Dolby AC-3 解碼器轉換域上之聲道運算

Abstract

杜比AC-3是目前HDTV 音訊標準且廣泛被應用在DVD影片上。當AC-3標準制定完成之後, AC-3定義兩種不同長度的轉換(包含一個512點的長轉換和兩個256點的短轉換)。 其目的是為了提供較好的音訊品質當音樂訊號變化很快時。 然而, 不同長度的轉換增加製作轉換域上聲道運算的複雜度, 例如混音。 因為每個聲道可能被512點的長轉換或是256點的短轉換編碼, 聲道運算不可能將兩個藉由不同長度轉換所編碼的聲道加以混合。 在本篇論文中, 我們將提出一個轉換矩陣, 這個轉換矩陣能將短轉換之轉換係數轉變成為長轉換轉換係數。因此聲道運算是可以在轉換域上被製作。此外, 本篇論文還提出數個快速演算法來計算這個轉換。本篇論文將分析不同演算法, 考量其在計算複雜度、計算準確度、和計算規則上的優缺點。

Dolby AC-3 is current HDTV audio standard and is widely used on DVD films. While the draft was frozen, it defines two different lengths of transforms (one 512-point transform and two 256-point transforms) to provide better audio quality when signals are changed in time. However, the different lengths of transforms increase the difficulty in realizing channel operations, such as downmix and channel reduction, in the transformed domain. Since channels can be encoded by 512-point or 256-point transforms, channel operations cannot mix those channels encoded by different lengths of transforms. In this thesis, we propose a conversion matrix that can convert the transformed coefficients of two 256-point transforms to the transformed coefficients of 512-point transform. Therefore, channel operation can be performed in the transformed domain. Besides, this thesis proposes several fast computing algorithms to compute this conversion matrix efficiently. This thesis also analyzes those fast algorithms according to computational complexity, numerical precision and regularity of computational structure