Machine-Learning-Based Generative Optimization Method and Its Application to an Antenna Decoupling Design
Hao Huang, Xue‐Song Yang, Bing‐Zhong Wang
Abstract
A machine-learning-based generative optimization method using masked autoencoders (MAE) is proposed and applied to multiobjective antenna decoupling structure design. The machine learning method contains <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$k$ </tex-math></inline-formula> -means algorithm and MAE neural network structure. The <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$k$ </tex-math></inline-formula> -means is used for label-free classification of decoupling structure samples, and MAE is used for the intelligent optimization design of decoupling structures. By applying the machine-learning-based method, MAE optimization models for designing decoupling structures are obtained. An antenna decoupling example using neutralization line is selected to validate the effectiveness of the proposed optimization method. Measurement results show that the neutralization line designed by the proposed method improves the antenna isolation by at least 6 dB, that is, <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$S_{21}$ </tex-math></inline-formula> reaches below at least -18 dB between 3.5 and 9.7 GHz, while requiring little manual intervention during the optimization progress.