Abstract

The generalized signal-to-noise ratio (GSNR) is a typical metric for quality of transmission (QoT), which in wideband systems may not be uniform without adjustment after multi-span transmission. In this study, we utilize a symmetric neural network structure including reverse and forward models to optimize the GSNR by controlling the transmitter launch power and the equalized power after the wavelength-selective switch (WSS) in both C- and C+L-band systems. The proposed digital twin (DT)-assisted method enables accurate GSNR optimization and adaptable adjustments based on physical conditions. Initially, a DT model is established using a neural network to accurately predict the GSNR in a five-span system, which is directly incorporated into half of the optimizer as the forward model. The optimal power is then derived from the middle layer between the inverse and forward models of the optimizer, thereby achieving convergence towards the target. For long-distance transmission, a cascaded model with the same structure is employed to obtain equalized power by configuring the transmitter and WSS in a ten-span system. Furthermore, we examine the optimization performance in dynamic links under various loads and achieve optimal results for C+L-band systems with a mean absolute error (MAE) of less than 0.75 dB. This study provides a promising solution for accurate margin provisioning, efficient network planning and optimization.