Stochastic Optimization And Control In Heterogenous Networks

With the development of Internet and wireless technologies and the available of mobile devices, heterogeneous network is the trend of the development and require-ment of future wireless systems. Heterogenous network has been studied and researched widely by academic and industrial, for example.3GPP and ITU have started working on heterogeneous network, and treat it as the key technology in the next generation cellular systems. Heterogenous network also bring us a lot of challenges. like the diversity of QoS requirements, complex cross-tier interference, time-variable system states. To ad-dress these challenges, we consider some different scenarios in heterogeneous networks. By leveraging stochastic optimization theory, we present online resource allocation and scheduling algorithms. The detail of our work is summarized as follows:1. Machine Type Communications(MTC) have many applications such as in smart grid, health monitoring/alerting and intelligent transportation. Due to unique characteristics different from the traditional Human-to-Human (H2H) communi-cations, numerous design challenges have to be tackled in M2M. A critical issue is that of dealing with the massive accesses of MTC devices while guaranteeing the desired Quality of Service (QoS). In this paper, we group MTC devices into clusters based on their QoS requirement. By utilizing the time-controlled M2M feature, we formulate the problem of massive access with QoS guarantee as a queuing problem. We derive an overflow probability estimation model by ap-plying large deviation principle, and then propose an online measurement based adaptive massive access management for deciding the amount of resources allo-cated for each clusters. The proposed method makes the decision based on the observed traffic workload without any prior knowledge of its statistics. We con-duct the simulations in multi-cluster scenarios with distinct maximum tolerable delay and corresponding acceptable probability in LTE system. The simulation results show the adaptability of the proposed massive access management regime to the variations in the traffic workload, thus capable of accommodating dynamic propagation conditions and unknown traffic characteristics.2. The channel conditions change over time in wireless networks, especially in heterogeneous networks. The throughput of wireless links always have unpre-dictable fluctuation. The randomness of wireless channels, which may cause frequent interruption or fluctuation of video playback, is challenging for wire-less video streaming. In this dissertation, a smoothness constraint based dynamic layer switching strategy is proposed for wireless scalable video streaming over a variable bitrate channel. It aims for optimizing long-term quality of users" experi-ence via maximizing video quality subject to a constraint on playback smoothness (i.e., interruption and fluctuation). The proposed algorithm does not require a pri-ori knowledge of the channel dynamics, and is capable of operating online with the currently available information. We further characterize the attainable perfor-mance of the proposed algorithm by theoretical analysis as well as experiments.3. This dissertation considers multi-user scalable video streaming in heterogeneous networks in order to address the challenges of time-varying channel conditions, stringent QoS requirements of video applications as well as complex cross-tier interference between macrocell and femtocells. Dynamic video layer selection and resource allocation are invoked to enable the adaptation of the scalable video streaming service to the dynamics of channel quality and interference prices. We formulate the problem into a constrained stochastic optimization problem which characterizes the trade-off between the perceivable video quality and the mone-tary cost of the interference. Since the time scale of resource allocation is much s-maller than that of the video layer selection, we decompose the original long-term average goal into two instantaneous optimization subproblems by leveraging the technique of Lyapunov drift and optimization. By exploiting the special structure of these subproblems, low-complexity algorithms are derived for dynamic video layer selection and resource allocation, which relies on the currently available information rather than any statistical prior knowledge. We further derive the an-alytical bounds on the theoretic achievable performance. Experimental results are presented to illustrate the performance of the proposed solution.