寬頻語音的多重敘述編碼系統

Abstract

在數位通訊系統中，多重敘述編碼是一種有效提升冗息保護資訊的機制，我們在該架構下提出基於位元與索引層級的疊代訊源通道解碼演算法。首先探討訊源量化後的殘餘冗息與通道編碼資訊的交叉運用，系統設計藉由納入多重敘述間的相關資訊加強其解碼的可靠性，其效能隨著疊代解碼次數的增加而遞增。為了克服位元間獨立假設的限制，我們更進一步將解碼演算單位由位元改良為索引層級。於實驗模擬中使用自迴歸訊號源與寬頻語音，成功驗證不同敘述間的相關訊息可以有效提高解碼效能。

For digital communications, multiple description coding (MDC) is an efficient technique to enhance the channel robustness. Based on the MDC structure, we develop iterative source-channel decoding (ISCD) techniques at bit-level and index-level. Our work began with the exploitation of the residual source redundancy and the parity check information of channel coding. By taking into account the correlation among all descriptions, the ISCD schemes have higher decoding reliabilities and the performance gain increases with increasing iterations. Moreover, in order to solve the mismatch between the bit-independence assumption and real environment, we propose an index-based decoding algorithm. Experimental results on AR-source and AMR-WB coded speech indicated that the correlation among multiple descriptions yields significant performance gain.