用於分層串流之同步滑動無限制比率碼:模型化及最佳化

Abstract

在異質網路中, 伺服器端必須傳送不同的串流檔案給不同需求的客戶端。但相同的檔案必須準備不同串流，的這對伺服器端是一種很重的負擔。如果這時候我們可以將一個串流分割成數個部分，而客戶端的使用者只要接收他們所需的部分的話，這將大大的減少伺服器端的負擔。而這種方法就是「可調式視訊編」。

在現實的網路環境中，封包的遺失或損壞是不可避免的。當封包遺失或損壞時，重傳是最一般的做法。但是，在某些環境中重傳是不被允許的。因此，我們必須嘗試使用已經接收到的封包去回復這些有問題的封包，這種方法叫做「通道編碼」，而「噴泉編碼」也其中一種通道編碼。

在本篇論文中，我們將會研究如何整合可調式視訊編碼和噴泉編碼,同時提供「非相等錯誤保護」,保護可調式視訊編碼串流中比較重要的部分。此外我們還提出了兩個預測的模組。其中一個模組是用來預測在我們提出的保護方法中，訊息區塊解碼失敗的機率是多少。而另一個預測模組是預測在可調式視訊編碼解碼結束後有多少張圖可以成功的撥放出來。藉由使用這兩個模組，最終我們可以產生擁有最大PSNR 值的最佳化串流。

In heterogeneous networks, the server transports the streaming to the clients that have different requirement. It is a heavy load that the server has to prepare the different streaming for the same video. If we can cut a streaming into several parts, and the client users can receive the parts according to their requirement. It can reduce the load of server. The method is named SVC (Scalable Video Coding).

In real network environment, packet loss or damage is unavoidable. When packet loss or damage takes place, retransmission is one of the usual solutions. But, retransmission is not allowed in some environment. Therefore, we have to try to use the packets that we already receive to recovery the lost packets. The method can be done through “channel coding”, and the fountain codes is one kind of the channel coding approaches.

In this paper, we are going to research on the integration of SVC and Rateless codes. Meanwhile, unequal error protection scheme will be proposed to protect the important part of SVC streaming. In addition, we will propose two prediction models. One of the prediction models estimates the decoding failure probability of each message block in our proposed protection method. Another prediction model estimates the amount of frames that can be displayed successfully after SVC decoding. By using these two models, we finally can produce the optimized streaming that has the highest PSNR.