Measures and Methods for Optimal ELINT Signal Detection and Classification

The warfighter is presented with a complex and dynamic spectrum environment cluttered with radio signals from commercial users, allied forces, and adversaries.  It is imperative that they be able to recognize and react to signals in real time.  This makes signal classification, and the cognitive ELINT processing applications built on top of it, a critical task for achieving mission objectives. Traditionally, ELINT algorithms must have a high signal-to-noise ratio with specific clutter models to accurately process data. Technology continues to expand while data is increasing in both dimensionality and environmental complexity. Machine learning models have been developed for detection and classification for ELINT applications; however, the success of these models relies on the assumption that the deployed environment is similar to the training environment. Deviations of this nature can cause a model’s performance to degrade, and cluttered environments lend itself to noisy, multi-modal data with lower information content.  Additionally, machine learning models often operate as a black box with little understanding as to the reasoning why a decision is made and with little to no consideration for uncertainty in the input data. Graf Research Corporation (GRC) and Virginia Tech National Security Institute (VT-NSI) have continued an extensive partnership to develop methods for detecting differences between these two environments for ELINT applications.