[time-nuts] 10MHz Rubidium reference source for frequencycounter

Tom Van Baak tvb at LeapSecond.com
Mon Nov 3 09:43:06 EST 2014


Hi Karen,

Ok, you are making very good progress. Thanks for the update.

Your new ADEV plot, "Morion OCXO ADEV Freq mode SmartFreq Sample_1s InterpCalib_On.jpg" is now looking more like I would expect; it is showing the expected trade-off in measurement speed and measurement resolution. I see you are getting below 5e-12, past about tau 10 seconds. There are some periodics in the measurement which might go away with different trigger levels or better isolation or much larger frequency delta between the external ref OCXO and external input OCXO.

Attached is CNT91-error.gif -- the uncertainty calculation as found in chapter 8 of the CNT-91 manual. It is worth reading the details of this chapter to gain further understanding of the measurement process.

The single-shot spec for this counter is 50 ps. But from the equation you can see why they are able to claim 1 ps rms resolution for sufficient averaging (up to 1000 measurements per second). This works because the CNT-91 is an extremely quick continuous time-stamping counter with massive internal memory (something like 750,000 readings in internal RAM). And it has true back-to-back (zero deadtime) period / frequency measurements, which are available over GPIB.

The question is if these GPIB capabilities are compatible with TimeLab, which may just treat this as "dumb" frequency counter instead of a high speed timestamping counter. I will continue to test my own CNT-91 here to see if I can improve on the noise floor. I will loan John my CNT-91 if it looks like a change to TimeLab is necessary; to make best use of raw high-speed timestamp-based measurements.

One other thing to try is the ADEV function on the front panel. The CNT-91 is one of the few counters that calculates Allan deviation, not just mean and standard deviation. With this, you do not get a fancy log-log ADEV(tau) plot but the number reported will be valid for the gate time (tau) that you have selected. The advantage of this is that you don't need any GPIB adapter or computer connection or CNT-91-compatible PC software. So maybe try tau 1, 5, 10, 20, 50 and write down what it reports.

The other question is the validity of the ADEV calculation when the counter's "smart" features are turned on. The good news is that this should not impact your goal to compare multiple oscillators against each other; since that is essentially a pair-wise, relative comparison. But it may affect the absolute accuracy of the calculations themselves. Its a subtle point, which you are welcome to ignore for now.

Yes, it is possible that some of the OCXO that you are testing are near or even better than the short-term measurement ability of the CNT-91. In this case either you can be very happy to have such good OCXO, or you can develop a new method to compare oscillators at high resolution at short times.

Time-nuts -- Have a look at Karen's intro page (http://www.qsl.net/ra3apw/ra3apw/). As I understand it, his project is to use the high frequency output of a ublox NEO-7M to discipline a MV89 with a James Miller-style analog PLL. Some links:

http://www.jrmiller.demon.co.uk/projects/ministd/frqstd0.htm
https://sites.google.com/site/g4zfqradio/u-blox_neo-6-7
http://www.ra3apw.ru/proekty/ublox-neo-7m/
(perhaps someone can post an English translation for us)

Thanks,
/tvb

----- Original Message ----- 
From: "Karen Tadevosyan" <ra3apw at mail.ru>
To: "'Tom Van Baak'" <tvb at leapsecond.com>
Sent: Monday, November 03, 2014 4:24 AM
Subject: RE: 10MHz Rubidium reference source for frequencycounter


Tom,

Yes, it is my fault - my set up for TI mode was not correct. 
I need some time to read/understand more how to realize it in my condition as I never used this mode (TI) before.
BTW, it makes me read the CNT-91 operating manual ... :-)  really clever device ! 

According your advice I went back to frequency mode and set gate / sample time to 1 second for CNT-91 and TimeLab.
Then I switched on "Smart Frequency" and "Interpolator Calibration" option on CNT-91

1. Setting -> Misc -> Smart Meas -> Smart Frequency -> ON
Smart Frequency (valid only if the selected measurement function is Frequency or Period Average)
By means of continuous timestamping and regression analysis, the resolution is increased for measuring times between 0.2 s and 100 s.

2. Setting -> Misc -> Smart Meas -> Interp Calib ON
Interpolatator Calibration - By switching off the interpolator calibration, you can increase the measurement speed at the expense of accuracy.

Please see the result attached with and without these options. It seems to me it's a measurement limit of my set up.

Many thanks for your help.

Karen
 
> -----Original Message-----
> From: Tom Van Baak [mailto:tvb at LeapSecond.com]
> Sent: Monday, November 03, 2014 12:39 AM
> To: Karen Tadevosyan
> Subject: Re: 10MHz Rubidium reference source for frequencycounter
> 
> Karen,
> 
> Ah, that's not quite what John Miles meant by using TI mode instead of
> Frequency mode. What you have accidentally done is measure the risetime,
> not the phase or frequency or ADEV...
> 
> Later today I will power up my CNT-91 and find the best settings for you.
> 
> Meanwhile, could you go back to frequency mode (both the counter, and
> TimeLab) and set the gate time / sample rate (both the counter, and
> TimeLab) to 1 second? That will give the CNT-91 more data to work with.
> Previously your gate time was very short (0.04 second); maybe too short.
> 
> Thanks,
> /tvb
> 
> 
> ----- Original Message -----
> From: "Karen Tadevosyan" <ra3apw at mail.ru>
> To: "'Tom Van Baak'" <tvb at leapsecond.com>
> Sent: Sunday, November 02, 2014 12:57 PM
> Subject: RE: 10MHz Rubidium reference source for frequencycounter
> 
> 
> Tom,
> 
> Thank you for the reply.
> When I switched from Frequency to TI I changed mode in TimeLab and in
> CNT-91.
> I didn't use a second input.
> My set up is:
> Time Interval: A to A
> Rise Time: A
> Fall Time: A
> TIE: A
> 
> A - the first port of CNT-91.
> 
> Thanks,
> Karen
> 
> > -----Original Message-----
> > From: Tom Van Baak [mailto:tvb at LeapSecond.com]
> > Sent: Sunday, November 02, 2014 11:47 PM
> > To: Karen Tadevosyan
> > Subject: Re: 10MHz Rubidium reference source for frequencycounter
> >
> > > Mode has been switch from Frequency to TI - result is much better now.
> > > It was the main improvement in my measurement. Thanks a lot!
> > > Picture of ADEV OCXO result -
> >
> > Karen,
> >
> > When you switched from Frequency to TI, was that just a mode change in the
> > TimeLab menu, or did you also change the mode of the CNT-91?
> >
> > In frequency mode there only one CNT-91 input, but in TI mode there must
> > be two inputs (start, stop). So the question is, what frequency source did you
> > use for the second input?
> >
> > Thanks,
> > /tvb
> 

-------------- next part --------------
A non-text attachment was scrubbed...
Name: Morion OCXO ADEV Freq mode SmartFreq Sample_1s InterpCalib_On.jpg
Type: image/jpeg
Size: 351428 bytes
Desc: not available
URL: <http://www.febo.com/pipermail/time-nuts/attachments/20141103/d735021e/attachment-0001.jpg>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: CNT91-error.gif
Type: image/gif
Size: 28655 bytes
Desc: not available
URL: <http://www.febo.com/pipermail/time-nuts/attachments/20141103/d735021e/attachment-0001.gif>


More information about the time-nuts mailing list