【摘要】 新近对局域网和广域网上大量突发业务流量的监测结果表明，采用自相似建模表征业务到达过程的长时间相关特性具有较高精度，其中Hurst系数是表征业务突发特性的重要参数，因此在一定的观察时间内对突发业务的Hurst系数进行快速、准确的估计是高速宽带网络（如ATM）实施流量控制和缓冲资源分配的前提．本文提出一种基于多分辨率采样和小波分析的Hurst系数快速估计方法，对严格二阶自相似模型下Hurst系数的估计为最大似然估计．此方法还可对不同观察时隔下业务源Hurst系数的一致性进行检验，而且比传统估计方法大大减少了对突发业务流进行离散采样的计数样本总量．采用分形高斯噪声和真实突发业务数据的仿真结果均表明，本文所述方法比传统的R/S、方差时间分析等估计方法具有更高的估计精度，而且在样本量不足时具有更好的鲁棒性．本文的方法可望应用于ATM网络的业务量管制和拥塞控制．更多还原

【Abstract】 Recent measurement studies show that the burstiness of packet traffic in IAN as well as WAN is associated with long-range correlation that can be efficiently modeled as self-similar arrival process in terms ofaccuracy. Since Hurst index is the key value of this model replesenting the burstiness of traffic source , efficient estimation of Hurst index to the given accuracy is the basic step of flow control as well as buffer management in high speed broadband networks(e. g. ATM) with self-similar traffic. In this paper, we propose a fast Hurst index estimation method based on multiresolution sampling and wavelet analysis. The proposed method is the maximum likelihood estimator for strict self-similar model and can be used to check the consistency of Hurst index in different observed range of time. Simulation results based on fractal Gaussian noise and real traffic data reveal that the total number of count samples of the burst traffic for parameter estimation is peally reduced compared to the traditional methods such as R/S statistics and variance-time analysis. The proposed approach also shows more accuracy and robustness than traditional methods when lacking of sample data.Thus our method can be applied to the application of traffic enforcement and congestion control in ATM networks.更多还原