Design of efficient architectures for discrete orthogonal transforms using bit level systolic structures

Abstract

Discrete orthogonal transforms (DOTs) are important in many applications, including image and signal processing. Novel 1D and 2D bit-level systolic architectures are presented for the efficient implementation of DOTs for image and signal processing. The authors describe the design methodology of the techniques based on the Baugh-Wooley (1973) algorithm, and the associated design, including a case study of an FPGA implementation. They also discuss the efficiency of implementations which have O(N/sup 2/) and O(2nN) as the area and time complexities for 2D structures, respectively, and O(N) and O(2nN) as the area and time complexities for 1D structures, respectively (where N is the transform length and n is the word length). Furthermore, it is shown that the architectures are parameterisable and that the area required by the designs can be predicted for different values of N and n. A comparison with existing and similar structures has shown that the proposed architectures perform better.