Joint QoT Estimation and Soft-Failure Localization using Variational Autoencoder

We propose joint QoT estimation and soft-failure localization leveraging the latent space of a variational autoencoder trained on optical spectra. The framework shows F1-scores of 0.989 for soft-failure detection, 0.996 for identification and 0.908 for localization. The QoT estimator reaches an R2-score of 0.998 and a MAE of 0.17 dB.