Classification prediction analysis of RSSI parameter in hard switching process for FSO/RF systems

This paper presents a new approach when dealing with availability and reliability of wireless optics by employing several prediction methods for RSSI parameter in hard FSO/RF switching methodology. Despite the fact that FSO link signal and its overall communication is strongly affected by weather conditions (fog, heavy rains, snow, etc.), the strength of RSSI will finally result in link availability or a signal fade. Eventually, there is almost unlimited number of weather states which have either positive or negative impact on RSSI and FSO link, respectively. However, the way that nature and surrounding world are organized is behavioral patterns which can be found everywhere. Finding appropriate pattern is mostly matter of scale of observed group of parameters. Machine learning, data mining or big data are interchangeable terms used in pattern finding processes. We have decided to leverage the power of machine learning algorithms to predict RSSI parameter base on measured weather parameters in this paper. First chapter is dedicated to a deep analysis of negative effects encountered when light propagates through an atmospheric channel and deteriorate optical signal. In following, we paid attention to practical design and assembly of a monitoring device. The further discussion evaluates the methodologies of so called soft and hard switching between FSO and RF (Radio Frequency) link. The primary attention was paid to the hard switching technique and to relations between the received signal strength indicator with the parameters influencing atmospheric channel. Selected machine learning methods and their modifications (AdaBoostClassifier with DecisionTreeClassifier, RandomForestClassifier, GradientBoostingClassifier) were applied in this analysis in order to estimate received optical power parameter based on series of weather parameters. Such a process involves classification and regression methods. This paper presents finding out an optimal scheme for input matrix as an essential for machine learning training process. Considered time and calculation requirements are compensated by employing automation tools in overall process. Finally, we conclude and evaluate the outcomes of machine learning models in FSO/RF hard switching process.