Services on Demand
Journal
Article
Indicators
- Cited by SciELO
Related links
- Similars in SciELO
Share
Ciencia, docencia y tecnología
On-line version ISSN 1851-1716
Abstract
TORRES, María Eugenia and SCHLOTTHAUER, Gastón. Physiological conclusions and fractal estimators. Cienc. docencia tecnol. [online]. 2007, n.34, pp.177-205. ISSN 1851-1716.
In this paper we compare three different methods used to estimate Hurst exponent and we analyze their efficiency while they are applied in data series of different lengths. We analyze synthesized fBm time series, pure and with superimposed sinusoidal trends. We show that the three methods here discussed, DFA, wavelets based and discrete variations, no only are highly dependent on the signal length, but also on the order or number of moments (polynomic, wavelet regularity or discrete variations). For large enough data length (higher than 212), the methods based on wavelets and discrete variations have shown to be the less biased and the more stable ones for simulated fBm signals. We show that DFA method, the most widely used among the biomedical community, is the one that provides worst estimators, displaying ambiguous results while applied to real biological signals of different lengths and for different parameters in the estimator. More over, nor DFA neither the other two methods provide reliable estimators if physical or physiological conclusions are wanted. The obtained results indicate that more caution should have to be taken while trying to derive physiological conclusions from the Hurst exponent estimations obtained from real data.
Keywords : Long range dependency; Self similarity; Hurst exponent; Physiology.