以節省記憶體為基礎之可程式化可變長度解碼器設計

Abstract

H.264/AVC是最新的視訊壓縮標準，與MPEG-2，H.261及H.263相比H.264提供了更好的壓縮效率與壓縮品質。而在這些視訊編碼中最重要的可變長度編碼是一種無失真的高效率編碼，其原理是給比較常發生的資訊較短的編碼長度，反之就給較長的編碼長度，因此在長度不定的狀況下要相對應到固定長度之記憶體位址勢必要花費許多不必要的記憶體空間，以及現今多種視訊標準中都採用了霍夫曼可變長度編碼，其中若要符合多種視訊標準，其多個霍夫曼編碼表將是浪費記憶的瓶頸。

從系統設計的角度，這篇論文提出了用記憶體程式化的方法來符合MPEG-2與H.264之雙模的視訊標準。在演算法方面，我們從霍夫曼編碼方法中著手，採用了霍夫曼編碼中自已本身就有分群組的資訊，來分每一張霍夫曼表的群組，以降低所需記憶體的需求量，再採用在霍夫曼樹叢中反向提供定址位址的概念來把浪費的符號記憶體需求量最小化。在硬體架構設計方面，為了符合多種視訊標準與使用單一SRAM記憶體來降低多張霍夫曼表要分多塊記憶體的額外負擔，我們採用了在記憶體位址中分群組之方法來達成僅使用單一SRAM記憶體即可代表多張霍夫曼表的方法。 最後本論文利用C++語言証實了此演算法可使MPEG-2與H.264所有的可變長度編碼其所有的符號記憶體可達到90.22%的平均利用率，而且利用了UMC 0.18製程合成MPEG-2與H.264雙模的可變長度解碼器在操作頻率200MHz下，邏輯閘與記憶體總數31.12K，不僅可以滿足大部份的應用需求，而且此演算法與主要的硬體架構也可以應用於未來的可變長度解碼器系統之中。

H.264/AVC is the newest video coding standard. Compared to MPEG-2, H.261 and H.263, it provides more efficient compression ratio and quality. However, one of the most important variable length coding is the high efficient lossless coding. The principle is to assign high frequency information to shorter coding length, and vice versa. Therefore, in the situation of uncertain length, it must waste more unnecessary memory space for assigning to fixed length memory address. Furthermore, in the existing various video coding standards, they all adopt Huffman variable length code. It will be a bottleneck in memory consuming to conform to various video standards. In the system point of view, this thesis provides a programmable method to conform MPEG-2 and H.264 dual mode video standards. In the aspect of algorithm, we undertake the method of Huffman coding and adopt the grouping information that naturally exists in its Huffman coding. It groups every Huffman coding tables and reduces the memory requirement. Moreover, a reversed addressing concept in the Huffman cluster is adopted to minimize the wasted symbol memory. Furthermore, in the aspect of architecture design, a single SRAM memory is used to reduce the overhead of memories. A memory address grouping method is adopted to achieve that using single memory represents many Huffman tables. Finally, this thesis proves that the proposed algorithm can achieve 90.22% symbol memory utilization of MPEG-2 and H.264 via C++ tools. Moreover, the MPEG-2 and H.264 dual mode design is synthesized using UMC 0.18 process under operation frequency of 200 MHz. The logic and memory gate counts are 31.12K. The algorithm and main architecture not only satisfies the most of requirements and can be adopted for advanced applications.