Behavioral modeling and digital predistortion of Power Amplifiers with memory using Two Hidden Layers Artificial Neural Networks

Abstract

This paper presents a novel Two Hidden Layers Artificial Neural Networks (2HLANN) model for behavioral modeling and linearization of RF Power Amplifiers (PAs). Starting with a feedback loop principle model of a PA, an appropriate neural networks structure is deduced. This structure was then optimized to form a real valued and feed-forward 2HLANN based model capable of predicting the nonlinear behavior and the memory effects of wideband PAs. The validation of the proposed model in mimicking the behavior of a Device Under Test (DUT) is carried out in terms of its accuracy in predicting the output spectrum, dynamic AM/AM and AM/PM characteristics and the normalized mean square error. In addition, the 2HLANN model was used to linearize two 250 Watt peak-envelope-power Doherty PAs (DPAs) driven with 20 MHz bandwidth signals. The linearization of these DPAs using the 2HLANN enabled attaining an output power of up to 46.8 dBm and an average efficiency of up to 47.5% coupled with an Adjacent Channel Power Ratio higher than 50 dBc. When compared to a number of previously published behavioral and DPD schemes, the 2HLANN model demonstrated an excellent modeling accuracy and linearization capability.