[Beta_Carotene]

David Nebenzahl wrote:
So using my original construct ([S + N] - N = S), there's no way that
using XOR instead of subtraction is going to get you there.

You have to remember that what you wrote is not what I wrote.

A1 gives us a raw file with only binary fluctuations corresponding
to the state of the diaphragm. No bits dedicated to exponents, or
signs, or anything like that. (If you will recall, Input #1 hangs
off the boom at about 2 feet or so. It picks up all kinds of stuff
(ambience), and not just signal. Some of that "ambience" is certainly
undesirable.)

A2 gives us a raw file just like above, but from about 10 feet away,
in a cross-direction. Again, there is a whole lot more "ambience"
(most of which we ought to be able tolerate, though a case could be made
for throwing up your hands in disgust because of the wind whooshing
around). The point of A2 is for a reference level against which to mix A1.

A3 is a lot farther away, maybe 30 feet or more, and is pointed
directly *away* from A2; it is pointed at the only source of automobile
noise - through a small gap in the trees where the dirt road runs.

As for geographical locations, the cast, A2, and A3 are all in a
straight line with each other. But A2 and A3 involve microphones
pointed directly away from each other.

Halfway through the shoot, a car drives by, and disappears into the
distance. Let's say its boom box radio was blaring, too. Its noise is
logged in the mic on A3. I will concede that there are times when
there is just nothing you can do about noise. Like a dump truck, for
instance. The noise is so loud that it washes out A3, A2, and A1 - in
that order, and in that sequence - and even though there may be 20 to 30
feet between each of the mics, the sound propagates so fast that the
rate of change exceeds the amount of bits dedicated to the samples per
second. We don't really know how many bits are dedicated to each
fraction of a second. it could be 80 but it is probably greater than 8,
and I doubt greatly the bits are compressed, or resolved into floating
point numbers. I suppose somebody here was suggesting we only needed 8
bits of resolution to 80, or even $80. Well, that's a value judgment.
I would just as soon go for 80, as I wouldn't have to worry about loose
carries floating around. As you know, a loose carry represents an
overflow or underflow condition. The reason I avoid ADD or SUB is that
they involve the Carry flag, whereas the Boolean operations of OR, XOR
don't touch it.

Here is the problem: the gain could be turned up on A3, resulting in
an unusually "loud" file full of awful screeching automobile noise when
the car drives by. Not all of the mics have the same gain. They all
have to be set by the technician assigned to that sort of thing.
Anyway, everything is preset because the technician walks away, and
leaves well enough alone, and what do you know, the cast is delivering a
moving and sentimental exchange of romantic whispers. At this point,
the amplitudes are way different between A3, A2, and A1. They have to
be matched before any of the Boolean operations can be performed on
them! That's called "normalizing" - throwing away the bits that are out
of range - introducing a regrettable inaccuracy, as the more bits you
throw away, the worse the integrity of the file, and the less good you
are going to get out of it. And we can't just throw the bits away
without listening to the files, and making value judgments, scrutinizing
again and again, as to what is Signal and what is Noise. There is no
way to get rid of the technician at this point. There's nothing
automatic. But having a file of raw Noise is useful because you can
extract it by performing an XOR operation. It's done all the time with
video data.

The formula I suggested, was Noise = A3 XOR A2, assuming the files
have first been adjusted so they shae the same magically correct degree
of amplitude. I have no idea how you arrived at that magically correct
degree of amplitude, short of having listened to it in the privacy of
your studio. Anyway, after that is done, A1 XOR Noise (one or more
times, according to how nicely it sounds to your ear) is a whole lot
cleaner sounding than A1 alone.)