Machine learning-based broadband GaN HEMT behavioral model applied to class-J power amplifier design

A novel, broadband, nonlinear behavioral model, based on support vector regression (SVR) is presented in this paper. The proposed model, distinct from existing SVR-based models, incorporates frequency information into its formalism, allowing the model to perform accurate prediction across a wide frequency band. The basic theory of the proposed model, along with model implementation and the model extraction procedure for radio frequency transistor devices is provided. The model is verified through comparisons with the simulation of an equivalent circuit model, as well as experimental measurements of a 10 W Gallium Nitride (GaN) transistor. It is seen that the efficiency prediction throughout the Smith chart, for varying fundamental and second harmonic loads, across a wideband frequency range, show excellent fidelity to the measured results. Device dc self-biasing is also modelled to allow prediction of power amplifier (PA) efficiency, which is shown to be highly accurate when compared with corresponding measured data. Finally, a class-J PA is constructed and measured across the frequency with a large-signal input tone. The resulting measured and modelled values of key PA performance figures are shown to be in excellent agreement, indicating the model is suitable for broadband PA design.