Dynamic Resource Allocation for Asynchronous Uplink OFDMA systems (DRUS)
Final Report Abstract
In this work, we address the important networking question: how to efficiently stream low-latency video contents using modern streaming technologies like DASH2 over mobile networks using OFDMA1 like LTE. In this context, we jointly consider video adaptation and wireless resource allocation to efficiently enhance the performance of multi-user and low-delay live streams that compete in a bottle neck link. The advantage of this approach is twofold. First, exploiting the information about the link in the lower layers, the client application can select suitable video bit rates that satisfies the resource constraint and maximize the quality of experience (QoE). Second, given the characteristics of video contents, the precious radio resource can be assigned only to users who can use most out of it. This work distinguishes itself from others by, above all, the in-depth investigation of the fundamental difference in the time scale of video adaptation—in seconds—and wireless resource allocation—in milliseconds. The work is divided into two parts, considering two adaptive streaming paradigms based on either non-layered or layered video encoding. While the former has been widely deployed (e.g. in YouTube and Netflix), the former is expected to be more popular in the near future. In the case of non-layered video encoding, we face two challenges, which are, first, how clients can select an appropriate video bit rate that fits the future link capacity, and, second, how the network assures the delivery of the selected qualities. While selecting too high video qualities leads to video stalls, selecting too low one underuses the spectrum; both result in low QoE. Our proposed solution for such systems consists of two components. (i) The first component—a video quality selector—foresightedly optimize the video qualities before the actual delivery. We deal with the aforementioned trade-off by proper optimization formulations. The selected qualities can fairly maximize individual QoE and do not exceed the future link capacity. Additionally, we leverage the history channel state information to improve the estimation of link capacity and, hence, the performance of the video quality optimization. Moreover, we tailor that framework for two use cases, depending on whether the client skips or waits for (i.e. re-buffer) segments that miss its deadline. (ii) Given the quality selection from the first component, the second component—a resource allocator—strives to deliver the requested video qualities. To cope with the time scale difference, a sequential process of adapting the resource allocation to the instantaneous wireless channel states is performed to gradually match the user’s throughput with the demand. We consider both the downlink and the uplink. In the uplink, we incorporate the mitigation of multiple access interference (MAI) caused by the imperfect synchronization among users into the resource allocation. The optimal resource allocation assures the mutual damages caused by the MAI are minimized, and the frequency and multi-user diversities are exploited. In the second part, we exploit the unique feature of the layered video encoding, where the video content is formatted into layers in the hierarchical manner, and propose a novel cross layer design that tightly integrates the problem of enhancing video quality in the sequence of resource allocation. Particularly, the resource allocation in this case is optimized to target, not the throughput demand as in the non-layered encoding system, but directly the incremental gain of QoE. To that goal, we introduce, via a proper mathematical transformation, a simple but yet efficient coupling function that assesses the impact of the instantaneous resource allocation scheme to the long-term QoE goal. Then, we formulate the optimization problem to fairly maximize QoE perceived by multiple users, and convert it to the a sequence of quality driven resource allocation optimization using the introduced coupling function. At each resource allocation step, resources are allocated to users according to their utility determined by QoE constraints and quality fairness. By doing that, the proposed system can instantly react to the channel fluctuation and better assure the QoE goal. In this case, the optimal resource allocation assures the mutual damages caused by the MAI are minimized, and the frequency and multi-user diversities are exploited.
Publications
- “Quality driven resource allocation for adaptive video streaming in OFDMA uplink,” in the Proc. of the 26th Annual IEEE International Symposium on Personal, Indoor, and Mobile Radio Communications (PIMRC), 2015
H. Le, A. Behboodi, and A. Wolisz
- “Cross-Layer Approach for HTTP-Based Low- Delay Adaptive Streaming in Mobile Networks,” in the Proc. of the 18th IEEE International Symposium on a World of Wireless, Mobile and Multimedia Networks (WoWMoM), 2017
H. Le, K. Miller, A. Behboodi and A. Wolisz
(See online at https://doi.org/10.1109/WoWMoM.2017.7974322)
