Purdue Teams up on $15M IARPA grant to advance frontiers of Radio-Frequency Machine Learning Security
Prof. Shreyas Sen’s group is part of Expedition Technology-led team including Riverside Research that wins Multi-Year Contract on IARPA SCISRS Program
Shreyas Sen, an Elmore Associate Professor with Purdue University’s Elmore Family School of Electrical and Computer Engineering, will participate in an Expedition Technology-led team that will allow further exploration on applications of Machine Learning on Radio-Frequency Communication and Unintended Emanation Signals for improving System Security. Sen leads the Sensing, Processing, Analytics and Radio Communication (SPARC) lab at Purdue that works on Hardware Security as one of the key focus areas.
Expedition Technology (EXP), a leader in the development of innovative solutions with national impact for the defense and intelligence communities, today announced it has received a multi-year award worth more than $15 million from the Intelligence Advanced Research Projects Activity (IARPA). EXP, along with its subcontractors, Purdue University and Riverside Research, will perform work under IARPA’s Securing Compartmented Information with Smart Radio Systems (SCISRS) program.
The SCISRS program aims to develop smart radio techniques that can automatically detect and characterize radio frequency (RF) signals potentially associated with attempted data breaches. SCISRS will elevate the abilities of the Intelligence Community and Department of Defense to safeguard information and data generated, stored, used, transmitted, and received in secure facilities and beyond.
Purdue’s contribution is twofold: Electromagnetic (EM)-emanations and Radiofrequency Physically Unclonable Functions (RF-PUF). For EM-emanations, Sen’s group has been a pioneer in developing Machine Learning based on Physical Side-Channel Signals. In 2019, Purdue developed the first Cross-Device ML Attack for Power traces (X-DeepSCA) and recently in 2021 has extended this concept to develop the first Cross-Device ML Attack using EM emanations from cryptographic Integrated Circuits (IC) (EM-X-DL). The deep expertise of how to use ML algorithms for EM emanations will be leveraged to advance the frontiers of detection on unintended EM emanations under the IARPA SCISRS Program.
Purdue first proposed the idea of RF-PUF in 2018, using unique multi-dimensional signatures inadvertently imparted by Radio Transmitter’s on Radio Communication Signals to uniquely identify each transmitter or a group of transmitters. This is achieved by Machine Learning running on the receiver-end and is suitable for an asymmetric network as RF-PUF does not require any additional hardware on the resource-constrained transmitter side. Recently, by using Dynamic Irregular Clustering (DIRAC) with incremental learning, the confidence level of the classification is increased as more RF signatures are encountered. Such ML based identification techniques will be explored under the SCISRS program to enable strong identification of known and anomalous radio transmitters.
“I’m really excited to team up with Expedition Technology and Riverside Research on this high-risk high-impact IARPA project as it will allow us to really explore and advance the frontiers of Radio-Frequency Machine Learning Security” said Sen.