[time-nuts] Measurement of frequency of HP 8720D option 1D5 oscillator after switch on

Tom Van Baak tvb at LeapSecond.com
Sat Oct 4 08:24:00 EDT 2014


Thanks for the raw data. Attached is my plot. Your data looks ok to me.

If you're interested here are some random comments about your 8720D-1D5-oscillator-frequency.csv file:

- Thanks for including the description of the data as comments in the file. I can't tell you how many times I get data files from people and you have to play "20 questions" before you know what the columns are, or what the units are, or what the sampling interval is, etc.

- Thanks for including the elapsed time (column 6) as part of the data set. This is the preferred way to determine the sampling interval; 6.3 seconds in this case.

- You can see the finite resolution of the counter. For example, looking at these five lines:
we can see it's a 7 digit measurement. Notice the last two digits are always 00. For a warm-up test, where the oscillator is changing by tens of ppm over several minutes, 7 digits is sufficient. So no worries there.

- You have 355 data points. Notice that the last 319 measurements are all 9.99999400. This is partly due to the fact that the OCXO has warmed up by this point. But this is mostly due to the fact that your OCXO is now more stable than the counter has resolution.

- Out of habit my plot uses ppm units rather than Hz. I converted your frequency measurements to relative frequency error. For example, a frequency of 10.00013100 MHz the error is (10.00013100 - 10) / 10 or 1.31e-5. The five frequency lines above simply become:
and this shows that even though it's a 7 digit counter there are only 2 or 3 digits of resolution (the DUT being quite close to 10 MHz).

- By the time you get to all the 9.99999400 MHz readings, the counter is down to 1 digit of resolution, e.g., -6e-7.

- Most people plot points and draw interpolated lines between them. The Stable32 plot I attached shows a staircase. This is partly due to the fact that the raw data is quantized in both the x and y axis (sample quantization is 6.3 s, frequency quantization is 1 Hz). It is partly done to convey that frequency is inherently an average across some interval rather than a point measurement.

Again, none of this has any impact on your original question about your VNA timebase, but I thought you might be interested in what's hidden in the data. Now imagine the fun you could have making 12 digit measurements...


----- Original Message ----- 
From: "Dr. David Kirkby (Kirkby Microwave Ltd)" <drkirkby at kirkbymicrowave.co.uk>
To: "Discussion of precise time and frequency measurement" <time-nuts at febo.com>
Sent: Saturday, October 04, 2014 3:38 AM
Subject: Re: [time-nuts] Measurement of frequency of HP 8720D option 1D5 oscillator after switch on

> On 4 October 2014 08:01, Magnus Danielson <magnus at rubidium.dyndns.org> wrote:
>> Tim,
>> The shape does not look exactly what I would expect from a less than
>> critical damped oven, it looks a little to peaky, but maybe it is the
>> resolution of the graph that fools my eye.
> I stuck the raw data as an attachment, so that should not fool your
> eye.  It is only a few kB as compressed, and not that big when
> uncompressed.
> But be careful in trying to do too much with the data after it has
> "settled down", as I have no idea of the accuracy or stability of the
> time base in the "HP 70310A precision frequency reference" which is
> used to measure the VNA. I have no idea when it was calibrated. But it
> had been powered on at least 24 hours. I don't even know the
> specification of that - it might be better or worst than the VNAs
> oven, but at least the VNA has been regularly calibrated.
> What is a bit odd is one of the tests of the VNA performed by
> Agilent/Keysight is the frequency accuracy at 20 GHz. Although when
> calibrated by Agilent a year ago there was some error, this time it
> was spot on 20.0000 .. GHz. But when the 10 MHz oscillator was
> measured, there was some error. Of course, there will be some errors
> in the VCO / PLL and the timebase could have drifted between the two
> measurements, so such differences are not entirely unexpected.
> The 3 cal certificates I have for this are here, in order of date
> (most recent first)
> 1) Keysight a few weeks ago
> http://www.kirkbymicrowave.co.uk/cal_certificates/Keysight-standard-calibration-with-uncertainties-for-8720D-vector-network-analyzer-16-09-2014.pdf
> 2) Agilent just over a year ago
> http://www.kirkbymicrowave.co.uk/cal_certificates/Agilent-standard-calibration-with-uncertainties-for-8720D-vector-network-analyzer.pdf
> 3) Techmaster (lacks any useful information), about 2 years ago
> http://www.kirkbymicrowave.co.uk/cal_certificates/Techmaster-8720D-vector-network-analyzer-cal-certificate-EXPIRED.png
> Although the Techmaster one tells you nothing useful, the Agilent and
> Keysight ones do give frequency information, including the
> uncertainty, which has actually increased since it was calibrated by
> Agilent.
> Anyway, at least I know what I have in this VNA- not that it makes a
> huge different to know, but sometimes I am just inquisitive.
> Other than invest in a GPS controlled TCXO, I am not going to do
> anything to improve this accuracy, and if I am honest, I don't really
> need any extra accuracy.
> Dave
-------------- next part --------------
A non-text attachment was scrubbed...
Name: 8720D-1D5-oscillator-frequency.gif
Type: image/gif
Size: 26530 bytes
Desc: not available
URL: <http://www.febo.com/pipermail/time-nuts/attachments/20141004/fff742ee/attachment-0001.gif>

More information about the time-nuts mailing list