[time-nuts] time-nuts Digest, Vol 38, Issue 6
bruce.griffiths at xtra.co.nz
Thu Sep 6 18:05:50 EDT 2007
> Hi Bruce,
> I appreciate the time you put in providing feedback. Your points are well noted and some of them will be applied in any subsequent design.
> Regarding throwing away the sample while the OCXO is stabilizing, I don't agree with your statement. While it might not hurt to leave it in there, I don't see the harm done by removing it. The averaging period lasts for a few hours, so doing an average of "known stable" samples to me is the way to go. I don't see the information contained in that dropped sample as valuable, as it would be if the time constant were fast. You are probably looking at it from the angle of a continuous frequency/phase adjustment as an analogue loop and discrete PC would do it...
Theory says otherwise, you have increased the noise by around 40% by
discarding half the information.
The information you have discarded is actually useful in reducing the noise.
In a well designed system the EFC voltage should only change by a few
(1-2??) least significant bits between updates.
Digital control systems in general can only approach the performance of
an ideal analog control system.
However in some cases an analog control system may be impractical due to
the long time constants required etc.
Try consulting a text on digital control systems.
> Regarding DAC vs PWM, I agree with your statement. The decision I took was to try to decrease the number of components (and cost) while still achieve a decent performance. Of course you and I have a different definition of "decent", but that's OK. ;-)
> The PCB is a double-side one. Even though I would have preferred to pour a ground plane on an entire layer, once routing was over, there was not much gain left in doing so. Only small isolated islands of copper would have shown. A four-layer PCB would have been better but much more expensive.
> I wish I could make a good sigma-tau measurement, but I don't have easy access to a known good reference and a very good T.I. counter. My HP 5328A would not cut it, or would it? How would a HP 53131A do? We have an HP Phase noise test set (entire rack) at work, it might be possible to use it...
A 53131 should be capable of a resolution approaching 5E-10/T where T is
the gate time. So its resolution is marginal for reliably characterising
a the frequency stability of a source at the 1E-9 level for gate times
of less than about 10 seconds. The stability of its time base will
directly affect the results, ideally a quiet stable reference frequency
should be used. The frequency offset of such a source is unimportant (as
long as it is sufficiently stable) when measuring frequency stability.
Not having a stable reference (even another 10811) certainly limits your
options in measuring frequency stability.
If you use a mixer to generate the difference frequency between the
source under test and a stable source (a good 10811 should suffice for
averaging times of up to 100 sec or even more.) offset by a few Hz, then
the resolution of your 5328 is more than sufficient to resolve
fractional frequency fluctuations smaller than 1E-12. The mixer IF port
has to be suitably terminated amplified and low pass filtered before
connecting it to the 5328 input. Since you dont require state of the art
performance the zerocrossing signal processing circuitry need not be
particularly elaborate however some low noise slope amplification is
advisable. If you have a couple of stable frequency references then you
can use the dual mixer time difference method.
> Bert, VE2ZAZ
More information about the time-nuts