Experimental Investigation over 400 km C+L Band Optical Testbed for Quality of Transmission Estimation Using Machine Learning

Multiband transmission (MBT) has appeared as an immediate solution for the network operator to solve the optical fiber capacity crunch problem. However, the quality of the transmission (QoT) in MBT becomes susceptible to nonlinear impairments (NLI) due to the presence of interchannel stimulated Raman scattering (ISRS). This paper considered the scenario of $\mathrm{C}+\mathrm{L}$ band optical transmission, as the additional transmission over L-band becomes the first stepping stone towards the commercialization of MBT. We have experimentally investigated the signal transmission over the $400 \mathrm{~km} \mathrm{C}+\mathrm{L}$ band optical link while using different band-specific amplifiers and transceiver modules. Two launch powers per channel, -1.5 dBm and 0 dBm, have been considered for both 200G DP-QPSK and 400G DP16QAM signal transmission. Multiple real and dummy channels have been multiplexed to generate a 100 GHz spaced WDM spectrum with 96 total channels. The positions of the real channels are tuned across the $C+L$ band spectrum, and the optical power profile across different channels is collected for individual scenarios using the optical spectrum analyzer (OSA). In addition, the QoT has been monitored in terms of signal-to-noise ratio (SNR), Q-factor, and bit-error-rate (BER). In order to leverage the advantage of machine learning (ML) techniques for next-generation optical network designing, these monitoring data have been used, and random forest (RF)-based regressor models have been utilized for QoT estimation. Reported results indicate that the estimation error lies within 1 dB for SNR and Q-factor regressor models, whereas the BER estimator achieves a maximum estimation error of 0.009 units.