フルプログラマブルエントロピー符号器の実現法と性能評価

抄録

Entropy coding algorithms, such as Huffman coding, frequently use bit operations with variable word lengths and although the pipeline processors of existing DSPs can perform numerical processing efficiently, the bit operations in entropy coding are a heavy burden because of a mismatch between these operations and the pipeline processors. We propose a new approach to implement a full programmable entropy CODEC with a combination of a high-level design LSI-CAD system and field-programmable gate arrays. This method simultaneously provides high performance for entropy coding and full programmability. The effectiveness of the proposed approach is demonstrated by using it in designing the Huffman encoder of a JPEG image coding algorithm.