Knowledge-Guided Machine Learning for Broadband Pixelated Patch Antenna Design

A knowledge-guided machine learning assisted (KG-MLA) method is proposed for the design of a broadband pixelated patch antenna. Differing from the traditional MLA design scheme, the proposed KG-MLA utilizes prior knowledge to select optimization pixels for sample collection based on a typical E-shaped patch antenna. The reduced sample space ensures a significant decrease in training samples, facilitating easier convergence of the surrogate model and enhancing the optimization efficiency. Specifically, the wideband E-shaped patch antenna is first pixelated, and sample collection points are selected based on prior knowledge. The collected samples are then binarized into a 0-1 matrix. A convolutional neural network (CNN) is employed to establish a mapping between the binary matrix and the S-parameters, thereby constructing a surrogate model. Using the trained surrogate model, the bandwidth is efficiently optimized via the particle swarm optimization (PSO) algorithm. The simulation results indicate that the reflection coefficient of the optimized patch antenna remains below −10 dB across the frequency range from 1.92 GHz to 2.63 GHz, corresponding to a bandwidth of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">${31.2\%}$</tex>.