Radio Frequency Machine Learning (RFML) Laboratory
Abstract
In recent years, the Hume Center for National Security and Technology (ÒVT HumeÓ) at Virginia Tech has established itself as a leading innovator in advancing Radio Frequency Machine Learning (RFML) research in support of our national defense partners. RFML is the application of state-of-the-art deep machine learning concepts to wireless communication and electronic warfare applications. In the past 4 years, VT Hume has been awarded RFML research efforts totaling >$10M in funding from our commercial and government sponsors in key RFML innovation areas such as: signal detection and estimation (including LPI/LPD detection) [1-3], signal format identification (e.g. modulation recognition) [3-8], specific emitter identification [9-11], dynamic spectrum access [12], enhancements and robustness [4, 13-17], distributed RFML [18-19], and adversarial machine learning [16, 20-23] (including both vulnerability analysis and exploitation). Additionally, in support of VT HumeÕs primary mission to Òcultivate the next generation of national security leaders by developing and executing curricular, extracurricular, and research opportunities to engage students,Ó we are currently providing resources and faculty mentorship for multiple graduate students in RFML research towards fulfilling their degree requirements and on average 20-30 undergraduate research assistants per semester in RFML-based research. At present, these research efforts have led to four successful MasterÕs degrees [11,13-15] and over 10 publications (with five currently in peer review) [1, 6-10, 12, 17, 19, 21-23]. This DURIP proposal requests funding for a state-of-the-art GPU-based deep learning training system and data storage solution in order to enhance our capabilities for executing on these RFML research opportunities of critical importance to our government sponsors and in order to increase our ability to engage the rapidly growing student base with interest in this critical research area. This equipment would benefit our work in two primary ways. First, it would decrease the data transfer and training times of our deep learning solutions by multiple orders of magnitude (in some cases from weeks to hours). Second, it would allow investigation of much deeper neural network architectures to tackle much more challenging and sophisticated RFML problems. We believe the requested DGX-2 machine learning system and Flashblade data storage system fits perfectly the stated purpose of the DURIP program, enabling university investment in resources and equipment that no individual program at VT Hume could afford.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Jul 09, 2020
- Source ID
- W911NF2010130
Entities
People
- William Headley
Organizations
- Army Contracting Command
- United States Army
- Virginia Tech