Steepest descent algorithms for optimization under unitary matrix constraint

被引:139
作者
Abrudan, Traian E. [1 ]
Eriksson, Jan [1 ]
Koivunen, Visa [1 ]
机构
[1] Aalto Univ, Signal Proc Lab, SMARAD CoE, Dept Elect Engn, FIN-02015 Espoo, Finland
基金
芬兰科学院;
关键词
array processing; optimization; source separation; subspace estimation; unitary matrix constraint;
D O I
10.1109/TSP.2007.908999
中图分类号
TM [电工技术]; TN [电子技术、通信技术];
学科分类号
0808 ; 0809 ;
摘要
In many engineering applications we deal with constrained optimization problems with respect to complex-valued matrices. This paper proposes a Riemannian geometry approach for optimization of a real-valued cost function J of complex-valued matrix argument W, under the constraint that W is an n x n unitary matrix. We derive steepest descent (SD) algorithms on the Lie group of unitary matrices U(n). The proposed algorithms move towards the optimum along the geodesics, but other alternatives are also considered. We also address the computational complexity and the numerical stability issues considering both the geodesic and the nongeodesic SD algorithms. Armijo step size [1] adaptation rule is used similarly to [2], but with reduced complexity. The theoretical results are validated by computer simulations. The proposed algorithms are applied to blind source separation in MIMO systems by using the joint diagonalization approach [3]. We show that the proposed algorithms outperform other widely used algorithms.
引用
收藏
页码:1134 / 1147
页数:14
相关论文
共 54 条
[1]  
ABRUDAN T, 2005, C REC 39 AS C SIGN S
[2]  
ABSIL P, 2006, NA200601 U CAMBR
[3]   Riemannian geometry of Grassmann manifolds with a view on algorithmic computation [J].
Absil, PA ;
Mahony, R ;
Sepulchre, R .
ACTA APPLICANDAE MATHEMATICAE, 2004, 80 (02) :199-220
[4]   Natural gradient works efficiently in learning [J].
Amari, S .
NEURAL COMPUTATION, 1998, 10 (02) :251-276
[5]  
[Anonymous], 2002, Adaptive Blind Signal and Image Processing: Learning Algorithms and Applications
[6]  
[Anonymous], 1992, FUNCTION THEORY SEVE
[7]  
BRANDWOOD EH, 1983, I ELECT ENG P F H, V130, P11
[8]   DYNAMIC-SYSTEMS THAT SORT LISTS, DIAGONALIZE MATRICES, AND SOLVE LINEAR-PROGRAMMING PROBLEMS [J].
BROCKETT, RW .
LINEAR ALGEBRA AND ITS APPLICATIONS, 1991, 146 :79-91
[9]   LEAST-SQUARES MATCHING PROBLEMS [J].
BROCKETT, RW .
LINEAR ALGEBRA AND ITS APPLICATIONS, 1989, 122 :761-777
[10]   BLIND BEAMFORMING FOR NON-GAUSSIAN SIGNALS [J].
CARDOSO, JF ;
SOULOUMIAC, A .
IEE PROCEEDINGS-F RADAR AND SIGNAL PROCESSING, 1993, 140 (06) :362-370