盲源分离的单源主导方法研究
文献类型：学位论文
| 作者 | 刘琨 |
| 学位类别 | 博士 |
| 答辩日期 | 2008-06-02 |
| 授予单位 | 中国科学院声学研究所 |
| 授予地点 | 声学研究所 |
| 关键词 | 盲源分离 单源主导 单源主导区间检测 衰减与时延估计 信源数目估计 信号稀疏性 |
| 其他题名 | Research on Single-Source Dominant based Blind Source Separation Method |
| 学位专业 | 信号与信息处理 |
| 中文摘要 | 盲源分离（blind source separation, BSS）也称盲信号分离（blind signal separation, BSS），是一种在传输信道未知情况下，从两个或两个以上传感器接收到的多源混叠信号中分离出各个传输信道的输入源信号的信息处理技术。由于在人机口语对话、多媒体应用以及安全监控等诸多领域有较强的应用需求，盲源分离成为国际上信号处理研究领域的热点之一。过去几年里，对于传感器数量不少于信源数的盲源分离，即盲源适定或超定分离，已经有了比较成熟的办法，而对于传感器数量少于信源数量的情况，即盲源欠定分离或病态盲源分离，依然是个难题。 本论文从盲源混叠模型、混叠信号特征提取、信源数目估计、混叠矩阵参数估计和针对语音信号的信源恢复几个方面进行了深入研究，并以语音信号为对象，采用会议室实际场景模型对本文提出的盲源分离单源主导方法进行了系统详实的实验验证，总结讨论了超定程度、欠定程度、传感器数目、噪声等因素对恢复语音信号质量的影响。本论文主要的创新贡献如下： （1） 提出信号时频域上单源主导的概念及其性质，大大扩充了稀疏性概念涵盖的混叠信源范围，更加接近多信源混叠的客观实际条件，在理论上大大弱化了盲源分离对信源稀疏性的严苛要求； （2） 提出并证明了单源主导区间的检测准则，由于具有较强可操作性可以直接作为工程实践中的检测度量； （3） 设计实现了包括单源主导区间检测、信源数目估计、混叠矩阵复参数估计、针对语音的信源恢复在内普遍适用的盲源分离方法，能够自动处理传感器数目大于等于信源数目的盲源超定适定分离问题和传感器数目小于信源数目的盲源欠定分离问题. |
| 英文摘要 | Blind Source Separation(BSS)，also called blind signal separation，is to recover the unknown independent sources from several observed mixed signals according to the signal’s statistical characteristic without any knowledge of the sources and channels. BSS is an important technique in information and signal processing and is meaningful in speech interaction, multimedia and security fields. Therefore, more and more research works are conducted on BSS worldwide. Previously, a lot of works was done with conditions that the number of sources is not more than that of sensors, called (over-)determined blind source separation. However, it is still difficult to handle the problem with conditions that the number of sources is more than that of sensors, called under-determined blind source separation. In this paper, problems arising from source-mixturing model, mixture-feature extraction, source number estimation, mixture-complex-matrix estimation and source recovery are investigated. And according to the comprehensive experiments conducted with speech-source-mixturing model of conference room, the proposed method with conditions of under-determined, determined and over-determined, and with different signal-to-noise ratio are validated. The innovative contributions are summarized as following: 1. Putted forward the concept of single source dominant (SSD), extending significantly the scope of mixtures that BSS applying to than that based from the concept of signal sparsity. 2. Developed and shown two properties of single source dominant interval (SSDI), resulting in two easy-to-use measures to detect SSDI from mixtures. 3. Designed and implemented a universal BSS method, consisting of SSDI detection, source number estimation, mixture-complex-matrix estimation, and source recovery. The proposed method is capable to perform BSS adaptively with conditions of under-determined, determined and over-determined. |
| 语种 | 中文 |
| 公开日期 | 2011-05-07 |
| 页码 | 90 |
| 源URL | [http://159.226.59.140/handle/311008/406]  |
| 专题 | 声学研究所_声学所博硕士学位论文_1981-2009博硕士学位论文 |
| 推荐引用方式 GB/T 7714 | 刘琨. 盲源分离的单源主导方法研究[D]. 声学研究所. 中国科学院声学研究所. 2008. |
入库方式： OAI收割
来源：声学研究所
浏览0
下载0
收藏0
其他版本
除非特别说明，本系统中所有内容都受版权保护，并保留所有权利。