Author Topic: Building your own voltage reference - the JVR  (Read 127247 times)

0 Members and 2 Guests are viewing this topic.

Online MegaVolt

  • Frequent Contributor
  • **
  • Posts: 909
  • Country: by
Re: Building your own voltage reference - the JVR
« Reply #300 on: October 05, 2021, 10:17:35 am »
drift of 130 nV per day (4.75 ppm/year), without any indication of relaxing. Deviations of the daily averages from linear drift ("noise") are 430 nVrms (0.043 ppm), while the setup is good for 100 nV or better.
To which scheme do these values apply?
 

Online dietert1

  • Super Contributor
  • ***
  • Posts: 2021
  • Country: br
    • CADT Homepage
Re: Building your own voltage reference - the JVR
« Reply #301 on: October 05, 2021, 11:15:14 am »
That was made in January and has been running since then in a TEC "oven".
https://www.eevblog.com/forum/metrology/building-your-own-voltage-reference-the-jvr/msg3433032/#msg3433032.

The circuit was unusual in that it used a diode for TC compensation. Later i also got the original proposal working to adjust TC by drain current and/or source drain voltage. Maybe i can try a JVR as a low noise external reference for an ADS1263 ADC i am experimenting with. In that case anything between 2.5 V and 5 V will work.

Regards, Dieter
 
The following users thanked this post: MegaVolt

Online dietert1

  • Super Contributor
  • ***
  • Posts: 2021
  • Country: br
    • CADT Homepage
Re: Building your own voltage reference - the JVR
« Reply #302 on: November 02, 2021, 12:27:10 pm »
Since July i have the JVR i made in January in a MUX setup together with two DIY LTFLU references and now after nine months of continuous drift it appears to settle. Nine months of patience with no signs of relaxation. It drifted about 4 ppm in total. Appears to be a good build. The rms noise of the daily averages is 0.04 ppm. It runs in a TEC oven at 24.5 °C.
Maybe a bit early to say, but this observation looks completely different from the LTFLUs that settled in a slow exponentional relaxation with a time constant of about 4800 h.

Regards, Dieter
 
The following users thanked this post: exe, SilverSolder

Online iMo

  • Super Contributor
  • ***
  • Posts: 4675
  • Country: nr
  • It's important to try new things..
Re: Building your own voltage reference - the JVR
« Reply #303 on: September 21, 2022, 09:50:58 pm »
Hi, let me refresh this page a little bit - I've found in my junkbox several jfets (at least i think so)
DN349 (dual in metal, Siliconix?)
NF510 (single in metal)
Both types with 73 date code (50y old next year!)  :clap:
I cannot find any datasheet.. No idea whether they are suitable for experiments with jvr, however..
Any hint on it?
« Last Edit: September 21, 2022, 11:26:31 pm by imo »
 

Online Messtechniker

  • Frequent Contributor
  • **
  • Posts: 774
  • Country: de
  • Old analog audio hand - No voodoo.
Re: Building your own voltage reference - the JVR
« Reply #304 on: September 22, 2022, 06:48:10 am »
Quick look around yielded:
DN 347 was used in the B&K 2607. Probably something specially selected by B&K.
The NF 510 is probably equivalent to Linear Systems LC 4393 and 2N 4393.
as  taken from DSA00411750.pdf and LSM_LS4393_SOT-23.pdf

« Last Edit: September 22, 2022, 06:51:22 am by Messtechniker »
Agilent 34465A, Siglent SDG 2042X, Hameg HMO1022, R&S HMC 8043, Peaktech 2025A, Voltcraft VC 940, M-Audio Audiophile 192, R&S Psophometer UPGR, 3 Transistor Testers, DL4JAL Transistor Curve Tracer, UT622E LCR meter
 
The following users thanked this post: iMo

Online Kleinstein

  • Super Contributor
  • ***
  • Posts: 14080
  • Country: de
Re: Building your own voltage reference - the JVR
« Reply #305 on: September 22, 2022, 07:03:39 am »
The duals are usually low threshold types and are thus not that suitable for a JVR. The 2N4393 is also a more low threshold type. With the inherent scattering of the parameters one may still get some 2 V.
These types are more suitable for an amplifier or switch.
 

Online iMo

  • Super Contributor
  • ***
  • Posts: 4675
  • Country: nr
  • It's important to try new things..
Re: Building your own voltage reference - the JVR
« Reply #306 on: September 22, 2022, 10:14:33 am »
Thanks guys! I've been just measuring the group of NF510s - the voltage at the resistor (a trimmer I play with) for a group of samples varies from 4.6V to 5.8V while playing with the drain currents from 150uA to 1.3mA. First experiments with heating up and cooling down show pretty stable currents, like with a sample  (trimmer set to the 1.3mA current)
24C 1.307mA
80-100C (burned my finger) 1.310mA
PS: with 3k9 25ppm resistor therm isolated and away of the transistor
24C  1.493mA
>100C (heated up 7secs with iron) 1.495mA
and returns back with zero "visible" hysteresis..
Now, how to find the zero TC..  :D
« Last Edit: September 22, 2022, 11:46:57 am by imo »
 

Online iMo

  • Super Contributor
  • ***
  • Posts: 4675
  • Country: nr
  • It's important to try new things..
Re: Building your own voltage reference - the JVR
« Reply #307 on: September 25, 2022, 06:11:32 pm »
Hmm, when trying to find the NF510's zero TC I get this curve.. Why this shape??
Measured via 34401A (100PLC) and 10T 5k Bourns WW as the resistor, heating up by "finger" method, powered by an epoxy LT1021-10..
Also I tried to look at a longer run with the Bourns set to "zero TC" (difficult to find the zero, indeed) and I got 13uV/C (2.3ppm/C).
The noise I saw at the level of the dmm noise (450-550nV rms in quiet periods), the lowest I've ever seen here..
« Last Edit: September 25, 2022, 06:32:04 pm by imo »
 

Online Kleinstein

  • Super Contributor
  • ***
  • Posts: 14080
  • Country: de
Re: Building your own voltage reference - the JVR
« Reply #308 on: September 25, 2022, 06:41:07 pm »
The trimmer resistor is a bit suspect. It would be a good idea to have at least the main part of the resistance / current from a reasonable stable fixed resistor.

Getting to 13 µV/K for the drift is not that bad, given that one usually gets a kind of parabola shape. So a maximum in the voltage at some temperature and quite some 2nd order TC.
The +8 C step may bring you from one side to the other for quite some temperature range. The heating with the fingers may not be good enough to handle the TC that changes with temperature.

The drain voltage also has some effect. Depending on the part and voltage the suppression of DS voltage variations is not that great. It could help to generate the drain voltage from the source votlage with a little gain (like x 1.2 or x 1.33) with also reasonable stable (e.g. 50 ppm/K or better) resistors. With the extra reference for the drain voltage one kind of gets the difference of the 2 ref votlages to enter.
 

Online iMo

  • Super Contributor
  • ***
  • Posts: 4675
  • Country: nr
  • It's important to try new things..
Re: Building your own voltage reference - the JVR
« Reply #309 on: September 25, 2022, 06:49:02 pm »
I did a longer run with the WW pot replaced by a mix of three 15/25/50ppm resistors such I got them close to the WW resistance (WW=3116ohm, resistors=3090ohm) and I got 24uV/C. In both cases the the stuff closed into a box with foam inside during the run. I will repeat the run with WW trimpot and different voltage setting this night with a bigger temp difference..
« Last Edit: September 25, 2022, 06:51:54 pm by imo »
 

Online Kleinstein

  • Super Contributor
  • ***
  • Posts: 14080
  • Country: de
Re: Building your own voltage reference - the JVR
« Reply #310 on: September 25, 2022, 08:10:56 pm »
I think it would really help to approximately record the temperature and thus get voltage versus temperature curves and just the difference between 2 temperature points.
A larger temperature difference for the voltage difference makes is more tricky to keep the average temperature fixed. The 2nd oder TC is quite significant, so the TC depends on the temperature.

So far I have not done test with many different resistors usually just an iterative process to come close to zero TC. My fast tests for this were more with small temperature changes, but this may not work with the 34401.

 

Online iMo

  • Super Contributor
  • ***
  • Posts: 4675
  • Country: nr
  • It's important to try new things..
Re: Building your own voltage reference - the JVR
« Reply #311 on: September 26, 2022, 06:30:40 am »
Yep, the previous finger method was jumping around the max thus I got weird numbers, the 8C jump is too much.. Below the max at aprox 27C when scanning with temp. The box TC there is (from 24 to 28.5C) 14.5uV/4.5C=3.2uV/C, or 0.56ppm/C.
« Last Edit: September 26, 2022, 06:36:31 am by imo »
 

Online iMo

  • Super Contributor
  • ***
  • Posts: 4675
  • Country: nr
  • It's important to try new things..
Re: Building your own voltage reference - the JVR
« Reply #312 on: September 27, 2022, 06:32:07 am »
And finally a longer run of the NF510 jfet with 3100ohm (3x1k+100, 50ppm), at room temperature, schematics as above.
 

Online Kleinstein

  • Super Contributor
  • ***
  • Posts: 14080
  • Country: de
Re: Building your own voltage reference - the JVR
« Reply #313 on: September 27, 2022, 06:43:41 am »
Much of the noise in the curve can be from the 34401 DMM. The LM399 reference alone usually has quite some noise (especially popcorn type jumps by some 4 µV) and the ADC of the 34401 adds some more white noise to this.  With so many points, it looks like the measurement was done quite slow. With rather new parts and a new soldered circuit there is also the possiblitiy of drift for the JFET and the resistors and maybe also the DMM.
 

Online iMo

  • Super Contributor
  • ***
  • Posts: 4675
  • Country: nr
  • It's important to try new things..
Re: Building your own voltage reference - the JVR
« Reply #314 on: September 27, 2022, 07:04:29 am »
Yes, the most noise comes from the dmm, sure. Below the running standard deviation made upon 100 last samples during that measurement. The peaks are most probably popcorn shots from the 399, or EMI or EMF of connectors. With JVR_DUT you would need a much better setup.
 

Online Kleinstein

  • Super Contributor
  • ***
  • Posts: 14080
  • Country: de
Re: Building your own voltage reference - the JVR
« Reply #315 on: September 27, 2022, 07:17:43 am »
There is a way to look at the JVR noise with better resolution when using the difference of 2 similar references. This allows to use a lower range (e.g. 100 mV) at the DMM an maybe an extra amplifier, so that the main noise is 2 x the JVR. One still does not know which one is causing the noise, but it is mainly reference noise.
 

Online iMo

  • Super Contributor
  • ***
  • Posts: 4675
  • Country: nr
  • It's important to try new things..
Re: Building your own voltage reference - the JVR
« Reply #316 on: September 29, 2022, 09:32:21 am »
The same NF510 sample bootstrapped with an OP07 creating its Vdrain. TC0 set to aprox 25C. Below a scan around 25C (up and down), 34401A 100NPLC, the data processed via a moving median filter (n=11).
The max looks sharper than with the LT1021 source, could be the resistors around the OP07 play the role as well (125ppm one).
« Last Edit: September 29, 2022, 11:46:13 am by imo »
 

Online Kleinstein

  • Super Contributor
  • ***
  • Posts: 14080
  • Country: de
Re: Building your own voltage reference - the JVR
« Reply #317 on: September 29, 2022, 11:44:39 am »
The measurements are really stretching the 34401 performance. Especially when relatively slow and measured only for such a small range there are many parts that can contribute.
With the limited meter it would make more sense to measure over a larger temperature range, like 15 to 30 C. This gives more change in the voltage and eaier interpolation.

I would not expect the resistors to have much effect of the form of the maximum. At least in my tests (with 2N4391) I saw not much difference in the 2nd order TC part (factor before T²) with different drain voltage.  A linear TC in the gain would mainly add some linear term to the reference and thus shift the position of the maximum, but not make it steeper.
The resistors (the 2 for the voltage gain and the one to set the reference current) do effect the voltage. AFAIR the 2 resistors for the gain were a little less (but not very much) critical than the one for the current.
 
The following users thanked this post: iMo

Offline miro123

  • Regular Contributor
  • *
  • Posts: 202
  • Country: nl
Re: Building your own voltage reference - the JVR
« Reply #318 on: September 30, 2022, 10:24:30 am »
Hello Voltnuts experts,
I've read this interesting thread. I have one beginner questions.
How those discrete JVR scores against monilitics from lets say Analog Devices  ADR44x series.
I have no experiance with ADR44x but
1. datasheet is promising
2. They are automotive certified - mostly means that datasheets is met for all devices
3. D grade has already excellent long term stability

BR
Miro
 

Online Kleinstein

  • Super Contributor
  • ***
  • Posts: 14080
  • Country: de
Re: Building your own voltage reference - the JVR
« Reply #319 on: September 30, 2022, 11:08:51 am »
The ADR44x use a similar reference inside, but with 2 FETs. This could allow to reduce the 2nd order effect. So the TC can be low over an extended temperature and not just in a small range.

The JVR performance depends quite a bit on the resistors and likely some luck with the JFET choise. There is no real need for close temperatur coupling between the parts - so no real need for a monolytic solution. The discrete JVR needs additional effort for trimming to low TC and the absolute output votlage is scattering quite a lot (e.g. 2N4391 may range from some 4 to 8 V). The large scattering between JFETs also means an individually matched resistor - so nothing to buy the parts for, solder and done.  Noise wise the 2N4391 based solution seems to be a bit lower noise than the ADR44x, when looking at the voltage noise relative to the absolute voltage.

For the long term drift there is quite some uncertainty though the test so far look good for the JVR.  The plastic case of ADR44x make them likely sensitive to humidity and could be the reason for not that great drift specs (50 ppm /1000h and 70 ppm hysteresis). JFETs are still availabe in a metal case.  It can still be expenside to get hermetic sealed resistors.

The really low drift vesion is the ADR45xxD and this one seem to be available next year.
 

Online iMo

  • Super Contributor
  • ***
  • Posts: 4675
  • Country: nr
  • It's important to try new things..
Re: Building your own voltage reference - the JVR
« Reply #320 on: September 30, 2022, 11:12:32 am »
Not a voltnut here, but I spent some time messing with references in epoxy packages (around 70 pieces incl ADR431 REF50xx LT1021-x here). Not worth of spending a single cent when targeting something with "voltnut" in mind.. The hysteresis makes them almost unusable for something like >=5digits..
PS: as I reported in a different thread, my epoxy 2xLT1021-10+OP07buff shows <1ppm/C short term TC (2 chips selected for an opposite TC), but the hysteresis after a month powered-off is something like 60ppm..
« Last Edit: September 30, 2022, 11:19:53 am by imo »
 

Online dietert1

  • Super Contributor
  • ***
  • Posts: 2021
  • Country: br
    • CADT Homepage
Re: Building your own voltage reference - the JVR
« Reply #321 on: October 07, 2022, 06:48:09 am »
Looking once more into how to make a zero TC JVR with Ids < 1 mA and low Uds. The schematic shows a variant i am testing right now with good results.
After some tuning steps for the Uds voltage generator as shown in the table i got a flat top at 23.1 °C with +/- 1 ppm in 18 .. 28 °C interval.
Noise as measured by a K2700 in a temperature chamber is about 400 nV RMS, i.e. sub-ppm.
After a four day log there is no indication of drift. Limit would be about +/- 15 ppm/a right now (will improve rapidly).

Regards, Dieter
« Last Edit: October 07, 2022, 07:17:11 am by dietert1 »
 
The following users thanked this post: BravoV, Kleinstein, iMo

Online iMo

  • Super Contributor
  • ***
  • Posts: 4675
  • Country: nr
  • It's important to try new things..
Re: Building your own voltage reference - the JVR
« Reply #322 on: October 07, 2022, 10:55:07 am »
What if I use the second jfet (in a dual package) instead of the 1N4148?
PS: there is a missing dot at Vref (to the opamp's output) in your schematics, imho.
 

Online Kleinstein

  • Super Contributor
  • ***
  • Posts: 14080
  • Country: de
Re: Building your own voltage reference - the JVR
« Reply #323 on: October 07, 2022, 11:23:10 am »
A diode connected JFET could work instead of the 1N4148. However most of the dual JFETs are low threshold, as they are made for use as an amplifier. There are a few higher threshold types (e.g. the 3456 uses a pair in it's input amplifier for bootstrapping), but they are rare.


The diode instead of a resistor shift the TC to a more negative value and allows to work with a small current for the FET. I would not consider some 1-1.5 mA for the 2N4391 too high. The question for me is more if it helps agains the curvature in the curve.
What does the curvature parameter in the table mean ?
 

Online dietert1

  • Super Contributor
  • ***
  • Posts: 2021
  • Country: br
    • CADT Homepage
Re: Building your own voltage reference - the JVR
« Reply #324 on: October 07, 2022, 11:34:20 am »
Yes, it's great to use the second JFET for TC adjustment (on-chip sensor). What would be the reference voltage?

No, there is no connection. The OpAmp does not output the reference voltage but the drain voltage of the FET. I am just showing how i connected the K2700.

But your question contains a good proposal: Will also try to adjust the circuit for zero TC on the drain node. Then the OpAmp can already serve as buffer.

A curvature of 2 ppm means that on the extremes the output voltage deviates -2 ppm from the maximum, so one can spec Vref +/- 1 ppm within the 18 to 28 °C range.

Regards, Dieter
 


Share me

Digg  Facebook  SlashDot  Delicious  Technorati  Twitter  Google  Yahoo
Smf