[time-nuts] Characterising frequency standards

[Magnus Danielson]

Steve Rooke wrote:
> Hi Mark,
>
> 2009/4/13 Mark Sims <holrum at hotmail.com>:
>   
>> Hello Steve,
>>
>> Try this...  take Tom's sample data set,  run the numbers.  Then,  using a good random number generator,  make another data set by randomly throwing out half (or more) of the samples (to simulate a non ZDT counter).  Run the numbers again.  See how they change.  This should give you a good idea of how using a standard counter would affect your adev numbers.
>>     
>
> But randomly throwing out data points would introduce ZDT.
It would introduce dead-time, it would not introduce zero dead-time 
(ZDT). Dropping every second sample of a phase/time-error series can 
maintain the zero dead-time property, but you loose the resolution for 
higher taus.
>  The whole
> point I was making was that the data set is well defined the "missing"
> data occurs every other sample therefore tau0 = 2 x (sample period of
> each sample).
>   
You can reduce the dataset size that way if you had phase/time-error 
samples and attain twice the tau, yes.

The downside is that you also reduce the degrees of freedom in the 
dataset and thus the statistical precission. With a large enought 
dataset this may not be much of an issue.

Cheers,
Magnus