One of the characteristics of a self-similar time series is that the logarithm of its spectral density
as 
, with C a constant. This is used as the
basis for another method of estimating H [Beran 1993]. First,
the periodogram
is calculated, with 
and
typically 
. A multiple linear regression [Press 1992] is
then performed to find the coefficients 
in
which best fit 
to 
for 
. In 7, the logarithmic term represents
long range dependence and the remaining terms represent the short range
dependence in the time series. It is clear from 7 that
from which it follows, by comparison with equation 5,
that 
. The multiple linear regression
algorithm can also give the standard deviation on 
.
