Implementing ML & AI for Automatic ELINT Identification

What AI-enabled suite of tools could enable the IC to increase the pace and quality of threat-processing and threat warning?  What are more robust ways to process data and decrease data-load on operators? From the most recent National Defense Strategy, there is a renewed focus on peer adversaries, along with the growing interest of incorporating machine learning techniques to aid operators in an increasingly clustered and contested electromagnetic environment. The dense electronic intelligence (ELINT) environment in these countries while performing strategic reconnaissance missions for the Air Force has highlighted the gaps in our automated equipment’s capacity to distinguish between land-based tracks and air-based tracks. While operators can eventually make the distinction between the two, the time necessary to conclude the difference between a Surface to Air Missile (SAM) or a Ship (surface track) vs an Airborne Interceptor (AI) would likely result in massive blue-force loss in a wartime scenario.

• Baird, Lt. Col. Michael D., "Implications of Artificial Intelligence Integration into Intelligence, Surveillance and Reconnaissance Operations," AFGC thesis, 2020, 40 pgs. 

  • Focuses on resolving the data-load problem by shifting the Intelligence, Surveillance, and Reconnaissance (ISR) enterprise from a manpower-intensive framework to a human-machine teaming approach. The author notes that employing AI on autonomous platforms allows sensors to conduct automated target recognition and target discrimination directly at the point of collection. This capability significantly reduces the amount of raw data that must be transmitted back to processing centers, solving bandwidth issues in contested environments and ensuring analysts only receive usable, filtered information to increase the pace of threat warning.

• Beckett, Lt. Col. Gary P., "Leveraging Artificial Intelligence and Automatic Target Recognition to Accelerate Deliberate Targeting," AWC SSP, 2020, 25 pgs. 
  • Explains how Artificial Intelligence and Automatic Target Recognition (ATR) can accelerate threat processing by mitigating the intelligence community's data-load challenge. ATR software processes sensor data to automatically locate and classify targets based on characteristics like shape, velocity, and radio frequency signatures, distinguishing between different threats instantaneously. By pairing ATR with machine learning and deep learning for data fusion and predictive analysis, the Intelligence Community can process massive amounts of multi-source data far faster than human operators, significantly reducing the human burden in the Processing, Exploitation, and Dissemination (PED) cycle.
• Bojanic, Maj. Oliver, "Advancing Tactical Command and Control: A Comparative Case Study Analysis in Harnessing Artificial Intelligence for Enhanced Efficiency and Optimization," AFGC thesis, 2023, 71 pgs. 
  • This research highlights how predictive models and Deep Learning algorithms can revolutionize Tactical Command and Control (Tac C2) by relieving operators from the fatigue of manual threat processing and picture building. The paper illustrates that an AI-augmented system can instantaneously evaluate raw sensor and radar data to autonomously determine an opposition platform's type, flight profile, and employed tactics before a human operator even grasps what is happening. By adopting a "human-on-the-loop" framework, the AI shoulders the burden of time-consuming, repetitive data fusion tasks and filters out noise, giving controllers actionable tactical solutions rather than overwhelming them with raw data. This capability guarantees that operational units maintain rapid, highly accurate situational awareness across a complex battlespace without succumbing to data overload.
• Boukhris, Capt. Hicham, "Artificial Intelligence Applied to Electronic Warfare to Recognize and Classify Radar Signals," SOS AUAR, 2021, 7 pgs. 
  • Answers the question by proposing the integration of Artificial Intelligence into Electronic Warfare to autonomously recognize and classify radar signals in complex, noisy electromagnetic environments. The paper highlights the use of Deep Convolutional Neural Networks (CNN) for Pulse Repetition Interval (PRI) classification, which can take raw signal sequences and bypass manual feature extraction to accurately classify signals. Additionally, it introduces a method using a Non-Negative Matrix Factorization Network and an Improved Artificial Bee Colony Algorithm to extract and fuse conventional radar parameters—such as carrier frequency, pulse width, and angle of arrival—providing a robust way to distinguish between agile emitters and decrease the cognitive load on operators.
• Brode, Michael C., "Battle Management Automation: Balancing Technological Adoption and Trust with Risk," SAASS 2025, 95 pgs.  
  • This paper emphasizes the necessity of adopting automation for combat identification to handle the rapidly shrinking decision space brought on by complex aerial threats and dense signal environments. Reflecting on historical friendly-fire tragedies caused by operator task saturation and misidentification—such as the USS Vincennes incident—the author argues that AI and automation can drastically improve threat processing by reliably synthesizing aircraft kinematics, locations, and electronic signatures (IFF). By delegating the routine and mathematically heavy tasks of tracking and identifying aircraft (distinguishing between civilian, friendly, and specific enemy tracks) to automated systems, battle managers are freed from cognitive overload. This allows operators to focus their mental bandwidth strictly on the complex, final engagement decisions, effectively minimizing the risk of blue-force fratricide or civilian casualties in a clustered battlespace.
• Conway,Keith O., "AI/ML and Electronic Warfare," ACSC, 2023, 13 pgs. 
  • This paper addresses the challenge of data saturation in the electromagnetic operating environment (EMOE) by advocating for the use of artificial intelligence and machine learning (AI/ML) embedded in advanced radar warning receivers (RWR) and self-protection jammers. The author explains that AI/ML techniques—such as Deep Neural Networks and Convoluted Neural Networks—can measure and compare intricate signal metrics like pulse repetition intervals, frequency, and angle of arrival to rapidly categorize and prioritize incoming threats. By automating the analysis of these raw signals, the AI can instantly distinguish between a surface-to-air missile (SAM) and an enemy fighter air intercept radar, displaying only the most relevant, high-priority threat data to the pilot. This autonomous filtering prevents extraneous data from overloading human operators, vastly accelerating the threat-warning and electronic attack kill chain to avoid blue-force losses.
• Harms, Lt. Col. Brent N., "Data Fusion as Software Solution for 2018 OIR Lessons Learned and JADC2," AWC SSP, 2020, 35 pgs. 
  • Addresses the challenge of resolving target ambiguity in noisy electromagnetic environments by highlighting software solutions like Talon Thresher. This cloud-based data fusion tool leverages machine logic, Bayesian inference, and Kalman filters to rapidly process raw sensor data—such as Radar Warning Receiver (RWR) signals—and decipher details like Pulsed Repetition Frequency much faster than a human analyst can. By evaluating kinematic and contextual behaviors (e.g., assessing if a track acts like a formation or is moving laterally to align an Inertial Navigation Unit for a radar-guided missile shot), the system's behavior-based identification provides operators with a higher resolution and confidence level regarding a target's identity and intent. This reduces the data-load on the operator by shifting the tedious correlation of raw data to the machine, allowing the human to confidently distinguish between complex tracks and focus on decision-making.
• Hart, Maj. Mariko, "Vulnerabilities and Challenges of Integrating AI into Future Air Force Intelligence Systems," ACSC RTF, 2020, 22 pgs. 
  • Examines the "Sensing Grid" (Sensor Integration) concept, an architecture designed to leverage Deep Neural Networks (DNNs), Computer Vision, and cognitive modeling to handle the speed and scale of intelligence analysis required for great power conflicts. To robustly process data without overwhelming the operator, the paper advocates for the integration of "visual analytics" and "semantic interaction". These AI-enabled interfaces allow analysts to interactively tune analytical processes and test hypotheses against massive datasets, translating the analyst's dialogue with the data into automated model adjustments. By relying on the Air Force Research Lab's Processing and Exploitation (PEX) programs for "fast sensemaking," this approach aims to streamline data extraction and comprehension so that human analysts are not overloaded by raw data streams.
• Kramer, Capt. (No First Name), "BBP on Artificial Intelligence Augmenting Electronic Attack," SOS AUAR, 2022, 3 pgs. 
  • Proposes directly countering the challenges of signal-dense environments against peer adversaries by augmenting airborne electronic warfare officers with AI, machine learning, and modified Automatic Target Recognition (ATR) software. The paper notes that ATR—traditionally used for identifying objects like airborne radar signals, buildings, and vehicles—has the proven potential to be repurposed for analyzing enemy communications and dense electromagnetic signals. By implementing this AI-enabled software on existing airborne platforms like the EC-130H and RC-135V, the military can automate intelligence analysis, fuse data between assets, and utilize predictive algorithms to identify adversarial firing orders or troop movements, which decreases the operator's analytical burden and drastically shortens the non-kinetic kill chain.
• McLamb, Capt. Elizabeth E., "Integrating Artificial Intelligence to Joint All-Domain Command and Control for the 2030 Fight," SOS AUAR, 2020, 21 pgs. 
  • Explains how the Advanced Battle Management System (ABMS) utilizes a specific suite of AI applications to decrease operator data-load and accelerate threat processing. Specifically, the fuseONE application operates on the cloud at the tactical edge to rapidly process, correlate, and fuse ISR and weapons system data from diverse domains. Meanwhile, the AI/smartONE application applies complex machine learning algorithms to this real-time intelligence to rapidly generate and assess Courses of Action (COAs) at a rate that exceeds human capacity. By shifting operators from being "in the loop" to "on the loop," this suite of tools reduces the cognitive burden on analysts and enables the rapid, decentralized decision-making necessary to maintain an information advantage in contested environments.
• Rogers, Capt. Ilya K., "Data Annotation with DBSCAN and GMM Unsupervised Clustering using Flow Cytometery Slides," SOS AUAR 2021, 9 pgs. 
  • Demonstrates how unsupervised machine learning algorithms can be utilized to autonomously separate noise from actionable intelligence in visual or electromagnetic imagery. The paper illustrates how combining Density-based spatial clustering of applications with noise (DBSCAN) with Gaussian Mixture Modeling (GMM) allows a computer to independently group data points into distinct clusters based on density and distance, without requiring prior human labeling. This computational technique can effectively process "Big Data" to identify hidden patterns—such as detecting drone swarms or analyzing space debris—which reduces the manual data annotation workload on operators and drastically improves decision-making speed by highlighting only the most critical and relevant signals.
• Sutton, Capt. Kelvin J., "Improving Radar Tracking of Highly Maneuvering Targets Using Advanced AI Strategies Such as Hybrid Traffic Modeling and Neural Networks in Conjunction with Kalman Particle Filters," SOS AUAR, 2023, 5 pgs. 
  • Addresses the challenge of accurately tracking and distinguishing highly maneuvering targets by combining hybrid traffic modeling and machine learning with particle filters. Because traditional Kalman filters struggle with the non-linear motion of such targets, the author suggests augmenting particle filters with deep learning neural networks to learn motion patterns directly from data. This AI-assisted predictive modeling allows the tracking system to handle multi-modal distributions, adapt to target behavior in real-time, and significantly improve radar tracking accuracy over traditional methods, ultimately helping operators differentiate between various target trajectories much faster.
• Wilcoxon, Chloe, "BBP On Artificial Intelligence Aided Electronic Warfare," SOS AUAR, 2023, 5 pgs. 
  • To overcome the human-intensive and slow nature of traditional electronic support measures (ESM), this paper proposes soliciting industry to develop specific AI algorithms—such as Convolutional Neural Networks (CNN) and deep Q-networks (DQN)—to be integrated directly into aircraft electronic warfare suites. The author explains that these AI models can process massive amounts of raw sensor data to instantly extract discriminant, static features of incoming inputs, such as pulse repetition interval (PRI) modulation. This enables the system to "fingerprint" and classify entirely new or ambiguous radar signals in real-time, effectively distinguishing between different hostile emitters. By shifting the analytical burden to the machine, the system provides on-demand threat indications and warnings without overwhelming the human operator, ensuring rapid platform survivability in highly dynamic and contested electromagnetic spectrums.
• Yae, Capt. Jung H., "Accelerating Mission Data Reprogramming Process Using Machine Learning," SOS AUAR, 2024, 2 pgs. 
  • Focusing directly on the dense signals intelligence (SIGINT) and electronic warfare environment, this paper advocates for the implementation of Machine Learning to automate the currently unsustainable and heavily bottlenecked mission data reprogramming (MDR) process. The author details how ML models can be trained on vast amounts of historical joint forces SIGINT data to rapidly preprocess signals, calculate precise parameters, and perform modulation identification. This automation helps seamlessly isolate background noise from meaningful data, empowering the system to autonomously identify and explain the specific nature of intercepted electromagnetic emissions. By utilizing a human-in-the-loop approach for final verification rather than primary data sorting, the military can drastically decrease the cognitive load on operators and expedite the tactical decision-making process required to maintain spectrum superiority.