串流視訊品質於網路頻寬時變下之最佳化

Abstract

近年，拜無線寬頻網路之賜，視訊串流逐漸成為一大眾化的網路應用。不過，由於無線網路的不穩定性，視訊串流的品質可能受到時變通道頻寬的影響而大幅下降。從視訊串流系統架構的角度來看，現有的可適性技術大致可分為codec-based和network-based兩大類技術。在codec-based的技術中，視訊編碼演算法和壓縮參數可根據頻寬的變動而有所改變。而network-based的技術則是利用不同的傳輸方式來適應頻寬的變化。在以往的研究中，大多以兩者各自的特性來設計實驗參數和分析方式。所以在同一個系統中，若同時使用兩種技術，可能無法客觀且正確地分析或預測其結果。因此，為了解決此問題，我們在此論文中提出了一個統一的最佳化架構。利用這個架構，我們能將現有的兩大類技術，分別建構成子串流控制技術和傳輸控制技術的最佳化模型。接著，我們分別討論子串流控制技術和傳輸控制技術各自在時變頻寬下的表現。子串流控制能減少視訊串流在低頻寬時的品質失真，但由於使用基本的傳輸方式，導致頻寬無法做最有效率地使用；另一方面，傳輸控制技術利用可變的傳輸時間或頻率，可得到有效地頻寬利用，然而，當使用單一串流編碼和有限的播放暫存器的雙重影響下，亦無法得到最佳的表現。從實驗結果的分析中可得知，當我們只應用其中一類技術時，兩者都有各自的極限，但事實上如果能同時使用兩種技術，這些缺點可被彼此互補。因此，在論文的最後，我們亦考慮結合子串流控制技術和傳輸控制技術的最佳化可能性。實驗結果顯示在網路時變頻寬下，結合之最佳化的確可使視訊串流品質達到更好的表現。

In recent years, thanks for the deployment of wireless broadband networks, video streaming becomes a popular application. However, due to the unstable characteristics of the wireless network, the fluctuation of channel capacity impacts the quality of video streaming. From the perspective of the video streaming system, techniques used for addressing the problem of capacity variation can be classified as codec-based and network-based techniques. Codec-based techniques adapt to capacity variation by adjusting video encoding algorithms and parameters. Network-based techniques use different transmission methods to adapt to capacity variation. Conventionally, these two classes of techniques have been designed and evaluated separately without consideration of each other. The lack of fair performance comparison of individual techniques makes it difficult to explore their performance tradeoffs and the optimal strategy for using these two classes of techniques at the same time. Hence, in this thesis, we proposed a unified optimization framework to address this problem. Under this framework, we first model codec-based and network-based techniques as sub-stream control and transmission control mechanisms. We then analyze and compare the performance of sub-stream control and transmission control mechanisms when they are operated in capacity-varying scenarios. Sub-stream control mechanisms reduce the distortion when the capacity is low, but they cannot use the capacity efficiently because they use baseline transmission. Transmission control mechanisms adapt transmission time or frame rate so they can use capacity more efficiently. However, they cannot perform optimally due to the single sub-stream coding and the limited play-out buffer size. Finally, we consider the joint optimization of sub-stream control and transmission control and analyze the performance benefits of joint optimization. The optimization results show that joint optimization of sub-stream control and transmission control indeed has performance benefits in capacity-varying wireless networks.