Linealización digital de transmisores mediante redes neuronales no lineales

This paper presents an overview on how the artificial neural networks (ANN) are applied to digitally linearize modern transmitters. The use of nonlinear ANNs is intended to either assist or replace the traditional crest factor reduction (CFR) and digital predistortion (DPD) building blocks, and benefit from their inherently good approximation capabilities and reduced hardware complexity when targeting complex transceiver scenarios such as those present in 5G. There is not a universal procedure to set up the best ANN given a specific application. However, in this paper some design considerations which have been experimentally validated in the literature will be summarized both considering single-antenna and multi-antenna transmitters. Finally, some principles in the selection of ANN parameters for nonlinear modeling will be showcased by using asimulation test bench that employs measured data from a strongly non-linear GaN PA operated with wideband signals.
Agradecemos el soporte y financiación recibidos por el Gobierno de España (MICINN) y FEDER en el marco de los proyectos TEC2017-83343-C4-2-R y RTI2018-099841-B-I00, y por la Generalitat de Catalunya bajo las ayudas 2017 SGR 813 y 2017 SGR 891.
Peer Reviewed
Postprint (published version)