【作者】 王飞；

【导师】 廖晓峰； 郭松涛；

【作者基本信息】 重庆大学 ， 计算机科学与技术， 2012， 博士

【摘要】 随着无线网络应用的日益普遍以及复杂多媒体业务的不断涌现，无线网络的业务量急剧增加。由于无线网络所能提供的传输能力大多是有一定限度的，如何将有限的无线资源，以合适的方式分配给不同的用户或业务，以满足他们对无线资源的需求就成为一个必须要考虑的问题。然而，无线网络资源的稀缺性，无线网络信道状态的不稳定性，无线网络中不同类型业务的不同服务质量(Quality ofService，简称QoS)需求，以及无线资源分配过程中体现的不公平性，都给无线网络资源分配带来了很多挑战。基于网络效用最大化(Network Utility Maximization，简称NUM)理论，本论文主要研究了无线Ad Hoc网络与无线蜂窝网络的资源分配。论文主要从以下几个方面展开了研究：①对于支持实时业务的无线Ad Hoc网络，研究了其中的速率和功率分配问题。由于实时业务具有严格的QoS需求，因此，需要在充分考虑它们QoS需求的基础上进行资源分配。首先，对于信道慢衰落的网络，给出了一个基于NUM的资源分配模型。在这个模型中，我们充分考虑了实时业务对于QoS度量，即端到端时延、缓冲区溢出引起的丢包率、以及数据流可靠度的需求。其次，对于信道快衰落的网络，通过允许网络经历一定的信道快衰落引起的拥塞，提出了另外一个模型。在这个模型中，除了考虑上述QoS度量外，我们还考虑了信道快衰落引起的丢包率。虽然这两个模型都是非凸的，通过适当的数学变换并利用拉格朗日对偶法，我们仍然给出了若干分布式算法。最后，通过与一个已有模型相对比，实验结果表明本论文提出的模型更适合处理实时业务的资源分配，它们可以使得实时业务的QoS需求得到很好的满足。②研究了瑞利快衰落环境中，无线Ad Hoc网络中带有链路中断约束的速率和功率分配问题。此时，由于信道快衰落，无线通信可能会发生中断，以至于大量数据包被丢弃，从而数据流目的节点的接收速率远小于源节点的发送速率。为了处理这种情形的资源分配，作者提出了一个基于NUM的模型。在这个模型中，为了更公平地分配资源，作者假设效用函数是目的节点接收速率的函数而非源节点发送速率的函数。同时，为了充分考虑数据包的丢失，作者考虑了通信链路的中断概率。而且，通过考虑信道快衰落的统计特性，给出并利用了一个近似平均信道容量。虽然所提出的模型是非凸的，通过变量替换并使用拉格朗日对偶法，仍然给出了一个分布式算法。并且，由于提出的模型充分考虑了信道快衰落的统计特性，链路功率可以不随信道的快衰落状态而变化。通过与基本的NUM模型相对比，实验结果表明提出的模型可以更好地处理快衰落环境中的资源分配。③对于弹性流与非弹性流共存的信道慢衰落的无线Ad Hoc网络，研究了其中的动态速率和功率分配问题。在充分考虑弹性流和非弹性流的不同QoS需求的基础上，给出了一个基于NUM的随机最优化模型。这个模型的目的是在动态分配链路功率和数据流服务速率的基础上，最大化网络性能并满足不同数据流的不同QoS需求。由于允许非弹性流的效用函数取任意非凹函数，所提出的模型是一个NP-hard问题。为了求解这一非凸优化问题，基于随机对偶理论和粒子群优化(PSO)方法，给出了一个动态速率和功率分配算法。通过使用这个算法，可以在不知道网络状态分布的基础上动态分配功率和速率。并且，当信道条件变化的随机性非常大时，该算法可以提供一个很好的近似解。实验结果表明所提出的算法可以有效地利用网络资源以最大化网络性能，同时可以很好地满足不同数据流的不同QoS需求。④对于上行通信采用OFDMA多址接入方式的无线蜂窝网络，研究了其中的子载波和功率分配问题。为了将子载波在具有不同信道条件的用户间公平分配，并将每个用户的功率在该用户使用的子载波间有效分配，给出了一个带有公平性考虑的最优化模型。这里的公平性是通过赋予每个用户一个效用函数，并设置分配给每个用户的子载波数量下限而得到保证。特别地，允许效用函数取非凹、非可微函数，以便所提模型也适合于实时业务的资源分配。为了求解这一非凸优化问题，作者提出了一个基于蚁群优化(ACO)的算法。利用该算法，可以公平有效地分配子载波和功率。实验结果表明，相比于其它算法，所提出的算法在资源分配的公平性方面具有更好的性能。更多还原

【Abstract】 With the popularity of wireless networks and the emerging of complex multimediaapplications, wireless network business has increased dramatically. Since generallythere is a limit on the transmission capacity of wireless networks, how to allocate thelimited wireless resources to different users or applications in a suitable mannerbecomes a key point to wireless resource allocation. However, wireless resourceallocation faces a lot of challenges, such as the scarcity of wireless resources, theinstability of wireless channel states, the different QoS requirements of differentapplications, and the unfairness embodied in wireless resource allocation. In this thesis,on the basis of NUM, we mainly study resource allocation for wireless ad hoc networksand wireless cellular networks. The contributions of this thesis are listed as follows:①We study the rate and power allocation problem for wireless ad hocnetworkswhich support real-time applications. Since real-time applications have strictQoS requirements for packet delay, packet loss, and reliability, we need to considerthese QoS metrics when allocating resources. First, we present a NUM-based model fornetworks with slow-fading channels. In this model, for real-time applications, wesufficiently consider their QoS requirements for end-to-end delay, packet loss due tobuffer overflow, and flow reliability. Second, for networks with fast-fading channels, wepresent the other model by allowing networks to experience a limited amount offading-induced congestion. In this model, besides packet loss due to buffer overflow, wealso consider packet loss owing to channel fast fading. Although the above two modelsare both non-convex, we still construct several distributed algorithms by applyingappropriate transformations and the Lagrangian dual method. Finally, by comparingwith an existing model, simulations verify the validity of our models on tacklingresource allocation for real-time applications.②For Rayleigh fast-fading ad hoc networks, we address the problem of rate andpower allocation with link outage probability constraints. At this time, due to the fastfading of wireless channels, wireless communication may be interrupted, so that a largenumber of packets are discarded and the receiving rate at the destination node is muchlower than the transmission rate at the source node. In this thesis, we give a NUM-basedmodel to designate this scenario. In this model, to allocate resources in a more fair way,we assume that the utility function is a function of the receiving rate at the destination node. In addition, we take into account the link outage probability to sufficientlyconsider packet loss owing to channel fading. Meanwhile, by taking account of thestatistical characteristics of channel states, we give an approximate average link capacity.Although our model is non-convex, we still construct a distributed algorithm byapplying appropriate transformations and the Lagrangian dual method. Since our modelsufficiently considers the statistical variation of channel states, the power updates neednot follow the fast-fading states of wireless channels. In comparison with the basicNUM framework, simulation results demonstrate that our model can better deal withresource allocation in Rayleigh fast-fading environment.③we study the problem of dynamic rate and power allocation for wireless ad hocnetworks with slow-fading channels, where a mixture of elastic and inelastic traffic issupported. A stochastic optimization problem incorporating the different QoSrequirements of the two types of traffic is formulated, which aims to maximize thenetwork performance by dynamically allocating link powers and flow service rates.Since the utility functions of inelastic flows are allowed to be any non-concavefunctions, the proposed original problem is NP-hard. In order to solve this non-convexoptimization problem, we propose a dynamic rate and power allocation algorithm basedon the stochastic duality theory and the particle swarm optimization (PSO) approach.This algorithm provides a good approximation to the optimal solution when thevariation of channel condition of each link gets larger. Besides, by using this algorithm,flow rates and link powers can be dynamically allocated without the need for thedistribution of network states. Simulation results show that our algorithm can efficientlyutilize network resources to improve the network performance.④We address the subcarrier and power allocation problem for uplink OFDMAcellular networks. First, an optimization framework with fairness is formulated, whichaims to fairly allocate subcarriers among users with different channel conditions and todistribute the transmission power of each user over the assigned subcarriers. Here, thefairness of resource allocation is guaranteed by associating each user with a utilityfunction and putting a lower limit on the number of subcarriers assigned. In particular,utility functions are allowed to be non-concave and non-differentiable so that ourframework can be suitable for resource allocation for real-time applications.Furthermore, an algorithm based on the ant colony optimization (ACO) is proposed,according to which subcarriers and powers can be fairly and efficiently allocated.Simulation results show that our algorithm outperforms several other algorithms in terms of the fairness of resource allocation.更多还原