Cascaded Neural Network Module for Digital Predistortion Under Various Operating Conditions
Ke Tang, Cuiping Yu, Yuanan Liu
Abstract
This letter presents a cascaded neural network (NN) module behind a traditional NN digital predistortion (DPD) model to track the power amplifiers’ (PAs’) behavioral change under various operating conditions. Using the various operating conditions to control the weights of the cascaded NN’s last hidden layer, the DPD model can achieve excellent linearization performance without coefficient update in large-scale operating conditions. Since the number of controlled coefficients is relatively small, the proposed technique has higher performance and lower running complexity than existing NNs. The experiments were carried out on a commercial PA. Compared with the state-of-the-art NN models, the experimental results show that the proposed technique can improve linearization performance with lower complexity.