节点文献
多媒体视频流量建模预测及相关问题研究
Study on Multimedia Traffic Modeling Prediction and Related Issues
【作者】 付立;
【导师】 冯玉珉;
【作者基本信息】 北京交通大学 , 通信与信息系统, 2008, 博士
【摘要】 宽带通信网络技术的快速发展,为多媒体业务的更广泛应用奠定了坚实的基础,而其结构设计、性能优化和参数选择等问题的解决,对多媒体业务的理论建模和实时预测等提出更高的要求;同时,充分考虑多媒体业务特点的网络资源管理机制和路由算法又能够保证多媒体业务的服务质量。本论文从提供业务的角度出发,研究了多媒体视频业务的建模及预测问题;并针对多媒体业务的特点,提出了改进的无线多媒体网络中的呼叫接入与切换控制策略;最后用渗流理论分析了本文提出的对等网络中的路由算法。本论文的主要工作及创新点在于:提出了一种快速VBR视频流量的混沌特性检测算法,该算法在确保检测结果准确的前提下,首先对VBR视频流量的GOP数据进行预处理,然后,选用速度最快的简化方法进行视频混沌特性的快速检测。通过实际数据的仿真验证,该算法在判断准确性、实时性等方面均表现良好。另外,从混沌理论出发对VBR视频流量的GOP序列进行分析,考虑到视频流量的GOP序列可以视为含有噪声的混沌时间序列,从而可将视频GOP序列分解为混沌特征项和噪声项分别进行分析和建模。在此基础上,提出了一种基于混沌理论的VBR视频流量GOP序列建模方法。在建模过程中,先使用混沌序列拟合视频流量序列,而后利用滑动平均MA过程预测残差项。在预测误差、自相关函数和自相似性等方面,该模型取得了较好的拟合效果。与高阶AR模型相比,该模型同时兼顾了流量自相关函数的长时相关性和短时相关性,并保持了与实际流量几乎相同的自相似性。从实时预测的角度出发,针对传统视频流量模型在预测时对先验统计特性过度依赖的缺点,提出一种新的动态步长LMS算法(DLMS)。在VBR视频流量的GOP所表示的时间尺度上预测视频流量,由于该算法不需要过多的关于被预测流量的先验统计特性,且可以根据预测误差的反馈,动态改变步长的大小,计算量小,适合于在线实时预测,能够有效改善预测性能。在此基础上,提出了一种集成的视频流量预测方法。该方法使用本章前一节提出的动态步长LMS算法和基于混沌时间序列预测理论的加权一阶局域算法共同对VBR视频流量进行预测,对两种算法的近期预测效果进行比较,根据比较的结果选择其中一种算法的下一步预测结果作为该集成方法的预测结果输出。仿真结果表明,该集成方法的预测效果好于其中包含的两种算法,且具有良好的实时性。针对多业务移动通信系统的特点和现有接入策略中存在的不足,提出双动态门限保护信道的呼叫接入(DDT-GC)策略,该策略通过动态调整门限值,合理地对资源进行分配,采用二维马尔可夫过程对系统进行建模和性能分析,并与固定门限值的保护信道策略及不使用预留信道的策略进行了比较。数值仿真结果表明,DDT-GC策略既保证了系统的服务质量,又极大地降低了系统的切换中断率和呼叫阻塞率,同时提高了系统资源的利用率。提出了一种基于缓存的动态门限切换控制策略(DDT-Q算法),该算法基于共享信道的方式,考虑到非实时业务对于时延的敏感性较低,对溢出的非实时业务进行缓存,通过呼叫缓存及动态门限的不同设置,降低呼叫被阻塞的概率;同时兼顾了非实时业务的速率可变特性,动态改变其服务带宽,利用三维马氏链对其进行性能分析,与DTR-Q、DTBR算法相比,该算法的平均信道利用率、呼叫强拆率及阻塞率等性能有明显地改善。通过分析无结构对等网络中现有的路由算法:发现为避免广播风暴,其文件搜索总是被限制在某局部范围内。为此提出一种无结构对等网络中基于临界概率的路由算法,该算法使用基于临界概率的广播方法发布搜索报文,在避免广播风暴的同时,能够完成全局资源的查找。最后,使用渗流理论对该算法的覆盖效果进行了理论证明,结果表明,使用该路由算法进行文件搜索时,其搜索范围能够覆盖无结构对等网络中所有节点的概率为1。
【Abstract】 The rapid development of the information techniques by wideband networks has laid a steady foundation for more extensive use of multimedia services. To solve the problems of wideband networks’ structure designing, optimizing performance and selecting parameters requires much higher demands on multimedia service theoretical modeling and real-time prediction. On the other side, to maximize the quality of multimedia services requires the full consideration of the special characteristics of multimedia network services by network resource administrative organs and routing mechanism.From the standpoint of service providers, this dissertation studies the problems of the modeling for multimedia video service and the prediction. Moreover, it proposes how to improve the schemes of wireless multimedia network in call admission and handoff management according to characteristics of multimedia services. Finally, the routing algorithm proposed in this thesis is analyzed by using percolation theory. The major achievements and innovative points in this dissertation are outlined as fellows:A novel and fast method of detecting algorithm for Variable Bit Rate (VBR) video traffic with chaotic characteristics is proposed. By this method, the real data pretreatment and fast calculations are adopted under the premise to ensure the correctness of the detected result. By calculation with simulated actual data with algorithm, this method shows good performance in correctness and real-time capability. By considering the conclusion that GOP series of VBR video traffic includes chaotic time series with noise, we can divide the GOP series of VBR video traffic into two parts: chaotic characteristic part and noise part. Based on this analysis, a novel model for the GOP series of VBR video traffic based on chaotic theory is proposed. In the process of modeling, the algorithm uses chaotic series to approach the target series and then predicts the remaining part. The model’s performance is acceptable in the prediction of error, auto-covariance function and Hurst effect. Compared with AR model, the proposed model considers the long-range dependence and short-range dependence characteristic of video traffic at the same time. The predicted traffic remains approximately the same hurst effect as the actual video traffic.In order to predict the video traffic in real-time circumstance, a novel Dynamic step-size Least Mean Square algorithm (DLMS), which does not depend heavily on the transcendental statistical parameters like some traditional traffic prediction algorithm, is proposed. The proposed algorithm can predict the video traffic on the long-time scale of GOP. The algorithm can adjust the step size by the predicted error from the feedback and does not need any transcendental statistical parameters. So it is fit for real-time prediction. Simulation result shows that performance of the algorithm is improved significantly. Based on it, a combined video traffic prediction algorithm which can utilize the strongpoint of both two algorithms is proposed. The method for prediction uses both the DLMS algorithm and the weighted, one-rank and local-region algorithm based on chaos theory. The next step would be taking the predicted result of the algorithm which shows better performance. Simulation results indicate that the proposed combined algorithm is better than the two individual algorithms in predicted result and in real-time capability.Some research and discussion were available in the current literatures, in which the system resource utilization can be improved by adopting efficient call access control scheme. However, this is inadequate. A Dual Dynamic Threshold Guard Channel (DDT-GC) scheme is proposed in this paper, which is targeted at the characters of the multi-service mobile communication system and the shortcoming of the existing access control scheme. By dynamically adjusting the value of both thresholds, the system resource utilization can be improved with a better overall system award. Two-dimensional Markov process traffic models for cellular mobile systems are formulated and analyzed. It is verified by the numerical simulations that the DDT-GC scheme can provide the necessary service guarantee. At the same time, it can decrease the handoff dropping probability and forced termination probability, while improving the system resource utilization.A novel multimedia traffic call admission control algorithm based on call buffer in wireless networks is presented, aiming at different service bandwidth demand of multimedia application, which is characterized by Dual Dynamic Threshold access scheme with Queuing for VBR (DDT-Q). The algorithm gives attention to the variable rate character of none real-time traffic and can adapt its service bandwidth through different settings of the thresholds and the call buffer size. Theoretical performance analysis is made by using three-dimension Markovian chain, and it is also shown that compared with DTR-Q and DTBR algorithm, the performances of the presented algorithm such as average channel utility, call dropping probability and block probability can be improved significantly.By analyzing the routing algorithm in Peer-to-Peer network, the resource searching in the P2P network is always limited in certain local region to avoid broadcast Storm. Therefore, a critical probabilities based on mesh structure P2P network routing protocol is proposed. The algorithm broadcast message by critical probabilities is based on this method. The cover range of the algorithm is proved by using the percolation theory. The proved result shows that the probabilities of searching range cover all the nodes equal to 1 by using the given routing protocol.
【Key words】 Multimedia Video Traffic Modeling; Video Traffic Prediction; Call Admission Control; Percolation Theory; Chaos Theory; Peer-to-Peer Network;