[time-nuts] How does one actually do Allan variation graphs?

Magnus Danielson magnus at rubidium.dyndns.org
Sun Nov 14 15:48:10 UTC 2010


On 11/14/2010 09:41 AM, John Miles wrote:
>
>> I’ve looked at Wikipedia and I am as lost as when I started.
>>
>> Could someone walk me through the process step by step and also
>> tell me what test equipment is required?
>>
>
> Besides the pointers at www.leapsecond.com , I've collected a few links at http://www.ke5fx.com/stability.htm that may be helpful.
>
> The first .PDF link on that page is my presentation from the Microwave Update conference a few weeks ago.  It was meant as an introductory "Stability Measurement for Radio Nuts" talk, discussing the state of the commercial art in light of what's available to hobbyists.
>
> The NIST links under "General timing and noise metrology", in particular this one ( http://tf.nist.gov/general/pdf/2220.pdf ) are excellent.
>
> If you have an HP 5370A/B counter and a GPIB interface you can do a lot of good measurement work.  With the appropriate software you can make conventional strip-chart style plots of frequency and phase, as well as ADEV and similar plots.  Unless you are a software nut you probably do not want to homebrew the necessary code to do this.  Most people don't use the same program for acquisition and plotting; a script or batch file does the job of reading the data from the counter and spooling it to a text file, while a program like Stable32 or Ulrich Bangert's (search on df6jb plotter) renders the graphics.
>
> My own app (TimeLab) is an exception, in that it attempts to do a good job at both data acquisition and rendering.  It's still under heavy construction.  Right now I'm rewriting all of the acquisition routines to support, among other things, the use of more than one GPIB counter at once.
>
> Given that you have an HP 5370 available, if you wanted a walkthrough, you could try something along these lines:
>
> 1) Get an NI or Prologix GPIB adapter, install per manufacturer's guidelines.
>
> 2) Download the current TimeLab beta.  You have two options here:
> 	http://www.ke5fx.com/timelab/setup.exe -- Graphically ugly but better tested
> 	http://www.ke5fx.com/timelab/setup_temp.exe -- Nicer looking, but more likely to have bugs, and some features have yet to be ported over to the new codebase.  Use this one for the instructions below.
>
> 3) Decide whether you want your HP 5370A/B to run in talk-only mode or addressable mode and set its DIP switch accordingly.  The software will work either way since it doesn't actually try to control the counter, but for a 5370 I'd use addressable mode unless you have a reason not to.
>
> 4) Set up a basic frequency measurement to begin with.  Feed a 10 MHz signal or whatever into the STOP jack, and hit FREQ and 1s.
>
> 5) In TimeLab, select Acquire->Acquire from HP 5370A/B, and then select the NI interface or the Prologix interface's COM port from the list.  Hit the "Monitor" button and you should start seeing the counter's frequency readings scroll by.  If not, find out why before going any further.
>
> 6) Hit "Start Measurement."  After a few readings have come in, you should see your ADEV plot start to take shape.
>
> 7) Hit the 'f' key to switch to a frequency-difference chart, or the 'p' key for a phase-difference chart.  The 'y' key will toggle the Y-axis between easy-to-read round numbers and full display range.
>
> You can get somewhat cleaner measurements from the 5370 if you use time-interval mode rather than frequency mode, but time-interval measurements require a 1-pps or similar source and some additional setup effort.

0) Essentially whatever source you have (crystal, Rubidium, Cesium, 
GPSDO) unless you haven't done it before, turn it on well in advance. I 
prefer days over hours. Locked crystals such as Rubidium, Cesium and 
GPSDOs will cancel the last part of the oscillator drift but depending 
on details performance may be more or less compromised by this drift. I 
think this is one of the practical details one should not miss.

I for one thinks that using a trigger signal such as the PPS or more 
preferably a higher frequency trigger is worthwhile, as you get a more 
stable rate of read-outs. Also, it gives a larger amount of raw data, 
allowing for the increased degrees of freedom and quicker convergence of 
estimator(s).

Do use TimeLab, I think it is a great way to get going. It's also fun to 
see the curve converge as more data comes in...

Cheers,
Magnus




More information about the Time-nuts_lists.febo.com mailing list