[time-nuts] Re: Assessing performance of a frequency counter

[Bob kb8tq]

Hi

To add a few more “gotcha’s” to the list:

1) A fast slew rate square wave will usually do better than a sine wave with the same peak amplitude. 
In some cases the difference will be several orders of magnitude. There is a YouTube video from 
Keysight showing this on the 532xx counters. 

2) There is at least one counter that delivers absolutely amazing numbers when the input is hooked to
its internal reference. There is no other way to get that sort of performance. Does this apply to anything 
other than a SR620? Who knows ….

3) Any OCXO “improves” as it warms up. If you are looking at an internal reference OCXO, leave it on
power for a couple weeks before running the test. Are weeks always required? It depends a lot on the
OCXO design and how long it’s been power off.

4) The confidence interval for ADEV improves with the number of samples. Anything much under 100
samples is going to hop around a bit if you look at today vs a bit later today. 

Yes, this could go on and on …..

Bob

> On Nov 8, 2021, at 12:39 PM, John Ackermann N8UR <jra at febo.com> wrote:
> 
> Hi Erik --
> 
> Unfortunately, your attachment did not make it through the list.  That sometimes happens when you put images "in line" or as HTML content. Best thing is to send them as a "real" attachment.
> 
> Your question breaks into two pieces: (1) what's the resolution or noise floor of the counter, and (2) how good is its internal oscillator?  I'm not sure which piece you are most interested in.
> 
> 1.  To characterize the counter itself, the easiest test is to feed the same signal into both the counter timebase and its input (e.g., if there's a "10 MHz Out" on the rear panel, connect that to the input). That will show you what the counter is capable of -- what's its resolution, how stable is the gating, etc.
> 
> 2.  To characterize the internal oscillator, there are two things to consider:
> 
> (a) is your reference oscillator stability/accuracy better than the counter's over the ADEV "tau" range you're considering?  Ideally you'd like to *know* that the reference is at least 10x more stable than the device under test. Otherwise you may find yourself measuring the reference rather then the DUT.
> 
> (b) what's the resolution of your measurement system (see point 1)?  For short measurement periods, the counter is likely to be the limiting factor because it has a fixed resolution and internal jitter that impact short measurements more than long ones.  If you think about it in the time domain, a "perfect" time interval counter with 2ns resolution will have a measurement floor of 2e-9 at 1 second (2ns / 1 second). At 10 seconds, it's 2e-10 (2ns / 10 seconds), and this slope continues until you hit some ultimate noise floor.   In the frequency domain, the number of digits per second implies the same thing.  Most test-bench counters probably have a resolution of around 1e-8 to 1e-9 in 1 second.  Some of the best are in the 11s, but to get better than that at 1 second averaging you are getting into esoteric gear.
> 
> Based on all that, ADEV measurements of two good oscillators are probably going to be counter noise limited at tau (measurement interval) of less than 100 seconds.  Your 0.1 and 0.05 seconds measurements are almost certainly limited by the resolution of the counter.
> 
> I've ignored some subtleties and gotcha's in the above, but hopefully it gives you an idea of how to take and interpret measurements.
> 
> John
> ----
> 
> 
> On 11/8/21 10:53 AM, Erik Kaashoek wrote:
>> Hi all time-nuts.
>> Hopefully one of you can provide some advice on how to proceed.
>> The ambition is to assess the short term (less than 1 hour) performance of a frequency counter.
>> As I understood this can be done by plotting the Allan Deviation of a repeated measurement of a very stable source.
>> Below plot shows the Allan deviation of a 1 hour measurement of the 10MHz output of an OCXO versus the internal reference of the counter with a 1 and 0.1 second gate time.
>> To be able to better see the gate time impact, short measurements with a gate time of 0.05s and 5s have been added.
>> Temperature variations during the measurement have not been recorded but there where some more variations during the 1s gate time measurement.
>> Are there any conclusions to draw from this plot?
>> Or should I do a different measurement, or use different representation?
>> A pointer to a web page or document describing how to do this type of assessment would be most welcome.
>> Erik.
>> _______________________________________________
>> time-nuts mailing list -- time-nuts at lists.febo.com -- To unsubscribe send an email to time-nuts-leave at lists.febo.com
>> To unsubscribe, go to and follow the instructions there.
> _______________________________________________
> time-nuts mailing list -- time-nuts at lists.febo.com -- To unsubscribe send an email to time-nuts-leave at lists.febo.com
> To unsubscribe, go to and follow the instructions there.