Cross-layer Based Replication Strategy For Streaming Media Of Vod Systems Over Wireless Mesh Networks

Video on-demand systems (VoDs) consume considerable bandwidth and storage resources of network due to high data rate of numerous huge sized videos. However, the resources of wireless media are often limited in wireless mesh networks (WMNs). The lossy channels and multi-hop delays in WMNs make VoDs difficult to deliver videos at high quality of experience (QoE). Importantly, QoE, which is perceived by users, is characterized by many subjective and complicated aspects of visual quality such as high hit rate, continuous playback, high quality playback in terms of spatial artifacts, and smooth playback. Furthermore, the users involve unusual behavior when viewing a video. They may move backward or forward for interested video contents, stop to access other videos with more popularity or/and higher definition, depending on their processing capacities or display resolutions. These in turn pose a set of technical challenges for VoDs over WMNs to gain high streaming performance, i.e. high QoE and resource savings. In this thesis, to overcome these challenges, we propose some novel joint techniques by applying three key methods:cross-layer, replication, and rate allocation.Firstly, we can deduce that the QoE at application layer is affected not only by the behavior and display resolution of users, but also by:1) the source encoding scheme,2) the transmission errors and delays from lower layers, e.g. network, medium access control, and physical layers, and 3) the available resources of network. For high QoE and resource savings, the relationship between the QoE and all the mentioned effect aspects should be analyzed in cross-layer approach, before delivering the videos. In particular, we propose a flexible framework of cross-layer architecture for replication strategy (CRS) to assist VoDs over WMNs. This CRS enables us to achieve different sophisticated objectives of providing high streaming performance in the sequel.Based on the framework of CRS, we study the mathematical relationship between the average reconstructed distortion (ARD) of received video frames and the skewed rate-distortion (RD) of each frame, under the effect of loss probability and intra popularity associated with users'behavior. In this way, we find a closed-form of optimal replication densities for a set of video frames so as to minimize the ARD for high quality playback. Unlike other pioneering replication strategies in which the optimal densities are found by solely relying on the skewed intra popularity for high data accessibility and minimum server's workload, our obtained optimal densities further exploit the skewed RD of each frame for high quality playback. The optimal results also guarantee smooth playback aspect of QoE and small consumption of storage resource.In CRS for VoDs over WMNs, the frames of video segments with higher intra popularities are often replicated with higher densities. Therefore, they suffer less loss probability to be reconstructed in lower distortion than that of video segments with lower intra popularities. This causes fluctuation of reconstructed distortion, i.e. unsmooth playback, among received frames of video segments. In addition, the frames encoded at high bit rate consume large resources but not always support high quality over lossy wireless links. In this regard, if we take into account the relationship between the rate of video frames and the lossy characteristic of wireless links, we can achieve smoother and higher quality playback. To do so, the optimal replication density of each video segment is essentially generated based on its intra popularity, for continuous playback. We then find the optimal rate allocated to each frame of every segment by solving a minimum ARD optimization problem for high quality playback, under the constraint of reconstructed distortion fluctuation for smooth playback and the constraint of resources for saving.Finally, to cope with the problems of diverse bandwidth of wireless links and heterogeneous display resolution capacity of users as well as transmission errors and delays, we propose a scheme of layered multiple description coding with embedded forward error correction (LMDC-FEC) to packetize a scalable video coding (SVC) segment into descriptions for streaming. We further elaborately allocate optimal rates to all layers of each segment and find optimal replication density for each segment to thoroughly optimize QoE and storage resource by solving a joint constrained optimization problem using genetic algorithms (GAs).The simulation results are shown to demonstrate that our proposed CRS with/without rate allocation significantly enhances the streaming performance of VoDs over WMNs.