Machine Learning for Spectrum Defragmentation in Space-Division Multiplexing Elastic Optical Networks

Abstract

In Elastic Optical Networks with Space Division Multiplexing, the dynamic allocation and deallocation of frequency slots can generate spectrum fragmentation, which increases the blocking of requests for lightpath establishment. In this article, we introduce a reactive algorithm and a proactive one that can jointly reduce spectrum fragmentation. We introduce a novel defragmentation approach based on an unsupervised machine learning technique to rearrange a fragmented spectrum by clustering lightpaths. A Routing, Modulation Format, Core, and Spectrum Allocation algorithm uses information on the clustering of lightpaths to establish new lightpaths for incoming requests. Results show that our approach can reduce the blocking of requests and spectrum fragmentation.