Harnessing machine learning to probe dielectrics in next generation telecommunication and automotive radar applications

Abstract

In the last decade, technological advancements in signal transmission, particularly in fifth generation (5G) and emerging sixth generation (6G) wireless technologies, have grown at an exponential rate. While 5G provides faster data speeds and improved broadband capacity, 6G is projected to operate in the terahertz (THz) frequency range, achieving transmission speeds up to one terabit per second (Tbps). However, several challenges such as transmission loss and heat generation, particularly in ceramic dielectric materials (DEMs), persist in the implementation and execution of these technologies. This study aims at addressing these challenges using machine learning (ML) and generative reinforcement learning (GRL) in predicting optimal fabrication parameters for DEMs and antenna design. By integrating genetic algorithms (GA), we demonstrated the optimization of the synthesis parameters of DEMs such as magnesium silicate (Mg2SiO4) and created databases to accurately predict desired dielectric properties and antenna configurations. Six comprehensive databases were created to predict the optimal dielectric properties and antenna configurations. These databases encompass variables such as sintering temperature, dielectric constant, and dopant concentration, enabling highly accurate predictions of material performance. Our results underscore the transformative potential of ML-driven approaches in expediting the fabrication processes of DEMs and advancing the field of next-generation wireless communication technologies.