10V Reference (I Did It My Way) No1

[blackdog]

Hi Volt-Nuts, :-)

I got several emails from people on this forum why I dont start my own topic over "Voltage References"
Your Wish Is My Command 

But, please bear with me, I'm a dyslexic monkey, it takes a lot of time, to make reasonable sentences in Englels.
And yes, Google is my Friend.

OK enough blah blah...

The first design that I want to show here, I already posted in another topic.
It is designed approximately 3 to 4 years ago by me.
I've been playing for over 30 years with voltage references and the last 5 years more intensive.
But only in the last view years, I was able to buy the instruments, to test voltage references.

I still have no Fluke 732a reference, therefore I myself am trying to build a voltage reference with good specifications.
i have not the resources like Fluke to test my reference design.
But i have a fresh calibrated HP3458A and two new KEYSIGHT 34461A DMM's.

This is an older design of mine, but there are good decisions,
such as multiple safeguards against the misuse.

Let us begin the upper left, where the power supply is made for the 4x LT1021.
The power supply is built around a LM10 IC, which has a 200mV reference, I scale to approximately 8V. R8 and C3 filter the noise of this voltage, this is a low pass filter of about 5Hz.
The alert viewer falls on, that R 1 and R 8 has the same value, in order to avoid bias errors.

At the output of the opamp, which is pin 6, there is a 47 Ohm resistor (R2) which, together with R1, C1 and C2 also create a low pass filter.
The effect of this circuit is that there are only about 5uV RMS noise is present on the 8V power supply for the LT1021. The bandwidth of the measurement was 20Hz to 22Khz.
The 8V power supply is clean, there is no visible popcorn or schutgun noise.
The circuit might be a bit overdesigned.
It is necessary to have a good quality capacitor for C3, as it is not DC offset corrected, like C1.

Next part, mixing the references...
To make the noise and drift lower of the already good LT1021, I use four of them in parallel mode.
R4, R5, R6 and R7, together with the capacitors C4 and C5 create a low pass filter, having a frequentie of about 3.9 Hz.
This filter is  a "bootstrapt", R10 of 10K, ensures that the upper 330uF only sees a very low DC voltage, the leakage current is than very low.
Also, use good quality capacitors.

Gain of 2 amplifier.
I used a selected LT12ACN opamp, i want to test with a LTC2057 or a Max44241.
The problem is, there is not much info, about about the low frequentie noise, especially the current noise is not properly specified...
I'm always fighting with the values of R12 and R17 (10K, down 0.01% Vishay S102) and the current noise.
The lowest noise and the greatest stability, you need low value for these resistors.
Low value for these resistors also means, that the dissipation in these resistors will go up.

D1 and D2 is for protection, normaly there is no voltage over the inputs, so no leakage.
To trim the Reference i use small 0.1% film resistors in parallel with a 10-turn potentiometer.
Perhaps it is better to inject current through a resistor a high value in the -input of the opamp?
Less connection @ the -input.

Buffer
The output of the opamp goes to a 1x buffer.
As a result, the opamp has its maximum open loop gain.
At the output of the buffer are 4 resistors of 100 ohms, C11  and 330UF again make of a low pass filter.
C9 and R17 are also involved in this filter.

Protection
The reason for the 4x100 Ohm is this, it is an inexpensive way to ensure that the reference remains intact, when power is put on the output. Bad Boy!!!
L1 keeps HF away from de output of de buffer.
F1, fast blow 150mA, u can choose a lower value, if you want.
D5 PGKE18CA protects IC3 a buffer IC like EL2001, BUF634, LT1010 etc.



And this is a high resolution picture
http://www.bramcam.nl/Diversen/Quad-LT1021-10Vref-01.png

Please shoot @ it, make it better ect. ect.
Make it a community project if you want 

My version is housed in a 42C heathed BimBox, but about ovens, i wil show you later my designs.

Kind regarts,
Blackdog (Bram)

[nctnico]

I got a grazy idea: how about putting a (thin) reference circuit board on a piece of aluminum (with thermal conductive rubber in between) and keep it at a constant temperature say 45 deg. C. That would eliminate temperature differences between components (including the reference chip).

[JohnnyBerg]

Hey Blackdog,

Nice to see you on this forum. I am looking forward on your posts and pictures of your references!

[T3sl4co1l]

Curious use of TVS and FBs...

Why use a bidirectional TVS in parallel with a diode?  The unidirectional (-A) version is probably much larger (lower series resistance) than a 1N4001, let alone a 1N4007.  Likewise, D5 missed the opportunity to be placed at the output, where it will absorb the surge and the series resistors will prevent the buffer from exploding; as shown, it might still blow the buffer (I'd have to check the datasheet).  It should also be unidirectional, for similar reasons.  Surge protection on the output, with all that capacitance there, is probably not a big deal, in the end, either way.

As for the FBs, they're in series with ground (doesn't really matter?), a 100u inductor (may matter at HF, but generally not a big amount), or a 10k resistor (no effect at basically any frequency).  So, they're probably not doing much..

Why not also filter / regulate the amp/buffer supply?  PSRR is good, but not fantastic.  I suppose you could even go overboard and bootstrap the amp/buf with the ref voltage itself... who needs an outboard regulator when you have an even better inboard one?

Bringing together multiple references is often a good idea (if not often used).  Same goes for amplifiers in parallel (not op-amps -- due to differing Vos, but JFETs in parallel for a low noise buffer / preamp is a classic).  Noise/error goes down as sqrt(N) for N sources; but, only for independent error.  So temperature drifts will remain.  As always, keeping convection currents away from the refs will help keep low frequency noise down.

Speaking of large capacitors, with all those beefy 330uF caps around, some diodes to allow them to discharge into the supply (if it collapses) might help.  The good old "back-diode around the 7805" trick.

Tim

[quarks]

Looks very good to me.
Thx for sharing

[3roomlab]

sorry i have noob question

what does C4 do if thru R3/6/7, C4 will seem to see evened out voltages on both ends? EMI / noise / spike reduction ?

[T3sl4co1l]

sorry i have noob question

what does C4 do if thru R3/6/7, C4 will seem to see evened out voltages on both ends? EMI / noise / spike reduction ?

Leakage reduction strategy: C5 takes "most" of the voltage (minus what gets dropped through R3, which is larger than the other resistors, to have less effect on the refs*), so that C4 gets 0V and almost exactly 0 leakage.

*Except, if the voltage drop is small, the value of R3 will have little impact, because... that's the point, 100% of C5's leakage is drawn from the refs.  So it's exactly as good as having C4 straight to ground, except with its leakage supplied by only one ref (more or less).

What should be done is, instead of using C5 and R3, use a unity gain (not gain-of-2) amplifier, and use its output as the "guard ring" voltage: for the bottom end of C4, for traces around the board, stuff like that.

Good catch.

Tim

[EEVblog]

Didn't the EL2001 buffer go obsolete a decade or two back?

[janaf]

If crazy, I'm one! Got copper plate and large silicone pad in the drawer :-)

You can get single-layer PCBs made on aluminium base material. Price $60 setup plus 0.02usd/cm2
http://www.ebay.com/itm/290575172294

I got a grazy idea: how about putting a (thin) reference circuit board on a piece of aluminum (with thermal conductive rubber in between) and keep it at a constant temperature say 45 deg. C. That would eliminate temperature differences between components (including the reference chip).

[blackdog]

Hi,

I will try to explain my decisions for the construction of this 10V reference.

T3sl4co1l
First, good remarks!
I dit not make good labnotes, yes i know, a beginners mistake...

Why use a bidirectional TVS in parallel with a diode? My bad, corrected :-)

The ferrite bead and the 100uH keeps the HF out of de reference, this is one purpose of these components.
The second one is, to keep the peak current down (100uH helps).

By misuse, the fuse, 100uH coil, R19 / R22 and D5 to keep the energy from the sensitive circuit.
If you use for the resistors R12 and R17 a LT5400 (Nice LT part) the voltage over these resistors are Max 80V.
I fault conditions you can reach this voltage because the inductions of your cables who are connected to the voltage reference.

T3sl4co1l, I already told you, good comments! 2 points ;-)
To ensure that the peak voltage is not too high on the output, in fault conditions, i have placed an extra TVS across the output.
I have not told you how the reference is used, normaly is powerd by the mains supply.
The battery is only used to keep the sensitiv parts under power (reference + heather)

About the 7805 diode trick, I thought of this, but am afraid of leakage currents...


The schematic is fed from a good regulated power supply, built around a good reference IC.


Dave
EL2001 a beautiful part, I'm also more than two decade's old, and still working fine 
Above the schematic i explained that you can use a BUF634 LT1010, etc. choose a buffer, it is not really critical part.



High Res Picture..
www.bramcam.nl/Diversen/Quad-LT1021-10Vref-02.gif


Kind regarts,
Blackdog

[EEVblog]

Gain of 2 amplifier.
I used a selected LT12ACN opamp, i want to test with a LTC2057 or a Max44241.
The problem is, there is not much info, about about the low frequentie noise, especially the current noise is not properly specified...
I'm always fighting with the values of R12 and R17 (10K, down 0.01% Vishay S102) and the current noise.
The lowest noise and the greatest stability, you need low value for these resistors.

If you were really keen you could parallel opamp to lower the noise a bit.

[JohnnyBerg]

Can one assume that R12 and R17 have the same temp. coef. ?
If they have the same temp. coef, are they thermally coupled?

What I mean is, as they drift both in different directions, lets say with 10 ppm/K, the drift due to this resistors will be much greater than the drift of the references?

[codeboy2k]

The power supply is built around a LM10 IC, which has a 200mV reference, I scale to approximately 8V. R8 and C3 filter the noise of this voltage, this is a low pass filter of about 5Hz.
The alert viewer falls on, that R 1 and R 8 has the same value, in order to avoid bias errors.

At the output of the opamp, which is pin 6, there is a 47 Ohm resistor (R2) which, together with R1, C1 and C2 also create a low pass filter.
The effect of this circuit is that there are only about 5uV RMS noise is present on the 8V power supply for the LT1021. The bandwidth of the measurement was 20Hz to 22Khz.
The 8V power supply is clean, there is no visible popcorn or schutgun noise.
The circuit might be a bit overdesigned.

The 8V regulator is quite good, very low noise.  My calculations show it does indeed have about 5uV RMS noise, but that's at the reference amplifier output.  The buffer amp adds another 15 uV RMS noise (which is still good!)

Reference:
Reference Voltage = 0.2 V
filter cutoff in the feedback loop = 1/(2*pi*RC) = 1/(2*pi*2k21*15uF)  = 4.8 Hz
reference amp gain = 1+(39k2/1k) = 40.2
reference output voltage = Vref * gain = 0.2 * 40.2 = 8.04 V
noise gain = reference amp gain = 40.2
spectral noise density @ 4.8 Hz = 45 nV/sqrt(Hz)
RMS voltage noise @ BW 4.8 Hz  = 45 nV/sqrt(Hz)  * sqrt(4.8 Hz * 1.57) =  124 nV RMS
RMS voltage noise gained up by noise gain = 124 nV * 40.2 = 4.98 uV RMS

Op-Amp:
Source impedance = 2k21
filter cutoff in the feedback loop = 1/(2*pi*RC) = 1/(2*pi*2k21*1nF)  = 72 KHz
opamp gain = 1
noise gain = 1
from figure 8: input referred noise @ 72 KHz is approx. 45 nV/sqrt(HZ)
RMS noise @  BW 72 KHz  = 45 nV/sqrt(Hz) * sqrt(72e3 * 1.57) = 15 uV RMS


Total RMS noise = sqrt ( 4.98uV ^2  + 15uV ^2  ) = 15.8 uV RMS

Then, the LT1021-5 has a PSRR of about 94 dB according to the charts (and 12 ppm/V max according to the tables)

So this 15 uV RMS noise on the power supply rail will equate to about 12 ppm/V * 15 uV = 0.00018 ppm of voltage change at the output of the LT1021-5.  It's hardly noticeable, and you could even use a much noisier regulator and still be fine (which is why you admit it might be over-designed)

If I made a calculation mistake, please point it out and I'll update my math.

Thanks!

[blackdog]

Hi, :-)

I made some changes to the schematic, but that wil come later...

Dave,
More opamps is also more current noise...
It's really hard to determine what is the best solution.
I have at present no measurement setup to properly measure from 0.1 to 10Hz noise
And it takes a lot of time!

codeboy2k
You are not far of the real world, about the noise calculations :-)
Yesterday i made a small setup, to measure the noise of the 8V regulator.
The test setup of the LM10, yes i know, breadboards en references, bad choise, but its not about DC stability in this test...


RMS Noise measurend by the Audio Presisoin, Portable One Plus.
About 5uV RMS, 22Hz-22Khz bandwith.


This is a picture of de analyser output, don't mind de scoop scale, the analyser is in autorange mode.
Look at the trace, no popcorn or shutgun noise, 1.2 Seconds data...


JohnnyBerg
The whole reference is in a 42C oven, temp drift is not so important,
uneven aging of these resistors is important, since the oven does not help to reduce it.

Kind regarts,
Blackdog

[codeboy2k]

Measured real-world noise is always going to be better than max noise.

My numbers came from the datasheets, and I think the datasheets try to be conservative when they list noise numbers. They don't ever want someone to say they measured 15 uV noise when the datasheet said it would only be 5 uV noise

But still nice to see that it comes in better than the datasheet specs.

 

PS: Can you measure the noise at pin 1 (the ref-amp output) over the same spectrum ?

[blackdog]

Hi codeboy2k,


I dit the measurement you ask for.
The noise on pin-1 is higher than your calculations, about 190uV RMS :-)


I had to change the powersupply to batteries, because of the high commonmode signal at my workbench at the moment.


I hope this measuremt helps you.

Kind regarts,
Blackdog

[codeboy2k]

Wow.. thanks!  That's the gained up reference noise, before the 5Hz filter.

Pin 3 would be the filtered noise, input to the final op-amp.

[Andreas]

Hello,

instead LT1021 with metal can housing
one could use the (today easier to get) LT1236AILS8-5 with hermetically ceramic housing.
But you would need some dead bug style mounting to a standard pcb to reduce humidity influence.
From behaviour the LT1021 and LT1236 are very similar.
Except that the LT1236 is available with low TC and at the same time with high accuracy.
(The LT1021 comes either with low TC or with high accuracy but not both).

With best regards

Andreas

[blackdog]

Hi andreas,

I cant say anything negativ about you remarks 

I already started making changes to the schematic, i wil post it later this week.
I propose diffferent 5V reference, like the LT1236, LTC6655, LT1027, APEX VRE310A, ect.


Kind regarts,
Blackdog

[Andreas]

Hello,

best of all would be my "famous" dead bug in hole mounting method which can be seen here:
https://www.eevblog.com/forum/projects/oshw-24bit-adc-measurement-system-for-voltage-references/msg474504/#msg474504

With best regards

Andreas

[blackdog]

Hi Andreas,

I do not want to polute this topic, but how about this...
Because i like to play with references 

A ceramic resistor BPR10, 50 Ohm.


The oven controler on the BPR10 resistor, nice blue WLK 30 glue!


The reference schematic.


Dead bug style, mounting, the reference is the LT1236-LS8


Kind regarts,
Blackdog

[Andreas]

I already started making changes to the schematic, i wil post it later this week.
I propose diffferent 5V reference, like the LT1236, LTC6655, LT1027, APEX VRE310A, ect.

VRE310A is a 10V reference.
With VRE3050A I have no good experiences (too noisy, much hysteresis)
LT1027 is only available in plastic. (its a pity: the chip seems to be first class of unheated references).
LTC6655 also has much hysteresis. The only advantage is the low minimum supply voltage. (if one needs it).

So LT1236AILS-8 or AD586LQ would be my choice.
What I ever wanted to do but had no time up to now:
use the (chip - temperature dependant) voltage at the trim pin of the LT1236 for temperature compensation.

When using glue you might introduce already some mechanical tension to the package.
You have to try this.

With best regards

Andreas

[codeboy2k]

When using glue you might introduce already some mechanical tension to the package.

The adhesive might shrink a little, but if it's a ceramic package, shouldn't it be able to handle any mechanical stresses better than the plastic packages ?

and 2nd) even if there was a mechanical stress, would it be ok to trim the output 10.0000000 AFTER the glue dried? or is there a risk still?

[Andreas]

The adhesive might shrink a little, but if it's a ceramic package, shouldn't it be able to handle any mechanical stresses better than the plastic packages ?

and 2nd) even if there was a mechanical stress, would it be ok to trim the output 10.0000000 AFTER the glue dried? or is there a risk still?

Better: yes. Perfect: no.

E.g. in the first notes to LT1236AILS8-5 LT stated 12ppm humidity sensitivity for a 30% rH change. This is only due to the swelling of the epoxy pcb used where the LS8-ceramic package is soldered. On the current datasheet the rH spec is deleted (because it is only related to the pcb material used).

Even with my "in hole" mounting method. I can get output voltage shifts of 5-7uV when I apply mechanical tension to the pcb. And there is only one wire connected directly to the pcb.

With best regards

Andreas

[Alex Eisenhut]

Didn't the EL2001 buffer go obsolete a decade or two back?

Wow, élantec. Yeah, 20 years.