Prediction of cut-off frequency based on Taguchi artificial neural network framework for designing compact spoof surface plasmon polaritons printed lines

In this paper, a novel approach for designing compact spoof surface plasmon polariton (SSPP) based printed transmission lines (TLs) using a Taguchi artificial neural network (T-ANN) is proposed. The challenge in determining the cut-off frequency of SSPPs lies in the absence of a closed-form expression relating it to the geometrical parameters of a planar dielectric substrate with a thin metallic strip. Typically, the cut-off frequency of conventional SSPP structures is highly dependent on factors such as the dielectric constant, metal strip length, unit cell length, and strip width. To address this, we employ a T-ANN-based methodology to accurately predict the cut-off frequency using the geometrical parameters of the SSPP structure. The T-ANN is trained with a dataset consisting of geometrical parameters and their corresponding cut-off frequencies obtained via full-wave electromagnetic simulations. The trained model is then utilized to optimize the SSPP unit cell parameters, aiming to achieve a desired cut-off frequency within a compact design framework. The MSE (mean square error) and validation $R^2$ scores of 8000 different data sets are 0.00134 and 0.99 respectively with normally distributed residuals. A comparative analysis between the T-ANN-predicted and full-wave simulated cut-off frequencies for 20 different design parameter sets demonstrates close alignment. The validation dataset converges within 20 epochs, confirming that the model avoids overfitting. Furthermore, a transmission line is designed based on the T-ANN-predicted parameters, and a prototype is fabricated. The performance of the design is validated through simulated and measured S-parameters.