Tracking satellite trajectories is a critical step for ensuring stable global communication and
accurate data transfer between ground and low orbit satellites. One of the primary purposes of highly
accurate tracking is to ensure ground dishes are pointed in the exact direction necessary to maximize
Gain-to-Noise Temperature. Thus, a predictive tracking system is needed to steer ground dishes with a
minimum accuracy of one tenth of a degree. This project specifically focuses on taking various inputs
from a wide range of scenarios and training a machine learning algorithm to predict PID parameters
based on previous orbit trajectories.
This project is the continuation of a previous team’s work. This specific team will utilize
machine learning (ML) techniques to efficiently produce control system parameters of an improved RF
simulation model that can be used with Viasat systems. There is also a hardware team, that will aim to
improve upon existing models. Both teams will work in close conjunction with one another.
The project involves using the existing Simulink model, which was left behind by the previous
team. As the project progresses, the hardware team will focus on increasing the fidelity of the model,
and the ML team will be consumers of the improved model. The ML team will then develop and
utilize scenarios representative of the ground station functionality to generate data with the Simulink
model. This data will be used to automatically populate a database that serves as an input to an
advanced ML algorithm.
The ML model will analyze the data being fed into it and will predict necessary corrective
actions from the parabolic dish’s position. Ideally, this analysis will allow the team to effectively
produce accurate PID parameters for the dish-satellite system and reduce angular position and tracking
errors.