DRL-based fragmentation- and impairment-aware resource allocation algorithm in C + L band elastic optical networks

Abstract

Efficient resource allocation and management can maximize the utilization of spectrum resources in C + L band elastic optical networks (EONs). To minimize spectrum fragmentation after resource allocation, it is vital to carefully design routing, band, modulation, and spectrum allocation (RBMSA) algorithms. However, the relentless pursuit of spectrum efficiency can degrade transmission quality, particularly due to inter-channel stimulated Raman scattering effects that exacerbate physical-layer impairment in C + L band EONs. To address this issue, we categorize lightpaths based on their generalized signal-to-noise ratio (GSNR) and propose a fragmentation- and impairment-aware RBMSA algorithm. Considering the dynamic arrival and release of requests that continuously alter the spectrum state of the network, we employ deep reinforcement learning (DRL) for adaptive resource allocation, state sensing and decision-making. Simulation results demonstrate that the proposed algorithm improves the GSNR of lightpaths and effectively reduces network blocking probability compared to traditional heuristic algorithms and DRL algorithms with simpler reward settings.