A QoS-Aware Routing Protocol for High-Speed Optical Networks Using Reinforcement Learning
Abstract
This paper proposes a QoS-aware routing protocol for high-speed optical networks that leverages reinforcement learning (RL) to adapt path selection to dynamic traffic, physical-layer impairments, and service-level objectives. The RL agent observes network state (e.g., residual bandwidth, path length, OSNR margins, and historical blocking) and selects routes that jointly optimize blocking probability, end-to-end latency, and differential delay while respecting wavelength/slot continuity constraints. We design state, action, and reward formulations that capture QoS priorities and implement training/inference with lightweight overhead suitable for online deployment. Simulation results on realistic network topologies show reduced blocking probability, improved tail latency, and enhanced QoS satisfaction compared with shortest-path, fixed-weight, and heuristic baselines.