Blind Estimation of Row Relative Degree Via Constrained Mutual Information Minimization
Blind Estimation of Row Relative Degree Via Constrained Mutual Information Minimization
复制标题
DOI：
10.1007/11679363_44
复制
发表时间：
2006-03
期刊：
影响因子：
--
通讯作者：
J. Even;Kenji Sugimoto
中科院分区：
文献类型：
--
作者：
J. Even;Kenji Sugimoto
本文研究了一种对无穷远不可逆系统的行相对度进行盲（输入信号未知）估计的方法。所提出的方法使用盲信号反卷积方案：将称为解混器的系统应用于观察到的信号并进行更新，以最小化互信息。关键的一点是，去混频器被限制为双真，而系统在无穷远处不可逆，因此无法实现反卷积。但是，行相对度可以通过两个步骤获得：i）最小化分离器输出处的互信息。 ii) 使用所获得的输出的二阶统计量。尽管收敛性尚未得到证明，但大量的数值模拟表明了该方法的有效性。
This paper studies a method for blind (input signals being unknown) estimation of the row relative degrees of a system non invertible at infinity. The proposed method uses a blind signal deconvolution scheme: A system, called demixer, is applied to the observed signals and is updated in order to minimize the mutual information. A key point is that the demixer is constrained to be biproper whereas the system is not invertible at infinity, consequently deconvolution is not achievable. But, the row relative degrees can be obtained in two steps: i) minimizing the mutual information at the output of the demixer. ii) using second order statistics of the obtained outputs. Although convergence has not yet been proved, extensive numerical simulation shows the effectiveness of this method.