Title :
An algorithm transformation approach to CORDIC based parallel singular value decompositions architectures
Author :
Ma, Jun ; Parhi, Keshab K. ; Deprettere, Ed F.
Author_Institution :
Dept. of Electr. & Comput. Eng., Minnesota Univ., Minneapolis, MN, USA
Abstract :
Singular value decomposition (SVD) has become a standard linear algebra tool in modern digital signal processing. CORDIC based SVD algorithms are among the most popular SVD algorithms which exhibit good numerical properties. The speed of the sequential algorithms is however limited by the recursive feedback loops in the underlying signal flow graph. The critical loop computation time is proportional to the size of the problem which prohibits pipelined processing. This paper addresses the derivation of parallel architectures for SVD updating algorithms. An algorithm transformation approach based on re-timing and matrix associativity is presented to drive parallel SVD updating architectures. These architectures have critical loop computation time independent of the problem size and are suitable for CORDIC arithmetic based VLSI implementations.
Keywords :
VLSI; digital arithmetic; digital signal processing chips; parallel algorithms; parallel architectures; signal flow graphs; singular value decomposition; CORDIC arithmetic; SVD updating algorithms; VLSI implementations; algorithm transformation; critical loop computation time; digital signal processing; linear algebra tool; matrix associativity; numerical properties; parallel SVD architectures; parallel singular value decompositions architectures; re-timing; recursive feedback loops; sequential algorithms; signal flow graph; Arithmetic; Computer architecture; Digital signal processing; Feedback loop; Flow graphs; Linear algebra; Parallel architectures; Signal processing algorithms; Singular value decomposition; Very large scale integration;
Conference_Titel :
Signals, Systems, and Computers, 1999. Conference Record of the Thirty-Third Asilomar Conference on
Conference_Location :
Pacific Grove, CA, USA
Print_ISBN :
0-7803-5700-0
DOI :
10.1109/ACSSC.1999.831936