摘要
已有的大多数稀疏分量分析算法都是假定源信号是充分稀疏的,或经过小波变换、Fourier变换等后是充分稀疏的,该论文对源信号的稀疏性要求放宽了条件,提出了一种新的基于超平面聚类的欠定混叠稀疏分量分析算法。算法在观察信号向量集中寻找线性无关的向量组,经过分析位于同一个超平面的观测信号向量的数目,确定所有超平面的法向量,通过求解与聚类后的超平面法向量都正交的向量,辨识出混叠矩阵A的所有列向量。数值仿真试验表明了算法的有效性。
Most of sparse component analysis algorithms suppose that source signals are sufficiently sparse or by wavelet packets or FFT. The condition about sparsity of sources was loosened and a new underdetermined mixing sparse component analysis algorithm was proposed based on hyperplane clustering. The algorithm researched linear independent vectors in observed signals, determines all normals ofhyperplanes by analyzing the number of observed signal vectors and identifies all column vectors of mixing matrix A through finding the vectors that are orthogonal to the normals of hyperplanes. Numerical simulation illustrates the effectiveness of proposed algorithm.
出处
《系统仿真学报》
CAS
CSCD
北大核心
2009年第7期1826-1828,共3页
Journal of System Simulation
基金
广东省自然科学基金(07001797
8151009001000044)
广州市科技计划项目(2007J1-C0501)
关键词
稀疏分量分析
欠定混叠
稀疏阶
超平面聚类
sparse component analysis
underdetermined mixtures
sparse of level
hyperplane clustering
