor and indoor radio environments.

Figures (12)

• Figure 1: The schematic illustrates the input, output, and application of our proposed neural point field network framework for predicting wireless radio channel properties in large-scale environments.
• Figure 2: RayProNet: a neural point field framework for wireless channel modeling pipeline. (The symbols A - E represent the subsections in Section III.)
• Figure 3: Identifying n-nearest light probes: Each receiver locates its $n$ nearest light probes and retrieves radiance information from them.
• Figure 4: Identifying K-nearest points: Each light probe finds its $K$ closest points and encodes occlusion information.
• Figure 5: Multi-head attention: In Section III.C, a multi-head attention module is employed to aggregate the point cloud feature vector $l_{j,k}$ along with the K-closest direction $j$. This process generates a light probe feature. The attention module described in Section III.D follows a similar structure.