Transient Temperature Variation-Dependent RF Power Amplifier Model Based on LSTM

Transient temperature variation-dependent behavioral model for radio-frequency power amplifiers (PAs) is proposed with consideration of input power level dynamics. The input power dynamics can cause temperature variations inside the PAs. In time-domain observations, such transient temperature variations were observed to degrade the modeling accuracy of a conventional modeling method, especially immediately after an abrupt change in the input power. The proposed model is based on the long short-term memory (LSTM) neural networks with smoothed power inputs in the input layer. The smoothed power calculation is based on exponential moving average of the input signal in series, which represents transient power variations. Through simulations using actual measurement data, the advantages of the proposed model are demonstrated.