使用變動長度碼對獨立訊源符號做編碼,其所對應的符號格狀圖架構於2000年首次被提出[12]。在解碼時,利用此格狀圖及Viterbi演算法可估算出符號序列的最大事後(Maximum a posteriori;MAP)機率;此外,作者亦修改傳統BCJR演算法使其適用於此符號格狀圖上,如此即可藉由收到的位元序列,產生每個符號的事後機率(a posteriori probability;APP),並且利用此軟式輸出值可疊代地(iteratively)改善整合式系統(變動長度碼串接一迴旋碼)的效能。 在[4][5]中,不同於過去高維度之符號/位元格狀圖架構,作者提出一低維度(一維)符號格狀解碼架構,其宣稱解碼效能及解碼速度均優於過去文獻裡的結果。因此,本論文主要探討應用於變動長度碼之植基於格狀圖的解碼技術。 理論及模擬結果顯示,Yang[4][5]所提之演算法,其解碼效能等同於過去文獻裡的結果,且其計算複雜度會來得高。
In [12], a symbol-level trellis representation for variable -length encoded independent sources is presented. On this trellis, Maximum a posteriori (MAP) sequence estimation using Viterbi algorithm is possible. Besides, with a modified BCJR algorithm that can be applied to this trellis, one can generate symbol by symbol a posteriori probability (APP) values of the decoded sequence. The soft outputs were used in a serially concatenated variable length code and convolutional code to improve the performance. In [4][5], instead of using higher dimensional trellis structure, the authors proposed a symbol-level decoding algorithm with lower dimensional trellis structure for VLCs. They claimed that the performance and the speed of the proposed algorithm are better than those of existing algorithms. Therefore, we focus on exploring the trellis-based decoding techniques for VLCs in this thesis. Experimental results show that the performance and the speed of the algorithm either proposed in [4] or [5] is not improved at all.