CNS Core: Medium: When Next Generation Wireless Networks Meet Machine Learning

点击翻译按钮获取中文摘要

The demand on wireless networks is increasing at an exponential pace whereas the spectrum resources are severely constrained. Millimeter-wave (mmWave) spectrum offers orders of magnitude higher bandwidth than the microwave spectrum and forms an integral part of next-generation wireless networking standards such as 802.11ad/ay and 5G. Because of high propagation path loss, mmWave communication uses focused directional beams. Unfortunately, signal blockage and dynamic mobility disrupt beam alignment and limit the efficient use of the spectrum. The goal of this project is to build a new paradigm of learning based channel estimation and tracking, network resource allocation, and optimization schemes for mmWave networks operating in a highly dynamic and even non-stationary environment. This project has the potential to transform the way how the next-generation wireless networks operate, enabling more intelligent, adaptive, and context-aware network management solutions.The project consists of the following synergistic thrusts for the successful design and implementation of mmWave communication networks, followed by a comprehensive system-level validation. The first thrust exploits ideas in non-convex optimization to quickly sense the wireless channels and learn optimal beams. The fast channel sensing and beam tracking schemes recover sparse mmWave channels with minimal measurements, even under highly dynamic conditions experienced under high-speed mobility. The second thrust designs online learning based access point (AP) scanning and association schemes for seamless mmWave connectivity. A novel online framework is proposed to learn the statistics of APs and rank them accordingly. This a ranked list will enable the device to identify a suitable AP with high data-rate link fast before the connection is dropped, thus achieving ubiquitous connectivity. The third thrust studies collaborative and distributed resource allocation algorithms, enabling dynamic data sharing with time-varying network topologies encountered in disruptive applications such as autonomous driving and industrial robotics/IoT. Finally, exhaustive experimental validation of the designed algorithms will be performed across various environmental settings and topological conditions.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.