An Innovative Temporal Convolutional Network(TCN) Combined with Self-Attention Mechanism for Fiber Nonlinear Compensation

The nonlinear effects in optical fibers significantly impact the transmission quality of optical communication systems. In this paper, a novel Time Convolutional Network (TCN) nonlinear equalizer scheme, integrating self-attention mechanism and incorporating residual connections is proposed for compensating fiber nonlinearity. A numerical simulation system was constructed for validating the effectiveness of the scheme in nonlinearity compensation over a 1200km long-distance transmission with 130GBaud dual-polarization (DP)-16QAM signals. The numerical simulation results show that the proposed scheme achieves Q-factor improvements of 0.52 dB and 0.57 dB in single-channel and WDM multi-channel systems, respectively, compared to the high-complexity digital back propagation algorithm with 6 steps per span (DBP-6StPs).