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

[JD]

Before I propose an experiment that might better illuminate the matter of 
SNR, let's review what I have _not_ said, given all the seeming misreadings 
and side issues that have popped up lately.  This should be about simple 
science and engineering, after all.

I have not "dissed" WSPR, only mocked what I consider its misuse of the term 
SNR--but even then took pains to mention that the reading is still useful in 
its own specific way, even though it's not SNR in the sense an engineer or 
scientist would use it.  I have only addressed Argo, WSPR, and basic OPERA 
modes, because they all rely on FFT analysis in comparable bandwidths, 
differing mainly in what they do with the information before displaying it. 
I specifically did not mention WOLF because BPSK decoding is a fundamentally 
different process.

Now, a few recent points and the experiment I mentioned:

>>> Not true. The noise is analyzed in a bandwidth wider than 200 Hz - this 
>>> has been well known since the early days of WSPR.

If you re-read my post, that's what I said myself.  It's exactly what I'm 
complaining about.  The NOISE is measured in the wider bandwidth, which is 
_not_ the same bandwidth analyzed and tracked for SIGNAL.  That's what's 
wrong with calling it "SNR"...no matter how many times the user's guide 
beats us over the head with an irrelevant bandwidth.

>>> You are free to un-scale the measurements if you prefer it that way.

Problem is, the number is only truly scalable where noise power is uniform 
over the bandwidth.  If there is non-random noise (ie, other signals and QRM 
as one so often finds in the real world), that non-random component is not 
scalable on a simple bandwidth ratio.  This is where the idea of an 
arbitrary noise BW breaks down.

Random or non-random, whatever stuff in the remaining 2300 Hz bandwidth that 
WSPR "sees" while it's looking for signals is of no concern to the detection 
process, but it too gets counted in the "noise" measurement, thus inflating 
the number.  Saying we can define noise BW to mean anything we want is sort 
of an Alice in Wonderland approach to engineering.

Why does it matter?  Follow this closely as a simple thought experiment, or 
feel free to duplicate it in the real world:

1.  First find, generate, or imagine two simultaneous WSPR 2 signals of 
steady, nearly equal levels... not overly weak and not overly strong 
relative to the available noise, not too close together but still both 
within the 200 Hz detection band.  (One signal alone will demonstrate the 
basic point, but it'll be more meaningful  to have two, because a single 
signal won't give an "SNR" reading when it fails to decode.)
2.  Introduce an extraneous carrier outside the critical 200 Hz band but 
elsewhere within the 2500 Hz noise band.  Let it add maybe another 6 or 10 
dB or more to the existing total alleged "noise" without it ever having any 
chance of affecting detection.  At this point, it only further inflates the 
reported "SNR" of the signals being detected.
3.  Now move that extraneous carrier right over one of those signals of 
interest.  The total energy in the 2500 "noise" bandwidth remains exactly 
the same as in Step 2.  Exactly the same!  Remember that well.
4. Two minutes later, the reported "SNR" of the unaffected signal shows very 
nearly the same as before, if propagation is stable.  But guess what--now 
the primary signal doesn't get decoded at all.

Same noise level in the overall bandwidth...exactly the same...but a 
completely different outcome for the affected signal.

Anybody still want to argue that an "SNR" based on noise outside the 
transmission channel is interchangeable with an SNR based on noise inside 
the transmission channel?

73
John