Full‐Wave‐Data‐Driven Array‐Excitation Prediction via Closed‐Loop Learning With a Differentiable Forward Surrogate
Microwave and Optical Technology Letters · 2026-11 · 68 (11)
DOI: 10.1002/mop.70866 ↗Kurz gefasst
Fachartikel in der Zeitschrift Microwave and Optical Technology Letters, online erschienen am 10.11.2026. Verfasst von Jiawen Cao und 3 weiteren Autor:innen. Die Arbeit nennt 15 Literaturangaben.
Abstract (Originalsprache)
Full‐wave‐simulated array responses provide a data‐driven representation of the excitation‐to‐pattern mapping beyond the ideal array‐factor approximation. This letter presents a full‐wave‐data‐driven, surrogate‐assisted closed‐loop learning framework for excitation prediction from prescribed scan‐beam radiation patterns. A differentiable forward surrogate is first trained on a limited set of full‐wave‐derived excitation–pattern pairs and then used for efficient surrogate‐based data augmentation. During inverse training, the forward surrogate is frozen to reconstruct the radiation pattern from the predicted excitation, enabling pattern‐domain supervision on the main‐beam gain and sidelobe peak while retaining excitation‐domain supervision. For a 1 × 10 uniform linear array evaluated with 360‐point realized‐gain patterns, the proposed method achieves an average pattern root‐mean‐square error of 0.48 dB and a main‐beam pointing error of 0.14°, compared with 1.03 dB and 0.28° for a 1‐D ResNet‐18 baseline. Additional full‐wave‐level evaluation on 20 samples generated from the same array configuration and HFSS–MATLAB labeling workflow, but excluded from model training, shows lower mean RMSE and MSLL with comparable main‐beam pointing error. These results demonstrate an efficient data‐driven closed‐loop framework for excitation prediction in the considered full‐wave‐simulated array setting.
Zitieren
APA
Cao, J., Lu, J., Zhou, S., & Yang, Y. (2026). Full‐Wave‐Data‐Driven Array‐Excitation Prediction via Closed‐Loop Learning With a Differentiable Forward Surrogate. Microwave and Optical Technology Letters, 68(11). https://doi.org/10.1002/mop.70866
Vancouver
Cao J, Lu J, Zhou S, Yang Y. Full‐Wave‐Data‐Driven Array‐Excitation Prediction via Closed‐Loop Learning With a Differentiable Forward Surrogate. Microwave and Optical Technology Letters. 2026;68(11). doi:10.1002/mop.70866
BibTeX
@article{Cao2026,
author = {Cao, Jiawen and Lu, Jijin and Zhou, Shigang and Yang, Yuhang},
title = {Full‐Wave‐Data‐Driven Array‐Excitation Prediction via Closed‐Loop Learning With a Differentiable Forward Surrogate},
journal = {Microwave and Optical Technology Letters},
year = {2026},
volume = {68},
number = {11},
doi = {10.1002/mop.70866}
}
EUVIMED