Performance Comparison of Optical Networks Exploiting Multiple and Extended Bands and Leveraging Reinforcement Learning

Exploiting additional low loss bands of optical fibres is a promising solution to expand the capacity of optical transport networks. Recently, extended bandwidth bands (super bands) have been proposed, having a total bandwidth of 6 THz, instead of the regular 4.8 THz. We compare network performance for bands with regular and extended bandwidths when employing transparent and translucent network designs with and without reinforcement learning on the US-NET reference network topology. A total of four MBT scenarios are considered, namely super C, C+L, super C+L, and C+L+S1-band, where S1 denotes half of the S-band bandwidth. We show that the use of super bands and reinforcement learning significantly improves network capacity compared to the use of regular bands and traditional network design methods.