Nonlinear Impairment-Aware RMSA under the Sliding Scheduled Traffic Model for EONs Based on Deep Reinforcement Learning

• Published in: Journal of lightwave technology Vol. 41; no. 22; pp. 1 - 11
• Main Authors: Zou, Yucong, Cai, Xiaofeng, Zhu, Min, Gu, Jiahua, Wang, Yunwu, Zhao, Guo, Shi, Chenglin, Zhang, Jiao, Cai, Yuancheng, Lei, Mingzheng
• Format: Journal Article
• Language: English
• Published: New York IEEE 15.11.2023; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Blocking probability; Deep learning; Deep reinforcement learning; Heuristic algorithms; Impairment; Interference; Modulation; Nonlinear impairment; Optical communication; Optical fiber networks; Provisioning; Resource allocation; Resource management; Routing; Signal to noise ratio; Slide scheduling; Sliding; Topology; Traffic models
• ISSN: 0733-8724; 1558-2213
• DOI: 10.1109/JLT.2023.3299272

The elastic optical network (EON) can accommodate dynamic and diverse demands of next-generation applications by provisioning flexible lightpaths for them. Consequently, blocking probability of these requests can be reduced especially in sliding scheduled traffic, where requests are scheduled to start well after their arrival. In this paper, a nonlinear impairment-aware routing, modulation and spectrum assignment (NLI-RMSA) problem was investigated and modeled under a sliding scheduled traffic model. We proposed a novel dynamic resource allocation scheme in EONs based on deep reinforcement learning (DRL) to jointly determine the path, modulation format and scheduling time of the lightpath requests. Several designed methods consider the nonlinear impairment in the physical layer to accurately estimate the channel state and increase the feasibility of the allocation scheme. Simulation results on the 14-node NSFNET topology demonstrate that the proposed DRL-based method greatly outperforms the two baseline heuristics, potentially saving at least 38.7% more blocking probability than the baseline heuristics.