Machine-Learning Based Digital Doherty Power Amplifier

Abstract

This paper reports a new architecture of power amplifiers (PA), for which machine learning is applied in real-time to adaptively optimize PA performance. For varying input stimuli such as carrier frequency, bandwidth and power level, developed algorithms can intelligently optimize parameters including bias voltages, input signal phases and power splitting ratios based on a user-defined cost function. Our demonstrator of a wideband GaN Digital Doherty PA achieves significant performance enhancement from 3.0-3.8 GHz, in particular, at high backoff power with approximately 3dB more Gain and 20% higher efficiency compared with analog counterpart. To the authors' best knowledge, this is the first reported work of model-free machine learning applied for Doherty PA control. It explores a new area of RF PA optimization, in which accurate analytical models and tedious manual tuning can be avoided.