Development and Evaluation of ANN, RBNNs, and GRNNs Based Small-Signal Behavioral Models for GaN HEMT Up to 40 GHz

This paper conducts an extensive analysis of small-signal behavioral modelling of Gallium Nitride (GaN) High Electron Mobility Transistors (HEMTs) up to 40 GHz, utilizing Artificial Neural Network (ANN), Radial Basis Neural Networks (RBNNs), and Generalized Regression Neural Networks (GRNNs). The study focuses on enhancing accuracy, generalization capability and speed by fine-tuning hyperparameters through standard trial and error method. Additionally, the paper evaluates the developed models' ease of implementation, and fitting and error behaviors under diverse biasing conditions. The acquired results indicate an exceptional consistency between measured and modelled behaviors for ANN based models. Furthermore, RBNNs based models demonstrate subpar accuracy, whereas GRNNs based models exhibit inferior prediction accuracy compared to ANN but better than RBNNs based models.