CCF: Small: Online Learning and Exploitation of the Radio Frequency Spectrum with Sub-Nyquist Sampling

点击翻译按钮获取中文摘要

This project will close technical gaps to enable cognitive radio receivers to explore the radio frequency spectrum online, using the most advanced form of Analog to Digital conversion, referred to as Finite Rate of Innovation (FRI) sampling coupled with the most advanced learning techniques. We plan to use the well-established framework of the multi-armed bandit (MAB) problem, which models the situation of a cognitive radio agent that simultaneously attempts to acquire new knowledge and to optimize its decisions based on what it has previously learned. Our main contribution lies in combining this framework with this novel Analog to Digital receiver front-end, sampling rate below the so called Nyquist limit, adaptively tuning parameters in the sampling structure to sense spectrum opportunities over a much wider range of frequencies than was previously considered possible, and specifically further below what is attainable myopically, without adaptation. The outcome of our study is a cohesive system model for a cognitive sensors, endowed with a decision engine that can optimize not only what to sample but also how to sample analog signals, leveraging on its expected success in finding spectrum holes. The project will explore the complexity of the overall architecture and, ultimately, evaluate the potential benefits of a cognitive MAB-FRI receiver. By moving learning algorithms a step closer to manage directly the data-acquisition interface to the physical world, the research as broad implications in a variety of related sensing problems. The project will also include activities to engage students in classrooms presenting the basic mathematical tools used in this research and minorities in research projects that contribute to advance the broad field of adaptive systems.