Predicting the performance of radio over free space optics system using machine learning techniques

The efficiency and reliability of radio over free space optical communication may be significantly degraded by different weather conditions, thus leading to decrease and increase of signal to noise ratio (SNR) and bit error rate (BER) of the recovered signal respectively. Therefore, real time optical performance estimation is of utmost importance to enable the awareness of channel conditions and for achieving high quality of service. In this work, a terrestrial radio over free space optics (RoFSO) system has been investigated by employing 32-QAM and 64-QAM OFDM modulation schemes under rain, haze and clear weather conditions. Performance of the system is further estimated by employing artificial neural networks (ANN), k-nearest neighbours (KNN) and decision tree (DT) machine learning (ML) techniques with root mean square error (RMSE) and coefficient of determination (R2) as performance metrices. Atmospheric attenuation and different internal system parameters act as input features with BER and SNR of the received signal as the modelling targets. The best performing model has been found to be by applying an ANN approach with lower RMSE values for SNR and BER as 1.87 and 1.26 respectively. The value of R2 predicted by the model are 0.97139 and 0.97802 for SNR and BER data sets respectively.