Meta learner-based optimization for antenna efficiency prediction and high-performance THz MIMO antenna applications

Abstract

A number of classical machine learning approaches have been used to predict antenna efficiency. However, machine learning needs to be enhanced to predict more accurately. The stacked generalization approach has been shown to be capable of learning from features and meta features. In this paper, we propose a meta learner-based stacked generalization ensemble strategy that passes classical stacked ensemble output to an optimized multi-feature stacked ensemble. For the optimizer, a grid search strategy is employed. Applying an ANN model with a classical ML model as a base learner for predicting antenna efficiency leads to increased performance in terms of R2, EVS, MAE, and RMSE of 0.9998, 0.9998, 0.0001, and 0.0001, respectively, with MSE tending to zero. This improved accuracy significantly aids in designing our THz MIMO antenna, which resonates at frequencies of 4.99 THz and 8.831 THz with an extensive bandwidth of 5.941 THz. The MIMO configuration, utilizing graphene as the patch and copper as the ground, achieves an impressive isolation of -27.34 dB, a gain of 11.89 dB, and an efficiency of 88.90 %. The antenna exhibits exceptional diversity performance, with an Envelope Correlation Coefficient (ECC) as low as 0.007 and a Diversity Gain (DG) of 9.9654. Furthermore, the RLC circuit model precisely replicates the reflection coefficients of the MIMO antenna, affirming the accuracy and reliability of the predictive models. These results suggest that our optimized prediction model and high-performance MIMO antenna design are well-suited for advancing next-generation THz.