A Hybrid Deep Learning Approach for Performance Prediction in Optical Communication Systems Based on PON Scenarios

Area of Science:

• Optical Communications
• Machine Learning
• Network Performance Analysis

Background:

• Optical access networks are evolving towards higher capacity and complexity.
• Traditional physics-based models struggle with nonlinear and stochastic behaviors in modern passive optical networks (PONs).
• Accurate performance prediction is crucial for next-generation optical networks.

Purpose of the Study:

• To propose a hybrid deep learning (DL) framework for predicting key performance indicators (KPIs) in asymmetric 160/80 Gbps TWDM-PON systems.
• To enhance the robustness and predictive accuracy of performance evaluation methods.
• To provide a computationally efficient alternative to conventional optical simulation.

Main Methods:

• Developed a hybrid DL framework integrating Gradient Boosting Regression and Multi-Layer Perceptron models.
• Utilized an ensemble learning structure for improved prediction.
• Generated a synthetic dataset of 1000 samples simulating diverse transmission scenarios (distance, power, noise).

Main Results:

• Achieved strong agreement between DL-based predictions and conventional optical simulation outcomes.
• Demonstrated superior adaptability and reduced computational complexity compared to traditional methods.
• Obtained high coefficients of determination (R² > 0.94) and low error metrics, confirming framework effectiveness.

Conclusions:

• The proposed hybrid DL framework is a fast and reliable tool for performance prediction in next-generation optical access networks.
• This approach effectively captures complex network behaviors, outperforming traditional methods.
• Highlights the potential of DL for optimizing future optical network design and operation.