[time-nuts] [Fwd: Re: What is the best counter for a Time Nuts?]
Mike Monett
XDE-L2G3 at myamail.com
Fri Oct 17 00:58:07 UTC 2008
>> hasweb at has.org.nz wrote:
>> There is a 3575A manual for downloading at the Agilent ftp link.
>> If you read it you will find that the specified error with 10MHz
>> inputs is much larger: 10-20 degrees or so.
The calculation I gave is an upper bound on the stability. It relies
on the resolution. The accuracy is basically irrelevant for this
measurement. Once the rate of change exceeds about 1/10 degree per
second, it is impractical to record the data manually, and you have
to go to a different method.
[...]
>> The basic principle is sound. It should be very easy to make a
>> simple PWM circuit to measure phase accurately at 10MHz. You
>> mentioned the same thing earlier. Why not post the schematic so
>> we can take a look?
>> Mike Monett
>Mike
> I did some time ago, for some examples (other implementations are
> also possible) see:
> http://www.ko4bb.com/~bruce/LinearPhaseComparators.html
Those are all variations of an XOR. This is perhaps the worst
possible phase detector to use for precision applications. It has
very high ripple at lock, which generates unwanted sidebands in the
PLL oscillator output.
It requires signals 90 degrees out of phase, so you have to add
extra logic to lock to signals that are in phase. This adds jitter.
The output is not referenced to ground, but to some floating level.
This means very small offsets and drift will change the lock point.
It is also unsuitable for measuring phase angle. The slope of the
transfer curve for the basic XOR reverses at +/- 90 degrees. So you
can't use it to measure phase angle.
The AD9901 hase all the above problems but it rails at +/- 180
degrees instead. It cannot be used to measure phase angle either.
The standard phase/frequency detector (PFD) is better for precision
PLL's since the output signal is zero at lock. This means minimum
ripple to generate sidebands. It can be reset to a known state so it
can be started in phase with an incoming signal. This greatly
reduces the lock time.
It wraps at 360 degrees, so it is also better suited for measuring
phase angle.
These are only some of the issues in analyzing a phase detector for
a precision pll. Here is a quickie LTspice analysis showing the
phase detector transfer curves for the three detectors. I posted it
so people would not get their mailbox clogged with attachments.
http://silversol.net63.net/spice/phasedet/phasedet.htm
[...]
> The quickest and easiest is to use the AD9901 phase detector as it
> only has a few ns of non linearity at the end of the range. For
> higher resolution just add a sigma delta ADC.
The AD9901 is a very bad design. Probably the poorest I have ever
seen. It is unsuitable for measuring phase angle. Adding a high
resolution ADC will do nothing to help.
I notice there is very little in the way of analysis on your site.
This means it is impossible to distinguish one circuit from another.
What you really need to do is go through and put everything in
LTspice so people can look at the circuit behavior in detail.
Then you need to do a tolerance and sensitivity analysis to show how
stable the circuits are over long periods.
Then you need to do a noise analysis to show why some circuits are
better than others.
This would make your information much more valuable and interesting.
One of my big problems right now is trying to understand why the XOR
works so well in one of the GPSDO's in Tom's Allan variance pages. I
can't find it at the moment, but I need to take a very close look at
this design and see what is going on.
[...]
>Bruce
Best Regards,
Mike Monett
More information about the time-nuts
mailing list