OFDMA 網路下結合資源分配及調變編碼模式之視訊品質最佳化

Abstract

在本論文中，我們研究在下行 OFDMA 網路中以最佳化視訊品質為目標之資源分配問題。為了更有效率地使用資源，OFDMA 網路提供了彈性的動態調變編碼技術，讓基地台可以根據使用者的通道品質及傳輸需求，提供不同調變編碼模式之傳輸。傳統上的作法，大多分開處理資源分配及調變編碼模式的決策，而沒有考慮調變編碼模式對於視訊品質之影響。為了解決這個問題，本論文研究如何在有限的資源下，同時找出最適當的資源分配以及相對應最佳的調變編碼模式，以讓整體使用者的視訊品質有最好的表現。由於這個問題的求解複雜度過高，為了達到即時壓縮與傳輸視訊串流之目的，我們提出了一個基於匈牙利演算法（Hungarian algorithm）的解法，能在多項式時間內算出最佳或接近最佳的分配方式。首先，我們找出每位使用者在每個頻道上最佳的調變編碼模式。接著根據當下的通道品質，利用匈牙利演算法決定每位使用者最佳的頻道及調變編碼模式。對於未分配之頻道，我們循序地將其分配給能夠提高最多視訊品質的使用者。然而，經由此進一步的分配過程，原先最佳的調變編碼模式可能不再適用。此時透過我們預先設計的視訊品質門檻，演算法會自動的調整調變編碼模式，以達到視訊品質的提升。從實驗結果的分析中可得知，比起傳統作法我們的演算法最多僅需要增加 1.3% 的運算量，在不同的視訊上可得到 3%-13% 的增益。對於較為靜態之視訊，訊號品質較不佳的使用者最多可以達到 230% 的增益；反之，對於較為動態之視訊，訊號品質較佳的使用者最多亦可以達到 220% 的增益。

In the thesis, we investigate a distortion-optimized resource allocation scheme for downlink video transmission over multiuser Orthogonal Frequency-Division Multiple Access (OFDMA) networks. Video distortion in wireless networks is mostly caused by rate-controlled source coding and transmission errors due to packet losses. Both distortions are results from the selection of modulation and coding schemes (MCSs), each providing different throughput and reliability levels. Therefore, it is important to select appropriate MCS to determine the optimal resource allocation among users. To proceed, we formulate the resource allocation problem as a distortion minimization problem by joint consideration of sub-channel assignment and MCS selection, subject to the constraints of limited system resources. To solve the formulated problem, MCS for each user on each sub-channel is first determined that induces the least distortion. We then apply the Hungarian algorithm to optimally assign the subchannels for users with their channel conditions accordingly. As a result, the proposed algorithm distributes the remnant sub-channels sequentially to the user with the most reduction in distortion. During the process of remnant sub-channels allocation, the MCS selected initially may become unsuitable and is dynamically switched to a more appropriate one to meet the designated distortion threshold. Simulation results show that with at most 1.3% increase in the computational cost, the proposed algorithm achieves performance gain by 3%-13% in general and 16%-55% for users with SNRs which a better solution is obtained by joint consideration, compared to the resource allocation scheme without joint consideration of MCS.