Abstract Ultra-reliable optical communication links are required for mission-critical applications such as 5G/6G fronthaul, industrial automation, and data-center interconnects. In these applications the rare events that lead to communication failures are the limiting factor, rather than average performance. This work introduces a Monte-Carlo-driven digital twin framework for reliably designing coherent single-mode fiber optical links using Monte-Carlo simulation methods to compute outage probability for the given link parameters. The optical propagation is modeled using the nonlinear Schrödinger equation and solved using the Split-Step Fourier Method. The model incorporates chromatic dispersion, Kerr nonlinearity, attenuation, and variations in the noise statistics in the environment. Outage probability is defined as P out = Pr( Q < Q th ), where Q -factor is estimated via Monte-Carlo sampling. To reduce the computational cost of rare-event sampling, a neural-network surrogate digital twin is trained to model the relationship between link parameters, noise statistics, and outage probability. The fidelity of the model is assessed using MAE and R 2 metrics. The multi-objective optimization problem of maximizing outage probability while minimizing transmit power is solved using Pareto optimality principles. The results show a dramatic reduction in outage probability compared to designs based on fixed margins while providing a speedup of over two orders-of-magnitude compared to brute-force Monte-Carlo SSFM simulations. All simulations and models are available in an open-source and reproducible Python framework.
Patil et al. (Thu,) studied this question.