SBIR-STTR Award

The RF spectrum is a dense and dynamic radio frequency (RF) environment. Traditionally, the limited number and relatively slow evolution of signals in the environment allowed signal intelligence, electronic support, and electronic warfare systems to emphasize specific signals of interest (SOIs) with little risk of surprise. Increasingly, civilian, commercial, and national users of spectrum compete for bandwidth using a rich and complex array of evolving waveforms, radio technologies, and access schemes. It is no longer feasible to know in advance how, when, where, and with which signals an adversary may transmit. Instead, new and unusual anomalous uses of spectrum must be identified and evaluated alongside traditional SOIs. To address this need, we propose Osiris, an autonomous sensor processing framework able to detect and characterize previously unknown signals and spectrum use patterns, learn new signal detectors in the field, use RF fingerprints to sort and associate transmissions, identify emitter behaviors, correlate spectrum activities to build RF scene understanding, and optimize RF sensor tasking. Recent advances in the processing of RF spectrum using artificial intelligence and machine learning-based approaches have demonstrated their suitability to a wide range of RF processing tasks, from raw spectrum characterization to reasoning over large temporal, spectral and spatial scales. The objective of this program will be to develop and demonstrate a hierarchical RF pattern of life learning capability suitable to detect anomalies and other atypical uses of RF spectrum.

Benefit:
Osiris is a sensor-agnostic signal processing framework suitable for use in a wide range of new and legacy electronic support and electronic warfare systems, and can scale from small to large platforms. The pattern of life and anomaly detection capabilities developed in this SBIR are well suited to large US Navy surface combatants, submarine and airborne platforms. The autonomous processing can be used in unattended applications, or be guided by an operator. It is similarly applicable to a broad range of Air Force, Army, Special Operations, Intelligence Community and commercial remote sensing applications as well.

Keywords:
Machine Learning, Machine Learning, anomaly detection, spectrum awareness, Autonomy, Patterns of Life, radio frequency sensing

Expedition Technology, Inc. (EXP) is developing Osiris, a modular suite of complementary radio frequency (RF) spectrum awareness applications that simplify and accelerate the way electronic systems adapt to novel, threatening and evolving uses of spectrum. Osiris embodies a new approach to spectrum sensing and signal processing based as much on learning as from expert engineering. This approach enables powerful new ways to monitor and make sense of congested spectrum, identify anomalous activity, authenticate individual emitters, form high-level emitter associations, adapt to environment changes, empower users who may not possess detailed signal knowledge or expertise, and visualize spectrum activity. Osiris is composed of several underlying artificial intelligence/machine learning (AI/ML) algorithms developed under the DARPA RF Machine Learning Systems (RFMLS) program. For users of signal intelligence and electronic support systems, Osiris detects and characterizes important and anomalous spectrum events and activities by identifying and rapidly learning new signal types and spectrum use patterns empowering operators and analysts to parse congested but otherwise normal spectrum behaviors quickly and efficiently, in search of potential threats and important signals of interest. Agile, software defined radios that can rapidly change their frequencies and waveforms challenge traditional signal collectors that are commonly library based and tasked to search for known signals of interest. Incorporating a learned approach allows signal detectors to be derived automatically from collected signal examples, mitigating the need for experts to reverse engineer a signals structure and tailor a custom detector. When trained in the context of uninteresting background signals, learned signal detectors will simultaneously reject signals not of interest. This prevents unimportant signals from passing to downstream processing layers where they are typically processed in the same way important signals are, thereby reducing computation overhead. The expressive power of learned detectors can discriminate complex spectrum activity composed of many individual emitters and waveform types, with or without time dependencies, to efficiently detect previously unseen, anomalous signals and activities an ability traditional RF sensors do not possess. This new ability readily supports learning RF patterns of life and emitter behaviors, providing entirely new intelligence data products and insights.

Benefit:
For commercial users, the RF fingerprinting capability provides an independent and highly robust method of user authentication. By measuring subtle but repeatable hardware-induced distortions to waveform transmissions, Osiris can accurately identify known emitters in population sizes of many thousands of devices. This ability adds a difficult-to-spoof physical layer to cyber security methods, much like biometric techniques. More generally, Osiris supports key requirements of a cognitive demand access communications system by providing autonomous, real-time spectrum monitoring, network analysis and classification of known and unknown spectrum users. More efficient ways of utilizing the RF spectrum are essential for meeting forecasts for wireless communications needs, and demand-access spectrum sharing offers an order of magnitude improvement in spectrum utilization over the fixed channel allocation methods used since the dawn of radio. However, unless and until existing license holders have the spectrum management tools needed to ensure their own access, they will be unwilling to share. Osiris offers the potential to allow license holders to better characterize ad-hoc, unstructured use of spectrum, detect unauthorized transmissions, and identify specific emitters.

Keywords:
Unsupervised learning, RF Anomalies, Spectral Awareness, anomaly detection, AI/ML, Spectrum sensing