| 摘要: |
| 针对基于核函数的非线性盲源分离算法性能对核函数及其参数选择依赖性强这一问题
，提出采用批处理方法代替聚类和核主成分分析方法来构造低维近似子空间的正交基，以
改进基于核函数的非线性盲源分离算法对核函数及其参数变化的稳健性，并对这种改进的非
线性盲源分离算法进行了完整的分析。通过仿真实验，对分离信号与源信号求相似度，可以
看到提出的基于批处理的非线性盲源分离算法能够取得更稳健、准确的分离效果。 |
| 关键词: 信号处理 非线性盲源分离 核函数 聚类 批处理 相似度 |
| DOI： |
|
| 基金项目:国家自然科学基金资助项目(60772056) |
|
| Nonlinear Blind Source Separation Algorithm Based on Batch and Kernel Function |
| YU Hua-gang,GAO Jun,HUANG Gao-ming |
| () |
| Abstract: |
| To solve the problem that the performance of nonlinear blind source separation a
lgo
rithm based on kernel function is dependent on the kernel function a
nd its parameters, this paper proposes using batch methods to construct orthonor
mal basis for reduced dimension approximate subspace instead of clustering and K
PCA methods. This improved nonlinear blind source separation algorithm based on
batch and kernel feature space is investigated firstly, and then is used
to improve the robustness to the variety of the kernel function and its paramete
rs.
The simulation results illustrate that the algorithm based on batch is more robu
st and is relatively simple and effective. |
| Key words: signal processing nonlinear blind source separation kernel function clustering batch resemble
degree |
《电讯技术》编辑部 