Universal Signal Matching for RF Threat Classification

In response to Navy SBIR Topic N092-113, “Universal Signal Matching for RF Threat Classification,” Michigan Aerospace Corporation proposes a robust method to identify emitter Electronic Intelligence Notation types from sensor data streams based on state-of-the-art techniques in estimation and detection for radar signatures developed in the arena of Signals Intelligence for Specific Emitter Identification. Our approach combines advanced time-frequency analysis to generate feature vectors for individual pulses, with cluster analysis for de-interleaving. This provides the basis for identifying the frequency agility of pulses as well as the PRI agility of the radars. These features are then compared to a library using Ensembles of Decision Trees, which provide robust classification as well as known/unknown detection. The feature vectors of newly discovered emitters are added to the library and additional examples of previously-known emitters are added to supplement description of the higher-order statistics of the clusters of these emitters in feature space. RF data streams from observational platforms often contain instances of several emitters, multi-path artifacts, and receiver coloration. It is possible to obtain concurrent streams from multiple platforms. In this situation, we employ a technique like Blind Equalization Source Recovery to recover undistorted transmitted pulses from the plurality of sensors.