MPEG-4音訊壓縮中TwinVQ之探討

Abstract

本論文中將在MPEG-4 TwinVQ的架構上改善向量量化的兩個問題。第一個是增加向量量化的抗雜訊能力，另一個則是減少碼書搜尋的複雜度。索引指派是在不增加冗餘位元下增加向量量化抗雜訊能力的傳統方法，我們根據MPEG-4 TwinVQ的架構與碼書，設計出增加TwinVQ抗雜訊能力的通道編碼方法。在增加碼書搜尋速度方面，我們提出了兩個搜尋演算法以加快在有動態權重變化下向量量化的碼書搜尋速度。第一個方法是利用部分誤差搜尋法(PDS)及向量中元素有不同權重的特性來減少碼書搜尋的計算量，第二個方法是利用改進候選者刪選門檻來改進第一個方法，實驗結果顯示這兩個方法能在不損失重建訊號品質的情況下分別減少平均66%及73%的計算量。

This thesis considers two issues of vector quantization (VQ) based on MEPG-4 TwinVQ. One is the robustness against channel noise; the other is the computational complexity of codebook search. Index assignment is a traditional method for VQ to improve robustness. This thesis considers the index assignment method based on MPEG-4. On the codebook search, two fast search algorithms for dynamic weighted VQ are proposed. The first algorithm is based on partial distortion search method (PDS) which searches entries in codebook by the partial distortion according to the associated weights. The second algorithm further reduces the computational complexity by obtaining the precise threshold to skip the candidates with the partial distortion. Simulation indicates that the first algorithm and the second algorithm reduce the complexity by 66 percent and 73 percent respectively without degrading the reconstructed quality.