網路語音傳輸系統規劃之研究

Abstract

網路語音通訊的服務品質取決於諸多因素，主要是封包漏失､延遲與擾動。為了讓網際網路能提供更好的服務品質，傳送端利用前向錯誤控制來回復封包漏失，接收端則利用撥放暫存機制來補償延遲擾動。基於漏失與延遲之間的交互作用會影響系統建構，於是在硬體製作前應先作實地量測再據以規劃網路語音傳輸系統。本論文的首要部分著重發展一以軟體為基礎的量測系統，用以評估網際網路的各項服務品質因素。再根據聽覺評量並參考國際電信聯盟制定的標準建立不同因素所對應的音質損害，整合推導出單一能具體反應網路通話品質的音質評量指標。有別於隨機漏失模型，二狀態的吉伯特模型更能適切地描述封包漏失的時域相依特性，因此採用吉伯特模型來分析封包層級的前向錯誤控制效能。最後我們分別針對隨機與叢發性封包漏失模式，配合音質評估模型，完成前向錯誤控制的最佳化設計。

The quality of voice over IP (VoIP) is mainly affected by network impairments such as delay, jitter and packet loss. In order for IP networks to provide quality of service, playout buffer algorithms at the receiver compensate for jitter and forward error control (FEC) is used to mitigate the impact of packet losses. Due to the dependence of network condition on loss and delay tradeoffs, it is a prerequisite to establish speech trans-mission planning prior to the system implementation. The first part of this thesis concentrates on developing a software-based measurement system that evaluates the characteristics of VoIP. We then combine individual loss and delay impairments using ITU standardized E-model to predict the subjective quality of VoIP. We also propose the use of voice quality pre-diction model for perceptual optimization of FEC design. Packet-level FEC performance is analyzed in the case of two-state loss model, and compared with the random loss model, in order to demonstrate the bene-fits of Gilbert model that more closely characterizes the temporal de-pendency in packet losses.