Sun Jul 29 17:28:29 EDT 2012

Combine S/D mod + decorrelator

Issue: can SDM and decorrelator be combined in one or is it necessary
to place them in series?  What's the difference?

Combined, there seems to be an arbitrary element about the "spread" of
the energy mismatch, i.e. the tradeoff between decorrelation and
exactness.  Also, how does the mismatch effect noise shape?

Integrating first-order S/D:

  s'  = s  + in
  s'' = s' - out
  out = f s'

Here 'f' is a random generator using its stored energy as a
probability to generate a "1" bit.  Probability should approach 0 when
bucket approaches 1 from above, and falls off based on the depth of
the bucket.

Main question seems to be: is the bucket bounded?  I.e. does the
chance for 0 or 1 approach 100% at some point.

Seems best to make it symmetric: for generating a "1", s=0 is the
p=50% point, s=-1 is p=0% and +1 is p=100%.  Then spreading the bucket
is to spread this out more.