【作者】 史海滨；

【导师】 朱光喜；

【作者基本信息】 华中科技大学 ， 信息与通信工程， 2008， 博士

【摘要】 随着通信技术、传输技术的迅猛发展,客观上使下一代通信网络同时承载语音、数据、多媒体等业务成为必然。已经有越来越多的用户利用网络进行工作、学习、娱乐,网络的服务质量(QoS)也越来越成为用户和运营商共同关心的问题。一方面,用户不希望自己购买的网络服务质量受到侵害,另一方面,网络运营商也希望自己的投入能达到最大的产出。对网络QoS的研究一直与网络的发展相随相伴,QoS的实现与资源管理密切相关。流量模型是资源管理赖以研究的基础。传统的流量模型都是假设随到接入的,这与实际系统的接入方式并不相符,实际系统的时间几乎都是被时隙化的。模型与实际的差异会对模型精度造成多大影响常常是流量模型研究中被忽视的问题,相关的文献就作者所知目前还没有见到。本文就传统的间断泊松模型(IPP)对这种误差做了研究,从数学根源上阐明了误差产生的原理。对IPP模型经过时隙化接入后的动态流量均值以及传统IPP模型经时隙接入后的误差近似公式做了推导,仿真表明考虑了时隙接入影响的IPP模型的动态流量均值与仿真值非常一致,相比传统IPP模型具有更高的精度,误差近似公式与仿真结果也高度一致,表明这两者的理论结果是可信的。本文的IPP流量均值误差公式还表明,只要接入的延时足够短,或者时隙足够小,时隙对传统流量模型精度的影响也很小,基本可以忽略。这个结论为我们放心使用传统流量模型提供了理论依据,也使得后面对流量模型的研究可以不再考虑时隙化接入带来的影响。实施呼叫接纳控制必须先已知网络流量。流量信息通常有两种获取方式,一是借助流量模型,二是对流量进行实地测量。CAC常用的流量模型有五种,但是这五种模型在流量描述和计算复杂度上互有利弊,缺少一种计算上相对简单同时又能描述动态流量分布的瞬时流量模型。本文提出了一种增强型ON/OFF流量模型。它是在传统ON/OFF流量模型基础上,通过修改传统模型假定的恒定流量为随机流量、ON和OFF状态以固定概率相互转换为ON状态持续时间分布,以及增加一个突发到达时间的概率分布作为已知条件得到的。该模型克服了传统ON/OFF模型不能描述流量随时间动态分布的缺点。本文通过推导给出了该模型在任意时刻的流量分布式,以及任意时刻的流量均值表达式。本文还给出了突发到达过程服从泊松过程、突发持续时间服从指数分布时,系统在任意时刻的流量分布和均值计算公式。并且为了计算可行性,还推导出了公式中概率的简化计算式。之后对系统动态流量均值做了仿真,结果与理论值一致。在蜂窝无线通信中,用户的移动行为对小区性能有重大影响。由于人们的移动行为受太多因素的影响,现有的模型往往不但复杂,结果离应用也有很大距离。本文提出了一个蜂窝无线通信系统的一般用户移动模型,它将小区中的呼叫分为本地呼叫、本地切换到外地的呼叫、外地切换到本地的呼叫三类,当每类呼叫的小区信道占用时间、带宽分布、到达过程都已知时,对小区带宽占用均值的预测就可以利用第三章的结果。这个一般移动模型用带宽作为模型的资源参数,克服了许多模型以信道数作为资源占用单位这种不适合分析分组通信的缺点。在该模型下,本文接着提出了一种以概率方式进行CAC的策略,并给出了该策略下优化问题的约束条件和目标方程,最后做了仿真并分析了该CAC的利弊。为了解决多业务共享传输带宽时的QoS保障问题,IETF组织提出了区分业务的DiffServ QoS模型,其基本思想就是将系统带宽按业务类别分割,每类业务获得一份带宽配额。以往带宽分割的不足在于以带宽分配以信道为单位,无法实现“无级”方式的分割,这对分组交换网络是没有太大意义的。本文提出了一种近似“无级”的带宽动态分割策略。它利用了基于有效带宽的CAC策略得到的结果,即在带宽溢出概率小于给定值的QoS要求下,系统允许的最大用户数与容量之间满足的关系式。本文动态分割算法先保证业务带宽最小配额与当前用户数相匹配,然后用搜索的方式获得使所有业务的有效带宽和最大时的各带宽配额。业务带宽分割粒度取决于搜索步长,因而可以实现近似的“无级”。本文还给出了业务带宽服从指数分布时的具体优化方程,并对双业务下的带宽分割做了仿真。对点过程流量模型的带宽动态消耗做了研究。掌握带宽动态消耗过程可以更准确地进行接纳控制和资源管理,并能在异常情况下对资源告罄预警。在流量模型研究中,点过程是另一种流量模型,与此相关的数学工具是排队论和马尔科夫链,用它们的结果计算带宽动态消耗过程涉及大量的矩阵运算,难以满足CAC的实时性要求。本文从另一个角度对带宽消耗的均值与方差过程做了推导,并在高斯过程的假设下,给出了能以一定的概率预测在当前的消耗速度下带宽用尽的时间范围。仿真表明结果是比较准确的。更多还原

【Abstract】 With the rapid development of communication and transmission technology, it’s obvious that next generation networks have the capability to undertake voice, data and multimedia services simultaneously. More and more people are taking advantage of networks in work, study, entertainment. So, the Quality of Service (QoS) are becoming more concerned by subscribers and network owners, for subscribers will not like their charged Qos to be damaged and network owners want their profit to be maximized. The research on network QoS always accompanies the development of network. Because the application type, network capacity and structure continue to evolve, which in turn leads to the change of traffic and subscriber’s requirement, the algorithms on providing QoS also need to change. So, QoS was, is and will always be the hot problem for networks to be researched.QoS problem is not a self-existent problem, but has close relationship with resource management problems. Resource management problems usually described as solving an optimal problem restricted to some QoS conditions. Call Admission Control (CAC) which is an aspect of resource management has the same model. The research in this paper includes two parts. One is fundamental research on traffic model error, traffic model and mobility model. The other is application research on dynamic bandwidth allocation for multi-service.First, the error of traditional traffic model applied to time-division system was analyzed. Traffic model is the foundation that the resource management researches rely on. In traditional traffic model, calls were assumed to be accepted at the moment they arrived. However, it’s not true in real system because in real system time was slotted. However, how the difference between model and reality made effect was neglected by traffic model researchers. According to author, the related papers were not found till now. In this paper, this effect was studied and uncovered from mathematical root. Besides, based on Interrupt Poisson Process (IPP) traffic model, the mean traffic formula of slotted IPP and error approximate formulas of traditional IPP were derived and given. Simulation results showed the high consistence of mean traffic of slotted IPP with experiments, which indicated the slotted IPP had more precision than traditional IPP traffic model. The error approximation formula of traditional IPP model was also verified through experiments. It explicitly demonstrated that the traditional traffic model will be precise enough as long as the time slot is little enough. This conclusion provided a theoretic support for the easy using of traditional traffic model due to the trifle time slot in real system. Thus, the traffic model error will not be considered again in the future traffic model researches.On the execution of CAC, one must know the network traffic. The network traffic usually attained through two methods. One is from traffic model and the other is by measurement. There were five commonly used traffic models for CAC. They had different benefits and disadvantage each. However, there is a lack of such a traffic model which can describe the dynamic network traffic and has relatively simple calculation complexity. In this paper, a new traffic model, called enhanced ON/OFF traffic model, was put forward. Basing on traditional ON/OFF traffic model, it modified the assumed constant rate in traditional model into stochastic rate, the constant transition probability into persistence time distribution of state ON, and added burst arrival time distribution as known condition. As a result, it overcame the defect in traditional ON/OFF traffic model that the dynamic traffic couldn’t be described. The dynamic traffic distribution and mean traffic formula of enhanced ON/OFF traffic model were derived and given. Besides, based on the widely used Poisson process model for bursts arrival and exponential distribution model for call persisting time, the dynamic traffic distribution and mean traffic formulas were derived and given. And simplified formulas for probability calculations in above formulas were provided. Simulations to the mean traffic showed the validity of theoretic results.In cellular communication system, the mobility of subscribers has great effect on the performance of a cell. Many models had been built for the mobility of subscribers. However, owing to too many factors’affection, the mobility models were very complex, and the results may not applicable for real system. In this pater, a general mobility model was proposed for cellular communication system. The proposed mobility model assumed directly that the channel holding time distribution was known by beforehand measurements. It divided the calls in a cell into three types, namely local calls, calls which will hand off to neighbor cells, calls handed over from neighbors. When the channel holding time distribution of these three type calls were known, the occupied bandwidth of a cell can be predicted from the results of enhanced ON/OFF traffic model in chapter three. By using this general mobility model, the defect which existed in many previous models that the bandwidth allocation unit was not bytes but channels, which was not suitable for packet switching system, was overcome. Based on this general mobility model, a CAC scheme which accepts calls at a given probability was put forward, and the corresponding optimization equations were given. Simulations were conducted and results were analyzed on this CAC scheme in the end.In order to solve the QoS problem for all services sharing system bandwidth, the Differentiated Service QoS model was proposed by IETF, which partitions the system bandwidth into sub-bandwidths for different services. The defect of many previous bandwidth allocation algorithms was that the allocation unit was channel but not byte, so the partition granularity was too large and meaningless for packet switch network. In this chapter, a new bandwidth allocation scheme was put forward which had very small granularity for partition. It took advantage of a CAC formula based on effective bandwidth. The formula gave an equation for maximum users and system capacity under the restriction of bandwidth overflow probability. On the precondition that the minimum allocated bandwidth satisfied the current user number, the scheme tried to attain the quotas of all services’bandwidth by searching method. The object of the scheme was to maximize the sum of all effective bandwidths of all services under those bandwidth quotas. The partition granularity depended on the length of searching step, so it was controllable. As an example, and for the need of simulation, a group of concrete optimal equations were derived and given when the service bandwidth obeys to exponential distribution. Based on this, simulations of bandwidth allocation between two services were conducted.更多还原