Experimental demonstration of transport over fiber using 5G NR with performance enhancement using DPD

Abstract

The article proposes an experimental implementation with a performance comparison of 5G new radio (NR) multiband waveforms utilizing analog radio over fiber (A-RoF) link equipped with performance enhancement via digital predistortion (DPD) methods. An unprecedented convolutional neural network (CNN) is compared with our earlier proposed robust methods such as deep neural network (DNN), magnitude-selective affine (MSA) and generalized memory polynomial (GMP) methods. In circumstances when deep learning is not practicable due to high complexity, we have presented an integrative lightweight convolutional neural network (ILWCNN) approach which is stable over time without the need for DPD coefficient updates. The experimental bench consists of a multiband 5G NR standard at a carrier frequency of 20 GHz with a bandwidth of 50 MHz and a flexible-waveform signal at 3 GHz with a bandwidth of 20 MHz. A 10 km of single-mode fiber is used to transport the signals modulated by a dual-drive Mach Zehnder Modulator. The paper discusses the selection of reduced complexity and robust performance of the ILWCNN-based DPD that eventually outperforms the conventional methods. The proposed ILWCNN reduces the error vector magnitude (EVM) to 1.65% as compared to 4.1%, 3.1%, 1.9% and 7.4% of EVM for GMP, MSA, DNN and non-compensated cases respectively. Similarly, the adjacent channel leakage ratio (ACLR) of −28 dBc uncompensated link is reduced with GMP, MSA and ILWCNN to −40 dBc, −45 dBc and −47 dBc respectively. The suggested ILWCNN approach outperforms the compared methods in terms of ACLR and EVM attaining the 3GPP Release 17 requirements.