MPEG-4先進音訊編解碼器之增速及其在DSP平台上的實現

Abstract

MPEG-4先進音訊編碼(AAC)是非常有效率的音訊壓縮編碼技術。它是由ISO/IEC MPEG所制定的一套標準。 在本篇論文當中，我們首先分析MPEG-4先進音訊編碼器在DSP上的執行計算複雜度。發現心理聲學模式(psychoacoustic model)和量化及位元編碼(bit allocation)所花的執行時脈週期為最多，因此針對它們，我們在DSP上的實現利用比較快速的演算法主要加速之。 在DSP實現方面，為了加速先進音訊編碼器，我們針對DSP的架構使用了一些程式技巧，包括定點式資料型態、TI DSP的特殊指定群等等。除此之外，我們也參考了一些快速運行的演算法，並套用在原來的音訊編碼器之心理聲學模式及量化位元編碼上。經由這些的程式修改，最後的編碼器版本在DSP上的執行速度比原來的有了77.89%的改善幅度。並且我們也成功的把先進音訊編碼器及解碼器兩者實現在II(Innovative Integration) 所提供的Quixote DSP平台上。而在主端及客端的傳輸介面，我們採用了緩衝之區塊傳輸模式，此模式讓我們容易實現整個架構。最後經由我們的加速及系統實現，此先進音訊編碼及解碼器各自都可達到即時編解碼的效果。

MPEG-4 AAC (Advanced Audio Coding) is an efficient audio coding standard. It is defined by the ISO/IEC (International Standard Organization) MPEG (Moving Pictures Experts Groups) committee. In this thesis, we first analyze the computational complexity of the MPEG-4 AAC encoder program. We find that the PAM (psychoacoustic model) and the quantization and bit allocation module require the most execution cycles on DSP. Hence, we mainly propose methods to accelerate them on DSP. In order to speed up the AAC encoder on DSP, we use several DSP codes acceleration techniques including fixed-point data types, TI (Texas Instruments) DSP intrinsic functions and others. In addition, we accelerate the PAM and the quantization and bit allocation modules by fast algorithms for DSP implementation. Through these modifications, the final AAC encoder version has about 77.89 percent improvement. Furthermore, we also successfully implement both the AAC decoder and encoder on the II’s (Innovative Integration) Quixote DSP board. We adopt the burst block transmission mechanism for communication between the host and the target side. Finally, the speed of the AAC encoder and decoder on DSP implementation can achieve real-time operation.