DST-II Based Recursive Fixed-Coefficient and Multiplication-Free Architecture Design for Forward and Inverse MDCT/MDST Algorithms

• Main Authors: Hsiu-ChiHuang, 黃秀祺
• Other Authors: Ching-Hsing Luo
• Format: Others
• Language: en_US
• Published: 2014
• Online Access: http://ndltd.ncl.edu.tw/handle/56658104656351403175

碩士 === 國立成功大學 === 電機工程學系 === 102 === This thesis proposes a signed DST-II fixed coefficient kernel for implementing MDST and MDCT designs. MDCT is used in various audio codecs including Advanced Audio Codec (AAC (MPEG-4)), AC3, ATRAC1, ATRAC3, ATRAC3plus, CELT, MP3 (MPEG-1, 2, 2.5 Audio Layer III), Opus (Hybrid mode), Real Audio, Vorbis (Ogg), Windows Media Audio Standard, and Windows Media Audio Pro. In the recently developed 3D audio codec also known as Unified Speech and Audio Coding (USAC) by MPEG, both MDCT and MDST are used in one of the two forms of complex stereo prediction. Since MDCT and MDST can both be represented as a DST-IV transform, and DST-IV transform can be further derived into 4 transforms represented by DST-II, MDCT and MDST can both be realized by DST-II kernel. The Proposed DST-II structure requires only one multiplier, three adders and two delays in implementation. Additionally, the multiplier can be further replaced by fewer shifters and adders using CSD representation. A q factor is used to minimize quantization effect. To verify the DST-II kernel running in hardware description language, the node and coefficient bits required for the kernel are configured by MATLAB, and the structure of the kernel with configured word length is built in Altera’s DSP Builder. By running the Signal Compiler and Testbench tool, the signed DST-II kernel is programmed and verified in Quartus II and Modelsim. And the PSNR of extracted output values exceeds 90dB, which is suitable for audio implementation.