網路電話之智慧型漸進式撥放緩衝區演算法

Abstract

網路的語音傳輸會受到網路的延遲差異(Jitter)的影響，在接收端控制撥放延遲(Playout Delay)可以解決Jitter 及封包過多所造成的遺失(Overflow Packet Loss)的問題。 傳統的漸進式緩充區演算法(Conventional Adaptive Playout Algorithms) 只是用過去的網路延遲差異去預測撥放延遲。 他們並沒有利用語音壓解器(Codec)，及單雙工工作模式(Duplex Mode)的特性去最佳化撥放延遲。 所以我們提出兩種方法: 智慧型漸進知壓解器 (Codec Aware Adaptive Playout, CAAP)方法及智慧型漸進式單雙工模式(Duplex Aware Playout Adaptation, DAPA)方法。CAAP又可以細分為:簡單智慧型漸進式壓解器 (Simple CAAP, SCAAP) 方法和高級智慧型漸進式壓解器 (Deluxe CAAP, DCAAP) 方法。 兩者的設計觀念是，假設每一種壓解器對封包遺失(Packet Loss)有不同的容忍程度，低壓縮比的壓解器可以增加更多的撥放延遲及減少封包遺失。 SCAAP是用單一預測撥放延遲，但DCAAP用多預測撥放延遲去設計，所以DCAAP有較好的效果。而DAPA的設計觀念為，在單工模式(Half Duplex Mode)下可以增加撥放延遲去減少封包遺失的問題。 因為用不同的觀念，CAAP及DAPA可以單獨或一起工作，且它們的改善也是累加的。在我們語音效能衡量中，CAAP及DAPA比起平均延遲及差異(Mean Delay and Variance, MDV) ，可以有10%到16%的改善。 對於同步的雙方通話(Two-way Communication)因為目前沒有客觀的語音品質衡量 (Objective Speech Quality Measurement) 方法，所以我們同時提出一套模型化的衡量 (Model-based Measurement) 方法。

Internet voice transmission suffers from variance of network delay, called – jitter. In VoIP receiver, the playout algorithm can control the playout delay to eliminate the jitter and minimize the overflow packet loss. Conventional adaptive playout algorithms estimate playout delay from previous network jitters only. They are not aware of voice codec type, and communication duplex mode. Thus, we present two approaches - Codec Aware Adaptive Playout (CAAP) and Duplex Aware Playout Adaptation (DAPA). The CAAP has two sub-algorithms, Simple CAAP (SCAAP) and Deluxe CAAP (DCAAP). Both design concepts are that the different codec has different tolerance to packet loss, the playout delay using a low compression ratio codec can be increased more to reduce packet loss. SCAAP uses single estimated playout delay and DCAAP uses multiple estimated playout delays, consequently DCAAP has better performance. As regards the DAPA design concept, it increases more playout delay to reduce the packet loss in half duplex mode. Because of different design concepts, CAAP and DAPA can work either alone or together; and their improvements are accumulative. In our performance evaluation of speech quality, CAAP and DAPA outperform the Mean Delay and Variance (MDV) algorithm by 10% to 16%. No objective mechanisms for measuring the speech quality of two-way communication exist; a model-based measurement mechanism is also proposed.