Effective Processor Architecture for Matrix Decomposition

Abstract

The singular value decomposition (SVD) and the QR decomposition (QRD) are two prominent matrix decomposition algorithms used in various signal and image processing applications. In this paper, new CORDIC-like algorithms for fast SVD and QRD have been proposed to speed up those processes. The proposed algorithms are based on the Kronecker (tensor) product (KP) of rotation matrices. This product allows implementing multi-angular transformations simultaneously as integrated macro-operation with execution time similar to the CORDIC rotation. The proposed algorithms lead to appreciable improvements in terms of processor area-time.