Deep Convolutional Neural Networks for Device Identification

Mon, Nov 9, 2020, 4:30 pm to 5:30 pm
Electrical Engineering

Abstract: Network densification is poised to enable the massive throughout jump expected in the era of 5G and beyond. In the first part of the talk, we identify the challenges of verifying identify of a particular emitter in a large pool of similar devices based on unique distortions in the signal, or ‘RF fingerprints’, as it passes through a given transmitter chain. We show how deep convolutional neural networks can uniquely identify a radio in a large signal dataset composed of over a hundred WiFi radios with accuracy close to 99%. For this, we use tools from machine learning, namely, data augmentation, attention networks and deep architectures that have proven to be successful in image processing and modify these methods to work in the RF-domain. In the second part of the talk, we show how intentional injection of distortions and carefully crafted FIR filters applied to the transmitter-side can help in enhanced classification. Finally, we discuss how to detect new devices not previously seen during training using observed statistical patterns. We conclude by showing a glimpse of other applications of RF fingerprinting, like 5G waveform detection in large-scale experimental platforms and identifying a specific UAV in a swarm.

Bio: Kaushik Chowdhury is Professor and Faculty Fellow in the ECE department and Associate Director at the Institute for the Wireless IoT at Northeastern University, Boston. He was awarded the Presidential Early Career Award for Scientists and Engineers (PECASE) in 2017, the DARPA Young Faculty Award in 2017, the Office of Naval Research Director of Research Early Career Award in 2016, and the NSF CAREER award in 2015. He has received best paper awards at several conferences that include, Infocom, Globecom, ICC (3x), SenSys, ICNC, and DySpan. He is presently a co-director of the Platforms for Advanced Wireless Research (PAWR) project office and the Colosseum RF emulator. His current research interests span applied machine learning to wireless systems, networked robotics, wireless charging and at-scale experimentation for emerging 5G and beyond networks.

Join Zoom Meeting: