Otimização de projetos de antenas microstrip: simulação de células metamateriais SRR com machine learning

Abstract

The design of microstrip antennas is widely documented in the literature and presents good results when an antenna is manufactured. However, since the technology includes low efficiency and limited bandwidth, it is often necessary to optimize the design without altering its physical dimensions to meet the requirements of the application. One way to achieve this is given by materials engineering: an application of metamaterials. Metamaterials are materials with remarkable electrical characteristics. When applied to antennas, the electromagnetic radiation behaves differently than expected, improving its specs, hence the performance, without altering the size. However, the geometry of the metamaterial cells and their array makes the analytical development of the resultant electromagnetic field equations more complex, becoming unworkable. Telecommunication engineers deal with this problem by designing an antenna with a metamaterial array based on computationally expensive and time-consuming electromagnetic simulations. With the popularization of data science libraries such as scikit-learn, pytorch, tensorflow, and others, this work aims to allow the optimization of antenna designs through development of predictive machine learning models to beat the computational costs of a microwave simulator optimization.