Low-Pass Equivalent Behavioral Modeling of Rf Power Amplifiers Using Two Independent Real-Valued Feed-Forward Neural Networks

Abstract

Abstract-Feed-forward artificial neural networks (ANNs) can provide the adequate model required for the linearization of power amplifiers (PAs) used in wireless communication systems. A common characteristic of previously available ANN-based models for linearization purposes is the use of a single real-valued ANN having two outputs. The contribution of this work is to report the benefits of performing such behavioral modeling based on two independent real-valued ANNs, where each network has a unique output. The proposed ANN-based model is applied to the behavioral modeling of a GaN HEMT class AB PA, and its accuracy is compared to previous approaches in two different scenarios. First, in case of similar number of network parameters, it is observed that the proposed ANN-based model can reduce the normalized mean-square error (NMSE) by up to 1.3 dB. Second, in a situation of comparable modeling accuracy (NMSE = −40 dB), it is observed that the proposed ANN-based model can reduce the number of network parameters by up to 40% (from 62 to 38 real-valued parameters).