Deep Learning for Anomaly Detection in Satellite Signal Traffic: Securing Space-Based Infrastructure
Sai Karthik Garnepudi, John M. Watkins, M. Edwin Sawan, Geethalakshmi Shivanapura Lakshmikanth · 2026
Satellite Communication (SATCOM) is fundamental to global connectivity, supporting applications ranging from telecommunications and navigation to disaster management and defense. A critical challenge is protecting remote terminals that provide end-user access and continuously report link-state information used for resource allocation. These terminals are increasingly exposed to cyber threats, such as False Data Injection Attacks (FDIAs), in which corrupted telemetry can mislead the controller, degrade link quality, and compromise overall system performance. This work presents a framework for FDIA detection in SATCOM that combines communication-aware data generation with deep learning. Realistic multi-terminal datasets are generated in MATLAB using a 5G-Toolbox, producing physically consistent link Key Performance Indicators (KPIs) such as Signal-to-Interference-plus-Noise Ratio (SINR), Channel Quality Indicator (CQI), Transport Block Size (TBS), throughput, Block Error Rate (BLER), and Energy per bit. Attacks are modeled as stealthy manipulation of a target terminal's telemetry by altering reported channel-quality measurements, thereby propagating the impact to related KPIs. This produces bursty, realistic inconsistencies in the monitoring stream. A deep neural network MultiLayer Perceptron (MLP) classifier is then trained to distinguish normal telemetry from attacked telemetry using the reported KPIs. Experiments on held-out data show strong detection performance, achieving 98.98% accuracy (overall correctness), 85.23% precision (fraction of flagged attacks that are truly attacks), 100% recall (fraction of true attacks detected), and an F1-score of 92.02% (a balanced measure combining precision and recall). The model missed no attacks while maintaining a low falsealarm rate. Overall, the results suggest that pairing physically consistent telemetry generation with realistic FDIA patterns and deep learning-based detection can strengthen SATCOM terminal monitoring and support more resilient resource allocation in the face of evolving cyber threats.