摘要
探讨在感知器个数少于源信号个数时的盲分离问题,提出频域中的单源区间矩阵恢复方法,以实现时域检索平均法在频域中的扩展,与传统的聚类算法相比,该算法计算简单、精度高,在理论上能够无偏差地估计混叠矩阵。在源信号的恢复上,根据稀疏的原则,在仅m个源在频域中较大、其余源可近似为0的假设下,得出求解L1范数的简化方法。语音信号仿真实验展示了该方法的性能和实用性。
This paper discusses the underdetermined blind separation problem when the sensors is less than sources. Matrix Recovery in Single Source Intervals(MRISSl) algorithm in the frequency domain is proposed. It is an extension of Searching-and-Averaging Method in Time Domain (SAMTD). Compared with the traditional clustering algorithms, its computational complexity is low and it has the precisely estimated matrix. In theory, it can estimate the mixing matrix without any error. In the sources recovery, a simplified method to resolve the Ll-norm is obtained in term of the sparse principle assumes the only m sources is large or nonzero and the remained sources is smaller or zeros. Several sound sources experiments demonstrate its performance and practicality.
出处
《计算机工程》
CAS
CSCD
北大核心
2009年第12期269-271,278,共4页
Computer Engineering
基金
国家自然科学基金资助项目(60674033
60505005)
关键词
盲信号分离
欠定的盲信号分离
稀疏成分分析
单源区间
Blind Signal Separation(BSS)
underdetermined Blind Signal Separation(BSS)
sparse component analysis
Single Source Intervals(SSI)