As part of our GPSDO work Richard Mc Corkle and I tested multiple DAC's
using hardware I developed and Richard wrote the code. We limited to
affordable  and solderable. The LTC1655 was the clear winner because of linearity and
temperature, see attached. We tested dithering 20 bits and stacking two
for  coarse fine, storing the test data from using the LTC2400 ADC. Limited
our  choices to dither and bare use. We did this 5 years ago. Contact with
Richard  has sadly stopped. I am very concerned, however we continue to use his
contributions on several projects with very good results. We use the LTC
1655 on  Rb's because its resolution and range as is, is perfect.
Five years later I know no better alternative
Bert Kehren

In a message dated 7/16/2017 5:09:08 A.M. Eastern Daylight Time,
attila@kinali.ch writes:

On Sat,  8 Jul 2017 11:42:44 -0700
Tim Lister listertim@gmail.com  wrote:

Much simplified, it boils down to it being easier to measure  voltage
differences by averaging than keeping a voltage  constant.

E.g. in those >20bit ADC's you will usually find a  delta-sigma ADC,
usually 3rd to 5th order with a 1.5 to 5 bit ADC/DAC  inside. The ADC
and DAC can be laser trimmed to be in the order of 0.1% of  their
ideal values. With a few additional tricks you can get the most  of
the remaining non-linearity out. These tricks also help to  remove
errors due to DC-offsets in the signal path. But the  biggest
improvement comes from averaging over many "samples" to get  the
white noise out. If you look at the usual sample rates at  which
those ADC reach their "full" performance, it is around
1-30  (output) samples per second.

On the other hand, on a DAC you need to  keep the output voltage
stable. You can do the same delta-sigma approach as  with the ADC
with much the same result, but you have one big problem:
it  is not easy to build an analog low pass filter that has a corner
frequency  down at 10Hz. This means, you have to work at a much higher
frequency to  have a low pass filter that can be realized (let's say 1kHz
if you are  building a discrete filter, higher if it's integrated).
But that means that  you have several orders of magnitude more (white)
noise.

Additionally,  a lot of people expect to do a couple of 1000 samples
per second at least,  to have a usefull DAC. But that contradicts the
need to have a narrow band  low pass filter to get the noise out.

Attila Kinali

It is upon moral qualities that a  society is ultimately founded. All
the prosperity and technological  sophistication in the world is of no
use without that  foundation.
-- Miss Matheson, The Diamond Age, Neil  Stephenson

time-nuts  mailing list -- time-nuts@febo.com
To unsubscribe, go to
https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts
and follow the  instructions there.

Bert wrote:

Well, to name just the first one that comes to mind, how about the
LTC1650?  Like the 1655, it is available in SO and DIP packages.  Its
differential nonlinearity is >2x better than the 1655, it settles 5x
faster (4uS vs. 20uS to 1 LSB) and is nearly 100x (40dB) quieter
(30nV/sqrtHz vs. 280).  The 1650 has substantially lower glitch energy,
as well (1.8nV-S for the 1650, 12nV-S for the 1655).

Not all of the better specs are required for steering an oscillator at
GPSDO rates, but others are a distinct advantage.

I have not surveyed the field to see what other "SO or easier to solder"
DACs are available with better performance than the 1655, but I'm sure
there are others.

This is to take nothing away from the 1655 -- I have accounted for
thousands of them in my own designs, and think very highly of it.

Best regards,

Charles

Hi

One gotcha with any ADC or DAC is going to be the reference. There, you are in the same
“get what you pay for” dilemma. Stable and noisy, can do. Quiet and not very stable, can do.
Both stable and quiet, not so easy if you want it cheap.

Noise can also be the sigma delta ADC’s weak point. Even at slow rates, some of them need
a lot of averages to quiet down.

Does this or that design need this or that level of stability or noise? That depends a lot on the
approach used. In a GPSDO, cutting down on the EFC range is a great way to “cost reduce” the
rest of the circuit in terms of noise and stability.  Well made modern OCXO’s don’t drift a lot ….

Bob

On 7/16/2017 1:51 PM, Bob kb8tq wrote:

The reference initially used in the E1938A turned out to be too
noisy/unstable.  It was non trivial to find an upgrade.  The
HP Smart Clocks of 20 years ago were limited in their performance
by the reference used.

Has there been much improvement in references in the intervening
20 years?

Rick N6RK

Hi

They still don’t seem to have the hysteresis problem licked. Yes, you can do an oversized reference
and take care of the issue. More or less that’s what you would have done 20 years ago.

Bob

What about josephson standards?  After all, this is "Time Nuts" and we are
allowed to propose silly-complex solutions to simple problems if it
improves performance even a little.

But seriously I thought the issue of making a perfect voltage standard was
solved because the Volt is defined to be whatever the Josephson array
produces.    Yes expensive because to runs at nearly absolute zero.

On Sun, Jul 16, 2017 at 4:10 PM, Bob kb8tq kb8tq@n1k.org wrote:

--

Chris Albertson
Redondo Beach, California

HI

This is a limitation on an OCXO based GPSDO. That’s really the bottom line here.
It’s a limitation in an OCXO based part, but not in one based on an Rb or a Cs. If
the added component costs far more than a Cs, it’s not an answer.

Bob

This implies that in a Rb or Cs there is not a voltage reference source?

On Mon, Jul 17, 2017 at 3:49 AM, Bob kb8tq kb8tq@n1k.org wrote:

On 7/17/2017 10:54 AM, Azelio Boriani wrote:

Yes, that's right, there is no voltage reference with a material
effect on stability or accuracy.

Rick N6RK

On Mon, 17 Jul 2017 12:07:29 -0700
"Richard (Rick) Karlquist" richard@karlquist.com wrote:

How about the temperature and C-field control? Aren't those based
on keeping the output of a sensor stable compared to a voltage reference?

		Attila Kinali

--
You know, the very powerful and the very stupid have one thing in common.
They don't alters their views to fit the facts, they alter the facts to
fit the views, which can be uncomfortable if you happen to be one of the
facts that needs altering.  -- The Doctor