Machine Learning Based Design of Pattern Reconfigurable Antenna

Abstract

In this study, a Machine Learning (ML) is implemented to soft computation of the Reconfigurable Horn Bowtie Dumbbell (RHBD) antenna at operating frequency range from 26 GHz to 29.5 GHz for 5G applications. An adaptive learning rate approach is used to build a ML model on a 5-layer system utilizing a simulated database of 180 RHBD antennas. In the training stage of a hybrid method that combines the advantages of particle swarm optimization (PSO) with a modified version of the gravitational search algorithm (MGSA), the architecture frame and hyper-parameters of the ML model are optimized. A precise electromagnetic analysis platform is used to simulate 180 RHBD antennas with varying geometrical properties in terms of the resonant frequency in order to create the database for training and testing the model. The ML model is tested and validated using a fabricated RHBD antenna operating at 27.5 GHz. Then, three PIN diodes are placed in the gaps of the reflectors located at the back of the antenna, and by changing the state of these PIN diodes, it can be noticed that they have a significant and direct effect on the radiation pattern, as they are able to change the beamwidth from 10.7 o to 156.2 o . The suggested antenna makes it easier to create dynamic radiation patterns that may be utilized to reconfigure the coverage area as required in accordance with the spatial-temporal user and traffic variations in high mobility environments.

Design and Comparative Analysis of THz Antenna through Machine Learning for 6G Connectivity

“…In reference [28], the author introduced an MLdriven generative optimization technique employing masked auto-encoders to enhance multi-objective antenna decoupling structures, achieving a minimum 6 dB improvement in antenna isolation. In [29], the author presents an ML-based framework for computing resonance frequency in reconfigurable antennas and investigates beamwidth control using PIN diodes as a cost-effective alternative to simulations. In [30], the author introduces an ML-based method for predicting resonance and directivity in a quasi-Yagi antenna, showcasing remarkable accuracy in directivity predictions with minimal error.…”