SDN Control of Optical Amplifiers in Multi-Band Networks: The TDFA Case

Multi-band (MB) transmission is being investigated as a cost-effective solution for increasing the capacity of the existing optical networks, while postponing the deployment of new fibers. Thulium–Doped Fiber Amplifiers (TDFAs) represent a viable solution for meeting the continuous bandwidth demand by providing the amplification in the S-band. However, the operation of the optical amplifiers deploying multiple pumps, such as the TDFA, is characterized by the nonlinear behavior, depending on the pumping currents and wavelengths, input power, operating range, etc. In the context of controlling such amplifiers, the main challenge is selecting the pumping configuration that allows the proper operation (e.g., the constant target output power). The numerical optimization, traditionally performed by the analytical models, is demanding due to large number of the unknown parameters in complex differential equations. Moreover, the complexity of storing and managing pre-determined configurations in lookup tables increases when considering the complete spectrum of potential operating conditions. On the other hand, Machine Learning (ML) techniques have the potential to effectively overcome such issues by utilizing amplifier-generated data and enabling fast and efficient pumping selection. In this work, YANG model describing the MB optical amplifiers is proposed to enable the Software-Defined Networking (SDN) control. The experimental TDFA characterization is performed, and the results obtained are recorded in a lookup table. In contrast, we then propose a neural network (NN) – based approach for optimizing pumping currents of the adopted TDFA, allowing the SDN controller to enforce such configuration to achieve the desired target output power. The NN model learns the relationship between the target output power, input power, and wavelength, and the suitable pumping, resulting in the Root Mean Squared Error (RMSE) of 0.23 dB between the target and actual output power. Therefore, a reliable and efficient NN-assisted SDN control is achieved. Moreover, an additional NN is employed to predict the noise figure (NF) based on the wavelength, input power, desired output power, and pumping, achieving an RMSE of 0.12 dB between the predicted and measured NF. Finally, experimental validation of C+S transmission is demonstrated.