Optical Modulation Format Identification Under Hybrid OSNRs Using LIN Model
Wei‐Wen Chen, Meng Liang · 2023
Modulation format identification (MFI) is the key to the normal operation of dynamic and flexible optical communication networks. In this paper, a novel lightweight convolutional neural network with improved Inception (LIN) model is proposed and combined with neighborhood grayscale enhanced constellation diagrams to achieve accurate recognition of six optical communication signal modulation formats in different communication scenes. Firstly, 10G baud coherent optical fiber communication systems are constructed to obtain the constellation diagrams of different modulation formats, and the constellation diagrams are processed by the neighborhood grayscale enhanced algorithm. Next, the neighborhood grayscale enhanced images with optical signal-to-noise ratio (OSNR) of l7-21dB are used as input features, the enhanced images are trained by four deep convolutional neural network models (LIN, AlexNet, ResNet, and GoogLeNet), and the comparison shows that LIN has the highest accuracy at fixed OSNR values and the minimum generation costs. Finally, in the multi-service systems, neighborhood grayscale enhanced images of modulation formats under hybrid OSNRs are considered as input characteristics, and the four models are migrated and applied to the dynamic communication systems, as well as the comparison reveals that LIN also performs very outstandingly in the MFI of hybrid OSNRs, which has significant performance advantages, and the recognition accuracy reaches 98.85%. The simulation experiments demonstrate that LIN has strong generalization ability and application feasibility.