具抗錯力之以記憶體為基礎的雙模視訊標準可變長度解碼器設計

Abstract

本論文提出在支援雙模(H.264/AVC和MPEG-2)視訊壓縮標準解碼器下的熵解碼器中，增進解碼表格在記憶體中的使用效率。此熵解碼器採用多張解碼表格合併演算法並可程式化以包含不同視訊壓縮標準的熵解碼。首先利用單一表格將字碼分類到不同的群組，只把各個群組的最重要的資訊存於記憶體中，解碼的過程只需要用字碼的運算和群組資訊就可以算出符號在記憶體中的位址。在這樣的結構下，可以比傳統的方式節省記憶體的使用量。視訊壓縮標準的表格相當多張，因此把需要使用的表格的群組資料再做合併以更減少表格間的贅餘部份而減少記憶體空間。經由修改演算法可以讓熵解碼器以較少的空間及較高的效率來存放足夠的資訊。 此論文另外針對視訊無線傳輸的應用下，提出了防止錯誤傳遞的方式而且不用額外傳輸資料來輔助，也就是幾乎不會增加頻寬成本。這個方式可以利用通道解碼提供的位元可靠度來決定啟動防止錯誤傳遞的模組，而這個模組裡用記憶體儲存區塊預測資訊。用這些資訊可以預測出區塊邊界而讓可能出錯的區塊的解碼不會影響到接下來的區塊。這些資訊的比對是由以記憶體為基礎的可變長度解碼的方式來決定是否預測成功，在演算上可與傳統的以記憶體為基礎的可變長度解碼器結合，硬體實作上的複雜度不高。這個提出的方法因為限制了錯誤傳遞的範圍，可以大幅改善畫質。

This thesis proposed entropy decoder which can support dual-mode video format (H.264/AVC and MPEG-2) decoding and improved memory efficiency. The entropy decoder adopts multi-table merging (MTM) algorithm and is programmable. First, for a coding table, all codewords are separated into different groups and only the most significant information of each group is stored in memory. The decoding can be completed by looking up information needed and arithmetic computation such that the symbol address is known. Under this type of decoding, memory space can be reduced compared to conventional VLC decoder. For multiple tables, the redundancies between group information of each table are further exploited and only different parts are stored in the memory. By the modified MTM algorithm, the proposed can store information needed for the standards with higher memory efficiency and less memory space. This thesis also proposed a scheme which can stop error propagation without transmission of extra data helping stop error propagation, i.e., no additional bandwidth overhead. In this method, the stopping error propagation module is activated when the bit reliability coming from FEC is low enough. The error resynchronization information is stored in the memory and block boundary prediction is achieved with the information. Once the block boundary, the following block can be correctly decoded and error propagation can be restricted. The searching of information used group-based VLC decoding algorithm to determine if prediction is finished. In the algorithm level, the method is combined with conventional group-based VLC decoding; in the viewpoint of hardware implementation, the complexity is low. Owing to stopping error propagation from this scheme, video quality is improved drastically.