Fixed point pipelined architecture for QR decomposition

Abstract

Matrix inversion is an essential step in solving least squares problems and finds application in various communication systems and signal processing applications. A direct matrix inversion involves high computational complexity and hence it is performed using various decomposition techniques like Cholesky decomposition, QR decomposition (QRD), LDL decomposition. In this paper, fixed point architecture for QR decomposition based on Givens rotation algorithm is implemented using 2D systolic array architecture and LUT based Newton-Raphson method. The proposed architecture is implemented for 4×4 real matrices on 2 different platforms: Xilinx XC5VLX110T and XC6VLX240T.