The mission of this project is to develop an ML-enabled system for estimating the pass geometry of Low Earth Orbit (LEO) satellites that use non-cooperative or undisclosed signal formats. Because most LEO satellites are operated by private entities, their downlink specifications are often kept secret and published orbit data can be outdated, leaving physics-based methods without accurate initial estimates. To address this, the project reframes non-cooperative Doppler tracking as an applied machine learning task, using a Python-based system to estimate closest-approach parameters directly from delay-difference time series collected during a single simulated pass. Additionally, the system detects noise bursts within a pass, with manoeuvres and dropouts left as future work, to enhance the reliability of satellite monitoring and passive orbit validation.
Team
Prof. Ghaziasgar
Supervisor
Raeez Ahmed
BSc Hons Student
Prof. Thron
Co-Supervisor