Wow, some great information has been brought forth here in the last 24 hrs!
Thank you, Sinisa and Grant, for taking the time to run those tests and sharing
the results with us.
They show me that my conclusion that a faster-adapting O1 v1.xxx NR version is
now in the v2.xxx code is quite wrong!
Sinisa, your spectagrams surely reveal the O1 v1.373b5 NR behavior nicely!
Sinisa said:
> Both views show that it takes around 600 ms for NR to reach a steady state.
> Therefore it is unable to follow actual CW code elements, or different parts
> of speech.
Indeed, the original NR implementation is well descibed by the manual and
doesn't react/adapt fast enough to attenuate short quiet periods, this we
already know. It's v2.xxxx that is of interest. Although I wonder what it would
sound like if the adaption rate was increased by an order of magnitude or so.
What's interesting to note is that the rate, 600ms, appears to be about the
same
regardless of NR setting, though the rate is exactly what the setting is
supposed to change! It looks like agc is involved, so maybe the agc decay rate
is masking the change of the NR adaption rate? I'm wondering if "we" could
trouble you again to repeat the test but with lower input signal levels; below
AGC threshold.
...And also with V.2xxxx firmware, looking primarily to see if the NR is mostly
just a non-linear expander now, (like one-half of the old dBX audio noise
reducion for magnetic tape media), with only slight amount of low pass added as
seen in Grant's images.
Bummer, wish I had the benefit of owning an O1, so I can't play with it to hear
the difference, but appreciate those who do!
Sinisa said:
> Instead, NR adapts itself to the "average" signal spectrum.
NR here meaning O1 V1.373b5 NR. Not all NRs are alike :)
Maybe we better start using notations such as "O2 v2.032 NR", because they are
very different.
Grant/NQ5T said:
> There is no bandpass filtering action on 2.032 with a steady state signal.
> That's also obvious from the spectrograms. Some high frequency rolloff, and
> that's it.
>
> Is there some reason it would show up with a keyed signal and not a steady
> state carrier? Doesn't seem that should be the case.
>
Grant, my interpretation of your images is that Doug Smith's adaptive predictor
is long gone. (sigh) So what do we have now?
What was the noise level difference between having signal present and not
present with O2 v.2032 NR on(9)? If it's not too much trouble, can you set up
the same test but leave NR(9) on and record the spectrum with and without the
signal? That should help reveal the extent of the expansion behavior I've
mentioned above. The noise floor should be pushed down when the signal is off,
if it is what I think it is. And if Sinisa is able to repeat his test and zoom
in the timescale, we can get a feel for the gain change response time.
Another question to answer is "In v.2xxx, what does the NR setting actually
adjust?"
Ron Castro wrote:
> In the process, though, the baby might have been thrown out
> with the bathwater.
I'm not so sure.
If Bill Tippet, W4ZV, is correct, that the O1 v.1xxx NR "will not do anything
that adjusting DSP BW and Gain controls will not do faster and
more directly", then that NR wasn't really adding anything except convenience.
(don't flame me to harshly for saying that :-)) With the possibility that the
"new" NR is adding this non-linear expansion behavior that I keep going on
about
(sorry) then this IS adding something NEW. We can twiddle the BW and RF gain to
get the same results as with V1.xxx NR, and now can throw the new NR function
on
top of that. Maybe it doesn't help in all situations, and makes some worse, but
I find it helps in many! This is a good thing.
The problem is, IMHO, that we have had all get together like this to figure out
what the heck it really is and how and when will it help. T-T could have
provided more info about it - oh well.
But it sure has us talking... :)
73,
Lin
WB1AIW
_______________________________________________
TenTec mailing list
TenTec@contesting.com
http://lists.contesting.com/mailman/listinfo/tentec
|