Accurate and Efficient Pattern Synthesis Using Generative Adversarial Network for Series-Fed Microstrip Antenna Array With New Elements

TY - JOUR

T1 - Accurate and Efficient Pattern Synthesis Using Generative Adversarial Network for Series-Fed Microstrip Antenna Array With New Elements

AU - Chen, Kexin

AU - Bao, Zengdi

AU - Liu, Yitao

AU - Li, Yang

N1 - Publisher Copyright: © 1963-2012 IEEE.

PY - 2026

Y1 - 2026

N2 - Due to the unique geometry-dependent control of element amplitudes and phases, accurate and flexible pattern synthesis remains complex and challenging for series-fed microstrip antennas (SFMAs). In this communication, a modified physics-guided generative adversarial network (GAN) is developed to synthesize SFMA geometries for given pattern objectives. This synthesis framework offers higher efficiency and better synthesized patterns compared to traditional metaheuristic algorithms. Additionally, it does not require network pretraining. Moreover, new trapezoidal radiating elements acquiring ultralow reflection over a wide tuning range of coupling coefficients are proposed to maintain the traveling-wave mode, which is essential for accurate pattern control. Unlike other designs, these elements do not require additional reflection-canceling structures, thereby simplifying the antenna structure and synthesis process. Measured results of prototypes with a cosecant-squared (CSC2) pattern and a low-sidelobe pattern show excellent agreement with their respective pattern objective across 79–81 GHz, which validate the effectiveness of the synthesis method and the proposed element.

AB - Due to the unique geometry-dependent control of element amplitudes and phases, accurate and flexible pattern synthesis remains complex and challenging for series-fed microstrip antennas (SFMAs). In this communication, a modified physics-guided generative adversarial network (GAN) is developed to synthesize SFMA geometries for given pattern objectives. This synthesis framework offers higher efficiency and better synthesized patterns compared to traditional metaheuristic algorithms. Additionally, it does not require network pretraining. Moreover, new trapezoidal radiating elements acquiring ultralow reflection over a wide tuning range of coupling coefficients are proposed to maintain the traveling-wave mode, which is essential for accurate pattern control. Unlike other designs, these elements do not require additional reflection-canceling structures, thereby simplifying the antenna structure and synthesis process. Measured results of prototypes with a cosecant-squared (CSC2) pattern and a low-sidelobe pattern show excellent agreement with their respective pattern objective across 79–81 GHz, which validate the effectiveness of the synthesis method and the proposed element.

KW - Cosecant-squared (CSC)-shaped pattern

KW - generative adversarial network (GAN)

KW - low sidelobe

KW - pattern synthesis

KW - series-fed antenna

UR - https://www.scopus.com/pages/publications/105025106948

U2 - 10.1109/TAP.2025.3641763

DO - 10.1109/TAP.2025.3641763

M3 - Article

AN - SCOPUS:105025106948

SN - 0018-926X

VL - 74

SP - 2833

EP - 2838

JO - IEEE Transactions on Antennas and Propagation

JF - IEEE Transactions on Antennas and Propagation

IS - 3

ER -