We propose and experimentally demonstrate a method of optical signal-to-noise ratio (OSNR) monitoring and modulation format identification (MFI) using a binarized convolutional neural network (B-CNN) in coherent receiver. The proposed technique automatically extracts OSNR and modulation format dependent features from the signals’ ring constellation maps. A group of modulation schemes including nine quadrature amplitude modulation (QAM) formats are selected as transmission signals. The experimental results show that the MFI accuracy can reach 100% and OSNR monitoring accuracy can reach higher than 97.71% for the nine M-QAM modulation formats. Compared with float valued convolutional neural network (F-CNN) and multi-layer perceptron (MLP), B-CNN can reach the same performance in MFI. For OSNR monitoring, the performance of B-CNN is similar to MLP and slightly worse than F-CNN. Moreover, the memory consumption and execution time of B-CNN is much lower than F-CNN and MLP. Therefore, B-CNN is power and time efficient with little performance loss compared with F-CNN and MLP. It is attractive for cost-effective multi-parameter estimation in next-generation optical networks.

中文翻译：

---

基于二值化神经网络的低复杂度 OSNR 监测和调制格式识别

我们提出并通过实验证明了一种在相干接收器中使用二值化卷积神经网络 (B-CNN) 进行光信噪比 (OSNR) 监测和调制格式识别 (MFI) 的方法。所提出的技术从信号的环形星座图中自动提取 OSNR 和调制格式相关特征。选择包括九种正交幅度调制(QAM)格式的一组调制方案​​作为传输信号。实验结果表明，9种M-QAM调制格式的MFI准确度可以达到100%，OSNR监测准确度可以达到97.71%以上。与浮点值卷积神经网络 (F-CNN) 和多层感知器 (MLP) 相比，B-CNN 在 MFI 中可以达到相同的性能。对于 OSNR 监控，B-CNN 的性能与 MLP 相似，但比 F-CNN 稍差。而且，B-CNN 的内存消耗和执行时间远低于 F-CNN 和 MLP。因此，与 F-CNN 和 MLP 相比，B-CNN 具有功率和时间效率，性能损失很小。它对于下一代光网络中具有成本效益的多参数估计很有吸引力。