Evaluating the impact of snow on free space optical communication using artificial neural networks for BER prediction

Abstract

Free Space Optical (FSO) communication provides high bandwidth and secure data transmission, which may be of immense importance to substitute wired systems in zones that are not environmentally appropriate for fiber optics. But the performance of FSO is highly sensitive to climate conditions, especially snow precipitation, which has an adverse impact on the signal strength and disturbances in the link. This work aims to study the impact of snow on FSO communication by analyzing changes in BER across various snow visibility ranges. MATLAB simulations have been done to validate the impact of link distance and visibility in different snowy conditions, and BER is predicted with high accuracy using the ANN model. The ANN model in this study uses a feed-forward neural network with one hidden layer containing 10 neurons, and with link distance as the input, the model predicts BER across four different visibility levels as outputs. The mean squared error (MSE) of 0.0011537 is achieved by ANN, proving learning capability and ability of generalization without overfitting. Regression analysis provides additional evidence on the accuracy of the BER estimates for snowy FSO systems with correlation coefficients close to 1 in all data subsets.