Neural Network-Based Optimization for Bandwidth Enhancement of Millimeter-Wave Franklin Antenna With Proximity-Coupled Feed

This study aims to present the optimization of the Franklin microstrip antenna using a proximity-coupled feed. The initial model is established through mathematical calculations, followed by refinement using various machine learning methods, including Levenberg-Marquardt, Bayesian Regularization, Scaled Conjugate Gradient, and Adaptive Neuro Fuzzy Inference System (ANFIS). The antenna performance evaluation comprises of a comprehensive technique that combines simulation, measurement, and development of R-L-C equivalent circuit models, as well as the application of machine learning methods. The HFSS-based simulation was conducted in the frequency range of 20-36 GHz, which was validated through measurement. The measured antenna was constructed on a substrate with relative permittivity <inline-formula> <tex-math notation="LaTeX">$\varepsilon _{r}$ </tex-math></inline-formula> of 2.2, thickness of 1.57 mm, and tan <inline-formula> <tex-math notation="LaTeX">$\delta $ </tex-math></inline-formula> of 0.0013. Among the applied methods, ANFIS outperformed the others, yielding a simulation and measurement bandwidth of 16 GHz, a simulation gain of 10.45 dBi, and a radiation efficiency of 97.7%. The proposed dimensions of the machine-learning-optimized antenna put it as a robust choice for the 5G technology, simultaneously providing high bandwidth and gain, as supported by both simulation and measurement results.