[time-nuts] Best counter setting for ADEV?

[Tom Van Baak]

> I was wondering why the resolution in TI mode is so much limited, since 
> I never had any problems measuring 5 or 10 MHz frequencies with up to 12 
> figures on that counter.

Limited? The question is not so much counting the figures but asking if the how much the figures count.

> As a quic 'n dirty test, I tried to measure ADEV with my 53131A in 
> frequency mode using a gate time of 5.1 s for max resolution, and found 
> the noise floor had shifted down to 3...4E-12 at 10 sec, and going 
> further down to 5E-13 at 1000 sec from where it was equal to the TI mode 
> noise floor.

In frequency mode, the hp 53131A/53132A counters use a clever internal averaging mode. It's mentioned in the manual. The readings it reports are a highly oversampled mean frequency. This is nothing to complain about, really. Typically, with a frequency counter you are only interested in a smoothed averaged result. And these hp and other modern frequency counters do this quite well.

> So, what is the best method to use for ADEV? What instruments and setups 
> are you using, and what works best for you? How to get the max out of 
> the given instruments?

An ADEV measurement is rather different from average frequency measurement. ADEV tries to tell you the variance, the deviations from the average frequency, as a function of tau. It's purpose is to measure the noise, not make the noise go away. So the more a frequency counter averages (in order to give you a smooth average frequency) the more it is actually suppressing the very variations that you are trying to measure. In order to gain "precision" these frequency counters are removing part of the "variance"; ADEV is completely ignores precision and is only concerned with the variance.

This is why one-shot phase meters or time interval counters give a more pure view of oscillator performance. Yes, they tend to report a little to a lot more noise -- but that's because there *is* more noise. An oversampling frequency counter takes the liberty to average away as much noise as it possibly can, to suppress the short-term variations, and present just a single value as the one true answer.

/tvb