Investigation on generalization when NN used as predistorter of PA with memory

Neural network (NN) can be trained as a predistorter which linearlizes nonlinear power amplifier with memory (NLPAWM). Direct learning architecture is proposed to make NN generalize well, for the train inputs sampled from square root raised cosine filter (SRRCF) can represent the actual inputs more accurately. Simulation and analyses confirm that this architecture is effective.