[time-nuts] Averaging effects

[ws at Yahoo]

Others said:
> There is a downside to this approach which should be understood, it will 
> also averaging out the white noise of the DUTs.
>The time interval counter method severely undersamples the phase noise 
>spectrum leading to aliasing effects.
> The measured ADEV depends on the associated filter bandwidth
> Filtering is tricky since you will both reduce the measurement systems 
> noise as well as the the DUTs noise.
> The aliasing effect is definitively there.
> Question is how to remove the system noise from the DUT noise

One way to avoid those trade-offs is to do the frequency difference 
averaging with a TPLL (Tight Phase Lock Loop) using a "proper" integration 
of it's output over the tau time period.
example at:     http://www.thegleam.com/ke5fx/tpll.htm

ws

******************
Magnus Danielson magnus at rubidium.dyndns.org

Hej Bruce,

On 12/27/2010 08:13 PM, Bruce Griffiths wrote:
> Tom Van Baak wrote:
>>> There is a downside to this approach which should be understood, it
>>> will also averaging out the white noise of the DUTs.
>>
>> Correct. A similar white noise effect can happen if you average
>> the raw data itself. See the plot at the bottom of:
>> http://www.leapsecond.com/pages/adev-avg/
>>
> Its a little more complicated than that.
> The measured ADEV depends on the associated filter bandwidth (typically
> for 1Hz sampling one uses a low pass (for the phase fluctuations) filter
> bandwidth of 0.5Hz or less).
> When one uses a time interval counter the counter input system noise
> bandwidth may be as high as 100MHz (5370A/B) or 500MHz or more (DTS2070)
> whereas the crystal oscillator buffer amp (principal source of OCXO
> white phase noise floor) may have a somewhat lower bandwidth. The time
> interval counter method severely undersamples the phase noise spectrum
> leading to aliasing effects.
> Averaging of this type creates a low pass filter that will reduce the
> system noise to a large extent whilst not greatly affecting the
> measurement as the equivalent filter bandwidth will still be much larger
> than 0.5Hz and the equivalent filter response is far from ideal.

True, but it is tricky since you will both reduce the measurement
systems noise as well as the the DUTs noise (which is what you intend to
measure).

The aliasing effect is definitively there.

Question is how to remove the system noise from the DUT noise, and I
know of only approach which really avoids it is cross-correlation, but
otherwise it is only various measures to remove and filter out the
signal from noise before it is folded in, i.e. conservative design measure.

Anyway, I wanted to play around with averaging to see how the filtering
effect behaves.

Cheers,
Magnus