Design and Implementation of a Highly Efficient

Since the discrete wavelet transform (DWT) is a kind of multi-rate transform, it is difficult to design an optimal computation-time architecture for the DWT. In this paper, we propose a highly efficient VLSI architecture for the 1-D DWT decomposition. This architecture contains two stages of systolic decimation filter banks to guarantee a high throughput and an optimal computation time. Using this architecture, N-point samples with J resolution levels can be computed in N clock cycles spending only JL registers, where L denotes filter length. Due to its regular structure, this architeawe can be easily scaled up with the tap size of the. filters and the number of octaves. The performance of the proposed architecture will be verified by the successful implementation of a 4-tap 3-octave DWT VLSI chip.