Estimação da relação sinal ruído óptica baseada em aprendizado profundo utilizando diagramas de constelação das modulações dp-mpsk e dp-mqam em receptores ópticos coerentes flexíveis

Abstract

The Optical Signal to Noise Ratio (OSNR) is a critical parameter for the operation, management, and maintenance of coherent optical communication networks. Given its importance, this work proposes and evaluates a deep learning (DL) based approach for OSNR estimation from constellation diagrams. For the development of the models, a diverse dataset comprising over 19,000 images was generated through the simulation of signals with DP m-PSK and DP m-QAM modulation formats under different visual configurations. In total, 15 representative state-of-the-art Convolutional Neural Network (CNN) architectures were evaluated, including the MobileNetV3, ConvNeXt, DenseNet, and EfficientNet families. The ConvNeXtBase architecture achieved the best predictive performance in the study, with a Mean Absolute Error (MAE) below 0.43 dB and a coefficient of determination (R²) above 0.98. The obtained results confirm the high potential of computer vision techniques for the continuous and non- intrusive performance monitoring of next-generation optical networks.