利用多層編碼配合DCCP能與TCP友善共存的網路電話


TCP-friendly VoIP By Scalable Codec over DCCP


林耿誠

   壅塞控制是網路管理的重大問題。目前的網路應用程式多半使用TCP或UDP 這兩個傳輸協議來傳遞資料。TCP具有壅塞控制機制可以隨著網路狀況調整資料傳 送速率,但其重送機制所導致的時間延遲不利於時效性的網路服務。而UDP的傳輸 速率多半是在傳輸之前先行設定,在資料的傳輸過程中不再改變,對於網路壅塞並 無任何調節作用,不利於網路之和諧共用。因此DCCP 這種具有壅塞控制機制的不 可靠傳輸協議被提出,期望取代UDP 成為不可靠傳輸的主流協議。網路電話大部分 使用UDP作為傳輸層協定,UDP不具壅塞控制機制,有礙於網路之和諧共用,而且 網路的品質也會因為網路壅塞而遺失封包導致品質受損,要達成網路電話在網路壅 塞時有壅塞控制能力,且能有好的通話品質,必須要根據網路的狀況調整封包傳送 速度或是封包大小。

  本研究利用多層編碼配合DCCP形成與TCP友善的網路電話。壅塞控制必須針對 網路中不同程度的壅塞作出適當程度的反應,因此我們改進語音編解碼器,設計出 多層語音編碼並設計出可以與之配合的DCCP,讓DCCP依照網路的狀況送出不同層 級的語音封包。本研究透過實際網路的實驗環境中評估以CBR over UDP、Flexible Bit-Rate以及Scalable Codec三種方式傳輸網路電話封包的效能。並也評估Scalable Codec VoIP與TCP同時存在於頻寬不足的網路中,對於頻寬競爭能力的表現。實驗 結果顯示在網路壅塞最嚴重的情況下可以和一般的CBR配合UDP的方法比較封包遺 失率達到40%左右的改善,語音品質評估指標MOS達到1.5分的改善,與Flexible Bit-Rate方法比較封包遺失率也達到8%以上的改善,MOS達到1分的改善,讓整體網 路狀況更為穩定,並因為封包遺失的減少讓語音品質提升。而在和TCP頻寬競爭實 驗中可看出,在頻寬不足的情況下,此研究提出的方法可以和TCP友善且公平的競 爭頻寬。ii TCP-friendly VoIP By Scalable Codec over DCCP

Congestion control is one of the major problems of network management. Most current network applications use either TCP or UDP to transport data. TCP is equipped with a congestion control mechanism but is not suitable for real-time multimedia applications due to its instability of delay time. On the other hand, UDP fixes its data rate and doesn't change it during the period of transmission even when the network is congested. Under this circumstance, DCCP, which is an unreliable transport protocol but has a congestion control mechanism, is proposed to replace UDP to support real-time network applications such as VoIP. Our previous study showed that a flexible bit rate CODEC to support VoIP over DCCP can effectively control network congestions while maintaining a good voice quality. However, it has an implementation issue yet to be addressed: it requires a bidirectional interaction between DCCP and CODEC.

  This thesis proposes to use a scalable CODEC approach to support flexible bit rate VoIP over DCCP. The CODEC sends the entire spectrum of input voice stream to DCCP. DCCP then selects the appropriate voice activation level to compose output stream according to the measured network status, which is feedbacked from the receiver side. The interaction between DCCP and CODEC is avoid. The proposed scheme was evaluated in a real local area network against two other protocols under various VoIP environments, CBR over UDP and Flexible Bit-Rate. Experimental results show that the proposed scheme can outperform CBR over VoIP in the most serious network congestion (under our lab configuration) by 40% in packet loss rate iii and 1.5 in MOS. It can outperform Flexible Bit-rate over VoIP by 8% in packet loss rate and 1.0 in MOS. Finally,the fairness test shows that our scheme can coexist with TCP with a fairness index higher than 95% even when network is congested.