軟性位元解碼在錯誤隱匿之研究

Abstract

在數位通訊系統的設計，針對通道編碼器無法消除之殘餘位元錯誤作進一步的隱匿處理有其必要性。本論文研究一種結合硬性位元判定和其可靠度的軟性位元解碼演算法，其關鍵在利用量化索引的事前消息且使用BCJR軟性輸出解碼演算法所提供的位元可靠消息。傳統的BCJR演算法以位元層級為基礎，其目標為計算每一個傳送位元的事後機率。為了改善錯誤隱匿功能，我們修改軟性輸出通道解碼的BCJR演算法，主要特色是同時利用位元層級和索引層級的相關性。自迴歸訊號源的模擬結果指出，基於索引的BCJR演算法和索引的一階事前消息的使用，對於軟性位元解碼有最好的錯誤隱匿表現。

In digital communication over noisy channels there is a need for concealment of residual bit errors which have not been eliminated by standard channel decoding. A soft-bit decoding scheme based on joint knowledge of a hard bit and its reliability is studied and applied to quantization of autoregressive sources over AWGN channels. This approach exploits a priori knowledge about the quantizer’s indexes and requires bit reliability information provided by the soft-output channel decoder using the BCJR algorithm. Conventional BCJR algorithm aims to compute the a posteriori probability of each transmitted bit and hence is based on the bit-level. For further improvement, we propose a modified BCJR algorithm which exploits bit-level correlation as well as index-level correlation in the soft-output channel decoding process. Simulation results indicate that for the soft-bit decoding, the combined use of a modified BCJR algorithm and the index’s first-order a priori knowledge performs the best for error concealment under noisy channel conditions.