Machine Learning Based Tool Chain Solution For Free Space Optical Communication (Fsoc) Propagation Modeling

To achieve a highly accurate free space optical communication (FSOC) propagation modeling and analysis for various complex environments, a novel machine learning (ML) based prediction model is proposed in this paper. FSOC attenuation caused by different physical factors such as atmospheric loss, geometric loss, weather conditions, turbulence effects. are first identified and analyzed. Then, a large number of corresponding training data are acquired using MODerate resolution atmospheric TRANsmission (MODTRAN) software. An effective and efficient deep neural network (DNN) is designed to implement the proposed model. The prediction error of the developed machine learning model was as low as 0.02% between the MODTRAN output and predicted machine learning output of transmittance. Finally, a tool chain based FSOC coverage prediction platform was developed by implementing the proposed model in the central processor. Our system is dynamically tailored to an operating status (numbers of factors, condition events, and configurations), which captures a minimum set of information in order to accurately and real-time validate the analysis of weather, terrain, and other conditions to achieve best propagation and coverage prediction for FSOC.