Title :
A bit level algorithm for subspace tracking
Author :
Götze, Jürgen ; Ali, Markus ; Nossek, Josef A.
Author_Institution :
Inst. of Network Theory & Circuit Design, Tech. Univ. of Munich, Germany
Abstract :
The presented EVD-based subspace tracking algorithm is highly suited for VLSI-implementation. It is based on the binary data representation of the CORDIC scheme combined with the use of approximate rotations. The different modifications of the original EVD-based subspace tracking algorithm can be interlaced, i.e. the number of rotations (sweeps) executed after each rank-1-update and the number of applied best CORDIC angles can be exchanged. The simulations show that it is usually advantageous to work with π = 2 and to adjust the number of rotations/sweeps executed after each update to the specific application, since for small rotation angles (generally present in tracking applications) π = 2 already yields an almost exact rotation
Keywords :
Jacobian matrices; adaptive estimation; direction-of-arrival estimation; eigenvalues and eigenfunctions; parallel algorithms; state-space methods; CORDIC scheme; EVD-based; Jacobi algorithm; VLSI-implementation; approximate rotations; binary data representation; bit level algorithm; eigenvalue decomposition; parallel implementation; subspace tracking algorithm; updating algorithm; Approximation algorithms; Covariance matrix; Data mining; Eigenvalues and eigenfunctions; Hardware; Jacobian matrices; Matrix decomposition; Multiple signal classification; Singular value decomposition; System identification;
Conference_Titel :
VLSI Signal Processing, VI, 1993., [Workshop on]
Conference_Location :
Veldhoven
Print_ISBN :
0-7803-0996-0
DOI :
10.1109/VLSISP.1993.404471