在變動位元速率串流下之平順演算法

Abstract

隨著網路頻寬的增加與視訊壓縮技術的進步，藉由隨選視訊、即時視訊轉播等服務，即時接收到高品質的視訊已慢慢變成可能而且可接受。VBR constant quality的視訊壓縮的方式可以得到較佳的品質，不過也會造成資料流突衝(bursty)的特性，它會造網路服務較大的負擔。這時我們就可利用平順演算法來改善這個問題，所謂平順簡單講就是在減少資料流突衝變異情況。最佳平順演算法目的是達成速率整體變異度最小，也就是整體速率標準差最小的一種平順演算法。本論文首先針對此演算法的方法作介紹與探討，並將它用在實際的video trace上作模擬，觀察客戶端緩衝區大小與統計特性的關係，像是峰值與標準差。根據模擬的結果再作討論與分析。其次我們引入每單位時間傳送最小單位精確度為位元組的情況，針對所會遇到的問題作修正，並用實際的video trace作模擬，最後並比較原本與修正過後表現的差異，並探討其差異的原因。

With the increase of bandwidth and the progress in image compression, it has become acceptable to watch real-time high quality video through the network.. VBR compression can gain better quality but it introduces burstier traffic. This will cause a burden to the network service, making the bandwidth usage inefficient. Smoothing algorithm can deal with this problem; it can make the traffic less burstier. The optimal smoothing algorithm can optimizes the traffic in terms of the variance. In this thesis, first we introduce and analyze this algorithm; then we simulate it on real video traces, discussing the relation between buffer size and statistical data, such as peak frame size and variance. Second, we introduce the minimum unit(byte) in transmission rate to this algorithm and solve the problems it meets with. Then we simulate this algorithm on the same video trace and discuss its characteristics. Finally, we compare the original algorithm and the modified one.