An adaptive power amplifier lineariser based on a multilayer perceptron

In this paper, a RF power amplifier (PA) linearisation method base on a multilayer perceptron (MLP) neural network is proposed. The method is an alternative to the digital baseband predistortion. The predistortion component is a single MLP with a dual output and single input. For the training of the MLP the response of the PA, in a training ramp, and the backpropagation (BP) training algorithm are used. After the training process, the MLP is capable of compensating both the amplitude-to-amplitude (AM/AM) and amplitude-to-phase modulation (AM/PM) that PA introduces. A method´s prototype has been constructed; this is according to the author´s knowledge the first published experimental PA predistorter that employs MLP. Measurement results reveal a considerable improvement in the adjacent channel interference (ACI) that reaches 25 dB.