A new nonlinear behavioral modeling technique for RF power transistors based on Bayesian inference

A novel nonlinear behavioral modeling technique, for transistor behavioral modeling, is presented in this paper. Compared with existing modeling techniques, the new approach is based on a fundamentally different theory, Bayesian inference (one of the core methods of machine learning). The new technique not only good at handling multidimensional modeling problem, it could also greatly alleviated the notorious overfitting issue through corresponded model extraction method. Both simulation and experimental test examples for a 10W Cree GaN transistor are provided. The new model provides accurate prediction throughout the Smith chart at different input power levels.