Bayesian Optimization-Based Algorithm to Improve the Quality of Transmission Estimation

Abstract

We propose a Bayesian Optimization-based algorithm to assess the optical parameters that are taken as input by the QoT computation tool. The method reduces the error in computed OSNR down to 0.07dB.

Field trial to assess Bayesian optimization for improving QoT estimation

“…We set up a learning process to optimize network parameters after each service deployment as follows: after the deployment of the k th service s k , the network parameters are optimized using the performance metrics of the services already deployed S D = {s j : j ∈ [1, k]} as objective func- The learning process described above is applied to the Bayesian optimization in the same way as [3] . We consider four uncertain parameters: the noise figure of amplifiers (NF), connector loss (CL), fused loss in the fiber (FL), and the power equalization indicator (PEI), which controls the output power of the Reconfigurable Optical Add Drop Multiplexer (ROADM).…”

“…In this paper, we improve our Bayesian optimization-based model proposed in [3] by using both SNR and power measurements in the objective functions. Moreover, we validate this new model by applying it in a live operational network as part of a field trial.…”

“…We assess the performance of our solution by computing the QoT estimation accuracy on S D and S N D . In the remainder of this paper, we refer to the error in the estimated SNR of deployed services S D as the training error and the error in the estimated SNR of non deployed services S N D as the validation error.The learning process described above is applied to the Bayesian optimization in the same way as[3] . We consider four uncertain parameters: the noise figure of amplifiers (NF), connector loss (CL), fused loss in the fiber (FL), and the power equalization indicator (PEI), which controls the output power of the Reconfigurable Optical Add Drop Multiplexer (ROADM).…”