[time-nuts] yet another GPSDO design, or so
lists at rtty.us
Tue Jun 29 16:54:52 UTC 2010
Some of the TI (Burr Brown) 16 bit parts are 1/4 lsb DNL on > 98% of the
transitions. Most of the time you have a "coarse" DAC that's at 18 bits.
Some of the errors are predictable and you can take them out with a simple
training process. You won't easily get 24 bits, but 20 is very achievable.
20 bits is a nice round million to one factor. A 1 ppm EFC becomes a 1 ppb
step if everything is perfectly linear. Even with a 4:1 slope ratio it's
still quite small. With a more restricted EFC of 2.5x10^-8 you are at
1x10^-13 with the 4:1 slope ratio.
From: time-nuts-bounces at febo.com [mailto:time-nuts-bounces at febo.com] On
Behalf Of Attila Kinali
Sent: Tuesday, June 29, 2010 12:42 PM
To: Discussion of precise time and frequency measurement
Subject: Re: [time-nuts] yet another GPSDO design, or so
On Tue, 29 Jun 2010 09:17:39 -0700 (PDT)
Stanley Reynolds <stanley_reynolds at yahoo.com> wrote:
> If we lower the size of each step to over lap more would this
> lower the error ? Software would adjust both converters at the
> cross over point so neither would change it's full range at this point.
> Two 12 bit converters would form one 18 or 20 bit converter.
Yes, you can do that. With good D/A converters, the non-linearity
is within 1-2LSB, so if a full range of the fine D/A is more than,
lets say 4LSB of the coarse D/A you can minimize the step in control
voltage, when you switch from one fine D/A range to the next. But
it will be always there, you cannot get completely rid of it.
There are two things that make this step problematic:
1) If it happens automatically, it will result in problems
when doing long term measurements using that GPSDO as reference.
When doing the switch manually triggered (ie the GPSDO says that
it needs to adjust the coarse D/A and does so only after the
user presses a button), then unattended operation over long periods
is not possible anymore.
2) Each adjustment of the coarse D/A is a non-linear operation. It is
basically a step function excitation of the control loop. With such
a non-linear element within the control loop, the software has to
take either precautions during the period of the non linearity or
the control loop has to be able to cope with such steps, without
starting to oscillate.
> I guess taken to the extreme we could interweave any number of
> converters so hopefully the errors would average out :-) 64 times
> 12 bit converters making a 20 bit converter each with an equal
> contribution at each step.
Actually, the less components you have, the better. Not only do
more components add more non-linearity (it is very difficult to
average multiple components if they arent R's and C's), the
whole circuitry involved produces also noise.
If you want to walk fast, walk alone.
If you want to walk far, walk together.
-- African proverb
time-nuts mailing list -- time-nuts at febo.com
To unsubscribe, go to
and follow the instructions there.
More information about the time-nuts