Deep neural network-based soft-failure detection and failure aware routing and spectrum allocation for elastic optic networks

Abstract

Soft failure with lower optical signal-to-noise ratio (OSNR) might reduce the quality of the supported services. When the soft failure is detected, the affected existing lightpaths are usually rerouted with alternative paths to avoid the use of the degraded link. However, rerouting without distinguishing between hard failure and soft failure may result in a problem of low utilization of network resources. Unlike hard failure, the degraded link under soft failure can still be used to deliver “shorter” traffic service if the transmission quality requirement can be met. To address this problem, a soft-failure detection method based on deep neural network (DNN) is proposed to detect and localize the failure in elastic optical network. Then, a soft failure aware resources allocation algorithm based on genetic algorithm (SFA-GA) is used for routing and spectrum allocation (RSA) in the network. Simulation results show that 99% accuracy of OSNR estimation can be obtained by the proposed DNN-based scheme for the degraded link under soft failure with estimation errors to be less than 0.5 dB. Based on the estimated OSNR evolution over time, the soft failure can be identified with the degraded link localized. Lower blocking ratio with higher network throughput can also be achieved by the proposed SFA-GA than conventional RSA algorithms. When soft failure exists in the network, the proposed SFA-GA can support the highest traffic load among all five algorithms at any given blocking ratio. At a blocking ratio of 0.01, the SFA-GA allows a traffic load as high as 280 Erlangs, which is about 1.5 times of the commonly used Dijkstra routing plus first fit for spectrum assignment. The traffic load can increase to over 400 Erlangs if a higher blocking ratio of 0.1 is allowed.