Efficient Multi-Objective and Mixed-Variable Pixelated Antenna Design Using Bayesian Optimization

Ferhat Can Ataman

doi:10.1109/ojap.2026.3741866

Efficient Multi-Objective and Mixed-Variable Pixelated Antenna Design Using Bayesian Optimization

2026-10-12

0

OA

AI

Georgios C. Trichopoulos

DOI：10.1109/ojap.2026.3741866

Original

Save

Abstract

En 中文

This paper presents a sample-efficient Bayesian optimization (BO) framework for high-dimensional pixelated antenna design with unified multi-objective and mixed-variable support. The framework combines Gaussian process (GP) surrogate models, numerically stable LogEI/LogEHVI acquisition functions, and graded bandwidth and field-of-view objectives. It supports the continuous, categorical, and binary variables encountered in practical antenna design. On a stacked pixelated antenna benchmark, the proposed method reaches the selected single-objective performance targets using fewer full-wave evaluations than the evaluated baselines and produces an approximation to the bandwidth–field-of-view Pareto front. Every reported antenna response and objective value is obtained from a full-wave Ansys HFSS evaluation; the GP surrogate is used only to guide candidate selection. The resulting framework provides a sample-efficient approach to pixelated antenna design.

Keywords:

Antenna design

Bayesian optimization

multi-objective optimization

mixed-variable optimization

pixelated antennas

surrogate modeling

Gaussian processes

Efficient Multi-Objective and Mixed-Variable Pixelated Antenna Design Using Bayesian Optimization

Abstract

Journal

Organization

Cited Papers