Operation Deep Sky is a lightweight intrusion detection system to protect satellite communications from jamming or spoofing. The system uses machine learning based anomaly detection and is designed to operate on small CPU-only devices such as Raspberry Pi or Arduino Uno--simulating the small compute in space.
In this study, we explore the efficacy of utilizing Arduino boards equipped with RF transmitters and receivers to simulate and differentiate between typical and anomalous radio signals, which are indicative of potential issues in satellite communications. Given the constrained computational resources available on satellites, it is critical to employ efficient algorithms that maximize performance with minimal compute overhead.
Our methodology involved modeling satellite communication signals under normal and anomalous conditions, transmitted via Arduino setups, to create a robust dataset for algorithm training and testing. We evaluated multiple classification algorithms to detect anomalies in these signals. Among the tested algorithms, the Isolation Forest algorithm emerged as the most effective, offering substantial accuracy while consuming minimal computational resources.
This approach not only demonstrates the feasibility of using Arduino-based setups for preliminary studies in satellite signal integrity but also highlights the potential of Isolation Forest in operational environments where computing power is limited. The implications of this research extend to improving the reliability and security of satellite communications by facilitating early detection and mitigation of signal anomalies.
Operation Deep Sky took First Prize Overall at the 2024 Space Coast Hard Tech Hackathon.
The hack was hosted by Florida Tech's Center for Advanced Manufacturing and Innovative Design. The hackathon challenged engineers to develop solutions for problems related to aerospace, defense, hardware, and manufacturing.
This is the hardware demo
IMG_3126.mp4
The project demos the hardware and software solution together to simulate and detect anomalies in satellite communications:
Simulation Environment: A dedicated setup comprising three Arduino boards was used to mimic satellite communication scenarios accurately. One Arduino board was equipped with an RF receiver, simulating the satellite's receiving module. The other two Arduino boards were fitted with RF transmitters, representing a ground station and an adversary, respectively. The model was specifically trained to recognize the signal patterns from the simulated ground station as normal. Consequently, when the adversary transmitter broadcasted its signal, the model identified it as an anomaly, effectively detecting potential jamming or interference attempts.
The Unsupervised learning Isolation Forest model was trained on simulated data generated using data_gen.py. The simulated data's characteristics came from real satellite transmissions published by SatNOGS.
The embedded system component consists of the following:
- Arduino Devices: Arduino boards equipped with RF transmitters and receivers were used to simulate satellite communications.
- Model Deployment: The trained model was deployed directly on the receiving Arduino board for real-time anomaly detection.
A GUI dashboard was developed using React and Node.js for visualizing the signals and representing what the system would be recognizing.
- Satellite 1: Shows normal signal from the ground station, used to train the model
- Satellite 2: Shows normal signal with occasional anomalies. Detected by the model but not characteristic of an attack
- Satellite 3: Shows a consistent string of anomalies indicating attack. The system would alert the satellite to deploy countermeasures
To get started with this project, follow these steps:
- Clone the repository:
git clone https://github.com/yen936/deep-sky.git - Install the required dependencies for the model and GUI components.
- Generate simulated data using
data_gen.py--current params are based on real sat data. - Train the Isolation Forest model on the simulated data.
- Set up the Arduino devices with RF transmitters and receivers.
- Deploy the trained model on the receiving Arduino board.
- Run the GUI dashboard to visualize the signals and monitor the system's performance.
This project is licensed under the MIT License.