Scholarlink

<p>This paper presents a generative adversarial network (GAN)-based design and optimization framework for microstrip patch antennas operating in the Terahertz (THz) band. Unlike conventional analytical or neural network models that require extensive parametric tuning, the proposed GAN learns to synthesise physically valid antenna geometries and predict their performance metrics directly from analytically generated and simulated datasets. The generator maps a 100-dimensional Gaussian latent vector to six key antenna design parameters–patch length, patch width, substrate height, dielectric constant, gain, and bandwidth–while the discriminator distinguishes real antenna configurations from generated ones. The framework integrates closed-form electromagnetic design relations during training, ensuring that the synthesised designs remain physically realizable and fabrication-feasible. Comprehensive evaluations demonstrate that the GAN-generated antennas achieve an average gain improvement of 1.7&#xa0;dB and a bandwidth enhancement of 8% compared to conventional deep neural network (DNN)-based and analytical baseline designs. The predicted return loss (<InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(S_{11}\)</EquationSource> <EquationSource Format="MATHML"><math> <msub> <mi>S</mi> <mn>11</mn> </msub> </math></EquationSource> </InlineEquation>) and radiation patterns of the generated antennas closely match analytical expectations, confirming their electromagnetic validity. When integrated into a reconfigurable intelligent surface (RIS)-assisted THz system simulation, the optimised antennas yield lower bit error rates and higher energy efficiency, illustrating their system-level impact. The proposed method provides a data-driven yet physics-informed approach to microstrip antenna design, offering superior adaptability, generalisation, and optimization efficiency over existing methods.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Design and Optimisation of Microstrip Patch Antennas Using Generative Adversarial Networks

  • 期刊论文

  • 发表日期：

    2025年12月22日

查看全文
  • Nitin Panuganti,
  • Pinku Ranjan,
  • Anupam Shukla

摘要

This paper presents a generative adversarial network (GAN)-based design and optimization framework for microstrip patch antennas operating in the Terahertz (THz) band. Unlike conventional analytical or neural network models that require extensive parametric tuning, the proposed GAN learns to synthesise physically valid antenna geometries and predict their performance metrics directly from analytically generated and simulated datasets. The generator maps a 100-dimensional Gaussian latent vector to six key antenna design parameters–patch length, patch width, substrate height, dielectric constant, gain, and bandwidth–while the discriminator distinguishes real antenna configurations from generated ones. The framework integrates closed-form electromagnetic design relations during training, ensuring that the synthesised designs remain physically realizable and fabrication-feasible. Comprehensive evaluations demonstrate that the GAN-generated antennas achieve an average gain improvement of 1.7 dB and a bandwidth enhancement of 8% compared to conventional deep neural network (DNN)-based and analytical baseline designs. The predicted return loss ( \(S_{11}\) S 11 ) and radiation patterns of the generated antennas closely match analytical expectations, confirming their electromagnetic validity. When integrated into a reconfigurable intelligent surface (RIS)-assisted THz system simulation, the optimised antennas yield lower bit error rates and higher energy efficiency, illustrating their system-level impact. The proposed method provides a data-driven yet physics-informed approach to microstrip antenna design, offering superior adaptability, generalisation, and optimization efficiency over existing methods.