平行串接迴旋碼之最大概度解碼演算法

Abstract

在本論文中，我們使用MLSDA這個以code-tree為基礎的演算法來做渦輪碼(turbo code)的解碼。不像在[3]裡面所用到的疊代式解碼演算法(iterative decoding algorithm)是一個次於理想的解碼演算法，對於渦輪碼的解碼來說，MLSDA是最大概度(maximum-likelihood)的演算法。我們以這個新的最大概度演算法來對長度短的渦輪碼做”error floor”現象[3]的研究。令人驚訝的是，”error floor”現象是起因於疊代式解碼演算法而不是碼的頻譜（code spectrum）本身。另外從模擬的結果也得知，疊代式解碼演算法在長度短的渦輪碼會有很差的表現。

In this thesis, we apply a code-tree based algorithm (MLSDA) proposed in [1] to the decoding of turbo codes. Unlike the iterative decoding algorithm proposed in [3], which is a suboptimal decoding algorithm, the MLSDA turns out to be a maximum-likelihood decoding algorithm for turbo codes. We also investigate the error floor phenomenon [2] for short turbo codes based on the new ML decoding algorithm. Surprisingly, the error floor phenomenon for the code is resulted by the iterative decoding algorithm not the distance spectrum of the code itself. By the simulation results, the iterative decoding algorithm proposed originally performs badly on these short turbo codes.