Science and Development Network (SciDev.Net)
One of the most popular scale exponent estimation algorithms is the Rescaled Range (R/S) analysis. The algorithm estimates the ratio of a range and standard deviation statistics in a window with increasing length. Firstly, through simulations with generated noise the authors analyze estimation precision of the algorithm and find that the algorithm overestimates and underestimates at lower and higher bounds of the estimation range correspondingly. Secondly, they analytically analyze the reasons for estimation errors. Based on spectral model of self-similar processes they find that the root of incorrect estimation is in the bias that persists in variance calculated for self-similar processes of limited length. Thirdly, they apply R/S algorithm to estimate scale exponents in WiMAX traffic, finally they compare the estimated exponent with the exponent obtained based on standard deviation statistics.