Memory-efficient Very Large Scale Integration Architecture of 2D Algebraic-integer-based Daubechies Discrete Wavelet Transform

Abstract

Low memory requirement and reduced latency have been two major concerns in the implementation of the 2D Daubechies discrete wavelet transform. In this study, a multilevel 2D Daubechies-4 (Daub-4) wavelet filter architecture based on an algebraic integer (AI) is investigated. This architecture can improve the traditional Daub-4 very large scale integration (VLSI) architecture design and reduce the number of adders in a 1D Daub-4 filter module architecture. The is because the AI-based multilevel architecture does not require any multipliers and can improve the efficiency of accurate calculations. In addition, to solve the problem of the large transpose memory generated by multimedia chip design, we examine the uses of N × N image inputs in the calculation of the Daub-4 filter by importing them into the interlaced read scan algorithm. This investigated architecture not only reduces the size of the transpose memory from N 2 to 10 or 21 (in the Daub-4 and Daub-6 modes, respectively) but also speeds up the sensing and reading of signals and the calculations. We also show that when a fieldprogrammable gate array is combined with the investigated system, it can enhance the implementation of 2D multilevel AI-based Daub-4 and Daub-6 VLSI architectures.