TY - GEN
T1 - A fully connected neural network approach to mitigate fiber nonlinear effects in 200G DP-16-QAM transmission system
AU - Catanese, Clara
AU - Ayassi, Reda
AU - Pincemin, Erwan
AU - Jaouen, Yves
N1 - Publisher Copyright:
© 2020 IEEE.
PY - 2020/7/1
Y1 - 2020/7/1
N2 - Fiber nonlinear effects in optical transmission are an important issue to be solved to ensure higher bit rate and denser spectral efficiency in metro/long-haul optical networks. Various kinds of digital signal processing techniques exist to mitigate fiber nonlinear transmission impairments, but they suffer from heavy computational resources and require the knowledge of system parameters. Artificial Neural Networks (ANN) have recently attracted a lot of interest by being agnostic to transmission parameters and driven only by data. Here, a fully connected ANN is proposed to mitigate nonlinear effects induced in optical fibers. It aims to find the inverse transfer function of the nonlinear transmission channel. Two positions for the ANN are studied at receiver side: after Multiple-Input-Multiple-Output (MIMO) or after Carrier Phase Estimation (CPE). We show that inserting the ANN after CPE leads to significantly higher Bit Error Rate (BER) improvement. The proposed solution is trained and tested over a numerically simulated single channel 200 Gbps DP-16-QAM signal. It is also tested in quasi-single channel condition for a 200 Gbps DP-16-QAM signal with experimental temporal traces generated over the transmission test-bed of the laboratory. It allows BER improvements over a large range of span input powers compared to purely linear equalization.
AB - Fiber nonlinear effects in optical transmission are an important issue to be solved to ensure higher bit rate and denser spectral efficiency in metro/long-haul optical networks. Various kinds of digital signal processing techniques exist to mitigate fiber nonlinear transmission impairments, but they suffer from heavy computational resources and require the knowledge of system parameters. Artificial Neural Networks (ANN) have recently attracted a lot of interest by being agnostic to transmission parameters and driven only by data. Here, a fully connected ANN is proposed to mitigate nonlinear effects induced in optical fibers. It aims to find the inverse transfer function of the nonlinear transmission channel. Two positions for the ANN are studied at receiver side: after Multiple-Input-Multiple-Output (MIMO) or after Carrier Phase Estimation (CPE). We show that inserting the ANN after CPE leads to significantly higher Bit Error Rate (BER) improvement. The proposed solution is trained and tested over a numerically simulated single channel 200 Gbps DP-16-QAM signal. It is also tested in quasi-single channel condition for a 200 Gbps DP-16-QAM signal with experimental temporal traces generated over the transmission test-bed of the laboratory. It allows BER improvements over a large range of span input powers compared to purely linear equalization.
KW - Coherent detection
KW - Digital signal processing
KW - Neural network
KW - Nonlinear fiber effects
UR - https://www.scopus.com/pages/publications/85092522794
U2 - 10.1109/ICTON51198.2020.9203197
DO - 10.1109/ICTON51198.2020.9203197
M3 - Conference contribution
AN - SCOPUS:85092522794
T3 - International Conference on Transparent Optical Networks
BT - 2020 22nd International Conference on Transparent Optical Networks, ICTON 2020
PB - IEEE Computer Society
T2 - 22nd International Conference on Transparent Optical Networks, ICTON 2020
Y2 - 19 July 2020 through 23 July 2020
ER -