Gabriele Galimberti

This study proposes a novel two-stage framework that integrates Digital Twin (DT) with Reinforcement Learning (RL) for failure prediction in optical communication networks. The framework focuses on optical amplifier failures, one of the most critical elements impacting network reliability. In the first stage, DTs are employed using the GNPy (Gaussian Noise model in Python) open-source optical network simulation framework to emulate amplifier behavior across diverse operational scenarios. This enables the generation of a comprehensive training dataset to train the Machine Learning (ML) model based on Long Short-Term Memory (LSTM). The LSTM model is able to achieve an accuracy of 98%. In the second stage, RL is applied to enhance the predictive accuracy to 99.5% and to improve the adaptability of the model. The LSTM model is rewarded for accurate predictions using external feedback, allowing it to iteratively refine its performance. This feedback-driven learning mechanism strengthens the robustness and decision-making capabilities of the model under dynamic network conditions. The proposed fine-tuning of the DT-based ML model using RL with external feedback enables proactive and adaptive fault management in optical networks.

This work presents a machine-learning framework for detecting and classifying independent and overlapping anomaly signatures in optical fiber infrastructure using state-of-polarization (SOP) dynamics. The framework exploits temporal variations of the Stokes parameters to characterize bending, tapping, shaking, and their simultaneous combinations. An experimentally collected dataset is used to evaluate multiple machine-learning models under realistic noise conditions representative of practical fiber monitoring environments. The results demonstrate that SOP-based features enable accurate identification of complex fiber disturbances, strengthening physical-layer monitoring and supporting the secure and reliable operation of next-generation optical networks.

This work introduces an advanced comprehensive framework for predictive maintenance by integrating Digital Twin (DT) with multiple Deep Learning (DL) models with the aim of predicting amplifier failures in optical networks. Using GNPy (Gaussian Noise model in Python), an open source framework, a DT is created to emulate network behavior under both normal and failure conditions, enabling the generation of synthetic datasets representative of amplifier degradation fault scenarios. These datasets are used to train DL models based on Convolutional Neural Networks (CNN), Long Short-Term Memory (LSTM), and Long- and Short-Term Time-Series Networks (LSTNet). A comparative analysis shows that all models exhibit strong performance, with LSTM achieving an accuracy of 99% and LSTNet, CNN models closely following at 98% and 96% respectively, demonstrating the ability of these DL models to identify complex temporal and statistical patterns in network telemetry data, facilitating accurate prediction of early failures. This integrated solution provides a scalable and data-driven approach for proactive fault management, improving operational capabilities in optical transport networks.

We present a thorough machine-learning framework based on real-time state-of-polarization (SOP) monitoring for robust anomaly identification in optical fiber networks. We exploit SOP data under three different threat scenarios: (i) malicious or critical vibration events, (ii) overlapping mechanical disturbances, and (iii) malicious fiber tapping (eavesdropping). We used various supervised machine learning techniques like k-Nearest Neighbor (k-NN), random forest, extreme gradient boosting (XGBoost), and decision trees to classify different vibration events. We also assessed the framework’s resilience to background interference by superimposing sinusoidal noise at different frequencies and examining its effects on the polarization signatures. This analysis provides insight into how subsurface installations, subject to ambient vibrations, affect detection fidelity. This highlights the sensitivity to which external interference affects polarization fingerprints. Crucially, it demonstrates the system’s capacity to discern and alert on malicious vibration events even in the presence of environmental noise. However, we focus on the necessity of noise-mitigation techniques in real-world implementations while providing a potent, real-time mechanism for multi-threat recognition in the fiber networks.

We present a thorough machine-learning framework based on real-time state of polarization (SOP) monitoring for robust anomaly identification in optical fiber networks. We exploit SOP data under three different threat scenarios: (i) malicious or critical vibration events, (ii) overlapping mechanical disturbances, and (iii) malicious fiber tapping (eavesdropping). We have used various supervised machine learning techniques like k-Nearest Neighbor (kNN), random forest, extreme gradient boosting (XGBoost) and decision trees to classify different vibration events. We also assessed the framework’s resilience to background interference by superimposing sinusoidal noise at different frequencies and examining its effects on the polarization signatures. This analysis provides insight into how subsurface installations, subject to ambient vibrations, affect detection fidelity. This highlights the sensitivity to which external interference affects polarization fingerprints. Crucially, it demonstrates the system’s capacity to discern and alert on malicious vibration events even in the presence of environmental noise. However, we focus on the necessity of noise-mitigation techniques in real-world implementations while providing a potent, real-time mechanism for multi-threat recognition in the fiber networks.

Relying on a two-measurement characterization, in this work a simple and effective gain profile model for dual-stage optical erbium-doped fiber amplifiers (EDFAs) working under full spectral load conditions is presented and validated. Starting from the model of an ideal EDFA, the gain ripple profile is determined as the target parameters of the amplifier vary using a linear combination of two contributions. Being a consequence of the absorption/emission curves of erbium in the two stages, the first represents a characteristic of the specific device and reflects also the residual impact due to the gain flattening filter (GFF), while the second parameter accounts for a feature of the considered amplifier model, scaling according to the set tilt target. The proposed model faithfully reproduces the dynamics of EDFA by varying the total input power and the target gain and tilt parameters, as shown experimentally on a set of 14 devices divided into 4 different models from 2 vendors, of which 10 in C-band and 4 in L-band. The obtained error distributions present an unbiased peak shape with variability between the 25 and 75 percentiles below 0.2 dB for the worst case in C-band and 0.1 dB in L-band, respectively.

In this work, a cognitive Raman amplifier controller using an evolutionary optimization strategy for both in-field device calibration and optimal pump power configuration design is presented. The conceived methodology allows to jointly characterize the physical layer properties of the optical fiber span, such as the fiber attenuation profile, the entities of the lumped losses due to mechanical connectors or splices, the polarization coefficients between each Raman pump pair and the Raman efficiency curve. The developed framework expects to work using optical channel monitors (OCMs) as telemetry device and it has been experimentally validated, obtaining outstanding results in terms of precision on a C-band setup.

Increasing traffic demands are causing network operators to adopt disaggregated and open networking solutions to better exploit optical transmission capacity, and consequently enable a software-defined networking (SDN) approach to control and management that encompasses the WDM data transport layer. In these frameworks, a quality of transmission estimator (QoT-E) that gives the generalized signal-to-noise ratio (GSNR) is commonly used to compute the feasibility of transparent lightpaths (LP)s, taking into account the amplified spontaneous emission (ASE) noise and the nonlinear interference (NLI). In general, the ASE noise is the main contributor to the GSNR and is also the most challenging noise component to evaluate in a scenario with varying spectral loads, due to fluctuations in the optical amplifier responses. In this work, we propose a machine learning (ML) algorithm that is trained using different ASE-shaped spectral loads in order to predict the OSNR component of the GSNR; this methodology is subsequently used in combination with a QoT-E in the lightpath computation engine (L-PCE). We present an experiment on a point-to-point optical line system (OLS), including 9 commercial erbium-doped fiber amplifiers (EDFA)s used as black-boxes, each with variable gain and tilt values, and 8 fibers that are characterized by distinct physical parameters. Within this experiment, we receive the signal at the end of the OLS, measuring the bit-error-rate (BER) and the power spectrum, over 2520 different spectral loads. From this dataset, we extract the expected GSNRs and their linear and nonlinear components. Through joint application of a ML algorithm and the open-source GNPy library, we obtain a complete QoT-E, demonstrating that a reliable and accurate LP feasibility predictor may be implemented.

Practical needs related to infrastructure management are driving optical network operators to include Raman amplification in order to improve the performance of long fiber spans. Compared to standard erbium-doped fiber amplifier (EDFA) management, Raman amplifiers require a greater degree of control and monitoring due to their distributed nature. Inevitably, this update leads to a key consideration; the introduction of additional telemetry devices with respect to standard EDFA photodiodes, resulting in an increase in required investments. In this work, we present an embedded controller architecture in combination with an ad-hoc probing procedure to manage Raman amplification within disaggregated optical networks, using only standard integrated equipment, allowing an efficient implementation without the introduction of optical channel monitors (OCMs). This proposal is validated using a fully representative experimental campaign, testing both the probing procedure on a single fiber span and the operation of a Raman amplifier using the extracted information.

This paper describes the implementation of a wavelength division multiplexing (WDM) network controller that accounts for the varying conditions of the network physical layer in real-time. A newly added software module is used to estimate the physical layer quality of transmission (QoT-E) in the form of a single transmission figure of merit for every line span in the network, i.e., the optical signal-to-noise ratio (OSNR) degradation. The estimated OSNR degradation is used to verify lightpath feasibility in the network, and produce optimal routing and wavelength assignment solutions in real-time, without the need for a separate network planning tool. The OSNR degradation is estimated without requiring detailed knowledge of the optical components deployed in the network, and consequently the approach is vendor-agnostic. The accuracy of the estimated OSNR degradation is assessed experimentally, and its impact on two network-wide performance indicators is studied through simulation.