[time-nuts] V standards

[Bruce Griffiths]

Mike
>   >Mike
>
>   > In testing  high resolution DACs (20+ bits) used in  GPSDOs  etc a
>   > high resolution DVM (or equivalent) with low noise and  good short
>   > term stability is useful if not essential.
>
>   > Monotonicity and  perhaps linearity together with good  short term
>   > stability are generally more important than absolute accuracy.
>
>   > Higher resolution is usually accompanied with higher accuracy.
>
>   > DVMs like the 3457A. 3456A etc dont have sufficiently low noise or
>   > good enough  short  term stability for testing  DACs  with  24 bit
>   > resolution.
>
>   >Bruce
>
>   I wonder if 20+ bits is even realistic for a frequency reference.
>
>   That would  give a LSB of 1/2^20=9.53e-7, or 1ppm, and  none  of the
>   voltage references  discussed will maintain this accuracy  over long
>   periods.
>
>   
24 bit resolution isnt required when using a 10811A class OCXO with a
total EFC adjustment range of about 1E-7.
Even 20 bits is a bit more resolution than actually required in this case.
However there are low noise OCXOs with EFC adjustment ranges of 1E-6 or
more.
In a GPSDO long term stability of the EFC DAC offset and gain isnt
critical except when the GPSDO is in holdover.
>   And what would a 24-bit DAC be used for? From  previous discussions,
>   high accuracy sources, like H-Masers are not adjusted. And  it would
>   seem silly to put such a high resolution DAC on a OCXO.  That leaves
>   Cesium, which  I understand are used in GPS satellites  and  do need
>   adjustment, but  I  don't   have   any   information  on  the tuning
>   sensitivity to figure the effect 1 LSB would have on the frequency.
>
>   I haven't had much luck finding a true 24-bit DAC. There  are plenty
>   of stereo dacs, but they can have gain drifts of 100ppm/C,  which is
>   useless for a reference.
>
>   
No you have to build your own.
The trick is doing it without requiring impossibly accurate and stable
resistors or resistor ratios etc.
Since the DAC update rate is relatively slow in such applications
(GPSDO) indirect techniques that are inherently monotonic can be used.
The FS725 rubidium standard has an internal 22 bit DAC.
Hydrogen masers usually include similarly high resolution DACs.
Such DACs are usually constructed by combining the outputs of 2 lower
resolution DACs with some overlap.
The drawback being the relatively large differential nonlinearity when
the MSDAC output changes.
This increases the settling time of the discipling loop in the vicinity
of such changes.
However such MSDAC output changes don't occur very often.
Another application for such high resolution DACs is in accelerator beam
steering.
In this case frequent calibration is used together with suitable
software to avoid such large differential nonlinearities.

>   With a  5V reference, a 24-bit DAC would give a LSB  of 5e9/2^24=298
>   nanovolts. If  I had to test one, and didn't have a  3458A,  I could
>   use a 3456A. It has a resolution of 100nV on the 100mV range,  so it
>   could verify the bottom portion of the DAC from zero to  100mV. Once
>   the lower  8 bits are confirmed good, the rest of the  DAC  could be
>   checked by  exercising each high-order bit singly,  then  in various
>   combinations with the other bits.
>
>   
I had such techniques in mind for checking the monotonicity.
Could also check against a KVD (e.g. Fluke 720A).
However the shipping charges are rather high.
It is possible to achieve a readout resolution of 100nV using a 3457A on
the 3V range.
However this is only available via the GPIB.
Similarly even a 34401A can achieve a resolution of 100nV on the 10V
range but only via either the GPIB or its serial interface.
However the 300nV LSB is uncomfortably close to the DVM noise level and
DVM drift isn't insignificant.
For this application the DAC transfer function only has to be monotonic
so testing requirements are somewhat relaxed.
>   Another method  would  be  to use two 24-bit DACs  and  a  AD8571 to
>   measure the  difference between them. The AD8571 could be  set  to a
>   gain of  100,  so  298nV becomes 29.8uV  which  is  well  within the
>   capability of a 3456A.
>
>   
The LTC1151 is perhaps a better choice in some respects in that it  can
use -15V and +15V supplies allowing a greater input voltage range
simplifying the input overvoltage protection.
Although the AD8571 power supply could be bootstrapped to achieve a much
larger input range.
Even lower drift and noise is possible if one builds one's own chopper
stabilised preamp.
>   The test  would  be  to  set  both  DACs  to  zero  and  measure the
>   difference in output voltage. It should be close to zero.
>
>   Next, set  the LSB of the reference DAC to 1. The 3456A  should read
>   close to 29.8uV.
>
>   Then set  the LSB of the test DAC to 1. The 3456A should  read close
>   to zero.
>
>   Follow this  procedure  with   each   bit   in  turn  to  verify the
>   functionality, then  test various combinations to check  for  two or
>   more bits  that  are  stuck  together.  The  LSB  could  be  used in
>   conjunction with  the  bit being tested  to  prevent  saturating the
>   AD8571.
>
>   The above tests are not as good as a dedicated test for  a precision
>   DAC, but might serve in lieu of spending $4k to $7k for a 3568A.
>
>   Regards,
>
>   Mike Monett
>
>   
Bruce