In a message dated 1/2/01 12:39:30 PM Eastern Standard Time,
[email protected] writes:
<< I am a bit sceptical about the "Grab" feature of Coherent - the longer
the grab length, the less random the output data seems to become, even with
just noise being fed into the demodulator input. >>
Are you sure it's just noise, Jim? After all, there are those who are
convinced they can extract spirit voices from blank magnetic tape and
terminated microphone inputs of audio amplifiers. Why not BPSK from "the
other side" as well? <grin>
Actually, I think you'd find this effect wouldn't show up if you were to use
plain ASCII. If Bill de Carle and/or Andy come along, they may wish to
amplify upon or correct my train of thought here, as I'm working from a very
elementary understanding of error correction techniques, but I believe the
effect is an artifact of error-tolerant coding methods.
When one defines an error correction protocol, the decoder is required to
make a decision about whether an error exists, based on the state of specific
bits or combinations of bits transmitted along with the payload data. If an
error is detected, a best-guess estimate is made of what the original
character might have been, given the pattern of corruption perceived. A good
correction algorithm will make educated guesses, which in turn will be based
in one fashion or another on probability.
Hence, with most any error correction scheme (if the output is not
deliberately muted at some threshold) when the input trends toward truly
random noise, the decoder will keep guessing at what output would "make
sense" for the nearest combination of legitimate bits. Since what "makes
sense" to the algorithm is based on probabilities defined by the programmer,
the decoder will trend toward certain symbols more than others.
I'm sure there are better mathematical terms for this, but the essence is:
if the input to COHERENT is truly random, the output is a "weighted
randomness." It's just as truly random as the input. All legal symbols will
eventually appear, and there will be no real order to them.
However, the chance of any one character appearing at a given instant is no
longer a simple 1-chance-in-x-to-the-power-of-y probability. It's more like
the randomness of energy from an incandescent lamp. Energy levels of
individual photons in the flux are definitely random, yet the spectral curve
peaks at some color temperature around which most of the energy will be found.
A suitable analogy for the Grab feature might be an optical filter that
happens to have the same response curve as the incandescent light source
being shone through it. The more such filters one stacks in front of the
light (that is, the greater Grab depth), the narrower the range of
wavelengths which will be visible to the eye. Ultimately, the light would
appear nearly monochromatic, just as the decoder would clearly begin to
output one character more often than all others; but it would never truly
become--pardon the expression--coherent.
73,
John KD4IDY
|
