Machine Learning-Aided Design of Defected Ground Structures for PRGW-Based MIMO Antennas

This paper introduces a scalable machine learning (ML)-based inversed design approach of a newly shaped Defected Ground Structure (DGS) for isolation enhancement of a 2port Multiple-Input Multiple-Output (MIMO) antenna based on Printed Ridge Gap Waveguide (PRGW) technology. The proposed approach employs an artificial neural network (ANN) to predict the dimension and position of the proposed DGS in terms of the isolation and gain. As a test case, the proposed approach is applied to a 2-port MIMO antenna design with very low antenna element separation of 0.36λ, enhanced isolation of 30 dB, an average high gain of 8 dBi, and a radiation efficiency of >85%. Following a unique data generation phase of 96 hours, the target objectives are computed in ≈ 1 minute with a low root mean squared error (RMSE) of 0.02. This is a significant improvement compared to conventional Genetic Algorithms (GA) and Particle Swarm Optimization (PSO) available in the commercial CST EM simulator, which require repeated computations of more than 74 and 46 hours, respectively. The proposed approach proved to be an attractive choice that can be adapted to design 4×4, 16×16 as well as more element MIMO systems for Internet of Space (IoS) downlink applications.