Machine Learning based Design Optimization of a GPS Antenna on a flexible substrate

Abstract

In this work, a machine learning optimized folded dipole antenna is presented for automotive GPS applications with wide band and good antenna performance characteristics. The optimization routine uses a regression based supervised machine learning algorithm, modified extensible lattice sequence (Mels) to develop a surrogate mathematical model from the physical antenna model. A Global Response Search Method (GRSM) optimization technique is used for optimizing the surrogate mathematical model to achieve the desired antenna performance such as high gain, low reflection coefficient, and wide bandwidth. The antenna is designed on a flexible Polyethylene terephthalate (PET) substrate to make it compatible with easier integration in a smaller lattice space. The designed antenna has a bandwidth of 182 MHz in L band with a gain of 1.9 dBi.