MPEG-2/4 低複雜度先進音訊編碼最佳化及雙核心處理器實現

Abstract

本篇論文主要針對AAC 編碼，提出一套最佳化演算法，使得運算複雜度與記憶體需求皆能夠降低以適於在行動裝置上的實現。在AAC 編碼最佳化中，我們移除了運算複雜度高的長短窗切換判斷，採取了簡化的聲響心理模型。 此外，在MDCT 的轉換上，套用了以FFT 為運算核心的時頻轉換，並精簡了FFT 運算過程中所需的記憶體；於TNS 與立體聲編碼上，亦以較為精簡的判斷來降低運算量。最後，為了減少記憶體使用，基於統計結果，我們化簡了做Huffman 編碼時所需要的編碼表。基於16 位元的定點DSP，我們於能量計算上採用動態格式的調整，以避免定點數運算超出所能表示的數值。我們將所提出的AAC 編碼器實現於一個雙核心的處理器上，並於雙核心處理器上實現兩種不同的軟體架構以達到錄音的功能。實現的DSP-based AAC 編碼器需要86 MIPS 及107KB 的記憶體而實現的ARM-based 錄音器可達到1X 速 的壓縮。

In this thesis, several optimized techniques in the AAC encoding process are presented in order to lower down the computational complexity and required memory. The decision of block switching is removed, and adopts a simplified psychoacoustic model. For the MDCT transformation, the fast MDCT with FFT as kernel computation is applied. Moreover, the memory requirement while performing FFT processing is reduced. Other modules, such as TNS, Mid/Side Stereo coding are also simplified. In order to minimize the memory usage, the Huffman tables are reduced base on statistics.So as to make sure the avoidance of the overflow computation and preserve the data precision, a simple dynamic scaling unit before energy calculations is applied. The proposed AAC encoder is implemented on a dual-core processor. Based-on the different software architecture, two solutions for recording system implementation are provided. The realized AAC encoder consumes 86MIPS and 107KB memory, and the recording system implemented by ARM can achieve at least 1X encoding.