CIF: Small: The Common Information Framework and Optimal Coding for Layered Storage and Transmission of Audio Signals

这项研究项目的前提是：i)可伸缩编码器的僵化层次结构代表了一个严重的缺陷，该缺陷对其实际部署构成了生存威胁，尽管随着异质网络的激增和设备能力的多样化，对这种技术的需求是迫切的；ii)要消除这一缺陷，需要从涉及公共信息的信息论原理出发，在其最基本的层面上重新考虑问题。根本原因是，实际数据源实际上从来都不是连续可细化的--实现粗略质量重建所需的信息不是较高质量重建所需信息的适当子集。因此，最优化需要将信息分成不同层共有的信息和每层“私有”的信息。研究沿着以下几条路线进行：推导用于分层编码的公共信息框架的理论基础，分析渐近和有限延迟性能增益，将理论见解转化为分层音频编码的实用框架，扩展具有记忆的信源，以及涉及采样分辨率、多声道音频和多视点视频的可扩展性。研究重点是消除可伸缩多媒体编码器次优的根本原因。这一限制的后果体现在业界的商业决策上，即避免现有的可伸缩编码器，默认使用浪费的(在存储和网络资源方面)各种质量级别的冗余、独立编码的替代方案。因此，除了对信息论的基本贡献外，成功消除这一缺陷还将显著影响多媒体存储和网络的效率，从而影响更广泛的相关高科技行业，包括多媒体内容交付、无线通信，以及满足快速增长的网络资源需求的关键努力。

The premise of this research project is: i) the rigid hierarchical structure of scalable coders represents a critical flaw that poses an existential threat to their practical deployment, despite compelling demand for such technology with the proliferation of heterogeneous networks and diverse device capabilities; ii) eliminating this shortcoming requires reconsideration of the problem at its most fundamental level, from information theoretic principles involving common information. The root cause is that real data sources are virtually never successively refinable -- the information required to achieve coarse quality reconstruction is not a proper subset of the information required for higher quality reconstruction. Hence, optimality necessitates splitting the information into what is common to different layers, and what is "private" to each layer. Research is pursued along several lines: derivation of a theoretical foundation for a common information framework for layered coding, analysis of asymptotic and finite delay performance gains, translation of theoretical insights into a practical framework of layered audio coding, extensions to sources with memory, and to scalability involving sampling resolution, multi-channel audio, and multi-view video.The research focus is on eliminating the underlying cause for suboptimality of scalable multimedia coders. Ramifications of this limitation are exemplified by the industry's business decision to avoid existing scalable coders and default to the wasteful (in storage and network resources) alternative of redundant, independent encoding at various quality levels. Thus, beside fundamental contributions to information theory, successful elimination of this shortcoming would significantly impact the efficacy of multimedia storage and networking, and thereby the broader sector of related high-tech industries including multimedia content delivery, wireless communications, and the critical effort to satisfy the rapidly growing demand for network resources.