適用於分區可重疊多權重通道編碼之機率分布優化研究

Abstract

在真實的網路傳輸環境下，資料遺失或損毀的情況發生頻率極高。針對這個問題，最常見的處理方式是將遺失的資料進行重傳，但在特定的傳輸條件下，這種作法卻不被允許，例如傳送具有即時性質的視訊串流資料。因此，接收端需要依靠已經接收到的資料去協助還原在傳輸過程中遺失的資料，這個想法目前已經透過「前向錯誤更正碼」或是「通道編碼」得以實現，而在通道編碼的範疇中，無限制比例編碼又是最常被使用的方法之一。

一個針對可調性視訊編碼進行改良的通道編碼結構我們稱之為「分區可重疊多權重通道編碼」。在這個編碼結構中，提供了針對分層資料進行保護強度控制的能力並在編碼的過程中依循傳統的無限制比例編碼，根據一個重要的關鍵要素「連線數機率分布」(Degree Distribution)來進行編碼，而連線數機率分布的選擇，將會決定編碼時的效率及解碼的能力。

在本研究中，我們對於分區可重疊多權重通道編碼下更好或更合適的連線數機率分布選擇產生高度興趣。首先藉由跳躍式的隨機行走模型，我們推導出解碼過程中保險的波動大小(Ripple size)。接著，我們提出預測模型用來預測分區可重疊多權重通道編碼下的波動變化量(Ripple variation)。結合以上兩者，我們制定出針對分區可重疊多權重通道編碼架構下的連線數機率分布優化目標函數，並使用基因演算法尋找出分區可重疊多權重通道編碼不同環境下優化的機率分布。最後，以資料解碼的平均還原率做為比較準則，和其他常見的機率分布進行效能上的比較。

Data loss or damage are happening frequently in real network environment. A usual solution for this condition is retransmitting the missing data. But this method is not available in specific environment like streaming video or some data have real-time property. Hence, we need to use the data already received to recovery the missing part. Such idea can be done through “Forward error correction” also called “channel coding”, and the rateless codes are famous method in channel coding approaches.

An improved coding structure of rateless codes for data having different priorities or streaming of scalable videos is called “Layer-aligned multipriority rateless codes”. This coding method provides controllable protection strengths for different layers of scalable data and the encoding procedure follows the traditional rateless codes with a key element called “degree distribution”, and the degree distribution is the most important component responsible for the efficiency and performance of the rateless codes.

In this paper, we are interested in finding better or more suitable degree distribution among layer-aligned rateless codes coding structure with different scenarios. We firstly derived the safety criteria of ripple size developed by a leaping random walk model. Then, we proposed the estimate function for the ripple size variation in layer-aligned multipriority rateless codes. Combining with this two, we formulate the objective optimization problem for degree distribution optimizing. After that, we apply the genetic algorithm to find the optimized degree distribution in layer-aligned multipriority rateless codes with different scenarios. Finally, we compare the performance between the optimized degree distribution and other common degree distributions in terms of data recovery rate.