An efficient compensation approach for fiber impairments in high-capacity system based on NARX neural network algorithms

This paper provides a unique and novel strategy for addressing and compensating fiber optics impairments based on advanced machine learning techniques known as Nonlinear Auto Regressive with Exogenous (NARX). This work focuses on increasing the performance of an optical transmission system by investigating the efficiency of nonlinear mitigation technique. The NARX method was implemented with 16 channels in single polarization-quadrature amplitude modulation (SP-QAM) formats, with a data rate of 125 Gb/s per channel over 5000 km. The suggested model’s efficacy is demonstrated by its ability to properly forecast the nonlinearity of optical fibers while accounting for signal distortions. In terms of the quality factor (Q-factor), the system improved by 11.31 dB and 10.19 dB for 16QAM and 64QAM, respectively, resulting in a reduction in bit error rates (BER) than hard decision forward error correction (HD-FEC).