Bob,
The minimal C code for (back-to-back or overlapping) ADEV is:
stride = overlap ? 1 : tau;
for (sum = n = i = 0; (i + 2*tau) < count; i += stride, n += 1)
sum += pow(phase[i + 2*tau] - 2 * phase[i + tau] + phase[i], 2);
return sqrt(sum / 2 / n) / tau;
See adev_lib.c under www.LeapSecond.com/tools/ and also adev0.c adev4.c adev5.c
The latter two display the number of terms used in the calculation, which
partly addresses your question.
/tvb
----- Original Message -----
From: "Scott Stobbe" <scott.j.sto...@gmail.com>
To: "Bob Stewart" <b...@evoria.net>; "Discussion of precise time and frequency
measurement" <time-nuts@febo.com>
Sent: Thursday, January 12, 2017 2:36 PM
Subject: Re: [time-nuts] General questions about making measurements with time
interval counter.
The pesudo code for the Adev is quite easy to interpret.
For a frequency record of N samples
For each tau=M samples
Reshape(N/M,M)
Mean
Diff
Rms
End
On Thu, Jan 12, 2017 at 4:11 PM Bob Stewart <b...@evoria.net> wrote:
> Hi Bob,
>
> OK, like Bugs Bunny, I'll venture out on the limb, cut the limb, and see
> whether I fall or the tree falls:
>
> Wouldn't it take 1801 samples to get 18 seconds at 100S tau? Maybe I
> didn't state that properly, but I think you get my meaning. Also, I've
> never actually taken the time to look at the formula or the code to see how
> the ADEV is calculated. But doesn't it use a sliding boxcar type of
> calculation? Or is that some other *DEV? My point is that for 1801
> seconds, aren't there a lot more than 18 samples put in the 100S bin? And
> I've probably stated that incorrectly, too.
>
>
>
> Bob
>
>
>
>
>
> From: Bob Camp <kb...@n1k.org>
>
> To: Bob Stewart <b...@evoria.net>; Discussion of precise time and
> frequency measurement <time-nuts@febo.com>
>
> Sent: Thursday, January 12, 2017 2:38 PM
>
> Subject: Re: [time-nuts] General questions about making measurements with
> time interval counter.
>
>
>
> Hi
>
>
>
> Keep in mind that when you do 1800 samples at 1 second, that data will
> only meet the
>
> 100 sample requirement out to tau = 18 seconds. Past that you are in the
> “under 100 samples
>
> region”.
>
>
>
> Bob
>
>
>
> > On Jan 12, 2017, at 2:32 PM, Bob Stewart <b...@evoria.net> wrote:
>
> >
>
> > Hi Bob,
>
> > OK, thanks for explaining. When you and others use highly technical
> terms like "small number of samples" it's not always clear to me what you
> mean. =) Ten samples? That's not enough for anything. Normally I run at
> least 1800 samples; at least if I plan to share them with someone.
>
> >
>
> > Bob -----------------------------------------------------------------
>
> > AE6RV.com
>
> >
>
> > GFS GPSDO list:
>
> > groups.yahoo.com/neo/groups/GFS-GPSDOs/info
>
> >
>
> > From: Bob Camp <kb...@n1k.org>
>
> > To: Bob Stewart <b...@evoria.net>; Discussion of precise time and
> frequency measurement <time-nuts@febo.com>
>
> > Sent: Thursday, January 12, 2017 1:03 PM
>
> > Subject: Re: [time-nuts] General questions about making measurements
> with time interval counter.
>
> >
>
> > Hi
>
> >
>
> > It varies from 5370 to 5370. You see a lot of plots that run out to 10
> samples or less. Anything below 100 samples
>
> > is risky in some senses.
>
> >
>
> > Bob
>
> >
>
> >> On Jan 12, 2017, at 12:25 PM, Bob Stewart <b...@evoria.net> wrote:
>
> >>
>
> >> Hi Bob,
>
> >> OK, what's a small number of data points? Attached is a screencap of
> captures for 25, 50, 75, 100, and 150 seconds. Yeah, at 25 seconds, the 1S
> tau is up at 4.56E-11, but it falls pretty quickly. I will mention that
> this particular 5370 is much better than my other one. So, maybe this one
> is an exceptional example?
>
> >>
>
> >> Just for grins, I also included a screencap of the phase points.
>
> >>
>
> >> Bob -----------------------------------------------------------------
>
> >> AE6RV.com
>
> >>
>
> >> GFS GPSDO list:
>
> >> groups.yahoo.com/neo/groups/GFS-GPSDOs/info
>
> >>
>
> >> From: Bob Camp <kb...@n1k.org>
>
> >> To: Bob Stewart <b...@evoria.net>; Discussion of precise time and
> frequency measurement <time-nuts@febo.com>
>
> >> Sent: Thursday, January 12, 2017 11:04 AM
>
> >> Subject: Re: [time-nuts] General questions about making measurements
> with time interval counter.
>
> >>
>
> >> Hi
>
> >>
>
> >> There is a big difference between RMS and single shot. Single shot, the
> 5370 is a very different beast.
>
> >> That’s not a big deal when you have a few thousand readings and it all
> averages down. Unfortunately
>
> >> we all love to do runs with a very small number of points and then draw
> conclusions from them. As the
>
> >> sample size goes down, you no longer have a 2 to 4 x 10^-11 beast, it’s
> more like 5X that.
>
> >>
>
> >> Bob
>
> >>
>
> >>
>
> >>> On Jan 12, 2017, at 11:31 AM, Bob Stewart <b...@evoria.net> wrote:
>
> >>>
>
> >>> Hi Bob,
>
> >>> Normally I see somewhere between 2E-11 and 4E-11 at 1S tau on my
> 5370A, as in the blue trace on the attached plot. Am I misunderstanding
> your meaning? Granted, I am clocking the 5370A with a GPSDO, but I believe
> I see about the same thing with the HP10811. This test was 1PPS vs 1PPS on
> two different units.
>
> >>> The plot also has a test run by Tom, in orange, using his H Maser and
> a Timepod to show how poor the 5370 is compared to the Timepod below about
> 60S tau. These are essentially apples vs apples tests.
>
> >>>
>
> >>> Bob
>
> >>>
>
> >>>
>
> >>>
>
> >>>
>
> >>> From: Bob Camp <kb...@n1k.org>
>
> >>> To: Discussion of precise time and frequency measurement <
> time-nuts@febo.com>
>
> >>> Sent: Thursday, January 12, 2017 8:27 AM
>
> >>> Subject: Re: [time-nuts] General questions about making measurements
> with time interval counter.
>
> >>>
>
> >>> Hi
>
> >>>
>
> >>> There are a number of ways to improve the resolution (and accuracy) of
> your data without spending
>
> >>> big piles of cash. They have been discussed here on the list many
> times over the last few years.
>
> >>> What I’m suggesting is that you dig into that ahead of taking data.
> You will dive into it eventually as you
>
> >>> look more and more at devices that are locked to some sort of stable
> reference internally.
>
> >>>
>
> >>> Ideally you would like a device with a floor 5X to 10X better than
> what you are measuring. For ADEV style
>
> >>> data, the 5370 is a 1x10^-10 sort of device single shot (so 1x10^-9 is
> the limit at 10:1). With a lot of averaging
>
> >>> (which is not something you do with ADEV) you can get about 5X better
> than that as a floor. In either case, it is getting in the way of any
>
> >>> readings that are much below 1x10^-9 at one second. A low cost XO can
> hit that level of performance.
>
> >>>
>
> >>> Bob
>
> >>>
>
> >>>
>
> >>> <ADEV.png>_______________________________________________
>
> >>> time-nuts mailing list -- time-nuts@febo.com
>
> >>> To unsubscribe, go to
> https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts
>
> >>> and follow the instructions there.
>
> >>
>
> >>
>
> >> <ADEVs.png><Phase.png>_______________________________________________
>
> >> time-nuts mailing list -- time-nuts@febo.com
>
> >> To unsubscribe, go to
> https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts
>
> >> and follow the instructions there.
>
> >
>
> >
>
> >
>
> > _______________________________________________
>
> > time-nuts mailing list -- time-nuts@febo.com
>
> > To unsubscribe, go to
> https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts
>
> > and follow the instructions there.
>
>
>
>
>
>
>
> _______________________________________________
>
> time-nuts mailing list -- time-nuts@febo.com
>
> To unsubscribe, go to
> https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts
>
> and follow the instructions there.
>
>
_______________________________________________
time-nuts mailing list -- time-nuts@febo.com
To unsubscribe, go to https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts
and follow the instructions there.
_______________________________________________
time-nuts mailing list -- time-nuts@febo.com
To unsubscribe, go to https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts
and follow the instructions there.
