Input Amplitude-Based Adaptive Tuning Neural Networks for Digital Predistortion of Doherty Power Amplifiers

In this article, a novel input amplitude-based adaptive tuning (AT) technique is proposed to improve the performance of the existing neural network-based digital predistortion (NN-DPD) models in Doherty power amplifier (DPA) systems. The changes in the behavioral characteristics of DPA are first discussed to explain why NN-DPD models designed for single power amplifier (PA) degrade in performance in DPA systems. Then, the AT technique leverages the input selection module (ISM) to help NN-DPD models adapt to changes in the operating states of DPA, while the parameter tuning module (PTM) aids NN-DPD models in accommodating the dynamic behavioral characteristics arising from active load modulation in DPA. Furthermore, the AT technique is applied to both the block-oriented time-delay neural network (BOTDNN) and the augmented real-valued time-delay neural network (ARVTDNN) for developing two adaptive tuning NN (ATNN) models, named AT-BOTDNN and AT-ARVTDNN, respectively. The experimental results demonstrate that both ATNN models achieve superior linearization performance with reduced computational complexity, and that the performance improvement mainly comes from the ISM while the remaining performance gains come from the PTM and other designs.