Satellite communication link anomaly identification assisted by AI/ML

Like all communication systems, satellite communication often encounters anomalies from a variety of impairments, such as weather and interferences. Timely anomaly identification and resolution is critical in improving the quality of service. With the advances in Artificial Intelligence and Machine Learning (AI/ML) tools, we explore how to leverage AI/ML to assist satellite communication anomaly identifications and resolutions. In this paper, we present our work on developing the AI assisted tool using an efficient convolutional neural network (CNN) model with multi-classification architectures, where various communication link parameters are classified in a single model. Optimizations are made to the activation function, pre-processing of input data, as well as the loss function for back-propagation. The model was trained using the hybrid data training approach consisting of both field data and simulated data. The model was tested using various link scenarios for both RF and lasercom links. Preliminary results showed 90% or higher confidence levels for most space-ground RF links and 70% or higher for lasercom space-ground links.