Pre-monitoring-assisted deep cascaded network for nonlinear equalization in coherent optical communication systems

Machine learning (ML) has emerged as a promising technique for nonlinear equalization in coherent optical communication systems. However, conventional ML-based equalization schemes typically rely on fixed network architectures and parameters, limiting the adaptability to dynamic link states such as varying launch powers and transmission distances. In this Letter, we propose a pre-monitoring-assisted deep cascaded network (DCN) for nonlinear equalization. A lightweight convolutional neural network (CNN) performs pre-monitoring by analyzing the received signal spectrum to identify link states, thereby selecting the optimal deep neural network (DNN) model for nonlinear equalization. Experimental demonstrations in a 28-GBaud PDM-16QAM system show that, at the optimum launch power for 250 km transmission, the proposed scheme achieves a Q-factor gain of 2.01 dB over linear equalization, with a computational complexity of 2885 real multiplications per symbol (RMpS).