Quality-aware resource provisioning for multiband elastic optical networks: a deep-learning-assisted approach

Multiband elastic optical network (MB-EON) technology can help to sustain exponential traffic growth in the optical backbone network. However, multiband operation creates high inter-channel stimulated Raman scattering, leading to a high nonlinear impairment (NLI) that may severely affect the optical signal-to-noise ratio (OSNR) of a lightpath. Additionally, the severity of NLI on the channel of interest depends upon the choice of allocated wavelength. Hence, appropriate channel allocation may cumulatively lead to a higher network capacity. This paper proposes a quality-aware resource provisioning scheme in the context of MB-EON that selectively chooses the available channels from different bands in order to achieve the maximum network capacity in the long run. A deep neural network-assisted quality of transmission estimator is considered to estimate the OSNR of a lightpath with accuracy of 99.65% and 0.012 dB variance in estimation error. The performance of our algorithm in the proposed scheme, namely, optical signal-to-noise ratio adaptive first–last-fit (OA-FLF), is analyzed for two geographically diverse networks, namely, BT-UK and the 24-node USA network, in terms of traffic admissibility, quality of established lightpaths, and contiguous aligned available slot ratio (CAASR), and compared with four state-of-the-art baseline algorithms: first fit, last fit, route adaptive first–last-fit, and distance adaptive first–last-fit. Numerical results indicate that the proposed algorithm outperforms all of the baseline algorithms in terms of traffic admissibility. Reported results show that, compared to the baseline algorithms, consideration of the effect of NLI before resource allocation in the OA-FLF algorithm can provide a maximum gain of nearly 30% in terms of traffic admissibility for smaller networks such as BT-UK, whereas, for longer geography such as the 24-node USA network, this traffic admissibility gain becomes close to 61% till 1% blocking.