• 研究论文 • 上一篇    下一篇



  1. (1. 西安电子科技大学 电子工程学院,陕西 西安 710071;
    2. 河南理工大学 计算机科学与技术学院,河南 焦作 454001)
  • 收稿日期:2007-09-30 修回日期:1900-01-01 出版日期:2008-12-20 发布日期:2008-12-20
  • 通讯作者: 张伟涛

Adaptive Least-Squares method for blind source separation with an equivariant property

ZHANG Wei-tao1;LOU Shun-tian1;ZHANG Yan-liang1,2

  1. (1. School of Electronic Engineering, Xidian Univ., Xi’an 710071, China;
    2. Dept. of Computer Sci. and Tech., Henan Polytechnic Univ., Jiaozuo 454001, China)
  • Received:2007-09-30 Revised:1900-01-01 Online:2008-12-20 Published:2008-12-20
  • Contact: ZHANG Wei-tao

摘要: 因为等变化自适应盲信源分离要求分离矩阵采用序列更新法则,所以从Cardoso 等提出的相对梯度出发,通过优化非线性主分量代价函数,提出了一种采用序列更新法则的自适应递归最小二乘盲分离算法. 由于算法实现了分离矩阵的序列更新,因此必然具有等变化特性. 在每一步迭代中对分离矩阵进行奇异值分解,然后将其投影到stiefel流形,加快了算法的收敛速度. 计算机仿真结果验证了算法的有效性.

关键词: 盲信源分离, 相对梯度, 等变化性, 序列更新, 奇异值分解

Abstract: Equivariant adaptive blind source separation algorithms require a serial update rule for the demixing matrix. From the point of view of relative gradient proposed by Cardoso, we optimize a nonlinear principal component analysis criterion. Then a modified recursive least-squares (RLS) algorithm for serial updating of the demixing matrix is presented. The equivariant property is guaranteed due to the serial updating rule used at each iteration. To speed up the convergence, the singular value decomposition is employed together with the stiefel manifold projection of the demixing matrix. Computer simulation results demonstrate the effectiveness of the algorithm.

Key words: blind source separation (BSS), relative gradient, equivariant property, serial update, singular value decomposition (SVD)


  • TN911.7