Core, Mode, and Spectrum Assignment Based on Machine Learning in Space Division Multiplexing Elastic Optical Networks

• Published in: IEEE access Vol. 6; pp. 15898 - 15907
• Main Authors: Yao, Qiuyan, Yang, Hui, Zhu, Ruijie, Yu, Ao, Bai, Wei, Tan, Yuanlong, Zhang, Jie, Xiao, Hongyun
• Format: Journal Article
• Language: English
• Published: Piscataway IEEE 01.01.2018; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Algorithms; Beams (radiation); Communications traffic; Core; Couplings; Crosstalk; Division; elastic optical networks; Estimation; Machine learning; Machine learning algorithms; mode; Multiplexing; Networks; Optical communication; Optical fiber networks; Propagation; Resource allocation; Resource management; Resource utilization; space division multiplexing; spectrum assignment; Time measurement; Traffic capacity; Training
• ISSN: 2169-3536; 2169-3536
• DOI: 10.1109/ACCESS.2018.2811724

Recently, network traffic has been growing exponentially and almost reached the physical capacity limit of single mode fibers. Space division multiplexing (SDM) is a promising technology to overcome the looming fiber capacity crunch. Especially, few-mode multi-core fibers (FM-MCFs) can aggregate multiple cores into one fiber and two or more modes can be transmitted in one core, which can greatly increase the capacity yet introduce crosstalk constraints including inter- and intra-core crosstalk. To our best knowledge, there is no accurate crosstalk calculation model study in SDM optical networks with FM-MCFs. To address this issue, we first introduce the machine learning into the crosstalk prediction phase and propose a novel crosstalk estimation model (CEM) exploiting the beam propagation method called CEM-beam propagation method (BPM)-machine learning (ML), which can be used to evaluate the crosstalk during the design for the resource allocation scheme. Then, a crosstalk aware core, mode, and spectrum assignment (CA-CMSA) strategy is presented. The simulation results for crosstalk estimation at the wavelength level indicate that the crosstalk at lower frequencies is less than that at higher frequencies. Thus, the lower frequencies are always the first choice in the spectrum resource assignment phase. In addition, for our specific training set, the Levenberg-Marquardt (LM) algorithm based on machine learning performs better on the training, including regression values measurement and time consumption. The simulation results of the proposed CA-CMSA scheme also show that the resource allocation algorithm based on LM can improve resource utilization without increasing total connection set-up time. Thus, it will be the best choice for the resource assignment in SDM networks with FM-MCFs.