Classificação automática de modulações em receptores ópticos coerentes flexíveis

Abstract

To keep pace with the increasing demand for higher transmission rates and improvements in internet connectivity, research is focusing on effective methods to optimize the use of optical networks. Such studies are crucial for developing systems autonomously gathering the necessary information to adjust transmission parameters. This includes choosing the most suitable modulations for the communication medium, deciding between single-carrier or multi-carrier transmission, and selecting the most efficient algorithms for signal regeneration and error correction. These approaches allow networks to automatically adjust to maximize performance and efficiency without direct human intervention. This dissertation investigates how machine learning techniques can be applied to automate parameter settings in optical communication systems, aiming to improve adaptability and efficiency by enabling autonomous adjustment of settings based on acquired information. Focusing on flexible coherent optical receivers, the study seeks to enhance system efficiency and adaptability by automating the process of modulation classification and OSNR value prediction, key elements for optimizing performance and reliability in optical data transmission. A back-to-back setup between simulated transmitters and receivers was used, generating $76,800$ signals with DP-BPSK, DP-QPSK, DP-8-PSK, DP-16-QAM, DP-32-QAM, and DP-64-QAM modulations across 51 different OSNR levels. Ensemble algorithms AdaBoost, CART Decision Tree, Gradient Boosting, Random Forest, and the Perceptron Multilayer neural network algorithm were employed for modulation classification and OSNR value prediction. The results show accuracies over 99% for modulation classification and OSNR prediction within a ±0.5 dB/0.1 nm error range with the Perceptron Multilayer model, demonstrating the viability and effectiveness of the proposed approach.