CAREER: Towards a Renassaince in Finite Blocklength Length Information Theory

在通信渠道的建模和性能表征中，信息论的数学框架对概念理解和技术发展都做出了重大贡献。尽管它在某些应用中取得了巨大的成功，但在许多应用中，特别是在延迟约束和许多网络场景中，经典信息理论提供的见解较少，或者可用的见解尚未完全集成到现有系统中。经典理论的缺点之一是它依赖于较长的编码块长度。我们相信，信息理论在现代应用中的应用可以通过“有限块长度”信息理论来加速，该理论足够通用，可以处理各种重要的通道模型，但不会太复杂和微妙，以至于系统设计者基本上无法使用。本研究围绕为现代通信系统和网络开发有限块长度信息理论的一般框架的活动展开。指导目标是根据称为互信息密度率的单个随机变量，而不是其期望，开发通信信道和编码策略的适当抽象。预期的抽象将使网络设计者能够表征性能和架构评估，并促进网络模拟中的数量级加速。技术方法是综合的，利用理论分析、计算机模拟和试验台实验。已经确定了将研究和教育结合起来的机会，预期会有大量的相互作用。

In the modeling and performance characterization of communication channels, the mathematical framework of information theory has contributed significantly to both conceptual understanding as well as technology development. Despite its magnificent success for certain applications, there are many applications, particularly in delay-constrained and many network scenarios, in which classical information theory has offered fewer insights, or otherwise the available insights have not been fully integrated into existing systems. One of the drawbacks of the classical theory is its reliance on long coding blocklengths. We believe that application of information theory to modern applications can be accelerated by a "finite blocklength" information theory that is general enough to handle a wide class of important channel models but not so complex and subtle so as to be essentially unusable by system designers.This research centers around activities for developing a general framework of finite blocklength information theory for modern communication systems and networks. The guiding objective is to develop suitable abstractions of communication channels and coding strategies in terms of a single random variable called the mutual information density rate, rather than its expectation. The anticipated abstractions will enable characterization of performance and architectural evaluation by network designers as well as facilitate orders of magnitude speedups in network simulation. Thetechnical approach is comprehensive, leveraging theoretical analysis, computer simulation, and testbed experiments. Opportunities for integration of research and education have been identified, and substantial cross-fertilization is expected.