Neural-Network-Based Adaptive Baseband Predistortion Method for RF Power Amplifiers

Abstract

An adaptive baseband predistortion method for RF power amplifier (PA) linearization is proposed and experimentally demonstrated. The predistortion component is implemented by a single-input dual-output multilayer perceptron (MLP). Both amplitude-to-amplitude and amplitude-to-phase distortion products are compensated by backpropagation training of the neural network including the response of the PA. Effects of modulator and demodulator imperfections on system performance are examined. Measurements on a system prototype reveal a significant linearity improvement that reaches 25 dB.