Learning Radio Frequency Patterns of Life for Near Real-time Anomaly Detection

As the electromagnetic spectrum becomes more congested, contested, and constrained, spectrum situational awareness will play an increasingly critical role in the fight against our adversaries. Environments such as airports and crowded littoral regions have multitudes of sources emitting diverse signals and waveforms for a variety of functions. Individual sources may appear and disappear following temporal or environmental patterns of life. It therefore becomes difficult to identify whether a current set of sources is following such patterns or represents a new, anomalous configuration. To maintain spectrum superiority, we propose to develop a capability to efficiently learn normal patterns of life in congested areas, and to detect deviant signals and emission features in near-real time. Metron, Inc., in collaboration with L3Harris Broadband Communications Systems, proposes to customize and extend its machine learning and anomaly detection algorithms and software to learn patterns of life in complex radio frequency (RF) environments and detect anomalies in near real-time. Key innovations of this proposed work include: 1) learning statistical normalcy models to characterize RF spectrum patterns of life; 2) customizing and extending Metron’s existing anomaly detection algorithms and software to alert on spectrum activity that does not match the patterns of life; and 3) generation, in near real-time, of comprehensive operator reports (no operator intervention required) detailing the anomalous features and their exact violations with respect to the learned patterns of life.  To generate the simulated environment for Phase I, we propose to leverage L3Harris’s Broadband Communication Systems (BCS) Radio-Frequency Modem Emulator (RFME) simulator. This simulator can generate diverse communication waveforms and signals that are commonplace to modern-day complex RF environments.