Machine learning aided optimization for balanced resource allocations in SDM-EONs

Abstract

A fine-grained flexible frequency grid for elastic optical transmission and space division multiplexing in conjunction with spectrally efficient modulations is an excellent solution to the coming capacity crunch. In space division multiplexed elastic optical networks (SDM-EONs), the routing, modulation, core, and spectrum assignment (RMCSA) problem is an important lightpath resource assignment problem. Intercore cross talk (XT) reduces the quality of parallel transmissions on separate cores, and the RMCSA algorithm must ensure that XT requirements are satisfied while optimizing network performance. There is an indirect trade-off between spectrum utilization and XT tolerance; while higher modulations are more spectrum efficient, they are also less tolerant of XT since they permit fewer connections on neighboring cores on the overlapping spectra. Numerous XT-aware RMCSA algorithms restrict the number of litcores, cores on which overlapping spectra are occupied, to guarantee XT constraints are met. In this paper, we present a machine learning (ML) aided threshold optimization strategy that enhances the performance of any RMCSA algorithm for any network model. We show that our strategy applied to a few algorithms from the literature improves the bandwidth blocking probability by up to three orders of magnitude. We also present the RMCSA algorithm called spectrum-wastage-avoidance-based resource allocation (SWARM), which is based on the idea of spectrum wastage due to spectrum requirements and XT constraints. We note that SWARM not only outperforms other RMCSA algorithms, but also its ML-optimized variant outperforms other ML-optimized RMCSA algorithms.