High performance VLSI architecture for the trellis coded quantization

Abstract

Trellis coded quantization (TCQ) is an efficient technique for encoding memoryless sources. Furthermore TCQ can be incorporated into a transform coding structure (such as the discrete cosine transform) for encoding monochrome and color images with fixed rate or entropy-constrained schemes. In all these cases an expanded codebook is partitioned into subsets used to label the branches of an appropriate graph (trellis). For a given data sequence, the Viterbi algorithm is then used to find the minimum mean square error path through the trellis. We present a generic architecture scheme that can be easily adapted to the different TCQ image compression methods. We also present a formal model that permits a regular and modular design solution that is optimal for a particular set of area and/or speed constraints.