[time-nuts] TimeLab and the Adev plot

Magnus Danielson magnus at rubidium.dyndns.org
Wed Feb 5 13:45:14 EST 2014


On 04/02/14 11:10, Azelio Boriani wrote:
>> In the second sentence, I make the point that when you take two sample values, at various taus, you really do not
>> average them but rather make their time stability contribution (trigger jitter and resolution) less important relative to
>> the tau between them. This is by itself not an averaging effect.
> Agreed that it is not an averaging effect, then taking two samples at
> various tau is, for example, taking samples at 2 seconds instead of 1
> second or using two samples for the actual tau under consideration?

Well, if you want to analyse the tau=2 s ADEV case, you need to use 3 
samples, with 2 s between them, which is exactly what the standard 
formulas use. If you have a systematic precision noise, that will fade 
with 1/tau and overshadow any actual DUT noise.

If you have a 1 s tau sample series, you then only use every n sample 
for your n*tau0 analysis. Remember that you should at least do 
overlapping accumulation for better degrees of freedom and hence tighter 
confidence interval.

The non-overlapping ADEV is only for historical reference.

Cheers,
Magnus


More information about the time-nuts mailing list