The radio spectrum that is usable for wireless communications is a reusable but finite resource. Exponentially growing demand by phones, computers, and the internet of things drives a need for significantly improved dynamic spectrum sharing and allocation. Spectrum allocation is slowly moving towards automated management, with improved efficiency and new wireless services enabled at each step. New spectrum systems still require controlled management of the spectrum. By comparison, this project investigates ways to completely automate the management of spectrum, leaving it to individual devices to learn how to collaboratively share spectrum.
The project team is an interdisciplinary research group that brings together the disciplines of "Big Data", machine learning (ML), telecommunications, and signal processing. Efficient opportunistic radio spectrum allocation requires intelligent dynamic agents that can make real-time decisions at the time scale of tens of milliseconds, and can only be achieved using sophisticated machine learning algorithms. A key ingredient for making informed spectrum allocation decisions is the accurate identification of radio signals such as radio and TV broadcasting, local and wide area data networks, cell phones, analog and digital voice, radar, etc., which have been contaminated by noise, fading effects, timing errors, and transmission channel distortion. The project focus is on the development and implementation of ML strategies for classification, strategies for wireless network configuration and multi-network collaboration. Building upon successful strategies that have been used for image and voice recognition, this project adapts deep learning (DL) algorithms such as convolutional neural networks, deep neural networks, and reinforcement learning networks for the classification of communication signals.
