-
Hey folks, I'm working on a precision voltage source platform. Not done yet, with much more left to do.
I'll probably make a handheld version of this, inspired by Ian's very excellent PDVS.
The architecture is modular, made of a DAC board, a reference board, analog supplies board and main board. The modularity is intentional and done for design flexibility as a top goal (reference, DAC, supplies can be replaced, improved or graded), and each board can be used separately in other designs by themselves as well.
Main board runs isolated digital supply, the digital electronics, battery charger, cal circuitry. A section of it will be dedicated to output buffering and output kelvin sensing. Main board, holds all other boards, on both sides, and is 100x70mm, smaller than most handheld instruments, a Fluke 57V is 150x70mm, Keysight U1272 is 170x80mm, Keysight U1733C is 140x70mm.
Reference board is based on LTC6655-5, it buffers and scales/inverts the reference to +10V and -10V. Reference board also contains an LDO supplied from the analog +15V supply.
The DAC board hosts some supply filtering, the AD5791 DAC IC, output buffer, and a small STM32 microcontroller which controls the DAC, keeps the cal constants, and communicates over UART.
The supplies board is based on ADP5070, with filters and with ADP7118 and ADP7182 as LDO post regulators.
Main board runs planned to run an STM32 micro, with an LTM8068 uModule isolated supply, for digital power.
The entire system runs on 3V, from which +15V and -15V is generated. The DAC is configured and can generate a 0 to +/-10V output. I plan to run the system from two independent 18650 cells, each independently powering the system. A charger will be built in to charge the cell that is not in use. Each cell should power the system for about 20hrs for a total of 40hrs. I do not plan to use an LM399 or LTZ1000 in a handheld version as simply they use too much power for handheld. However, the architecture can easily accommodate it with minimal or no changes to the mainboard and other modules.
If I ever experiment with an LTZ1000 version, it will be based on a 3458A reference module, with the buffering and scaling on the main board.
I plan to experiment at some point with an LT1533 regulator for lower noise. Of course the system can use any different supply scheme, like 2 x 9V batteries, and an 7815 regulator with an inverter to create the negative supply. Or external power from a dedicate low noise supply, like described above, but bigger, 2 x 7 x 18650 cells in an independent bank configuration. Such modification I'm considering for later not initial rev.
-
Very interesting,
Care to share your schematic? -
Very interesting,
Which one? Honestly I mostly follow the designs from Analog, from their two reference designs (the old one, and the new one, I find the old reference design more voltnutty, EVAL-5791). I need to clean them a little bit before sharing. :-)
Care to share your schematic? -
Using LTC6655 in LCC package would make more sense. LTZ1000 would make even more sense
-
Reference board is based on LTC6655-5, it buffers and scales/inverts the reference to +10V and -10V. Reference board also contains an LDO supplied from the analog +15V supply.
I do not plan to use an LM399 or LTZ1000 in a handheld version as simply they use too much power for handheld.
Why then a LTC6655? with > 5 mA current consumption and needing a extra voltage regulator to reduce the 15V to less than 12V?
By the way: the 1/f noise of the LTC6655 is dependant on supply voltage. At 10.2 V I measured 30% more 1/f noise than at 5.7V.
https://www.eevblog.com/forum/metrology/lm399-based-10-v-reference/msg2379519/#msg2379519
with best regards
Andreas
-
It’s not final, I’ll probably used a zener regulator for it in the end, as for LTC6655, do you have another recommendation?
-
Nice project niner_007, looking forward for a PDSV-like OSHW project
I am curious why one would combine AD5791 with LTC6655 and not LM399A or even LTZ1000?
At least the AD5781 is drop in replacement and would better fit LTC6655, but "only" 18bit resolution.
The new Ti DACx1001-Series - especially DAC11001A would be a lower noise alternative with slightly worse INL/TC/drift specs. -
The LTC6655 is not that bad. It takes quite some power, but also gets low noise.
If 10 V are wanted, a Ref102 (maybe 2 or 3 in parallel) may be an option too. It directly gives 10 V and has lower power, though higher noise.
The LTC6655 should be lower noise than a LM399 and still a bit lower power. A gain of 2 to get 10 V is also easier than some 1.45 needed for a 7 to 10 V step. The main drawback is the TC and long term drift.
I don't think one would really need 2 µCs. The calibration and similar part would not need much extra processing power. A single µC should be sufficient.
An idolated supply for the µC may cause more trouble than it help - good supply fitlering may be easier and lower noise, especially if the µC runs low power and maybe even goes to a deep sleep for DC output. -
Quote
The new Ti DACx1001-Series - especially DAC11001A would be a lower noise alternative with slightly worse INL/TC/drift specs.
Why would 7nV/rtHz matter against 7.5nV/rtHz for DAC11001A vs. AD5791B when the INL of the DAC11001A is +/-4LSBs versus +/-1LSB max. for the AD5791B?
The DAC11001A is only an 18-bit DAC in that regard...
The 5791A is at the same 1k-price level as the DAC11001A that has no option to upgrade if performance is key...
Cheers
Andreas -
Nice project niner_007, looking forward for a PDSV-like OSHW project
LTC6656 is a favorite
I am curious why one would combine AD5791 with LTC6655 and not LM399A or even LTZ1000?
At least the AD5781 is drop in replacement and would better fit LTC6655, but "only" 18bit resolution. Mostly battery life and warm up time were the concern for handheld operation. It would only be an option I consider with two independent battery banks topology within the handheld unit, were battery power is always used even when charging. The idea is that the architecture is modular, any reference could be used.
For non handheld operation LTZ1000 would be the way to go. -
I don't think one would really need 2 µCs. The calibration and similar part would not need much extra processing power. A single µC should be sufficient.
It’s not for processing power, it’s for added modularity. I could use the DAC board anywhere, it needs only the supplies and reference. -
There is one issue with plastic packages that is still vastly underestimated: Moisture.
You can get uncertainties in the order of 100ppm or more even with the best references like the LTC6655. A viable way to mitigate that problem is to store the circuit in an air tight box together with a desiccant which would need to be regenerated from time to time.
It won't matter to take the circuit out of that box for a couple of hours as it takes many houres to days for the package to soak enough water to influence the calibration. But it also depends on the ambient relative humidity...
LM399 or LTZ1000 are inherently immune against that uncertainty both by the heater and the metal can.
Cheers
Andreas
Edit: The calibration would have to be done in the dried state of course.
-
Why would 7nV/rtHz matter against 7.5nV/rtHz for DAC11001A vs. AD5791B when the INL of the DAC11001A is +/-4LSBs versus +/-1LSB max. for the AD5791B?
The DAC11001A is only an 18-bit DAC in that regard...
The 5791A is at the same 1k-price level as the DAC11001A that has no option to upgrade if performance is key...
1.1µVpp vs. 0.4µVpp, low frequency noise is more of a concern at DC than noise density at higher frequencies.
Then the AD5791A is an 18-bit DAC in that regard too
Why do you compare specs for AD5791B and than referring to price of AD5791A?
-
The numbers for the noise are not directly comparable: the AD number is for a +-10 V reference, while the Ti data are for a 10V/0 V reference and with an extra added buffer.
Chances are the DAC110001 would still be slightly low LF noise, but more like twice the 0.4 µV_pp. So the difference is not that large, more like very similar if the AD noise data also includes the reference inversion.
I think the point of comparing the data for the A and B version is, that that the cheaper B version is about comparable to the DAC11001A in price and performance. One still has the option to upgrade to the higher price and performance A version. -
There is one issue with plastic packages that is still vastly underestimated: Moisture.
not only the plastic package. It is also the Epoxy PCB which soaks moisture.
So the problem is also visible with those modern ceramics SMD packages. (LS8-package).
100 ppm is a bit on the high side.
Maybe this is true for a SMD package which is ~factor 2-3 worse in stability than a DIP package.
What I have measured is more like up to 0.5ppm / % rH for a DIP 8 package.
(it depends a bit on the manufacturer of the epoxy).Edit: The calibration would have to be done in the dried state of course.
And let at least 2 or more weeks for drying.
with best regards
Andreas
-
Quote
Then the AD5791A is an 18-bit DAC in that regard too
Kleinstein nailed it -I intended to make an apples to apples comparison - both AD5971A and DAC11001A are at 18Bit performance level and in the $30 ballpark. Comparing the DAC 11001A with the AD5791B would have been unfair as the B-version of the AD5791 is north of $60 (1k-price).
Why do you compare specs for AD5791B and than referring to price of AD5791A?Quotenot only the plastic package. It is also the Epoxy PCB which soaks moisture.
Yes, they drift as well - even with a U-cut around it.
So the problem is also visible with those modern ceramics SMD packages. (LS8-package).
Cheers
Andreas -
Personally I would trade away handheld-ability if it means better performance. Battery operation could still be useful as UPS if the stability calls for constant power.
If the user interface board is a separate module, we can experiment with different boards for different use cases, extend with USB or Ethernet, etc.
-
Quote
I think the point of comparing the data for the A and B version is, that that the cheaper B version is about comparable to the DAC11001A in price and performance. One still has the option to upgrade to the higher price and performance A version.
Oops - I missed that... in the Analog Devices world B is for Better.
Cheers
Andreas -
Quote
Then the AD5791A is an 18-bit DAC in that regard too
Kleinstein nailed it -I intended to make an apples to apples comparison - both AD5971A and DAC11001A are at 18Bit performance level and in the $30 ballpark. Comparing the DAC 11001A with the AD5791B would have been unfair as the B-version of the AD5791 is north of $60 (1k-price).
Why do you compare specs for AD5791B and than referring to price of AD5791A?
Thanks, now got it, the points have to be taken independently and then it makes complete sense.The numbers for the noise are not directly comparable: the AD number is for a +-10 V reference, while the Ti data are for a 10V/0 V reference and with an extra added buffer.
Chances are the DAC110001 would still be slightly low LF noise, but more like twice the 0.4 µV_pp. So the difference is not that large, more like very similar if the AD noise data also includes the reference inversion.
I am puzzled why would the output voltage noise scale with reference voltage?
Simplified this would mean output noise of resistive divider would scale with input voltage... -
Quote
I am puzzled why would the output voltage noise scale with reference voltage?
The reference noise is fixed in amplitude - if gain is applied, the noise will be amplified.
Simplified this would mean output noise of resistive divider would scale with input voltage...
Cheers
Andreas -
One of the good ideas I saw earlier, after I finish it I can contribute one to the metrology kit going around so people could play, if folks are interested.
-
How the LF noise changes with differente operation point is not so easy to predict. There are several noise sources that add up. Noise from ampliers would be about undependent from the reference amplitude. 1/f noise from the resistors, which is well possible at that level would scale with the applied votlage and the other point is noise from the reference scalung - here the data-sheets are not very specific on the details. At that level the noise measurment is also not easy, as there can be noise from the reference at a comparable level. So they may subtract reference noise - a method that can introduce errors it the model is not correct.
-
Just in case it wasn't posted before, here is something to read about AD5791.
The 20-Bit DAC Is the Easiest Part of a 1-ppm-Accurate Precision Voltage Source
Now that we have ADR1399/LTZ1000/ADR1000 I wonder if someone has finished one of the designs already?
I still have the evalboard, time to combine it with one of my references.
-branadic-
-
A couple of months ago, I had the opportunity to evaluate the full platform - EVAL-SDP-CB1Z (controller), AD5791SDZ (DAC) and EV-LTZ1000-REFZ (reference).
Figure 49 of the datasheet shows that the AD5791 is a segmented DAC with 6-bits of linear equal resistors and remaining 14-bits of R2R. I iterated through every setpoint in the MSB 9-bits and measured the resulting voltage (between -10V and +10V) using 3458A. This covers the 6-bit linear section, and the MSB 3-bits of the R2R section.
As shown in the attached charts for 2 different AD5791SDZ boards; the linearity of the 6-bit linear segment is good but when the R2R section is used, it degrades the linearity.
-
That pretty much looks like what is given in the datasheet Fig. 5, if we get rid of the additional curves.
-branadic-