A Novel Transfer Learning Approach for Efficient RF Device Behavior Model Parameter Extraction

This study introduces a novel mixture of artificial neural network(ANN) and polynomial modeling approach for efficient behavior modeling of nonlinear RF devices. It is achieved by employing the ability of knowledge retention and knowledge transfer of ANN model to merge the new working condition into the previous model, which makes the model more intelligent when dealing with increased complicated measurement data. In addition, a dynamic sampling techniques is utilized to accurately capture the key dynamic characteristics among different PA operating states to improve the knowledge transfer efficiency. The method was verified by modeling a 0.25um GaN HEMT device at different frequencies and DC bias conditions. The proposed method can be very easily implemented in CAD software hence suitable for large-scale device modeling scenario.