Performance Analysis of Transfer-learning Approaches for QoT Estimation of Network Operating with 400ZR

In the last decade, internet traffic has increased exponentially due to the expansion of bandwidth-intensive applications and the evolution of the concept of the internet of things. To sustain this growth in internet traffic, network operators insist on maximizing the utilization of already deployed network infrastructure to its maximum capacity to maximize the CAPEX. In this context, an accurate and earlier calculation of the Quality of transmission (QoT) of the lightpaths (LPs) is essential for minimizing the required margins that result from the uncertainty of the working point of network elements. This article presents a novel QoT-Estimation (QoT-E) framework assisted by Transfer-learning (TL). The main focus of this study is to present a detailed analysis of two major TL approaches, i.e., the Transfer-learning feature extraction (TLFE) approach and the Transfer-learning fine-tuning (TLFT) method, and demonstrate their effectiveness in minimizing the uncertainties in QoT-E in comparison with standard baseline models like Artificial neural network (ANN) and Convolutional-neural network (CNN). The Generalized signal-to-noise ratio (GSNR) is considered a char-acterizing parameter for the QoT of LP. The dataset utilized in this analysis is generated synthetically using the GNPy platform. Promising results are achieved by reducing the overall required margin and extracting the residual network capacity.