[time-nuts] Is it Hz or MHz ?

[Magnus Danielson]

On 06/02/2014 10:51 AM, Tom Van Baak wrote:
>> The value of 5E^-11 refers to the resolution that the precision can be
>> relied upon after taking into account all the factors that influence
>> it.  It means that there is an error that can be as much as +/-5 parts
>> per e^-11.
>
> No, it can be way more. A 5e-11 spec value is likely just RMS, or 1-sigma. Actual measurements will show significantly larger 2-sigma, 3-sigma, etc. value.
>
> Your "as much as" wording sounds more like a 6-sigma, or peak-to-peak spec.
>
> One must be very careful to match the measurement used to make the spec with how the device is intended to be used. For example, Magnus will likely tell us about MTIE, which is a perfect-storm, worst-case time error spec. It's very different from rough 1-sigma specs we usually talk about.

Indeed.

ADEV measurements give RMS-ish values, 1-sigma values, and is intended 
for random noiseforms. When listing the frequency stability of an 
oscillator you usually specify the 3-sigma value. However, doing that 
for ADEV measures the usual way would give you a false sense of what the 
confidence intervals is, since you need to look at chi-square 
distribution instead for similar limits.

Then, systematic deviations exists also while being locked in. Those 
will dominate as you slip out of lock. For systematics, the MTIE may be 
a better measure to consider.

Cheers,
Magnus