Network-Aware Distributed Machine Learning and Sensor Fusion for Spectrum System Intelligence
Abstract
Approved for Public ReleaseThe growth in volume of programmable wireless devices is beginning to cause significant challenges in interference mitigation for both commercial and government radio frequency operators. Current studies have mostly focused on developing spectrum sharing techniques across primary and secondary users, and while these efforts are essential, there are fundamental gaps in translating such methods into practical government and military systems. Todays status quo spectrum sharing does not necessarilyresult in agility and low latency. For example, even if there are abundant, shared spectrum resources available, the delay in determining efficient spectrum assignments can be large. In fact, identifying available spectrum in a region of interest is a relatively slow process vis-a-vis frequency resource grid sensing, gathering, learning, and adaptation across devices with heterogeneous capabilities. Thus, technologies for compressing and quantizing collected spectruminformation from distributed sensors with low latency isessential.The interference management problem becomes even further exacerbated in the presence of non-cooperative links and other adly those encountered in military settings. In these scenarios, intelligent functions like signal classification, anomaly detection,vice spectrum data has the potential of revolutionizing this intelligence, but studies of machine learning in the spectrum domain are only in their infancy.In this program, we will conduct fundamental research in spectrum intelligence for contemporary communication networks to realize these capabilities. Our objective is to establish the foundational methodologies for agile classification of spectrum usage scenarios and adaptation to adversarial threats in congested and contested environments, particularly those comprisedof heterogeneous wireless devices. Specifically, our approach will consist of research along the following three vectors: (i) compressed and quantized spectrum information fusion via spatially distributed, heterogeneous wireless sensors, (ii) real-time, distributed understanding and classification of spectrum usage scenarios based on network-aware learning techniques, and (iii) rapid detection and adaptation to non-cooperative wireless links based on inferred spectrum usage behaviors. Our program will enable new methodologies that are decentralized in nature, to avoid incurring the large overhead of data centralization, while enabling agile, dynamic spectrum intelligence.Our approach will also contain an extensive experimental component, through a testbed of software defined radio(SDR) devices operating in real-world spectrum environments. Measurements and results from these experiments will be shared and used to further refine our theoretical developments. We envision that our efforts will also be guided by databases of wireless signals emerging from the DARPA RF Machine Learning Systems (RFMLS) program, and any related programs being pursued by ONR. Close collaboration with DoD personnel will help shape the methodologies, testbed constFuture DoD networks will face demand and throughput challenges very similar to those seen in commercial systems. If successful, theproposed research will lead to Naval systems, and more generally DoD networks, that can use available spectrum to operate more efficiently in congested and contested wireless environments. Further, the integration of state-of-the-art machine learning techniques with spectrum sensing and sensor fusion could facilitate future spectrum sharing and operation in regions of the world without strongregulatory infrastructures.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- May 05, 2021
- Source ID
- N000142112472
Entities
People
- Christopher Brinton
Organizations
- Office of Naval Research
- Purdue University
- United States Navy