Feedforward Neural Network Enabled Optical Multi-Path Interference Mitigation for High-Speed IM-DD Transmissions
Yongfeng Qiu, Meng Xiang, Junjiang Xiang, Hailin Yang, Di Lin, Jianping Li, Songnian Fu, Yuwen Qin
Abstract
The performance of high-speed intensity modulation direct detection (IM-DD) transmission can be severely degraded by the optical multipath interference (MPI), due to the repeated reflections from polluted fiber connectors. Here, we propose a data-driven MPI mitigation scheme, utilizing a feedforward neural network (FNN) with the capability of powerful nonlinear mapping, after both the temporal and frequency domain MPI model are numerically investigated. Our simulation results reveal that, when the 28 Gbaud PAM-4 signals are transmitted over standard single mode fiber (SSMF), the proposed FNN can effectively mitigate the MPI and improve the transmission performance, under conditions of variable signal-interference ratios (SIRs) and various laser linewidths. Compared with recently reported non-data-driven MPI mitigation schemes, the FNN enabled MPI mitigation scheme can increase the SIR tolerance by 1.5, 1.8, and 0.8 dB, under conditions of 100 kHz, 1 MHz, and 10 MHz laser linewidth, respectively, in case the KP4-FEC threshold is considered. Meanwhile, we thoroughly study the simplification of the FNN enabled MPI mitigation scheme relying on the multiple-symbol-output (MSO) structure and weight pruning, leading to a computational complexity reduction by 32.4%, in comparison with non-data-driven MPI mitigation schemes. Finally, the performance of full-size FNN and simplified FNN enabled MPI mitigation is experimentally evaluated for the 28 Gbaud PAM-4 signals transmission over 10.81 km SSMF, observing a 2.2 and 0.9 dB improvement of SIR tolerance at the KP4-FEC threshold, respectively.