[Lowfer] WSPR noise bandwidth, was 74.5495 QRSS 60 as usual ...

[JD]

>>> I sincerely doubt Joe Taylor would attempt to 'phony-balony' anyone ... 
>>> if you're unfamiliar with his work you might try 'google-ing'
Joe Taylor Princeton university.

I've long been aware of Joe Taylor's background, and I don't casually argue 
with any Nobel Laureate's choice of words in most matters.  But...

>>> The only way anyone could be fooled is due to their lack of 
>>> understanding the concept of signal, noise and bandwidth ... and how 
>>> they are related to each other.

There could be a lot of people in that boat if they figure a purported SNR 
number is meaningful when the stated noise bandwidth is not related in any 
way to that needed for communication.  This and OPERA are the only 
communication tools I've ever used where we just pick an arbitrary noise 
bandwidth, totally unrelated to the signal as transmitted, and call the 
result SNR.  In 40+ years of measuring S/N ratio in all kinds of 
transmission systems for a living, I never encountered such a thing before. 
(Note, please, that I'm only quibbling over this specific term; more on that 
in the third paragraph below.)  You don't simply pick a number and everybody 
rallies around it.  If you're going to compare Signal to Noise and call it a 
Ratio, then they really need to be apples-to-apples measurements.

In ATSC digital television, for instance, I can't pretend my station is 
viewable at a level 26 dB below noise by choosing to average the noise over 
the entire VHF band.  If I don't deliver a signal to your receiver that is 
an honest 15 dB above all noise sources _within the same 6 MHz channel I'm 
using_ then you simply won't see a picture...the bit error rate goes through 
the roof and we'll fall off the infamous "digital cliff."  Likewise, neither 
Argo nor WSPR nor OPERA will detect anything unless the signal at the 
instant of measurement is some few dB above the noise in the 
few-millihertz-wide FFT bin(s) where it finds itself.  They all have that in 
common.  Viewed at the specific place and time where actual detection of 
signal takes place, there must always be more signal than noise if you 
expect valid output.  Always.  What you do with that signal afterward... all 
the processing and decision making needed to extract information from it... 
is where the differences are, and the necessary bandwidth for communication 
(a function, in part, of the expected throughput) becomes inextricably 
entangled in that process.

That's why a legitimate SIGNAL to noise ratio must relate both measurements 
to the signal's own bandwidth.  Garry mentions the Shannon theorem, the very 
heart of what I'm saying.  The signal-to-noise expression in Shannon's 
equation certainly does not assume some arbitrary bandwidth for noise!!  It 
explicitly specifies the same bandwidth for noise as the width of 
transmission channel.  When you talk "signal-to-noise," either to a 
practicing engineer or to an information theorist, that's the condition they 
always expect to be true.

Now, if you're simply comparing detection thresholds, then sure!  That's a 
very different matter.  Saying WSPR can pull a signal out of X dB of random 
noise in a 2500 Hz bandwidth, compared to Y dB for OPERA in the same 
bandwidth, is a perfectly legitimate comparison.  (Assuming the data 
throughputs are also similar, of course.)  Arbitrary bandwidth is entirely 
fine there, provided it is mutually understood.  I'm absolutely and 
completely not arguing against that.

But detection threshold is _not_ the same thing as SNR from a communication 
engineering standpoint.

It really should be called something else if one wants to be accurate and 
truthful.  That's the point I'm trying to get across.

73
John