| Summary: | 碩士 === 國立成功大學 === 電機工程學系碩博士班 === 98 === In audio coding and decoding flow, the major purpose of MDCT and IMDCT are to convert time domain into frequency domain and to convert frequency domain into time domain, respectively, and this step accounts great part of the computation in audio codec. Derived from previous experience, we can found that both MDCT and IMDCT can be found to a kernel type of DCT-IV, which can be employed a single hardware accelerator through this IDCT-II kernel to share the hardware resources. Fast algorithm based on a unified recursive IDCT-II is derived for MDCT and IMDCT, and we can reduce the kernel consideration by using pre-processing steps. Hence, the proposed design would reduce the hardware costs in implementation of MDCT and IMDCT on a platform of audio codec. The proposed algorithm takes (N2/64 + N/4) computational cycles for computing all output sequences, which achieve a great improvement than original definition algorithm that needs (N2/2) computational cycles. The proposed algorithm also has a significant improvement than other previous algorithm.
|
|---|
