Digital pre-distortion parameter optimization using Iterative learning control

Abstract

Network operators seek to minimize overall costs and energy consumption. In terms of power consumption, radio frequency power amplifiers inside transmitters are the main culprit. High efficiency is thus desired in power amplifiers but comes at a cost of reduced linearity. Today, a widely adopted technique to correct the nonlinearities of a power amplifier is digital pre-distortion, which is based on an inverse behavioral model of the power amplifier. This work explores the utilization of iterative learning control and machine learning in the optimization of digital pre-distorter model parameters. The main purpose is to find optimal model parameters and thus reach improved digital pre-distorter performance.

The conducted measurements included linearizing a gallium nitride Doherty power amplifier using an applied version of iterative learning control. After linearization, the input signal, which provides linear power amplifier output, was captured and saved. The captured optimal input signal was used as a target reference in the performed machine learning optimization that focused on adjusting the time delays within the digital pre-distorter's model.

Results of the optimization were tested with a real digital pre-distorter where a good and capable reference model was used as a benchmark. Compared to the reference, the optimized time delays accomplished a 0.7 dB improvement on the upper adjacent channel while the lower adjacent channel remained unchanged. The optimized model performed slightly better than the reference, demonstrating that the used method is a potential and efficient way to optimize the parameters of digital pre-distorters. The established workflow provides a solid foundation for future development.