Transfer Learning Optimized PA Behavioral Modeling over 2D Operation States

This work presents an efficient neural network-based power amplifier (PA) behavioral model across operating voltage (OV) and carrier frequency nonlinear states. This model can reduce the original training dataset size to 8.45% and the time to train to 2% when compared to a behavioral model (BM) trained for all the states. For proof-of-concept, this model is validated with laboratory data over a range of carrier frequencies from 2600 MHz to 3600 MHz, and OV levels from 3.5 V to 5 V.