[sdiy] (oops) EN #76 question

[Russell McClellan]

> My question was more about how the 8b circuit worked and if there is a mistake or oversight in the operation. A random source that can only hit 8 different levels seems a bit useless no matter how gaussian it is.

I just got out my copy of EN76 - this is a very specific DAC for a
random noise generator.  It's not a very standard technique.  There is
a long pseudorandom sequence of bits in a shift register loop, with
taps to the DAC.  The "DAC" is just a sum of the eight taps of the
sequence, all weighted exactly the same.  There are indeed only 8
possible output levels of the DAC (how many of the 8 bits are set to
"1"?).  If the sequence of bits was initially random, this will not be
white noise but rather gaussian noise where the midpoint is much more
common than either extreme.  That's why Bernie calls this the
"Gaussian" weighting.  There's nothing inherently "Gaussian" about an
equal weight-summer, that term only makes sense in the context of the
rest of the circuit.

I think the idea is you're supposed to use a reconstruction filter
after the DAC to get an analog signal, so the number of physical
levels the DAC is able to produce is not really incredibly relevant.
Certainly I would imagine 8 levels is enough.

-Russell