Shriharsh Tendulkar, LeoLabs; Darren McKnight, LeoLabs; Benedikt Reihs, LeoLabs; Harry She, LeoLabs; Erin Dale, LeoLabs
Keywords: Machine learning, object typing, classification, clustering, LEO, ground-based radar network, object characterization, Space Domain Awareness, Space Situational Awareness, LeoLabs, Radar, S-Band, Maneuverability, Artificial intelligence, Doppler, Residual
Abstract:
In this work, we show the use of doppler measurements from Leolabs’ global network of S-band radars for satellite monitoring — identification of loss of attitude control, configuration changes, and object characterization. We developed a pipeline that can provide low latency alerts (analysts prioritize high interest objects for further investigation.
With the ever-increasing population of space payloads and the advent of large constellations of satellites, there is an urgent need among space defense operators for status-monitoring and object characterization at scale. The expected number of active payloads is expected to grow to ~35,000 by 2030 with satellite constellations from multiple nations around the world. It is possible for hostile and non-cooperative constellation operators to hide dormant assets among a large number of commercial payloads or hide information about the maneuverability and operational condition of their assets. Accurate and timely identification of changes in object characteristics as well as identification of anomalous signatures among a large constellation are essential for maintaining defense and security of space and ground assets. High resolution optical imaging, either through satellite- or Earth-based platforms is the gold standard in identifying and characterization of payloads, but it is limited by the sheer volume of imaging tasks and the need for specific lighting and orbital conditions.
The measured radar reflection from a target is a complex summation of scattered signals from various scattering centers across the target. Rotation, tumbling, as well as configuration change (e.g. deployment of a solar panel) can affect the doppler measurements. Micro-doppler signatures, based on Short Time Fourier Transform processing of coherent radar data, have been well-studied; from detecting vibration, rotation, and tumbling in space targets to identification of humans/dogs/bicycles in automobile radars. However, these analyses need computationally intensive re-processing of raw radar data. Here we demonstrate the use of doppler residuals calculated from our routinely collected data to identify rotation/tumbling, and configuration changes in payloads. We calculate kinematic doppler residuals as the residuals in doppler measurements in a single radar pass after subtracting a kinematic model of the estimated center-of-mass motion through the pass. Our pipeline processes over ten thousand well-tracked objects in the Leolabs catalog daily and generates triggers for statistically significant changes in the doppler residuals.
We benchmark our analysis by comparing the residuals distribution for known stable targets (e.g. calibration spheres), compact targets (cube sats), as well as known tumbling and rotating objects. The root-mean-square of the doppler residual distribution is larger for known tumbling objects compared to those of stable objects.
We highlight incidents identified by our pipeline from Leolabs measurements beginning May 2024 with the deployment of a solar sail on NASA’s ACS3 spacecraft (configuration change), untracked debris impact on DELTA 2 rocket body (sudden tumbling), slow decrease in tumbling (gravity stabilization) of a large rocket body, and the start of tumbling on COSMOS 2553 verified through non-Earth imaging.
These machine learning and statistical tools add an independent signal stream to existing indicators of satellite tumbling such as optical light curve variability. Our goal is to enable analysts to monitor and characterize space objects on the scale demanded by the explosive growth of satellites in LEO. These tools are intended for focusing analysis and investigative efforts such as non-Earth imaging on specific objects of interest among large non-cooperative constellations.
Date of Conference: September 16-19, 2025
Track: Satellite Characterization