DRL-Assisted QoT-Aware Service Provisioning in Multi-Band Elastic Optical Networks

Abstract

Multi-band (MB) optical transmission is a promising solution to support the ever-increasing network capacity demand of 5G/6G applications by exploiting extra optical spectrum beyond the C- and L-bands, such as the L+C+S-band. The extensive spectrum resources and complex physical layer interactions in MB systems present challenges for traditional resource management solutions that are evaluated only for the C-band. Effective algorithms tailored for MB optical networks are needed to enable optical networks to provision services efficiently, thereby reducing service blocking and improving network throughput. In this study, we propose a deep reinforcement learning (DRL)-assisted framework for dynamic service provisioning in MB elastic optical networks. The proposed DRL framework aims to minimize long-term bit-rate blocking and includes several innovations. First, an accurate quality of transmission estimation model is employed to profile the performance of the supported modulation formats for each channel on pre-computed routes. Within the DRL agent design, a novel state representation incorporating both route-level and band-level features is designed to enhance the DRL agent's ability to perceive the network conditions. Moreover, a new reward function has been developed to enhance performance and accelerate convergence. Simulations are performed using a number of L+C+S MB systems with and without traffic grooming support. The results indicate that the proposed DRL-assisted framework can reduce bit rate blocking by an average of 35% to 85% compared to the existing heuristic methods from the literature while maintaining an appropriate inference time.