OLS-Based RBF Neural Network for Nonlinear and Linear Impairments Compensation in the CO-OFDM System

Abstract

For the first time, we propose and experimentally verify a novel low-complexity orthogonal least square (OLS) based radial basis function (RBF) neural network (NN) nonlinear equalizer for high-speed coherent optical orthogonal frequency division multiplexing (CO-OFDM) system. Its ability to compensate nonlinear impairments as well as linear impairments is comprehensively evaluated and compared with the linear equalizer. The impact of training overhead on system performance is also investigated. Results show that in a single-channel 40-Gb/s 16-quadrature amplitude modulation CO-OFDM system, with the training overhead of 4%, the maximum transmission distance is extended to 800 km at Q threshold of 8.7 dB, and RBF-NN outperforms the linear equalizer by 2.8 and 5.6 dB Q-factors after 800 and 600 km transmissions, respectively.

Keywords

• Coherent optical orthogonal frequency division multiplexing (CO-OFDM)
• nonlinear impairment
• neural network (NN)