X-parameters modeling based on LSTM and CG-BPNN for transistor

Abstract

In order to reduce the time of device parameter measurement or simulation and improve the efficiency of circuit design, X-parameters of gallium nitride high electron mobility transistor (GaN HEMT) are modeled based on long short term memory (LSTM) and double hidden layer conjugate gradient back propagation neural network (CG-BPNN) in this paper. Then, to verify the modeling efficiency of the two models, the harmonic balance experiments are carried out to obtain the three harmonics of the predicted data and expected data. Finally, the three harmonic errors of LSTM model are 0.801, 7.511 and 13.470 dBm, respectively, and the three harmonic errors of double hidden layer CG-BPNN model are 0.1117, 2.594 and 3.423 dBm, respectively. Through the above experiments, it is proved that double hidden layer CG-BPNN model proposed here can effectively model GaN HEMT with large-signal. The application in engineering is the demonstration of superior performance of the proposed CG-BPNN model in terms of accurate representation of X-parameters for transistor.