Enhancing Video Delivery in Wireless Network

Video and other real time multimedia applications are the largest consumer of wireless cellular data around the world. Satisfying the high throughout and strict delay requirements of these kind of applications has proven to be a challenging task. This problem has been tackled using the concept of adaptive video, in which the video is segmented in time, and users can adaptively choose a suitable quality representation for each segment. Adaptive video can also be implemented using the scalable extension of the video codec, where each segment is encoded into a base layer containing minimum quality and multiple enhancement layers that incrementally add to the quality. The main objective of this thesis is to investigate novel methods for improving the performance of wireless networks in terms of delivering high quality adaptive scalable video to mobile users.;We first focus on a simplified scenario with a single user in a wireless network streaming scalable video. Using a dynamic programming technique called Semi-Markov Decision Process, we determine the optimal order of retrieving different layers for each video segment, a procedure called Quality Adaptation. The objective of the quality adaptation is to provide high video quality whenever possible and intelligently decrease the quality (by delivering fewer enhancement layers) whenever the channel conditions degrade in order to avoid freezes in playback known as re-buffering. We derive the optimal quality adaptation for various environments and buffer limits in order to determine the effect of these parameters on the optimal policy. We then use a decision tree classifier on the decision process outcome in order to derive simple high level guidelines that describe the optimal quality adaptation policy and how it is affected by varying channel characteristics and buffering restrictions.;Next, we consider a multi-user scenario in which a base station has to perform optimal scheduling, i.e., share network resources among video streaming users. We argue that jointly optimizing quality adaptation and scheduling is not only impractical and overly complex, but it also has various negative business related aspects which makes it an unattractive design choice for content providers. However, current systems in which quality adaptation and scheduling are performed separately and independent of each other fail to provide satisfactory Quality of Experience (QoE). Therefore, we propose a novel system design in which the content providers first advertise the quality adaptation logic of their respective users to the base station and the base station optimizes the scheduling accordingly prior to the start of the streaming process. Then, we model the problem as a network of Restless Bandits and derive long term performance metrics through solving the Restless Bandit problem as a linear program. In order to implement the resulting scheduling scheme we propose two online algorithms. The first algorithm is designed for the case in which the base station is aware of the quality adaptation policies of each user. By studying the outcome of this algorithm we propose a simplified scheduling algorithm for the case in which the base station does not have such knowledge. By performing a thorough simulation study, we show that the adaptive scheme is simpler to implement than the joint method and improves the Quality of Experience compared to the performing scheduling and quality adaptation in a disjoint manner. We also show that our proposed scheduling scheme can help the network to support up to 30% more users for the same video quality. With a test-bed implementation, we demonstrate how our proposed schemes operate in a real world network scenario.;Lastly, we consider restrictions on copyright protected video delivery in order to design a novel video streaming technique called streamloading. Content owners aim at protecting their content and prevent illegal copies by limiting the amount of video that users can pre-fetch ahead of playback time and store on their device memory. Due to fluctuations in the wireless channel, this buffer limit can cause frequent re-buffering instances and degrade video quality. In streamloading, the user is allowed to pre-fetch enhancement layers indefinitely while base layers are streamed in a near real time fashion. Since base layers are required in order to decode enhancement layers, this delivery model does not violate the copyright restrictions imposed on the content. In order for streamloading to operate in a multi-user wireless network, we must devise optimal resource allocation and quality selection schemes. We tackle this problem by formulating it as a mixed integer program. Due to the complexity of the problem, we break the problem into two asynchronous sub-problems. The rate allocation sub-problem is solved by the base station at every time slot and the quality selection problem is solved by the user upon fully retrieving a previously selected segment. We then propose heuristic algorithms in order to solve both problems in an online fashion. Our simulation study shows that streamloading can achieve up to two times better resource utilization compared with conventional video delivery schemes. Also, we show that the online algorithms we proposed outperform baseline algorithms if applied to streamloading.