Has anyone here actually built the 20-bit DAC described in the app note?  I've been thinking about it and actually started laying out a PCB.  I am hesitant to pull the trigger, given that the DAC ICs that I was planning on using (LTC1595 in SO package) are $48 each and the design requires two of them.

Randy.

I think one reason it is expensive is because the simple "digital comparator" requires the composite DAC (the combination of the two 16 bits DACs considered as one device) to be monotonic. This is done by using two pretty good 16 bit DACs and having a significant overlap range to smooth out the differences.

If you can deal with longer settling time (milliseconds instead of microseconds), you can use a microcontroller to perform the digital comparison. Your code can correct for non-linearity in the composite DAC (via calibration, and/or real time closed loop correction) and reduce cost significantly. You may be able to use two 12 bit DACs for instance, at the expense of more software to develop and a slower response time, but much lower cost in production (you can buy a dual 12 bit DAC for a few $). You may need a sample and hold to reduce transients on the output while the circuit does its job, if that could be a problem with your application, or use a large (noise-free) capacitor :)

I built a breadboard for something similar to that a while ago using the 24 bit ADC in a Texas Instrument MSC1210 microcontroller and two available 8 bit DACs, to give me about 12 bits of linearity, as a proof of concept.TI advertises the MSC1210 as a smart ADC, because it integrates a
pretty good 24 bit ADC with an 8051 microcontroller and 32k of flash... The idea was to expand the concept later to 20 bits using two 12 bit DACs to drive the OCXO of a GPS disciplined oscillator. I never finished the work, but conceptually it was simple because a GPSDO does not need speed.

For metrology, where response time is not typically as critical as in industrial applications, or for something like a GPSDO, that may be a valid alternative.

Didier

----- Original Message ----

you might consider using the vref board out of the 3458a ala
http://maxmcarter.com/vref/ (mentioned here previously).
It already has the LTZ1000 and nice, low tempco parts on it, all it
needs is a power supply. I bought one and have been running it on  a
supply built from stuff lying around and has been pretty stable, at
least according to my 3457a. i've been meaning to build a nicer supply
for it, but have held off until i get a 3458a or a keithly 2002 or
something that i can actually see if there's any improvement with
various supply configurations.

-eric

On Thu, Nov 19, 2009 at 7:47 AM,  gsteinba52@aol.com wrote:

--
--Eric

Eric Garner

At 10:47 AM 11/19/2009, gsteinba52@aol.com wrote...

Schematic's fine for what it is, but they call for $40/ea resistors,
probably special order from a distributor. (VISHAY VHP-100 0.1%, ~0 ppm
tempco, 5 of 'em) which I would like to avoid the expense of...but you
can get Xicon 10ppm (288-0805 series) ones for <$1 from stock.

Beyond the basic app note is getting an accurate voltage (10 V?)
referenced to the low accuracy LTZ1000 (it's extremely stable, but not
very accurate). Again, the app note would have you use a custom
resistor array (and even that won't get you accuracy, unless special
ordered, custom matched to each LTZ1000 - $$$$).

Mike S wrote:

One solution if one requires a 10V reference is to use a DAC for fine
adjustment of the output of a gain stage.
Use the 7V output from the basic LTZ1000 circuit as the DAC reference.
The only problem being that the required adjustment range is about 0.7VV
to accomodate the 7.0 -7.5V output voltage range for the LTZ1000.
To achieve 1ppm adjustment resolution a DAC with 17bit or greater
resolution is required.

Since the DAC output need not respond rapidly to digital input changes
one can use something like an indirect PWM DAC that doesnt require any
precision components.
Such DACs can easily achieve >20 bit resolution and monotonicity
together with low drift.
However this requires using a microprocessor or equivalent with a PWM
output.
If one does add a microprocessor then one can also add a thermal
annealing algorithm to compensate the effect of powering down the
LTZ1000 after calibration.

Bruce

Vishay precision resistors are very, very good in my experience. You might
try Vishay directly as a source, even for small quantities. I used to get
them that way.

-John

==============

Bruce Griffiths wrote:

Another option if one prefers not to use a microprocessor is to use
coarse and fine multiplying DACs where the range of the fine DAC is
4LSBs of the coarse DAC.
During the lifetime of the reference only the input to the fine DAC need
be adjusted.
If parallel input DACs like the AD7945 or equivalent are used then the
output can be set by DIP switches or jumpers.

Bruce

Randy,

microcontrollers that feature pwm make excellent DACs if you put some clever
electronics around them. Have a look at

http://www.electronicsweekly.com/Articles/2008/10/30/44817/dc-accurate-32-bi
t-dac-achieves-32-bit-resolution.htm

for an example. I have build this circuit (with some resistor variations
because I needed to combine two 12 bit DACs into one 24 bit DAC) and have
found it to work like charme. I made stability measurements on the output
signal with my HP3457 (in high resolution mode over GPIB) which showed that
the voltage references low frequency noise (a LT1021 in my case) was to be
measured on the output down-divided by the pwm's on/off factor, i.e. the
output's signal noise stays the same RELATIVE to the output amplitude and is
basically the same as that of the reference itself.

Best regards
Ulrich