Ron Capik wrote:
"Floyd L. Davidson" wrote:
Ron Capik wrote:
"Floyd L. Davidson" wrote:

Distortion can *always* be counteracted by the introduction of
an "error signal" which is opposite to the distortion.
Therefore it would seem that distortion is necessarily a signal
in all cases.
...

I'd tend to say that distortion adds to noise side of the SNR, and
some can be corrected ...but *always* ? Let's say the distortion
is the result of clipping...

[ ...or maybe I've missed your point. ]

Later...

Absolutely always. Recall that distortion is a known condition
resulting from the communications channel itself. The output is
known *before* the signal is input. (E.g., clipping is not
arbitrary, and produces a very specific error signal.)

Which of course is something Shannon describes, and uses in
examples, in "A Mathematical Theory of Communication".
....

Seems I must have missed something.
From my reading of that paper it would seem that it is only the case
in a closed loop, discrete system when the (mythical) perfect observer
and error channel exist to generate said error correction.

That describes the theoretical "equivalent" implementation that
Shannon used to illustrate the point.

For a practical example, consider typical implementations of
equalizers to counter amplitude distortion. By measuring the
characteristics of the channel, and one time adjustment can be
made that corrects amplitude distortion. The equalizer
essentially introduces an equal and opposite error to the known
distortion introduced by other parts of the channel, with the
results that amplitude distortion is removed from the equation
(to the degree that the equalizer can actually match the
distortion).

With a continuous source Shannon noted: " ... Since, ordinarily, channels
have a certain amount of noise, and therefore a finite capacity, exact
transmission is impossible. "

From the subject line I would expect to be dealing with continuous
source "ordinary" channels.

(I'm not sure what you mean by a "continuous source" channel. I
more or less ignored your odd use of "discrete system" above, but
it suffers the same problem of being ambiguous in this context.
The two words should related to analog vs. digital, but I don't
think that's what you meant.)

Keep in mind that I merely said it "could" be done. I did *not*
say it was practical. Of course in many cases that is exactly
what is commonly done (e.g., with amplitude distortion as
described above), but in others it just is not practical for any
number of reason, one of which would be when enormous bandwidth
is required. For example, it would hardly make sense to reduce
quantization distortion with that method!

Regardless, the point is that distortion is a known change which
can always be predicted from the characteristics of the channel
when a known signal is applied to the input. The difference
between distortion and noise is that noise is external to the
definition of the channel, and cannot be calculated before the
fact. Hence there is no "known error signal" with noise, but
there is with distortion.

--
Floyd L. Davidson http://www.apaflo.com/floyd_davidson
Ukpeagvik (Barrow, Alaska)