项目名称: 矩阵联合块对角化的理论与算法
项目编号: No.11301013
项目类型: 青年科学基金项目
立项/批准年度: 2014
项目学科: 数理科学和化学
项目作者: 蔡云峰
作者单位: 北京大学
项目金额: 22万元
中文摘要: 矩阵(近似)联合块对角化是指求一个合同变换矩阵使得给定的多个对称矩阵在这个合同变换下同时(近似)成为具有相同结构的块对角矩阵。矩阵联合块对角化问题可以被看为三阶张量的一种分解。矩阵联合块对角化问题在许多信号处理的问题中都有应用,例如源定位,卷积盲源分离等。现有方法都是在假设块对角矩阵的结构是已知的条件下进行的,这就要求人们事先对问题有一定的了解。然而,在实际应用中,这个假设不总是成立。在这个项目中,我们将在不假设已知块对角结构的前提下,探讨研究矩阵联合块对角化问题(盲联合块对角化)。理论上,我们将利用多项式特征值问题的谱分解理论与多项式特征值反问题的理论,致力于建立矩阵盲联合块对角化问题解存在的充分必要条件,并基于此条件挖掘盲联合近似块对角化的理论。算法上,我们利用建立的理论,发展一类全新的算法,并使其高效、可靠、稳定。最终,通过实际问题数据检验我们的算法,并希望其能被一些软件所采用。
中文关键词: 张量分解;矩阵联合块对角化;独立成分分析;;
英文摘要: (Approximate) Joint Block Diagonalization ((A)JBD) of a set of symmetric matrices is to find a congruent transformation which makes the matrices are (approximately) simultaneously block diagonalized with the same zero pattern. JBD problem can be deemed as a decomposition of a third order tensor. AJBD problem arises in many signal processing applications, eg. source localization, convolutive source separation, etc. Currently, when dealing with AJBD problem, people assume that the zero pattern of the block diagonal matrix is given, which requires priori knowledge of the problem . However, this assumption is not always true in practice. In this project, we will focus in (A)JBD in the case when the zero pattern of the block diagonal matrix is unknown(hereafter we will call this case General (A)JBD), which would of course have great impact in pratical applications. Theotically, using the spectral decomposition theory on Polynomial Eigenvalue Problem(PEP) and inverse PEP theory, we shall devote to the establishment of the necessary and sufficient condition for the exsitence of the solutions to Blind JBD. Based on the necessary and sufficient condition, we shall then turn to the research on blind AJBD. Algorithmatically, using the established theory for blind AJBD, we shall propose a class of efficient, reliable and r
英文关键词: tensor decomposition;matrix joint block diagonalization;independent component analysis;;