【作者】 甄雁翔；

【导师】 徐惠民；

【作者基本信息】 北京邮电大学 ， 电路与系统， 2010， 博士

【摘要】 随着多种无线通信技术的迅速发展,其与有线网络的融合趋势也越来越明显。在无线网络环境中,由于具有较高的比特误码、复杂的信道衰落、突发的设备噪声,这些特点将导致频繁产生丢包,而传统的TCP协议把重复确认、超时所标识的数据包丢失均归因于网络拥塞,通过启动拥塞控制机制进行流量控制：加大重发间隔、缩小发送窗口来适应网络拥塞,这些方法适宜于链路质量良好的有线环境。对于有线／无线融合的异构网络环境,就势必降低带宽利用率,导致TCP性能恶化。因此,在有线／无线融合的异构网络环境下,明确而简便地区分网络拥塞和无线误码所引起的丢包具有重大意义。本文联合相对单向传输时延、平均Ack／超时次数比两种因素集参数,针对不同的网络环境,构造动态权重分布函数,进而采用Fuzzy综合评判模型对异构网络丢包原因进行区分,并对因素集隶属度函数的建立、初始化算法、动态权重的确定等进行了详细地描述。(1)Fuzzy综合评判丢包区分模型的建立本文运用Fuzzy逻辑更加准确地区分异构网络环境下的丢包原因,在变化的网络环境下,单因素评判不能达到令人满意的效果。为了进一步提高区分准确度,本文联合考虑两种因素,构建Fuzzy综合评判模型丢包区分模型,可以提供更合理的拥塞控制和错误控制。这一算法只对发送方的TCP协议进行修改,不改变中间节点的TCP协议,保证了TCP协议端到端的语义完整性、互操作性和向后兼容性,而且修改只涉及到传输层,与其他层协议无关,保证了协议层次间的独立性。(2)因素集隶属度函数初始化算法的研究各因素依据其隶属度函数对丢包原因进行独立判断,得到各因素对当前网络状况的评判值,其评判的准确度依赖于隶属度函数及其参数估算。本文对相对单向延迟、平均Ack／超时次数比在不同丢包模式下的统计概率分布进行了深入地研究,建立了模糊评判因素集的隶属度函数,并提出了因素集隶属度函数的初始化算法。(3)动态权重算法的研究模糊评判因素集的各加权系数是各评判集评判虚警与漏报概率的函数,加权系数反映了各评判集的可信度。在实际中,由于这些概率均未知道,而且所处网络环境等因素的影响处于时变状态,固定的加权系数就不可能使评判系统始终处于理想状态。本文对因素集的权重分配问题进行了深入地研究,提出了动态权重的确定算法,从而保证算法在环境变化时具有较好的评判性能。更多还原

【Abstract】 With the development of wireless application technology, the integration trend with the wired networks becomes more and more obvious. In the wireless network condition, owing to higher BER, complicated channel fading, burst equipment noise which characteristics can lead to packet loss constantly. All packet loss marked repetitive Ack or timeout are ascribed to congestion in traditional TCP would make flow control by starting congestion to control congestion control mechanism, such as increasing retransmission intervals, decreasing sending window and so on, but these methods are suited with the wired condition of well link quality. In the hybrid wired/wireless network, the bandwidth utilization would be decreased, and the meaningless deterioration of TCP performance would be caused. Therefore, the packet loss caused by wireless error or network congestion which should be distinguished correctly and simply that has the great significance.The relative one-way delay and average Ack-Timeout times ratio are combined in this paper, according to different network environments, by building the dynamic weight distribution function, fuzzy comprehensive evaluation model are adopted to distinguish the different packet loss reason. In addition, the construction of factor set membership function, initialization algorithm, and definition of dynamic weight are described in detail.(1)Establishment of fuzzy comprehensive evaluation packet loss distinguishing modelIn this paper, fuzzy logic is used to distinguish the packet loss reason more correctly in heterogeneous networks. In varying networks, single factor evaluation couldn’t reach the satisfactory effect. To improve the distinguishing accuracy, the two factors are combined to consider, fuzzy comprehensive evaluation model isestablished and more reasonable congestion control and error control are given. On the one hand, this algorithm only modify the sender TCP, not change TCP of the intermediate node, ensure the integrity, interoperability and backward compatibility of TCP end to end semantic; on the other hand, the modifications only refer to the transmission layer, is not relative with others layers, this can ensure the independence of inter-protocol layer.(2) Study on the initialization algorithm of factor set membership functionThe packet loss reason can be judged independently by each factor membership function, so the evaluation value of each factor would be got about the current networks, the accuracy of the evaluation depends on factor set membership functions and the parameters estimation. In this paper, the probability distribution of relative one-way delay and average Ack-Timeout times ratio in different packet loss pattern are studied deeply, then the fuzzy evaluation factor set membership functions are built, finally the initialization algorithm of factor set membership function is proposed.(3)Study on dynamic weight algorithmEach weight coefficient of the fuzzy evaluation factor set is the probability function of judgement set results, the weight coefficient could reflect the reliability of each judgment set. In the practical environment, owing to the unknown probability value, and the influence of network environment is in time-varying state,so the fixed weight coefficient couldn’t make the evaluation always in ideal state. In this paper, the weight distribution problems of factor set are studied deeply, then the determination algorithm of dynamic weight is proposed, so the well evaluation performance can be ensured in varying environment.更多还原