【作者】 刘彬彬；

【导师】 徐雄；

【作者基本信息】 上海交通大学 ， 电路与系统， 2009， 硕士

【摘要】 基于格型的矢量量化技术是一种极其重要的信号压缩方法，在音频信号处理的许多重要研究领域中，特别是移动通信中，格型矢量量化起着非常重要的作用。它的低计算复杂度和高性能使其可用于高保真低码率的移动通信系统。本文设计了一种适用于低码率音频编码的基于分裂表的格型矢量量化方法，这种方法具有低存储空间需求、低计算复杂度和高性能的特点。本文首先介绍了低码率变换域音频编码技术和其中的核心模块――矢量量化技术。其中格型矢量量化以其码本的高结构化和量化的高效性，受到了普遍欢迎。当今流行的格型矢量量化技术有经典的嵌入式代数矢量量化技术和以其为基础的Voronoi扩展技术。在现有理论的基础上，本文设计了一种新的基于分裂表的格型矢量量化方法。本文选取了经典的高斯格RE8作为基础格型，在这个基础上设计了一个具有23个特征码字的基础码本。并且设计了基于基础码本的量化编码方法，包括直接编码、2D8数据的奇化编码、常见码字的基于缓存的快速编码等方法。在解决格型矢量量化中的过载失真问题时，本文设计了一种基于分裂表的扩展方法。本文针对基础码本的特性，设计了与其相适应的分裂表，并基于分裂表设计了一级扩展编码和二级扩展编码两种编码方法。其中一级扩展编码适用于较小的过载矢量，二级扩展编码适用于较大的过载矢量。最后，本文阐述了格型矢量量化系统基于AVS P10编解码软件的实现，在实现的基础上测试了格型矢量量化系统的主观性能和客观性能，并进行了存储复杂度和计算复杂度的分析。测试和分析表明，本文所设计的基于分裂表的格型矢量量化方法与经典的基于Voronoi扩展的矢量量化方法相比，性能相当，但具有更低的存储复杂度和低得多的计算复杂度。本文所设计的格型矢量量化方法已经加入数字音视频编解码技术标准工作组的移动语音与音频编码标准AVS P10中。更多还原

【Abstract】 Vector quantization (VQ) based on lattice is a very importantsignal compression method. In many fields of audio signal processing,especially in mobile communication, lattice vector quantization (LVQ)plays an important role. LVQ has low computational complexity andhigh performance so that it can be used for mobile communicationsystem which is considerd to be of high-fidelity and low-bit-rate. Thisarticle applied to design a method of lattice vector quantization basedon split table (STLVQ), which is suitable for low bit rate audio coding.STLVQ has some good features, such as low storage requirements,low computational complexity and high performance.This article firstly introduced the low bit rate audio coding intransform domain and its core technology VQ. Among all the VQtechnologies, LVQ is efficient and attractive due to its highly structurednature of code book generated by certain algebraic rule. A method called embedded algebraic vector quantization (EAVQ) and anextension method called Voronoi extension are very popular in today’saudio coding technology.With the existing theory, this article has designed a new LVQsystem based on split table. We designed a base codebook with23leaders which are chosen from the Gauss lattice RE8. And we alsodesigned the encoding methods using base codebook, including directencoding, odd encoding of2D8data, and fast encoding of some normalvectors using cache.In order to resolve the overload distortion problem in LVQ, thisarticle designed an extension method utilizing split table. According tothe characteristics of the base codebook, we designed a split table andalso the encoding method based on it. The encoding method has twolevels, which are called the first-level extension encoding and thesecond-level extension encoding. The first one is applied to smalloverload vectors while the second one is applied to larger ones.Finally, this article described the software realization of the STLVQsystem based on the codec of AVS P10. Then we give the subjectivetests and objective tests performances of the new LVQ system, so as its storage requirement and computational complexity. Testing andanalysis show that STLVQ has almost equal performance with theclassic LVQ using Voronoi extension (VELVQ), but with less storagerequirement and much lower computational complexity.The method of STLVQ designed in this article has accepted byAVS group, and it has become a part of the mobile speech and audiocoding standard AVS P10.更多还原