Active Learning-Aided CNN-Based Entropy-Tunable Automatic Modulation Identification for Rate-Flexible Coherent Optical System

Flexible rate and real-time link monitoring are important tasks in the development of software-defined elastic optical networks (EONs). The tunable spectral efficiency characteristic of probabilistic constellation shaping (PCS) naturally provides a possibility to dynamically regulate the rate for future optical communication systems. In this work, we firstly propose an active learning-aided entropy-tunable automatic modulation identification (AL-aided ET-AMI) scheme based on convolution neural network (CNN) model for a PCS-based coherent optical system. An AL-based neural network allows monitoring of the link rate and signal-to-noise ratio (SNR) with tuning entropy or optical power fluctuation. The proposed AL-aided ET-AMI scheme is demonstrated over a 350∼550-Gbps line rate 10-km dual-polarized coherent optical communication system at entropies from 3.5 to 5.5. When the entropy tuning step is 0.1, corresponding to a rate tuning step of 5 Gbps at 50 Gbaud, the recognition accuracy can reach 98% with data aggregation (DA). When the fluctuation of SNR is 1 dB, the recognition rate can reach 87% at an entropy of 4.5 over 400 samples. The verifications show that our proposed AL-aided ET-AMI solution can monitor the rate and SNR performance of PCS-based high-speed rate-flexible optical links well. The solution provides a new perspective and tool for future optical systems and network monitoring.