Application of the NARX neural network as a digital predistortion technique for linearizing microwave power amplifiers

ABSTRACT This work presents a digital predistortion (DPD) scheme to linearize power amplifiers (PAs) using a recurrent neural network called Nonlinear AutoRegressive with eXogenous input model (NARX) neural network (NARXNN). The architecture of the NARXNN is based on a class of discrete‐time nonline...

Full description

Saved in:
Bibliographic Details
Published in:Microwave and optical technology letters 2015-09, Vol.57 (9), p.2137-2142
Main Authors: Aguilar-Lobo, Lina M., Loo-Yau, Jose R., Rayas-Sánchez, Jose E., Ortega-Cisneros, Susana, Moreno, Pablo, Reynoso-Hernández, J. A.
Format: Article
Language:English
Subjects:
Architecture
Digital
Dynamical systems
Evolution
long-term memory effects
Microwaves
NARX network
Neural networks
Noise levels
nonlinear systems
nonlinear systems, predistortion
PA linearization
Power amplifiers
predistortion
recurrent neural networks
Citations: Items that this one cites
Items that cite this one
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:ABSTRACT This work presents a digital predistortion (DPD) scheme to linearize power amplifiers (PAs) using a recurrent neural network called Nonlinear AutoRegressive with eXogenous input model (NARX) neural network (NARXNN). The architecture of the NARXNN is based on a class of discrete‐time nonlinear system named NARX. Its topology has embedded memory at the input and output of the neural architecture, which allows an efficient linearization of PAs. To show the benefits of the DPD with NARXNN, a commercial PA is fed with a long term evolution signal at 2.0 GHz with 10 MHz of bandwidth. Our experimental results show an adjacent channel leakage ratio improvement of 24 dB. © 2015 Wiley Periodicals, Inc. Microwave Opt Technol Lett 57:2137–2142, 2015
ISSN:0895-2477
1098-2760
DOI:10.1002/mop.29281