Title :
A parallel algorithm for generalized inverses of matrices, with applications to optimum beamforming
Author :
Withers, L.P., Jr.
Author_Institution :
E-Systems Inc., Falls Church, VA, USA
Abstract :
The author presents a parallel algorithm for computing the Moore-Penrose pseudoinverse of an m*n matrix in about 20 mn flops using a ring of n processors. The algorithm has a simple systolic structure and can produce pseudoinverses at a rate of about 20m flops each (after a 20mn latency), using a basket of n/sup 2/ processors. Communications between processors are very localized, so the algorithm appears to be scalable to very large matrices. How this algorithm applies to computing optimum beamforming weights is also discussed.<>
Keywords :
array signal processing; inverse problems; matrix algebra; parallel algorithms; systolic arrays; Moore-Penrose pseudoinverse; generalized inverses of matrices; optimum beamforming; parallel algorithm; systolic structure;
Conference_Titel :
Acoustics, Speech, and Signal Processing, 1993. ICASSP-93., 1993 IEEE International Conference on
Conference_Location :
Minneapolis, MN, USA
Print_ISBN :
0-7803-7402-9
DOI :
10.1109/ICASSP.1993.319132
