Improving Generative Inverse Design of Rectangular Patch Antennas with Test Time Optimization

TL;DR

This work addresses inverse design of rectangular patch antennas by learning a latent representation of feasible $S_{11}$ frequency responses via a $eta$-VAE and then mapping those responses to geometry with a conditional VAE augmented by adversarial disentanglement. Test-time optimization—through latent-space search and gradient refinement—improves design accuracy and enables manufacturability considerations without requiring more training data. The authors demonstrate that best-of-$N$ sampling and targeted latent optimization yield designs whose simulated EM responses closely match target curves, and show the approach generalizes to more complex geometries and design criteria. The framework thus offers a data-efficient, controllable path for patch-antenna inverse design with practical applicability to fabrication constraints and evolving design objectives.

Abstract

We propose a two-stage deep learning framework for the inverse design of rectangular patch antennas. Our approach leverages generative modeling to learn a latent representation of antenna frequency response curves and conditions a subsequent generative model on these responses to produce feasible antenna geometries. We further demonstrate that leveraging search and optimization techniques at test-time improves the accuracy of the generated designs and enables consideration of auxiliary objectives such as manufacturability. Our approach generalizes naturally to different design criteria, and can be easily adapted to more complex geometric design spaces.

Figures (4)

• Figure 1: Configuration of a Rectangular Patch Antenna fed via coaxial line through the ground plane.
• Figure 2: Overview of our two-stage generative inverse design framework. Stage 1 learns a latent representation of $S_{11}$ frequency response curves and finds in-distribution curves matching target responses. Stage 2 uses a conditional VAE to generate antenna geometries that produce the desired EM response. Red arrows represent test time optimization.
• Figure 3: Scaling performance as the number of curves (left) and the number of designs per curve (right) is increased. The shaded regions indicate variability across runs.
• Figure 4: Comparison of the idealised target $S_{11}$ curve $y^{*}$ (black dashed), the dominant-mode analytic resonance $f_{r,\mathrm{TM10}}$ (dotted vertical), and the simulated $S_{11}$ of designs $\tilde{x}$ generated with two test-time compute budgets. Blue = 1 curve $\times$ 1 design, red = 10 curves $\times$ 20 designs. Note: Generated geometries may exploit higher-order or coupled modes, so exact agreement with the analytic reference is not expected.