Title :
A digital predistortion technique based on a NARX network to linearize GaN class F power amplifiers
Author :
Aguilar-Lobo, L.M. ; Garcia-Osorio, A. ; Loo-Yau, J.R. ; Ortega-Cisneros, S. ; Moreno, Pablo ; Rayas-Sanchez, J.E. ; Reynoso-Hernandez, J.A.
Author_Institution :
Dept. de Ing. Electr. y Cienc. Computacionales, Centro de Investig. y de Estudios Av. del I. P. N., Zapopan, Mexico
Abstract :
This work presents a novel Digital Predistortion (DPD) scheme based on a NARX network, suitable for linearizing power amplifiers (PAs). The NARX network is a Recurrent Neural Network (RNN) with embedded memory that allows efficient modeling of nonlinear systems. Its neural architecture is very effective to model long term dependencies, such as the typical memory effects of PAs. To demonstrate the feasibility of the NARX network as a DPD system, a GaN class F PA with two LTE signals with 5 MHz of bandwidth is used. Experimental results show a distortion correction better than 10 dB.
Keywords :
III-V semiconductors; distortion; gallium compounds; neural chips; neural net architecture; radiofrequency power amplifiers; recurrent neural nets; wide band gap semiconductors; DPD scheme; GaN; LTE signals; NARX network; RNN; bandwidth 5 MHz; class F power amplifier linearization; digital predistortion technique; distortion correction; embedded memory; memory effects; neural architecture; nonlinear systems; recurrent neural network; Computer architecture; Gallium nitride; Predistortion; Recurrent neural networks; Training; NARX network; PA linearization; digital predistortion; long-term memory effects; nonlinear systems; recurrent neural networks;
Conference_Titel :
Circuits and Systems (MWSCAS), 2014 IEEE 57th International Midwest Symposium on
Conference_Location :
College Station, TX
Print_ISBN :
978-1-4799-4134-6
DOI :
10.1109/MWSCAS.2014.6908515
