Domain Knowledge Powered Machine Learning for the Classification of LOS/NLOS Signals for Dedicated-Spectrum SAGIN Networks

The demand for wireless technologies, such as mobile communication and the Internet of things (IoT), is growing exponentially. We develop a novel domain knowledge-injected machine learning algorithm for signal classification propagating over unmanned aerial vehicles (UAVs)-assisted space-air-ground integrated network (SAGIN) network operating in a dedicated spectrum. Importantly, we show for the first time that not all the attributes of the received signal are required to characterize the signal. Instead, we demonstrate that given the underlying machine learning model, there exists a set of few attributes that yields an accuracy higher than 90%. In particular, we show that the channel quality indicator (CQI) and the downlink coding and modulation scheme (DL MCS) are of paramount importance in the characterization of the signal. Bayesian analysis is utilized to minimize the loss function. Additionally, our extensive evaluation of sixteen models, particularly focusing on accuracy and testing time, highlights the efficiency of the proposed domain knowledge-injected learning algorithm, making it an ideal choice for UAV-assisted SAGIN networks requiring rapid data processing.