【作者】 林豪；

【导师】 任勇；

【作者基本信息】 清华大学 ， 通信与信息系统， 2002， 硕士

【摘要】 近年来，在宽带无线网络中向多媒体应用提供服务质量(QoS)保证已成为研究热点。其中，无线资源的分配与调度是提供QoS保证的关键，当前的研究集中在针对无线信道容量的时变和位置依赖特性，设计算法来有效利用宝贵的无线资源，支持用户的QoS要求，同时满足一定的公平性。其中有大量问题尚待解决，有的概念也没有统一的认识。以公平性为例，相关文献中采用的有：占有资源的公平性[1]、吞吐量公平性[2]、收益Utility公平性[3]和价格Price公平性[4][5]等。本文提出了分时间尺度(Time-scale Decomposition)的无线资源分配与调度策略和系统模型。（1）针对无线物理层，引入效率函数来表征在一定残余误码率BER要求和一定信道信噪比SNR情况下，上层应用正确传输的数据量和下层无线资源的关系。（2）针对多媒体应用，使用不同的Utility函数来表征不同的QoS要求，其优点有：一是可以同时表征业务对带宽和时延的要求；二是，可以表征某些应用的“软”QoS要求。（3）针对无线信道衰落的多时间尺度特性，将信道条件SNR分解到不同时间尺度上，在不同时间尺度上针对不同的衰落特性来设计不同的算法，完成不同的任务：在大时间尺度（帧）上进行资源分配，面向应用提供QoS保证，保持应用间Price公平性的情况下，寻求系统总收益Utility的最大化；在小时间尺度（时隙）上进行时隙调度，利用各用户信道容量的时变性，使各用户获得比平均信道条件情况下更高的吞吐量。该模型中的资源分配问题是一个非凸的非线性优化问题。论文定义了系统价格p的特征资源分配向量，提出使用该向量来求得最优解或次优解的算法，并给出了该算法求得最优解的充分条件。论文证明了：当该条件不满足时，次优解与全局最优解的距离（用百分比表示）不大于次优解中未分配的带宽占总数的百分比，不大于单个应用在临界状态时获得的无线资源占系统总量的百分比。最后，论文对该模型中的资源分配和资源调度分别提出了各自的低复杂度在线求解算法。计算机仿真结果表明该资源分配在线求解算法收敛于全局最优解，也说明了分时间尺度的资源分配和调度策略能够在提供QoS保证的同时，相对于纯调度策略，提高系统性能50%以上。更多还原

【Abstract】 In resent years, the research on how to provide quality of service (QoS) guarantees to multimedia applications in broadband wireless networks has been a hot spot. Radio resource allocation and scheduling is one of key functions to provide QoS guarantees in wireless networks. Current research focuses on design of efficient algorithms taking into account the special characteristics of the wireless environment such as time-varying channel capacity and location-dependent errors. These algorithms shall maximize the utilization of the wireless channels and guarantee QoS for the users, while providing certain fairness between users. Many questions remain open and even some definitions are ambiguous in this area. For example, there are several different definitions of fairness addressed in related works, such as time-fraction fairness [1], throughput fairness [2], utility fairness and price fairness [4][5].In this paper, a novel scheme of radio resource allocation and scheduling and its system model is proposed through the time-scale decomposition approach. (1) For wireless physical layers, the efficiency function is defined to quantify the upper layer throughput per unit of wireless resources while maintaining a certain maximum remnant bit error rate subjected to a certain SNR. (2) For multimedia applications, it is advantageous to use the different types of utility functions to express different type of QoS requirements, because it can express two major QoS metrics (bandwidth and delay) simultaneously and be used to represent adaptive QoS requirement. (3) For the wireless channels subject to several types of fading existing in different time scales, the dynamics of channel conditions are decoupled into two random processes with different mathematic properties in different time scales. Two algorithms in this scheme are proposed to dealing with each time scale: the resource optimizer allocates the resource to maximize the total revenue with price fairness and provide QoS guarantees to applications, and the slot scheduler exploits the time variability of<WP=5>channel capacities to provide higher throughputs to users than the ones obtained in mean channel conditions.The problem of resource allocation in this scheme is a non-convex non-linear optimization problem. By defining an eigen allocation vector, we propose an algorithm to obtain the global optima or sub-optimal solution. The sufficient condition to obtain the global optima is also presented. In this paper, it is proved that when the condition is not satisfied, the gap between the sub-optima solution and the global optima (expressed by a percentage of the global optima) is no more than the ratio between the left resources of the sub-optimal solution and the total resources, and is no more than the ratio between the cutoff resources obtained by one application and the total resources. Finally, the low-complexity online algorithms are presented for the radio resource allocation and scheduling in this scheme. The results of the computer simulations show that the online resource-allocation algorithm converges to the globe optima. The results also show that our scheme can obtain a 50% gain over the Only Scheduler scheme, while providing QoS guarantees.更多还原