Harris Mohamed, Kratos
Keywords: Space Domain Awareness, RF, Pattern of Life, Anomaly Detection
Abstract:
As space evolves into a critical infrastructure sector, it is becoming more congested and more contested. As a result, Space Domain Awareness (SDA) is significantly elevating in importance. In addition to Electro-Optical (EO), Radio Frequency (RF), and Ground-based radar positioning information, another critical SDA component is satellite payload behavior. Given these developments, this paper focuses on establishing the pattern-of-life for space objects using RF data.
While EO and radar have their respective strengths, they each also have critical shortcomings. For instance, optical observations can only be made at certain times of the day and in ideal weather conditions. Similarly, radar takes a vast amount of power to track objects in Geosynchronous Orbit (GEO). Furthermore, radar presents environmental and regulatory concerns given its active nature. Both approaches primarily focus on the orbital state of the object, while neither provide much insight into payload behavior. This leaves a significant information gap for the SDA observers using these systems.
These optical and radar-based system limitations can be overcome for objects that transmit RF energy because RF data can be collected in all weather conditions and at any time of day. The RF measurements collected from these objects can enhance the SDA mission by providing additional insight into the behavior and operations of both the space objects payloads as well as any signals they may be transmitting. Additionally, RF characterizations of downlinked signals (e.g., power, frequency, modulation type, data rate, etc.) can be taken at regular intervals for all observed signals. This provides SDA observers with a deeper understanding of the payload, to include determining intent, attribution, and detecting deviations in behavior.
To accurately perform pattern-of-life analysis on RF data, this paper will explore a variety of algorithms that broadly fall into three categories: statistical analysis, time series analysis, and unsupervised machine learning approaches. RF measurements collected from an object can be represented as both univariate and multivariate time series. First, classical statistics methods will be used to establish baseline performance by flagging anomalous measurements that fall outside 3 standard deviations. Then, we will apply state-of-the-art time series analysis algorithmssuch as Autoregressive Integrated Moving Average (ARIMA) and Ruptures[1] (a change point detection algorithm)to the RF time series. Finally, unsupervised machine learning algorithms, such as LSTM Auto-Encoder (LSTM-AE), will be applied to perform anomaly detection. The best of these approaches will be combined in an ensemble method used to collectively vote on anomalies. These approaches will be tested on real-world data and used to prototype a system that can autonomously alert operators of suspicious behavior on an objects payload.
In this paper, we convey the value that RF pattern-of-life measurements can provide to the SDA mission. First, we introduce the basic concept of operations for collecting and using RF measurements of a space object and its payload in addition to describing the various measurements that can be made. Then, we demonstrate how these basic measurements can be accumulated over time to build a pattern-of-life estimate for space objects as well as discuss approaches for using changes in this pattern-of-life to identify events relevant to SDA missions. Lastly, we evaluate the accuracy of statistical, time series, and machine learning algorithms to combine them into a pattern-of-life ensemble model. The pattern-of-life ensemble model will be applied to several examples. Nominal real-world data will be used to demonstrate the capabilities of the pattern-of-life ensemble model. Finally, the ensemble model will be applied to anonymous, real-world data that was captured during the jamming of commercial satellites used by Russia during the Russia/Ukraine conflict.
References
[1] Charles Truong, Laurent Oudre, Nicolas Vayatis. ruptures: change point detection in Python Jan. 2, 2018. [Online] Available: https://arxiv.org/pdf/1801.00826.pdf
Date of Conference: September 19-22, 2023
Track: SDA Systems & Instrumentation