Characterizing the worst-case quality of transmission in multi-band (C+L+S) elastic optical networks

The exponential growth of global data traffic has intensified the adoption of multi-band transmission (MBT) in elastic optical networks (EONs). However, defining the worst-case quality of transmission (QoT) for safety-margin dimensioning remains challenging because heterogeneous spectral configurations and the coupled effects of inter-channel stimulated Raman scattering (ISRS) and nonlinear interference (NLI) make maximum occupancy correspond to many distinct spectral configurations rather than a single one. This paper proposes a methodology to identify worst-case generalized signal-to-noise ratio (GSNR) configurations in C+L+S EONs. The Per-Band Homogeneous ( Hom. Band ) spectral-filling strategy is evaluated against a stochastic incremental-allocation model and a genetic algorithm (GA) that searches for near-worst-case conditions. Simulations over an 80 km standard single-mode fiber (SSMF) span using the GNPy library show that the conventional assumption of uniformly filling the spectrum with noise-like 50 GHz channels does not bound the sampled stochastic worst-case GSNR at any of the nine evaluated channel positions. In contrast, the Hom. Band strategy remains conservative relative to all sampled stochastic configurations. The GA finds configurations with GSNR at most 0.075 dB below the Hom. Band estimate, a difference that is small relative to practical engineering margins; Hom. Band therefore provides an empirical conservative estimate of worst-case GSNR and a near-worst-case predictor. Under the single-span scenario, worst-case GSNR is primarily limited by amplified spontaneous emission (ASE) noise and global Raman tilt. Requiring only a few thousand simulations per channel under test (CUT), the Hom. Band method provides an efficient predictor for worst-case dimensioning of single-span multi-band EONs.