基于泛函网络的VoIP业务流建模及语音质量评价研究

发布时间：2018-07-07 10:26

本文选题：Vo + IP　；参考：《西安电子科技大学》2014年硕士论文

【摘要】：近年来,IP网络规模不断扩大、网络用户数目逐年递增、网络业务种类日趋多样化,这些发展促进了数据传输网络和语音通信网络的逐渐融合,产生了新的业务——Vo IP。语音通信要求实时性、可靠性,因此对Vo IP网络的运营、维护和管理提出更高的要求,与此同时,掌握Vo IP网络的运行规律,对网络协议和设备的开发研究也有重要意义,同样也促使了Vo IP网络业务量和语音质量的研究。本文围绕Vo IP网络业务流建模及语音质量评价两方面展开研究。VoIP网络的业务量特性研究是网络测量、网络行为学、网络性能分析等研究方向的关键部分。VoIP网络中的业务量特性具有与传统的IP网络业务量模型极为不同的性质,所以我们力图构造有效的VoIP网络测量基础结构,获得VoIP网络的性能指标,掌握VoIP网络的业务性能,理解其运行规律,建立其行为模型。对于VoIP网络的语音质量评价方面,因为影响VoIP语音质量的因素很多,首先我们分类分析了不同因素对语音质量的影响,然后重点分析各种因素与语音质量之间的关系,最后给出新的高效地模型评价Vo IP语音质量。本文的主要研究工作和成果如下:(1)基于真实的网络,对Vo IP网络端到端的时延特征进行分析发现:网络中总排队时延的大小反映了网络“忙闲”的变化,排队时延具有自相似性、非平稳性和非线性的特点。根据真实网络中排队时延的特征,构造合理的网络结构,选取合适的泛函网络基函数。为了能够确切地表征时延的非线性特点,我们用泛函网络来辨识网络系统,并且为了充分反映时延的非平稳性,我们运用自适应算法来修正泛函网络中各神经元函数的参数。最后,用所搭建的实验环境检验所建立的模型和算法的有效性,在不同时间段进行验证分析的结果表明:泛函网络和自适应算法能够准确表征Vo IP网络业务流的大小。(2)虽然神经网络模型已经被应用到测评网络语音质量的领域,但是早期的方法依赖于代价高的主观测试结果,并且主观测试很难考虑复杂的输入条件,例如Internet中的变化因素。L.Sun提出一种新的神经网络模型,该模型是一种客观评价方法,但是当时还没有P.862标准,而且他使用的不是真实的实验平台。在本文中,提出的基于泛函网络的语音质量评价模型,使用最新的PESQ算法测得的值作为实测值,来训练所建立的泛函网络模型,并且使用真实的网络作为实验平台。本模型仅使用网络时延抖动标准差MSED和网络丢包率networkE作为输入,输出只有一个MOS值。用F检验得出由所建模型求得的方程是显著的,并且最后验证了所建模型是准确和有效的。
[Abstract]:In recent years, the scale of IP network has been expanding, the number of network users has been increasing year by year, and the types of network services are becoming more and more diversified. These developments have promoted the gradual integration of data transmission network and voice communication network, resulting in a new service, Vo IP. Voice communication requires real time and reliability. Therefore, higher requirements are put forward for the operation, maintenance and management of Vo IP network. At the same time, mastering the running rules of Vo IP network is of great significance to the development and research of network protocols and devices. It also promotes the research of Vo IP network traffic and voice quality. This paper focuses on two aspects of VoIP network traffic modeling and voice quality evaluation. VoIP network traffic characteristics research is network measurement, network behavior, The traffic characteristics of VoIP network are very different from the traditional IP network traffic model, so we try to construct an effective VoIP network measurement infrastructure. The performance index of VoIP network is obtained, the service performance of VoIP network is grasped, the running rule of VoIP network is understood, and the behavior model of VoIP network is established. For VoIP network voice quality evaluation, because there are many factors affecting VoIP voice quality, we first classify and analyze the different factors on the voice quality, and then focus on the analysis of the relationship between various factors and speech quality. Finally, a new efficient model for evaluating VoIP speech quality is presented. The main work and achievements of this paper are as follows: (1) based on the real network, the end-to-end delay characteristics of Vo IP network are analyzed. It is found that the total queuing delay in the network reflects the change of "busy and idle" network. Queuing delay is self-similar, non-stationary and nonlinear. According to the characteristics of queueing delay in real networks, a reasonable network structure is constructed, and the appropriate functional network basis functions are selected. In order to accurately characterize the nonlinearity of delay, we use functional network to identify the network system, and to fully reflect the nonstationarity of delay, we use adaptive algorithm to modify the parameters of neuron function in functional network. Finally, the validity of the established model and algorithm is verified by the built experimental environment. The results of verification analysis at different time periods show that functional networks and adaptive algorithms can accurately represent the size of Vo IP network traffic. (2) although the neural network model has been applied to the field of network voice quality evaluation, But the early methods depend on the costly subjective test results, and the subjective test is difficult to consider complex input conditions. For example, a new neural network model proposed by L. Sun in the Internet is an objective evaluation method. But there was no P.862 standard, and he wasn't using a real experimental platform. In this paper, the model of speech quality evaluation based on functional network is proposed. The new PESQ algorithm is used as the measured value to train the established functional network model, and the real network is used as the experimental platform. The model only uses MSED and networkE as input and only one MOS output. The equation obtained by F test is significant, and the model is proved to be accurate and effective.
【学位授予单位】：西安电子科技大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TN916.5

【相似文献】