Answers are grounded in this paper and cite their sources.
We propose and experimentally validate a low-dimensional optical performance monitoring approach based on fuzzy membership features extracted from received constellation diagrams in gridless Nyquist-wavelength-division multiplexing (Nyquist-WDM) systems. Fuzzy C-means (FCM) and Gustafson–Kessel means (GKM) are independently applied to ∼10k-symbol windows of received 16-QAM constellations. The resulting membership statistics quantify how the fuzzy association of received symbols is redistributed among constellation-derived clusters and are used to estimate spectral spacing in GHz and optical signal-to-noise ratio (OSNR) in dB, as well as to perform binary detection of spectral overlap and low/high-OSNR conditions. Blind and aided monitoring strategies are evaluated using K-nearest neighbors (KNN), support vector machine/support vector regression (SVM/SVR), and random forest (RF) models in 3×16 GBd and 3×32 GBd Nyquist-WDM systems, including back-to-back (B2B) and 270 km transmission scenarios. The best aided regression results achieved 0.463 GHz MAE and 0.606 GHz RMSE for spectral-spacing estimation, and 0.883 dB MAE for OSNR estimation. For binary monitoring, overlap detection reached 0.868 accuracy and 0.870 F1-score, while low/high-OSNR classification achieved 0.938 accuracy and 0.938 F1-score. SHapley Additive exPlanations (SHAP) analysis shows that aided models rely strongly on the auxiliary impairment variable, whereas blind monitoring mainly relies on low-order membership-rank statistics associated with the first and second ranks of the correct-symbol membership. These results indicate that fuzzy membership features provide a compact representation of constellation geometry for multi-task optical performance monitoring in adaptive gridless optical networks.
No takes yet. Share an insight, caveat, or question.
Zapata et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: