【作者】 李翠；

【导师】 巨永锋；

【作者基本信息】 长安大学 ， 交通信息工程及控制， 2013， 博士

【摘要】 IEEE802.15.4/ZigBee协议可以支持无竞争的GTS信道访问机制，因此基于GTS机制的无线传感器网络（Wireless Sensor Networks, WSNs）经过合理配置后能够提供有保证的高质量服务。通过对WSNs进行定性或定量的性能分析可以确定其配置参数是否合理。利用网络演算理论对WSNs进行定量的性能分析时，可以在不依赖复杂仿真实验平台的情况下，建立网络配置参数与性能界限之间的显式数学关系。论文基于确定型网络演算理论，系统地建立了GTS机制下WSNs的分批服务模型和网络演算模型，获得了比现有研究更为紧凑的性能界限。当这些性能界限用于指导WSNs配置时，可以节约更多网络资源。论文的主要研究成果如下：(1)针对现有研究给出的服务曲线定义比较抽象这一缺点，对三种服务曲线进行了详细分析，指出了它们之间的区别和剩余服务曲线的适用条件。(2)提出在对WSNs进行性能分析时，将激活时间序列后置于时分簇调度周期的末端，并假定各感应数据流猝发容忍在网络初始化完成后的瞬间同时产生。这种假定避免了猝发容忍都在节点GTS刚结束时发生这种低概率情况，可使最末端GTS对应节点的流入数据具有最大调度延迟，同时又避免节点服务曲线均取最大调度延迟这种不合理情况。(3)分析了GTS机制下节点的服务过程，明确了GTS长度内数据传输有效占用时间和空闲浪费时间的大小与分布规律，指出数据可以跨越同一个GTS内相邻的时槽进行连续传输。提出了更符合实际情况的基于TS的等效带宽和服务曲线，推导了基于TS的WSNs节点积压数据上界和单跳延迟上界。通过数值分析分别对基于TS和基于BI的等效带宽及WSNs节点性能界限进行比较。数值分析结果表明：节点等效带宽由数据帧长度、帧间间隔、占空比、最大重播次数等众多因素决定，而且等效带宽远小于名义带宽；同时，基于TS的方法求解的节点性能界限更符合实际情况。(4)证明了基于TS和基于BI两种不同方法求得的数据流输出函数上界是相同的，发现了输出函数上界与到达曲线的一致性、以及到达曲线的可平移性和不变形性这两个重要特性。提出了初始调度延迟、修正调度延迟、汇聚时刻、调度时刻、分段汇聚流等概念。以链式WSNs为研究对象，提出了分段汇聚流数量的求解方法，建立了链式WSNs的分批服务模型和性能分析演算模型。数值分析结果表明：与现有研究相比，本文方法能够大幅降低节点积压数据上界和端到端延迟上界；同时，对链式WSNs，现有研究所谓最坏情况，在本文演算模型中反而是最好情况。(5)建立了簇树WSNs的网络模型，分析了移动sink节点与簇树WSNs的连接方式对网络拓扑的影响。通过分析单层簇树WSNs的数据流传输与汇聚规律，发现通过建立“虚拟链式WSNs”可以建立簇树WSNs的分批服务模型和性能分析演算模型。数值分析结果表明：得益于分批服务模型的建立，本文方法求解的簇树WSNs性能界限比现有研究更为紧凑；采用合适的SD序列可以获得较优的网络性能。更多还原

【Abstract】 IEEE802.15.4/ZigBee standard supports collision-free and predictable access to thewireless medium through the Guaranteed Time Slot (GTS) Mechanism, thus Wireless SensorNetworks (WSNs) based on GTS mechanism could provide guaranteed service with highquality if they were configured well. Reasonable configuration parameters can be judgedthrough WSNs’ quantitative or qualitative performance analysis. When WSNs’ quantitativeperformance analysis is conducted by network calculus, the explicit mathematical relationbetween configuration parameters and performance bounds can be established, withoutrelying on complicated experimental simulation testbed. Based on deterministic networkcalculus and GTS mechanism, WSNs’ batch service models and network calculus models areproposed in the dissertation, which result in much tighter performance bounds. If these tighterperformance bounds were used in WSNs’ configuration, more network resourses might besaved. In detail, the main achievements of the dissertation are as follows:(1) The definitions of the three different service curves provided by the existing researchare too abstract to understand. Therefore, a detailed analysis is conducted to differentiate thethree curves’, with the applicable conditions of the left-over service curve are pointed out.(2) When WSNs’ performance analysis is being conducted, Active-Period (AP)sequences are proposed to be laid to the end of a Time-Division-Cluster-Schedule (TDCS),and all nodes’ burst tolerances arrive simultaneously as soon as WSNs have finished theirinitialization. The hypothesis avoids the low probability of burst tolerances’arriving just at theend of each node’s GTS, and makes the input data of the last node in the TDCS having themaximum scheduling time, meanwhile avoiding such cases as all service curves havingmaximum scheduling time.(3) Benefited from the analysis of a WSN node’s detailed service process for dataframes under the GTS mechanism, the size and distribution of data transmission efficient timeand ilde&wasted time in a GTS become explicit, with the fact that data can be transmittedcontinuously through neighbour time slots being pointed out. The equivalent bandwidth andservice curve which are based on TS method and much more factual are derived, whichresults in a node’s TS-based backlog upper bound and TS-based single-hop delay upper bound. Numerical analysis is conducted to compare the equilavent bandwidth and WSNs’ nodes’performance bounds both in the TS method and BI method. The numerical results illustratethat equivalent bandwidth is determined by many factors such as data frame length,inter-frame spacing, duty cycle ratio, maximum retries, and so on, and it is much less thannominal bandwidth. It is also shown that TS-based performance bounds are more factual.(4) In order to conduct multi-hop WSNs’ performance analysis, the expression of dataflow output upper bounds are analyzed at first. It’s found that the TS-Based and BI-basedoutput bounds are the same, and the output bound curve is the same as the arrival curve. Twoimportant properties, namely, the translational property of data flow arrival curve and theindeformable property of single flow arrival curve, are also discovered. Such concepts asinitial latency of service, modified latency of service, scheduling time of service, aggragatingtime of data flow and the segmental convergence flow are proposed. Then Chain WSNs aretaken as the object of study. The solution to the number of the segmental convergence flow isproposed, and batch service models and network calculus models of chain WSNs’performance analysis are established. Numerical analysis illustrated that, compared with theexisting research work, the method in the dissertation can result in greatly reduced upperbounds of nodes’ backlog and end-to-end delay. Meanwhile it’s discovered that the worst casein current research is not the worst one in the dissertation, on the contrast, it is surprisingly thebest one.(5) A network model of cluster-tree WSNs is defined, and the impact of linking typesbetween mobile sink nodes and cluster-tree WSNs is analyzed. Through analyzing the law ofdata flow’s transmission and aggregation in a simple single-level cluster-tree WSN, it’s foundthat multi-level cluster-tree WSNs’ batch service models and calculus models can beestablished, in the way of building “virtual chain WSNs”. Numerical analysis illustrated that,compared with the existing research work, the derived performance bounds derived in thedissertation are greatly reduced, and proper SD sequences can be chosen to derive goodWSNs’performance.更多还原