Performance Estimation of Outdoor Visible Light Communication System Over FSO Link Employing ML Techniques

In the present work, we propose and investigate the performance of an outdoor, free-space optical (FSO) link employing a visible light communication (VLC) system under various weather and atmospheric turbulence conditions. The Kim and Carbonneau models have been applied for calculating fog and rain-induced attenuation, respectively, to predict the performance of the FSO link in specific regions. A bit error rate (BER) of 10⁻¹⁰ has been observed in case of clear, rain, and fog climate conditions at transmission ranges of 980, 950, and 930 m, respectively, under no turbulence conditions. A dataset comprising different performance parameters, including range, attenuation, and laser input power, was used as input features for various machine learning (ML) techniques. The prediction accuracy of artificial neural networks (ANN), random forest (RF), decision trees (DT), k-nearest neighbors (KNN), and gradient boosting regression (GBR) ML algorithms was assessed using the coefficient of determination (R²) and root mean square error (RMSE) as performance indices. The ANN model achieved the best R² value (0.9942), while RF provided the optimal RMSE (2.78). Effectiveness of ML models in accurate prediction of the system performance has been validated, and the resultant system may be employed for performance monitoring of impairments in optical networks.