Description
A Master of Science thesis in Electrical Engineering by Anfal Alsayed Ali entitled, “Neural network based predistortion of radio frequency power amplifiers”, submitted in April 2021. Thesis advisors are Dr. Oualid Hammi and Dr. Usman Tariq. Soft copy is available (Thesis, Completion Certificate, Approval Signatures, and AUS Archives Consent Form).
Abstract
Over the last few decades, researchers have been able to successfully use neural networks for behavioral modeling and predistortion of power amplifiers. When used as predistorters, neural networks offer favorable results due to their ability to perform complicated tasks and model nonlinear systems. The main downside of neural network predisorters is their large number of coefficients which increases their complexity. This research proposes four hybrid neural network predistorter models that combine a real-valued focused time-delay neural network (RVFTDNN) with a memory polynomial, or a generalized memory polynomial. The aim of the proposed models is to reduce the complexity of the predistorter while maintaining its performance. The proposed models have been validated for linearizing three power amplifiers: a Doherty PA driven by a 1-carrier LTE signal with a 5MHz bandwidth, a Doherty PA driven by a 4-carrier (1001) LTE signal with a 20MHz bandwidth, and the PA provided by RF WebLab, at Chalmers University, Sweden. This latter device under test was driven by a 1-carrier LTE signal with a 20MHz bandwidth. Each power amplifier was linearized by the four proposed models. For every power amplifier, the models’ performance was evaluated by calculating the adjacent channel leakage ratio (ACLR) and the normalized mean square error (NMSE). The proposed models were successful in linearizing the power amplifiers and showed a significant improvement in performance when compared to a RVFTDNN predistorter of the same size. All in all, the proposed models successfully reduced the number of coefficients of the predistorter while maintaining the performance.