AC-3架構下之音訊壓縮:轉換,耦合,及擾動

Abstract

音訊壓縮的目的是降低須求的位元率而且能保持音訊的品質。AC-3標 準將被廣用於商品上。這本論文的主要論點將是架構在AC-3標準上探討三 個主題: 轉換, 耦合, 及擾動。 AC-3是頻譜上的壓縮系統, 但是時域 與頻域轉換的演算法須要相當大的計算量, 所以發展比較快速的轉換演算 法是有其必要。論文第一個主題中發展快速的轉換演算法經由兩個步驟。 第一步是將六個轉換公式藉由輸出入資料重排的方式表示成單一的轉換公 式, 此單一的轉換公式為一般的餘弦轉換。第二步驟將發展餘弦轉換的快 速演算法而且特別考量在X86系統上浮點運算器內的堆疊大小的限制, 以 達成降低計算量的須求。 第二個主題將探討耦合的技術。耦合的技術 利用人耳對於高低頻音訊的敏銳度及空間上的音感特性發現壓縮的空間。 利用耦合的技術可以突破兩個頻道位元率各128 kbps的瓶頸。論文中對五 種適合於AC-3架構下之耦合技術的策略及演算法提出討論。五種演算法中 有三種是新提出的。這三種新演算法是基於KL 轉換。經由主觀測式的結 果基於KLT演算法可以穫得較佳的音訊品質。 第三個主題將探討擾動 技術。擾動技術可以消除量化過程不悅耳的雜訊, 而且提供了改善音訊品 質的方法。論文中將經由實驗測式擾動技術的優點並且提供在AC-3架構下 的擾動技術策略及演算法。 The purpose of audio coding is to reduce the required bit rate, while maintaining the signal quality after decoding. The AC-3 standard is a widely used audio compression standard in many commercial applications. This thesis considers audio coding under the framework of AC3 through three design issues: time-domain-aliasing-cancellation (TDAC) transform, coupling, and dithering. The AC3 is a transform coding and the TDAC plays the role to transform a signal from time-domain into frequency domain. Since that the computing complexity of the TDAC is high, the development of fast computation algorithms for the TDAC has been a major concern for real-time applications. The first issue of this thesis develops a fast algorithm for the TDAC. The fast algorithm is derived through two steps. The first step converts the six formulae, which are for the three forward-inverse transforms used in the TDAC, into a unified formula known as discrete cosine transform through data permutation in the input and the output. For the discrete cosine transform, the second step develops a fast computing algorithm which takes special consideration on the stack size limitation in x86 CPU in addition to achieving the low complexity as traditional fast algorithms. The second issue of the thesis is on the coupling strategies in AC3. Nowadays, the CD quality audio sequence can be compressed into 2x128 kbits per second by independent coding of the left and the right channels. Coupling strategies can provide the design space exploiting the human hearing knowledge which indicates low human sensitivity of the stereophonic in high frequency components. With the use of coupling coding techniques, the compressed audio quality can be improved for the bit rate lower than 2x128 Kbits/s. This thesis presents five algorithms which are applicable to coupling strategies in AC3. We compare the five algorithms through theoretical analysis, objective quality measure and subjective quality measure. Among the five algorithms, three are newly developed. The three algorithms are developed based on the KL transform. Subjective and objective measurements show that the coupling strategy based on the KL transform can provide better stereo audio quality than the others. The third issue of the thesis is on the dithering signals. Dithering can avoid the undesired noise from quantization and provide the means to have better stereo signals. This thesis confirms the above two benefits through experiments and provides the strategies to apply the dithering in the framework of AC3.