Tackling FSO-WDM System Challenges with Artificial Neural Networks: A Comprehensive Analysis

Abstract

The integration of Free-Space Optical (FSO) communications with Wavelength Division Multiplexing (WDM) offers significant advancements in high-bandwidth, high-capacity systems. However, FSO-WDM systems face challenges due to atmospheric impairments and channel fading. Traditional mitigation techniques often struggle to address these complex and dynamic issues. Recently, Artificial Neural Networks (ANNs) have emerged as powerful tools for learning and adapting to system behaviors, providing novel solutions to enhance FSO-WDM performance. This study analyzes FSO-WDM systems, focusing on applying ANNs to predict channel attenuation accurately and enabling more stable transmission. An OptiSystem simulation was conducted across various transmission distances and climatic scenarios, with Q-Factor and Bit Error Rate (BER) as input features and channel attenuation as the target variable. Following preprocessing, the dataset was split into training, validation, and testing sets. The ANN model, implemented in MATLAB, consisted of an input layer, a hidden layer with 10 neurons, and an output layer. Performance was evaluated using Root Mean Square Error (RMSE) and R-squared (R²) metrics. The trained ANN model demonstrated an optimal mean squared error of 0.23439 and strong correlation between predicted and actual attenuation, with R² values of 0.99907 for the training set and 0.99745 for the validation set. These results confirm the model's robustness in accurately predicting channel attenuation across varying conditions.

Keywords

• Free-space optic (FSO)
• Wavelength division multiplexing (WDM)
• Artificial neural networks (ANNs)
• Channel attenuation prediction
• Root mean square error (RMSE)
• R-squared (R²).

References

1. Kebaili, R., Driz, S & Fassi, B. (2024). Tackling FSO-WDM system challenges with artificial neural networks: A comprehensive analysis. The Eurasia Proceedings of Science, Technology, Engineering & Mathematics (EPSTEM), 32, 304-310.