Predistortion of GaN Power Amplifier Transient Responses in Time-Division Duplex Using Machine Learning

Abstract

The extensive use of time-division duplexing in 5G and 6G poses a challenge to the linear operation of the power amplifiers (PAs) in radio base stations. Particularly with gallium nitride (GaN) technology, the PAs can produce strong transient behavior when resuming from an idle state, which degrades the first few transmitted symbols. This article proposes a novel machine learning technique to model and compensate the PA gain transient, based on a lightweight, low-rate recurrent model. Our RF measurements at 3.6 GHz examine the joint application of transient compensation and predistortion of short-term effects and show a successful mitigation of both types of distortion.