用於MPEG環場音訊差餘編碼之快速QMF-to-MDCT轉換

Abstract

MPEG Surround (MPS)為一低位元率多聲道音訊壓縮標準，其壓縮原理是透過降混音處理將多聲道合成雙聲道或是單聲道訊號。在此低位元率的壓縮標準中，差餘編碼提供了一個有彈性的切換機制，藉由傳輸降混音壓縮過程中的預測誤差，在解碼端達到低頻帶或是全頻帶高品質音訊的重建。因此，如何產生並解碼出正確的差餘訊號，成為一個相當重要的議題。差餘訊號藉由AAC編碼器傳輸並解碼出MDCT(Modified Discrete Cosine Transform)頻率係數提供給MPS差餘解碼器重建，為了降低計算上的複雜度，此解碼器並未使用原本的解析/分析架構，而提供了一個從MDCT到 QMF係數的近似快速轉換方法。 此篇論文，探討此快速轉換所造成的瑕疵，並提出新的快速編碼器架構，藉由改變差餘訊號的編碼架構，修正上述的問題。改進的效果將藉由頻譜和誤差計算呈現並驗證。

The MPEG Surround (MPS) is an important standard which defines the method to compress the multi-channel audio signals. In MPS, the residual coding with flexible rate control is an optional tool that encodes the prediction error between the downmix signal and the original signal for the target bit-rate in order to enhance the quality of the reconstructed signal. In the decoder defined in MPS standard, the residue can be decoded by the fast algorithm that maps the coefficients in the MDCT (Modified Discrete Cosine Transform) domain used in AAC to the QMF subband domain used in MPS to reconstruct the multi-channel signals. This coefficient mapping between different domains has greatly simplified the computation complexity required in the AAC encoder and the QMF analysis filterbank. However, the fast mapping algorithm cannot losslessly reconstruct the residue signal. In this thesis, the reconstructed loss caused by the fast mapping are investigated and two new fast mapping algorithms from the QMF subband domain to the MDCT domain are proposed for the encoder. From the simulation results, the proposed mapping algorithms show merits in terms of audio loss and computation complexity.