Poisson-noise removal in self-similarity studies based on packet-counting : factorial-moment/strip-integral approach

In this work we point out that some common methods for estimating self-similarity parameters -- involving packet counting for the estimate of statistical moments -- are affected by distortion at the finest resolutions and quantization errors and we illustrate -- using also a small sample of the Bellcore data set -- a technique for removing this undesirable effect, based on factorial moments and strip integrals. Then we extend the strip-integral approach to the approximation of the square of the Haar wavelet coefficients, for the estimate of the Hurst self-affinity exponent.