Physics-Augmented Graph Transformers for Patch-Antenna Forward and Inverse Design

2026-10-04

AI Summary

This study addresses the high computational cost of full-wave electromagnetic simulations in patch antenna design by proposing a mesh-native, physics-enhanced graph learning framework. The method formulates radiation pattern prediction as a signal reconstruction task over irregular surface meshes, leveraging GPS Graph Transformers and graph neural networks for feature extraction. It innovatively introduces Physics-Augmented Intermediate Supervision (PAIS) to predict complex surface currents, combined with a differentiable radiation integral consistency loss to enhance accuracy. Furthermore, efficient inverse design is achieved via a surrogate-filtered diffusion model. Experimental results demonstrate that the framework attains an MSE of 0.17 and a PSNR of 19.67 dB while exhibiting zero-shot generalization capabilities. Notably, the relative error in inverse design is reduced by 32% compared to nearest-neighbor retrieval.

Abstract

Full-wave electromagnetic (EM) simulation enables accurate patch-antenna analysis but is computationally expensive for large-scale forward prediction and inverse design. We present a mesh-native, physics-augmented graph-learning framework that treats radiation-pattern prediction as signal reconstruction on an irregular surface mesh. For the forward problem, a GPS graph transformer is trained with Physics-Augmented Intermediate Supervision (PAIS), an auxiliary node-level objective that predicts complex surface currents, the physical intermediate linking geometry to radiation. PAIS improves multiple GNN backbones at no inference-time cost, while shuffled-current and non-physical controls show the gain comes from physical correspondence. Direction-conditioned decoding and a differentiable radiation-integral consistency loss further exploit this structure. On an 80,000-sample CST benchmark, GPS+PAIS reaches MSE 0.17 / PSNR 19.67, generalizes to a PCA split, and transfers zero-shot to canonical patches. For inverse design, surrogate-filtered diffusion beats nearest-neighbor retrieval by 32% relative MSE.

Problem

• patch antenna
• forward prediction
• inverse design
• full-wave electromagnetic simulation
• radiation pattern

Innovation

• Physics-Augmented Graph Transformer
• Intermediate Supervision
• Patch-Antenna Design
• Diffusion Model
• Surface Currents