Complexity-Reduced Model Adaptation for Digital Predistortion of RF Power Amplifiers With Pretraining-Based Feature Extraction
Yue Li, Xiaoyu Wang, Anding Zhu
Abstract
In this article, we present a new method to reduce the model adaptation complexity for digital predistortion (DPD) of radio frequency (RF) power amplifiers (PAs) under varying operating conditions, using pretrained transformation of model coefficients. Experimental studies show that the PA behavior variations can be effectively tracked using a small number of “transformed” coefficients, even with large deviations in its output characteristics. Based on this discovery, to avoid reextracting all the original coefficients every time when the operating condition changes, we propose to conduct a one-time off-line pretraining stage to extract the common features of PA behaviors under different operating conditions first. The online model adaptation process will then only need to identify a small number of transformed coefficients, which can result in a drastic reduction in the computational complexity of the model adaptation process. The proposed solution is validated by experimental results considering varying signal bandwidth and output power levels on a high-efficiency gallium-nitride Doherty PA, where the computational complexity is significantly reduced and the system performance is not compromised.