RAS Team Proposal for Complex Emitter Behavioral Analysis Using Machine Learning

The WIFFLE-BM product, consisting of multiple Machine Learning (ML) techniques, is developed to model behaviors of complex radar with widely varying multi-agile waveforms. Behavioral models will enable inference of radar waveform function, intent and threat level, and enable prediction of future behaviors of observations from unknown waveforms. Statistical modeling and variety of ML techniques infer behaviors from waveform observations utilizing multiple modules at the i) receiver dwell and pulse level, ii) Waveform Component level to characterize RF, PRF, Scan agilities, and state changes, and iii) longer time Track levels to predict behaviors over time. In Phase I the behavioral model feasibility study addressing the four key questions in the solicitation is performed. Assumptions, limitations, and test approach/plan are developed to demonstrate effectiveness with quantitative performance metrics. In Phase II, the WIFFLE-BM prototype is developed and demonstrated using the defined e metrics and government furnished data.  Prototype source code (ARCEM-compliant) and report summarizing theory, implementation and performance are delivered. Multiple Phase III dual-use behavioral inference applications are pursued, including DoD cognitive EW platforms and commercial spectrum sharing and dynamic spectrum access to enable secondary users to maneuver to mitigate interference from primary users in shared DoD and commercial spectrums.