適用於MPEG解碼器之精簡係數二維反向離散餘弦轉換器的架構設計

Abstract

近年來，視訊壓縮技巧被廣泛應用於許多領域，其中二維反向離散餘弦轉 換(2-D IDCT)被公認為最有效率的壓縮技巧。在MPEG系統□量化都緊跟在 離散餘弦轉換之後，所以大部份經過量化的離散餘弦轉換係數都為零。因 此我們利用此特性發展出精簡係數二維反向離散餘弦轉換的演算法。這個 演算法不但非常有效率而且其架構很適合超大型積體電路的實現。在本篇 論文中，我們根據基本矩陣和其對稱的特性提出一個非常有效率的二維反 向離散餘弦轉換的演算法，同時利用餘弦函數積化和差的特性去減少乘法 的次數。因此本方法所需要的乘法數目是一般行列分解法的十分之一，而 其加法的數目也比一般行列分解法還要少。我們總共使用了 12500的閘數 去組成二維反向離散餘弦轉換的晶片，而其速度可達到八仟萬赫玆，因此 計算一個8x8區塊(block)的時間平均少於2.5us，且可以達到MPEG2的要求 。 In recent years, video compression techniques are widely applied and the 2-D DCT is recognized as the most effective technique. In MPEG, the DCT is postprocessed with quantization and most of the DCT coefficients are quantized to be zero. By using this characteristic; we developed a reduced coefficient 2-D IDCT algorithm. It is not only efficient for decoding but also its architecture is well suited for VLSI implementation. In this thesis, we proposed this two-dimensional inverse discrete cosine transform (2-D IDCT) algorithm based on base matrices and its symmetrical property, and then use the property of cosine product by sum-difference to reduce the number of multiplications. Thus the average number of multiplications with this algorithm is one-tenth comparing to the conventional row-column method. The average number of additions is also less than conventional row-column method. The IDCT chip totally uses 12500 gates; and thus is area efficient. The operation speed of IDCT chip is 80 MHz, and thus it computes one less than 2.5 us per 8x8 block that can satisfy the requirement of MPEG-2.