MLWiNS: Cross-Layer Integrated RF-Based Data-Driven Wireless Device Classification Framework for Spectrum Access Awareness

点击翻译按钮获取中文摘要

The recent proliferation of wireless devices and networks calls for new techniques capable of identifying devices that are transmitting over different bands of the wireless spectrum. These techniques play a key role in supporting spectrum access awareness applications that, for example, allow regulatory agencies to enforce their spectrum access policies and enable wireless network administrators to monitor for unauthorized network access. The goal of this project is to develop novel machine learning methods for automated wireless device identification that scale well with the massive and diverse numbers of emerging devices. In addition, the project provides educational activities that include graduate course development, teaching curriculum enhancement, and professional training of graduate and undergraduate students. The project will also provide research training opportunities for high school and minority students through Oregon State University’s Apprenticeships in Science and Engineering (ASE) programs.This project develops holistic solutions to cross-layer device classification that integrate all the different system component capabilities together, including transceiver hardware, wireless radio frequency (RF) domain knowledge, and deep learning. The first project thrust leverages features that go beyond conventional in-band impairments to include out-of-band distortions due to transceiver hardware imperfections. Using wideband receivers to capture in-band and out-of-band signals, these impairments can jointly serve as unique signatures for transmitters, thereby increasing classification accuracy and scalability. The second thrust integrates hand-engineered features with models acquired through RF domain knowledge and leverages input-output function mapping acquired through deep learning to assimilate meaningful representations from raw IQ data. The final research direction is to develop novel deep neural network algorithms and architectures that incorporate the domain-specific structure of the RF input data to increase device classification accuracy and scalability.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.