Tue Sep 4 22:13:26 CEST 2012

S/D conclusions

* From wave form pictures: nothing looks really special apart from
  low-denominator rational numbers.

* From analysis: it's clear why for low-denominator rational numbers
  there is no low-frequency content, and why there is some LF content
  for high-denominator rational numbers, but not (yet) clear that this
  follows the frequency slope from linear analysis (== inverse of
  integrator == high pass noise).

* The non-leaky 1st order accumulator is really a value -> frequency
  converter.  It has non-vanishing autocorrelation.  From this another
  suggestion for decorrelation can be suggested: integrator state
  increments/decrements with average 0.