基于小波变换的网络长相关业务Hurst参数估计

Long-Range-Dependent Traffic Parameter Estimation Using Wavelet Transform

小波变换具有多分辨率,也叫多尺度的特点,可以由粗及细地逐步观察数据,通过引入多尺度性可以进一步揭示网络中的流量特性。在小波理论框架下,重点分析了长相关过程在经过小波变换后的性质,即将原来在时间域内长范围相关的流量序列转化为小波域里尺度间不相关,尺度内短相关的小波系数,使得原先复杂的估计问题变为一个简单的线性回归问题。同时分析了小波估计方法的无偏性,小波函数的选择以及估计方法的时间复杂度等重要的问题。在分析过程中,小波函数的消失矩起到了关键的作用。通过实验证明,与其它估计方法比较,是在估计值准确和时间复杂度及适用范围上更简单、准确性高、计算量小以及对于数据集的大小并不敏感的方法。

There is now ample evidences that long-term correlations are present in network traffic which has the obvious multi-scale.Wavelet transform with the character of multi-resolution,can be used to gradually observer the signal by multi-scale.Using this multi-scale property can display the character of network traffic.This paper first introduces a wavelet-based tool for the analysis of long-range-dependence,then gets the estimator which is unbiased and efficient via the theoretic analysis.And also some other characters are obtained about the estimator,for example,the choose of the wavelet and the computational complexity.The number N of vanishing moments plays a central role in the analysis.By the experiment and compared with other estimator,the wavelet-based estimator is a natural,statistically and computationally efficient estimator which allows the direct analysis of very large data sets.The wavelet-based estimator will be used with priority.