You are here

Complex Emitter Behavioral Analysis Using Machine Learning


OUSD (R&E) MODERNIZATION PRIORITY: Autonomy; Artificial Intelligence/Machine Learning




The technology within this topic is restricted under the International Traffic in Arms Regulation (ITAR), 22 CFR Parts 120-130, which controls the export and import of defense-related material and services, including export of sensitive technical data, or the Export Administration Regulation (EAR), 15 CFR Parts 730-774, which controls dual use items. Offerors must disclose any proposed use of foreign nationals (FNs), their country(ies) of origin, the type of visa or work permit possessed, and the statement of work (SOW) tasks intended for accomplishment by the FN(s) in accordance with the Announcement. Offerors are advised foreign nationals proposed to perform on this topic may be restricted due to the technical data under US Export Control Laws. Please direct questions to the Air Force SBIR/STTR HelpDesk:


OBJECTIVE: Given sequences of observations of unknown radar waveforms, develop behavioral models to enable inference of radar intent and threat level, and to enable prediction of future behaviors. These models should generalize to arbitrary emitters without a priori knowledge.


DESCRIPTION: The proliferation of low cost, high performance computing hardware has enabled the development of increasingly complex radar systems. The agility and modularity of these new threats force electronic support (ES) systems to operate against a much wider threat parameter space where it is impossible to capture every radar system variant in a mission data file (MDF). Numerous techniques have been proposed to address the recognition of known agile multifunction radar systems [1] [2] [3] [4], and many methods for recognizing unknown observations have been proposed in the machine learning literature [5]. However, very few publications consider making useful inferences against unknown radar systems [6] [7]. An accurate understanding of the threat landscape is important when selecting and optimizing an electronic countermeasure response. Therefore, handling unknown signals remains an urgent challenge for ES systems.  If no MDF match for an observed signal is found, the observation is labeled as an unknown. Given a sequence of these observations over time, a range of useful inferences could be made, including the radar’s intent (e.g. search vs. track) and threat level to the ES host platform. The main objective of this effort is to use statistical modeling and machine learning techniques to construct behavioral models for these unknown radar waveforms to track and predict behaviors over time. Developing such models requires consideration of several questions.  What features are needed to construct effective behavior models? Traditional pulse descriptor words (PDWs) containing pulse time of arrival, frequency, pulse width, and amplitude have historically provided enough information for single pulse characterization and emitter identification. However, given the increased complexity of agile multifunction radars, it might be necessary to consider additional features at different timescales.  How can we best apply tools from statistics and machine learning to form behavior models? As this effort considers unknown waveforms with no a priori (MDF) knowledge, such a model should be generalizable to a wide range of potential threats.  Once a behavior model has been fit to a sequence of observations, how can these behaviors be associated with varying levels of threat? For instance, the behavior model might indicate that the radar is deploying tracking modes. Based on this knowledge, and considering other factors such as inferred distance to the threat, the ES system should infer an appropriate threat level.  Given a time history of behaviors, can the model be used to predict future behaviors? Predictive inference could reduce the reaction time of electronic warfare systems, reducing the time required to select and deploy electronic countermeasures.  The contractor will develop and evaluate a software prototype of the proposed modeling technique. The software prototype will need to be written to interface with the government-owned Advanced Research Concepts for Electronic Measures (ARCEM) test and evaluation framework to enable the government to conduct in-house verification testing. ARCEM provides the technical pipeline – technology maturation and staging – between AFRL and the 350th Spectrum Warfare Wing, which can be utilized for spiral transition of promising technologies. The government will provide data for the contractor to use in evaluating the developed behavior models, and will provide interface control documents for the ARCEM architecture. No other government materials, equipment, or facilities are required to successfully address this topic.


PHASE I: Conduct a study to evaluate the feasibility of the proposed solution (referencing (1) – (4) above). Phase I should document the proposed approach to behavior modeling, complete with a discussion of the assumptions made, limitations of the selected modeling approach, and a plan to demonstrate the model effectiveness given government furnished data with proposed performance metrics. The government will provide interface control documents for the ARCEM architecture to enable the contractor to plan their phase II implementation accordingly.


PHASE II: Develop and demonstrate prototype determined to be the most feasible solution during the Phase I study using government furnished data. Deliver ARCEM-compliant prototype source code and final report detailing the theory, implementation, and quantitative performance of the prototype.


PHASE III DUAL USE APPLICATIONS: The emitter agnostic behavior modeling described in this topic supports the vision of cognitive electronic warfare by enabling the host platform to make useful inferences about unknown emissions and formulate appropriate responses. Commercial applications include spectrum sharing and dynamic spectrum access, where predicting the behavior of primary users could enable a secondary user to rapidly maneuver to mitigate interference to primary users.



  1. R. Wiley, ELINT: The Interception and Analysis of Radar Signals, Artech House, 2006.;
  2. L. Cain, J. Clark, E. Pauls, B. Ausdenmoore, R. Clouse and T. Josue, "Convolutional neural networks for radar emitter classification," 2018 IEEE 8th Annual Computing and Communication Workshop and Conference (CCWC), 2018.;
  3. S. A. Shapero, A. B. Dill and B. O. Odelowo, "Identifying Agile Waveforms with Neural Networks," 2018 21st International Conference on Information Fusion (FUSION), 2018.;
  4. Z.-M. Liu and P. S. Yu, "Classification, Denoising, and Deinterleaving of Pulse Streams With Recurrent Neural Networks," IEEE Transactions on Aerospace and Electronic Systems, 2019.;
  5. S. Apfeld and A. Charlish, "Recognition of Unknown Radar Emitters With Machine Learning," IEEE Transactions on Aerospace and Electronic Systems, 2021.;
  6. A. Wang and V. Krishnamurthy, "Signal Interpretation of Multifunction Radars; Modeling and Statistical Signal Processing With Stochastic Context Free Grammar," IEEE Transactions on Signal Processing, 2008.;
  7. V. Krishnamurthy, K. Pattanayak, S. Gogineni, B. Kang and M. Rangaswamy,; Adversarial Radar Inference; Inverse Tracking, Identifying Cognition, and Designing Smart Interference,; IEEE Transactions on Aerospace and Electronic Systems, 2021.


KEYWORDS: Software defined radar; Cognitive electronic warfare; Radar intent inference; Behavior modeling; Probability and statistics; Machine learning

US Flag An Official Website of the United States Government