Design and optimization of four-port MIMO antenna using ML algorithms for wireless applications

Abstract

A compact multiple input multiple output (MIMO) antenna design combined with machine learning (ML) algorithms -driven Surrogate based performance optimization model for cutting-edge wireless communication applications is offered in this manuscript. The antenna design evolves from a simple rectangular patch to a modified arrow-shaped radiator patch with a defected ground structure. Then, four single element antenna arranged orthogonally with decoupling strips to make a small four-port MIMO system with less mutual coupling to achieve enhanced isolation, bandwidth, and diversity performance. The presented MIMO antenna occupies a size of 90 × 90 × 1.6 mm3 and resonates at 2.9 GHz covering operating frequency band 2.7–3.3 GHz. The antenna demonstrates desirable performance characteristics, including isolation better than 20.0 dB, and gain values surpassing 3.5 dB in the working band. It is suitable for sub-6 GHz wireless communication applications. To improve performance prediction, multiple ML algorithm models including K-nearest neighbor network (KNN), artificial neural network, decision tree, XGBoost, and random forest (RF) are employed. Among them, KNN delivers superior S11 prediction (MAE 0.03, R2 0.997), while RF excels for S21 prediction (MAE 0.11, R2 0.997).