An OSNR monitoring scheme for elastic optical networks with probabilistic shaping

Abstract

We introduce an optical signal-to-noise ratio (OSNR) monitoring method tailored for elastic optical networks employing probabilistic shaping (PS). The OSNR characteristics of PS signals are represented by three-dimensional density histogram matrices with dynamic power function factors and are identified through a lightweight convolutional neural network (CNN). The results show that the mean absolute error of OSNR monitoring can be reduced to less than 0.12-dB and 0.34-dB in back-to-back and optical fiber transmission settings for the four M-QAM modulation formats correspondingly. Additionally, we leverage transfer learning in conjunction with the CNN to facilitate OSNR monitoring in extended-distance scenarios. The results highlight the efficacy of transfer learning in rapidly adapting CNN architectures to varying transmission distances. It is anticipated that the proposed OSNR monitoring scheme shows potential for integration into future elastic optical networks.