WiFiUS: Collaborative Research: Sequential Inference and Learning for Agile Spectrum Use

扩展未来无线服务的一个关键任务是克服频谱短缺。目前，静态分配和刚性监管导致可用频谱资源利用不足。灵活使用频谱的目的是利用未充分利用的频谱。可用的频谱机会可能是不连续的，分散在大带宽上，并且由于无线传输的高度动态性，在局部和有限的时间内可用。这就需要了解如何以最小的延迟有效地发现、评估和利用时频位置变化的频谱资源。此外，以敏捷的方式访问已识别的空闲频谱至关重要。本项目将为频谱状态快速变化时的敏捷频谱访问设计顺序推理和学习算法。与块算法相比，顺序算法的主要优势是它们通常会显著减少决策延迟。该项目的总体目标是为敏捷频谱利用设计顺序推理和学习算法。特别是，该项目将采用先进的顺序推理和学习方法来完成以下三个相互关联但日益复杂和苛刻的任务：1)采用顺序强化学习和顺序推理算法来设计快速发现频谱机会的感知策略；2)设计序列算法，实现快速准确的频谱质量评估；3)建立、维护和开发网络运行区域的干扰图，并将其表示为空间势场。本研究可望在应用和理论两方面作出重大贡献。在应用层面上，本研究通过引入序列分析、机器学习和统计推断等新工具来设计频谱发现、评估和开发策略，有可能大幅提高频谱效率。在理论层面上，拟议的项目将推进序列分析的最新技术，并为最佳停止、控制和机器学习问题的一般方法基础提供新的方法。此外，还将开发利用空间势场、序列统计和先进传播建模的新方法和理论。

A key imperative to expanding future wireless services is to overcome the spectral crunch. At present, static allocation and rigid regulation lead to underutilization of available spectral resources. Flexible spectrum use aims at exploiting under-utilized spectrum. Available spectrum opportunities may be non-contiguous, scattered over a large bandwidth, and are available locally and for a limited period of time due to the highly dynamic nature of wireless transmissions. This fuels the need to understand how to discover, assess and utilize the time-frequency-location varying spectral resources efficiently and with minimal delay. Moreover, it is critical to access identified idle spectrum in an agile manner.This project will design sequential inference and learning algorithms for agile spectrum access when the state of the spectrum varies rapidly. The key advantage of sequential algorithms, as compared to block-wise algorithms, is that they typically lead to significantly reduced decision delays. The overarching goal of this project is to design sequential inference and learning algorithms for agile spectrum utilization. In particular, this project will employ advanced sequential inference and learning methods for the following three interconnected yet increasingly sophisticated and demanding tasks: 1) to employ sequential reinforcement learning and sequential inference algorithms to design sensing policies for rapid spectrum opportunities discovery; 2) to design sequential algorithms for fast and accurate spectrum quality assessment; and 3) to build, maintain and exploit an interference map of the area where our network operates and represent it as a spatial potential field. The proposed research is expected to make substantial contributions to both applications and theory. On the application level, the proposed research has the potential to substantially improve spectral efficiency by introducing novel tools from sequential analysis, machine learning and statistical inference for the design of spectrum discovery, assessment and exploitation policies. On the theoretical level, the proposed project will advance the state of the art in sequential analysis and contribute new approaches to the general methodological base for optimal stopping, control and machine learning problems. Furthermore, new methods and theory of modeling and exploiting knowledge of interference using spatial potential fields, sequential statistics and advanced propagation modeling will be developed.