Machine Learning Based Lightpath Classifier for Impairment Aware Resource Allocation in SDM-EONs

Abstract

Efficient resource allocation is needed for next-generation space division multiplexing-based elastic optical networks (SDM-EON). This paper presents a deep neural network (DNN) classifier for dynamic routing and core selection in SDM-EONs. It uses an advanced optimization algorithm to predict lightpath availability, considering the quality of transmission and resource availability. The DNN training accounts for spectral availability, and linear and nonlinear physical layer impairments experienced in SDM-EONs. In addition, a feature importance analysis is performed for different traffic loads. The D NN performance and blocking probability are shown to be superior by lowering blocking by more than a factor of 2 compared with the benchmark techniques, when tested on the 14-node NSFNET topology with unseen network states.