Physically Inspired Predistortion of RF Power Amplifiers with Artificial Neural Networks

Description

Mobile communication is rapidly growing. Increasing demands on capacity and bandwidth have to be addressed by future developments. This means higher signal requirements and bandwidth for transceivers in mobile basestations. Transceivers are the component with highest power consumption in a basestation. Especially analog components show different impairments and nonideal behavior with negative effects on energy efficiency and signal integrity. These effects can be analyzed and mathematically described to build a specific digital signal processing algorithm, which mitigates certain effects. This work treats impairments from machine learning perspective. IQ Imbalance of modulators as well as power amplifier nonlinearities are representive impairments with significant influence on the signal quality. These effects are trained to artificial neural networks (ANNs) for digital impairment mitigation. Furthermore it is shown that the ANNs are able to model different impairment effects with a single network and can be simply enhanced by further input parameters to mitigate dynamic effects. Physically inspired modeling of long term memory effects like thermal memory and charge trapping are a special focus of this work.