Study On Wireless Resource Allocation Based On Network Utility Maximization

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.