Adaptive baseband predistorter for radio frequency power amplifiers based on a multilayer perceptron

A radio frequency (RF) power amplifier (PA) linearisation method based on a multilayer perceptron (MLP) neural network (NN) is presented. The proposed method is an alternative to digital baseband predistortion. In this case, a MLP with a single input and dual output is capable of compensating the amplitude-to-amplitude (AM/AM) and amplitude-to-phase (AM/PM) modulation that PA introduces. In order for this to be achieved the MLP is trained using a part of the down-converted response and a variant in the performance error of the backpropagation (BP) algorithm. After the first time training, the system can be retrained so as to adaptively compensate condition variations. The derived simulation results reveal the system performance in the PA´s non-linearities.