Integrating Machine Learning into the International Space Station Antenna Management Software

The International Space Station Antenna Management (IAM) software tool is currently used on console by the CRONUS flight controller in NASA Johnson Space Center’s Mission Control Center (MCC) for International Space Station (ISS) operations. The IAM tool displays graphical line-of-sight imagery from the viewpoint of the S-band and Ku-band antenna systems to the current Tracking and Data Relay Satellite (TDRS), using real-time data as well as predict data. Flight controllers subjectively utilize the IAM tool to determine signal blockage by ISS structure that potentially results in "ratty" communications or an entire loss of signal (LOS). These determinations (or lack thereof) inherently depend on the experience of the flight controller. To limit the subjectivity of an LOS call, we introduce here a transformer neural network machine learning component to the IAM software tool that learns from past circumstances of signal blockage. We train the network on data from prior years of ISS S-band performance data. Each data instance encapsulates various dynamic physical parameters (e.g. antenna gimbal angles and solar rotary joint angles) of the ISS structure to predict the S-band digital automatic gain control (AGC). We report here current results, discuss operational implementation into the IAM software, and consider future applications, including onboard systems.