Advanced Space Mapping and Artificial Neural Network-Based Techniques for Microwave Device Modeling

Abstract

This thesis focuses on developing advanced space mapping and ANN-based techniques for microwave device modeling, addressing both microwave active and passive component modelings. Specifically, this thesis develops an advanced space mapping-based technique in combination with KBNN for modeling GaN HEMT devices. This thesis also develops advanced ANN-based transfer function mapping techniques for parametric modeling of microwave passive components. GaN HEMT model development can be time-consuming as the devices exhibit strong and sophisticated trapping effects that are hard to be accurately modeled. In the first part of this thesis, we propose to.develope separate mappings for different branches inside the existing device model, such that different behaviors in GaN HEMTs can be mapped separately. The KBNN drain current model is proposed, being used as part of the trapping mapping development. The proposed space mapping technique allows fast and systematic model development for GaN HEMTs to achieve an accurate large-signal model. The proposed technique also employs less training data, lowering the cost of data generation from measurements. Neuro-TF approach, a popular parametric modeling method, combines ANN and transfer function, where transfer function is used as space mapping between the ANN and EM responses. However, the existing neuro-TF approaches may have discontinuity and associated non-smoothness issues. In the second part of this thesis, we propose to systematically combine both pole-residue and rational formats of the transfer functions in the neuro-TF model based on whether the poles/residues have those issues or not. The proposed technique can obtain good model accuracy in challenging applications of large geometrical variations, addressing the discontinuity and non-smoothness issues. The existing neuro-TF approaches may also have a high-sensitivity issue. In the third part of this thesis, we propose to decompose the original rational-based neuro-TF model with high order of transfer function into multiple sub-rational-based neuro-TF models with much lower order of transfer function, decreasing the sensitivities of the overall model response with respect to the coefficients of the transfer function in each sub-neuro-TF models. The proposed modeling technique can achieve good model accuracy in challenging applications of large geometrical variations, addressing the high-sensitivity issue.