Behavioral Modeling of GaN Power Amplifiers Using Long Short-Term Memory Networks

This paper presents the formulation of a behavioral model for a gallium nitride (GaN) Doherty power amplifier (DPA) using long short-term memory (LSTM) networks. Implemented in TensorFlow, LSTM networks can construct the dynamic behavior with memory effects by learning the useful patterns in the time domain. The behavioral model is built using the measured in-phase and quadrature ($\boldsymbol {I}$/$\boldsymbol {Q}$) data of the DPA, under excitation by a 20-MHz LTE signal. A comparative study indicates that the LSTM model is capable of accurately capturing the AM/AM and AM/PM characteristics of the DPA, as well as achieving competitive accuracy when compared to Volterra-based models.