-
Hi,
In this topic I want to show you how i build a 10V reference circuit with 8 times an aged LT1021CH.
This Reference circuit i want to build in one of the ovens i made showd in this topic.
https://www.eevblog.com/forum/metrology/small-oven-controler-for-voltage-reference/
The schematic showd here is one of the first drawings, it is being designed with the knowledge I have gained by building other voltage references.
But first the block diagram, so that it becomes a bit clearer which way I want to go.
I want to trim the reference to +10V, this wil be my the first high stability reference with this trim function.
I would also like to apply my knowledge gained with the programming of an Arduino.
The lower LT1021 have a trimming resistor connected to pin 5 which goes to the TrimDAC.
I have to do some measurements to scale the trim current.
Temperature compensation of this trim connection like showd in the datasheet is not needed here, because the LT1021 references are in the 42C oven.
The trim resistors showd in this schematic are 10M, but I have to do some measurements to scale the trim current.
Most of my LT1021 are beter than the specs, so only a small current is necessary.
I chose a series of parallel circuit to make a 10V reference and at the same time reduce the noise.
This way I don't need expensive resistors to amplify the 5V to 10V.
The LT1021 ICs are switched as normal zener, the internal power source is not used.
But the ICs do need to be supplied with energy and that is done by the 3K resistors, these ensure that about 1mA will run through the zener.
For the best stability, the 13V need also be verry stable, the regulator at the top of the schematic uses the 10V output to make a decent +13V power line.
This schematic is wrong! I leave it here to show why it is wrong and how my development is progressing when building the they voltage reference.
You can't stack the 5V version of the LT1021 this way, this because it has a 6.3V Zener inside and it need minimum of 8.5V on the supply pin...
The output of the LT1021 factory goes to a 0.4Hz filter built up with two 330uF capacitors that have a low leakage current of the brand Nichicon.
The upper capacitor will have almost no DC voltage and therefore the leakage current through this capacitor will be extremely small.
One of my heathed references allready has this filter and for more than 4 years there are no problems with this setup.
So i wil try it also in this reference
After the 0.1 Hz filter, there will be a buffer stage that is also a second order low pass filter.
Also this time i use en J310 fet to onload the opamp, and the J310 can be used for at least 5mA output current.
The J310 makes it also a LDO setup. the +13V is therefore more than enough for the proper functioning of the output stage.
The opamp ADA4522-2 (A) has the same value for both inputs in terms of both resistance and capacity.
This wil keep the input currents balance in optimum stage.
The rest i wil explane later.
I have also a nother schematic with large parts the same setup, only some components wil be outside the oven,
i will show this later this week.
For now, shoot at it!
Kind regards,
Bram
-
This is a very nice concept and design. There is one issue that is easily fixed. The 5v version of the LT1021 will not work in shunt mode. So all the chips will need to be powered from the 13v line into pin 2. The upper references can still be stacked on the output pin (pin 6) of the lower voltage references. The 5v output of the upper chips will be pin 6..
The worst case supply current for the LT1021-5 is 1.5 ma so you may want to supply a little more current to the chips.
I stumbled across a characteristic of the ADA4522 last month that is not in the data sheets. With no load on the output of the opamp I had more noise than I was expecting. When I put a 100k load resistor (100ua) on the output the noise dropped in half. I ended up using a 50k load resistor (200ua) for margin. It had the same noise as the 100k load resistor. A 10k load resistor did not make it any better. A 1 Megohm load resistor did not help the noise. The load resistor probably puts the output stage into class A operation for better general operation.
Some audio opamps use this trick. The LT1115 datasheet recommends a 2ma load on the output for reduced distortion. So I would put in provisions for a load resistor between pin 4 and 1 of the output opamp.
Good luck. -
Some (perhaps) quick questions. I've never designed anything like this before....
1) Is there a reference on how to use the LT1021 without a voltage on the IN pin? Based on the datasheet, I have no clue what it'd do. The equivalent schematic isn't complete (e.g., it doesn't have the trim pin). It seems that you'll be biasing the current source above the zener in the wrong direction? Also, I think that you'd lose the temperature compensation of the "Q1" transistor?
2) I hear parallel voltage sources will reduce their noise by a factor of sqrt(n), but series voltage noise sources should add together. So, the noise of the references should be sqrt(2)/2 = 71% of a single LT1021-5's noise. So, the (.1 to 10uVp-p) noise of your parallel-series should be about 0.71*3 = 2.1uVp-p. Eight LT1021-10 in parallel should have a noise of 6/sqrt( 8 )=2.1uVp-p. It seems simpler to just use the eight in parallel? (EDIT: David Hess pointed out that the SNR ends up being the same in the parallel and series-parallel cases, so noise should not be the reason to choose one versus the other IF the device noise scales with the output voltage. However, in this case, the 10 Hz to 1 kHz SNR of the 10V reference is better, so it should be used. )
3) Move the +5Arduino ArduinoGND and the isolator out of the oven? There should be fewer feed-thrus after doing that.
4) The Trim of the LT1021 seems weird. The datasheet suggests a weird arrangement with a diode. I guess you use the 1M resistors to minimize the trimming, and to try to minimize the impact of the TC on the trimming? Yes, it helps that it's ovenized, but it's still not a good idea to degrade the TC....
5) Vref of your DAC isn't supposed to be above its Vdd. Chances are that it will be half of the time. Even worse, in its buffered mode, it can't be above Vdd-0.04V. In unbuffered mode you are shunting current from one of your references.
6) It's unclear where your ground comes in and out of the oven. Is the ground symbol inside the oven really a separate symbol from the one outside the oven? I'm not sure that the star grounding is correct. -
I may not quite understanding ... but LT1021 not a shunt, but it like a shunt connections 2 in series. ( pin 2 in, 6 out, 4 ground) LT has output 5, 7, 10 V. no reason to stack 5V gettin 10V. Then, with my little work with 1021 - B kinda expensive. and you need metal case not a plastic. and C,D version- close to 20ppm, 3ppm kinda optimistic, then all of them will drift together in one direction, so an absolute 10V would be no better then much less in parallel. for a filter .. it work, on safe side you can build 3 stages, even more safer.
ah yes i did dry to compensate with diodes - doesn't work or i did wrong ... so ... -
I'm not sure you'll get performance much better than a single ovenized LT1021-10.
If you hold the 1021 at a fixed temperature then the drift is primarily due to ageing. This can be accelerated somewhat by running at high temperatures, but I'd only age the whole shebang when it's built (unless you use sockets) - soldering has an annealing effect.
I haven't tried running the part as a two-terminal shunt - not sure you can?? Also, they are likely all to have the same drift ageing characteristics, so any anti-ageing mechanism will likely be negated unless you carefully match units. Is the 1021 really that noisy that you need the whole filter/buffer. The ADA4522 is certainly a nice part - however seebeck effect contributions from all the extra components in your design will dwarf the 'zero' drift nature of that part.
-
Hi,
First of all, thanks for the remarks, I will try to respond to your comments as best I can.
Almost all the sentences come from translation machines, so "bear with me", please...
chuckb
There are a number of things that are not quite clear to me in the datasheet of the LT1021CH that I am using.
It is said that the LT1021-7V and LT1021-10V can be used as a Shunt Zener, but not the LT1021-5V?
I will test that when I am back in my LAB with LT1021 ICs that are not yet aged
Thanks about the tip about the noise behaviour of the ADA4522.
Both opamps are designed as low pass filters with their -3dB points far below the switching frequency of the IC used.
So I don't think the noise will be a problem, but I will at least measure it to see if it is the case in this circuit.
pigrew
This will be the first time I've tried to connect the LT1021 in series, so I have no idea yet, if it will work.
As I indicated, this is one of my consept schematics, this schematics has not been tested only parts like the two regulators.
About the trimming and the TC, my point of view is, that there will be no TC!
I don't understand why I can use fewer wires if I place the i2c insulator outside the oven, mayby my blind spot?
I hadn't seen the small difference of 40mV in the datasheet between Vdd and the reference voltage.
This can be solved by giving the DAC its own 5V regulator and connecting the Ref to the Vdd, thanks for the remark.
The exact grounding "Mecca Point" has yet to be determined.
GigaJoe
I have more than 20 LT1021CH ic's that I've aged for more than three years and from this bunch I'm going to choose the most stable versions in this consept.
I am nog gonna buy LT1021 references :-)
This topic is to show how i will build a reference in a oven with the stuf what I have available, it will not be a "you can build it to project!"
But we can help each other to try and do it differently, perhaps one of you will be inspired by the way I approach this project and I will learn from your comments.
fcb
What will certainly improve is the noise coming from the LT1021 and this circuit.
If the total circuit ends up drifting too much(then I didn't do my job properly), I can compensate for this with my trimdac.
If I get out of my trim range, I throw the entire reference out of the window!
The LT1021 is a relativ low noise reference.
If i build this reverence well, the noise will be lower than a LTZ1000, we shall see and time will tell...
I have already a 4x LT1021-5V 10V reference in a oven, and its proven realy stable.
If the setup with te serial connected LT1021 is not working, i will go back to 8x parallel en two Vishay 10K resistors for the times two amplifier.
Thats one of my other conseps for my aged LT1021 references.
Aboud Mr Seeback, i know, the problem is also a bit the space in the ovens I have available. So a dual opamp in one case helps with the space.
It is complex and i need to think more aboud this.
When iám back in my LAB, i wil put alot of the components on a piece of circuitboard to see if it wil fit the proper way.
Maybe its is possible to fit two circuitboards in the oven...
Keep the remarks comming, it realy helps.
The reference wil be checkt on a regular basis with my Calibrated Agilent 3458A DMM and verry stable TEK 4050 DMM's.
And the Reference will only be build one time.
Kind regards,
Bram -
You're right about the shunt mode; I had completely missed it in the datasheet. Since the Zener voltage is 6.3 V or so, the 5V version shouldn't work in shunt mode. Note that the datasheet lists the Zener current as 1.7 to 10 mA, but your schematic uses only 1 mA.
The Arduino power connections wouldn't need feedhrus if the isolator was moved outside the oven. -
" Yes, it helps that it's ovenized, "
Would 60 C be adequate?
George Dowell -
Instead of using the BME280 which is very difficult to solder, why not a Si7021? It's a really nice humidity and temperature sensor.
-
Yeah, lt1021-5 doesn't work in shunt mode.
Why not use 10V version?
And why connect them all together as one reference?
If one (in a unlikely) event becomes defective or shows extreme drift it will affect the output without you knowing.
Maybe use 10V versions with separate output and adjustment and integrate a arduino controlled scanner-card?
Oven and power-supply can be shared of course. -
Hi pigrew
You hit the Jackpot, stupid me, i knew the LT1021 was based on a 6.2V Zener, it must be my age...
It can therefore never be used like the first schematic i showd here
I will lather this day change the schematic to a parralel setup.
GEOelectronics
The oven is designd for 42C
The Soulman
Why not use 10V version?
Because I have a lot of aged LT1021-5V references.
I can tell you that if one LT1021 will be defective I wil now!
And I always test with different dmm's to see if a reference is still correct.
Navarro
I will use a ebay BME-280 sensor already on a board, and i have a order list for also the Si7021, TSYS01 and the TSYS01 temp sensor, and another high res temp sensor.
Later I will decide which sensor I will use, on the display of the Arduino I will show both the conditions in the oven and outside the oven.
Because I also want to learn how good the oven is over time, I'm going to measure the temperature in the oven in different ways with high resolution sensors.
This is for now not in the schematic, only the basics, it's already complex enough.
It is always posible to measure every LT1021-5V with two ADS1115 ADC IC's to see is there is a problem with one of them, i keep it in mind :-)
Kind regards,
Bram -
With many good references in parallel it might be a good idea to have the option to check some of the references (could be series connected pairs) separate or to get an idea of the difference. This could be used to see of some of the refs are drifting faster than others.
If the refs have a significant different noise level it would be better to give the high noise ones a higher resistance, so they contribute less.
Depending on the adjustment range needed, one could skip the trim with some of the refs. So one could have both refs with and without the trimming. It could also reduce the number of very high value resistors that tend to be less stable.
The output amplifier has a rather low cross over frequency from using the JFET and the direct capacitive feedback. This is kind if good for very large capacitive load, but adds quite some output impedance at moderate frequencies. With a positive gate/source voltage a JFET can deliver more than it's Idss, that is specifies for zero gate voltage - so the current limit can be quite a bit higher.
The input current to the ADA4522 pins is about equal magnitude but opposite sign. So the voltage drop on the resistors will not compensate but add. So with AZ OPs it is generally not useful to add resistance for equal impedance.
-
Hi,
First of all, i chaged my first post to correct my stupid mistake
But let's go on, this is the latest schematic from today and of course not finished yet.
All 8 LT1021 references are now in parallel, and again the 0.4Hz low pass filter behind it.
Then a 2x amplifier stage is used as low pass filter which, among other things, ensures that the switching frequency of the selected opamp does not appear at the output.
I2C Power
There is also a 9th LT1021 present, which supplies the power for the i2c components (And DAC Reference).
Because I have many LT1021 5V references, it is also used here in a configuration that can supply a bit more power.
And probably I'm going to use the extended version with current limitation because in this schematic the power supply already comes from the 12V Power Line and it was connected to the 8.5V power line first. when using the 8.5V power line I have a LDO problem.
Power On main regulator
Again I use the 10V output to make the power supply for the LT1021 IC's.
The transistor at the output of the 8.5V power supply seems a bit strange.
It has a "pull up" resistor at the base and a zener diode to the output of the opamp.
This way I am sure that the power supply starts up.
When switched on, there is approximately 3V on the basis of the BD139-16 even if the opamp output would be "0V".
This is enough to start charging of the capacitor C12 and when this voltage rises the capacitor of 330uF will also be charged slowly.
Once there is some positive voltage at the + inputs of the opamp's , the gain of the opamps does the rest.
I've already used this way of setup in another piece of equipment.
Oven space
I thinkthere to many parts to put in my oven,
That is why an extra box has been drawn in the schematic to indicate which parts can be placed outside the oven, because they are not sensitive to temperature change.
The only sensitive parts good be the gain setting resistors for the 8.5V power line.
But 15PPM resistor wil doe the job here well, when I place them close to each other, and the high power supply rejection of LT1021 wil do the rest.
LT1021 testpoints
In the diagram I also made connections for measuring each LT1021 (TP-1 to TP-8).
However, I think there is no room in my oven to do this, and what wil I gain... if i measure the output of this 10V referense and its more of than 2PPM i wil know it!
I explanes already, I always check on different DMM's before using, but if you guys uses a part of this schematic for your inspiration, be my guest
If I'm going to apply it at all, it will be 2x an ADS1115 ADC pcb.
Output gain en trim
R100 and R101 set the gain of the output opamp.
Usually I buy a pair of 10K 0.01% Vishay resistors and select them so that the output is as good as possible at 10.00000V.
Then I need very little "trim".
As Kleinstein mentioned, if I am already close to perfect 10V output voltage, I don't need all trim connections of the LT1021.
That is why there is a red cross at each trim line of the LT1021 ICs.
LT1021 current use
I know that the 14 pieces of LT1021 that are aging in my LAB uses 10.1mA unloaded, this is slightly more than 0.7mA per LT1021.
The datasheet tels me that typical it will be around 0.8mA with no minimum specified.
LT1021 Noise
I have not yet taken any noise measurements on the aged 14x LT1021, just the drift of the 5V output.
I will certainly take the noise measurements, to see if there is a "noise diode" between them.
Like always, Shoot at It!
Kind regards,
Bram
-
I2C Power
There is also a 9th LT1021 present, which supplies the power for the i2c components (And DAC Reference).
Because I have many LT1021 5V references, it is also used here in a configuration that can supply a bit more power.
And probably I'm going to use the extended version with current limitation because in this schematic the power supply already comes from the 12V Power Line and it was connected to the 8.5V power line first. when using the 8.5V power line I have a LDO problem.
Hello,
I know you have a lot of spare LT1021.
But I would use a LP2950(A) (or LT1763) for creating 5V.
And I would take the DAC-reference from the (buffered) 5V reference summation point.
with best regards
Andreas
-
Shi....
I lost my last post through browser error... Grrr
-----------------------------------------------------------
Lets start again, Hi Andreas
As pigrew had already shown me, if you use the DAC reference input, it may not differ by more than +-40mV from the 5V power supply you are using for the DAC logic.
By supplying the DAC with a good 5V power supply and connecting the reference input to the Vdd,
I think I have overcome this problem and I have a stable DAC output voltage.
But if i see it the wrong way, please tel me.
I have several LP2950 types of regulators here, and maybe also a sample of the LT1763.
The LT1763 seems beautiful to me for this application and if I don't have a sample, I ask Analog Devices for one te test it.
Thanks and regards,
Bram
-
2) I hear parallel voltage sources will reduce their noise by a factor of sqrt(n), but series voltage noise sources should add together. So, the noise of the references should be sqrt(2)/2 = 71% of a single LT1021-5's noise. So, the (.1 to 10uVp-p) noise of your parallel-series should be about 0.71*3 = 2.1uVp-p. Eight LT1021-10 in parallel should have a noise of 6/sqrt( 8 )=2.1uVp-p. It seems simpler to just use the eight in parallel?
When the outputs of the references are averaged (parallel), the output voltage remains the same but the noise gets divided by the square root of the number of references. When the outputs of the references are added (series), the output voltage is multiplied by the number of references but the noise is multiplied by the square root of the number of references.
In either case, the signal to noise ratio remains the same. In the parallel case, the signal remains the same but the reference noise is decreased and in the series case, the reference signal is increased faster than the noise is increased.
-
The MCP4706 seems to be like many other Microchip DACs: a family of three, with a very good 8 bit, an okay 10 bit, and a poor 12 bit in the same 'family'. They add resolution, but keep the same accuracy throughout. Looking for example at zero scale error:
8bit: 0.13 LSB typ, 2.0 LSB max
10bit: 0.52 LSB typ, 7.7 LSB max
12bit: 2.05 LSB typ, 30.8 LSB max
Same story (but rather worse) on full-scale error.
Have you considered running the 5V logic at 5.5V? You would then have no issues with a 5.00V reference. I'm doing that in an ongoing design of mine (though I have more freedom to do so, perhaps, as the MPU is 3V3 so I need a voltage translator anyway, making it trivial to run the DACs at 5V5). -
As pigrew had already shown me, if you use the DAC reference input, it may not differ by more than +-40mV from the 5V power supply you are using for the DAC logic.
By supplying the DAC with a good 5V power supply and connecting the reference input to the Vdd,
I think I have overcome this problem and I have a stable DAC output voltage.
But if i see it the wrong way, please tel me.
Hello,
I read the data sheet different than you both:
If (and only if) you use the internal (nearly rail to rail) voltage reference buffer of the DAC
then:
this buffer is only able to output from
40 mV to VDD - 40 mV
due to lacking true rail to rail features.
otherwise your external buffer sees a 210k resistor string in unbuffered mode.
Of course you should stay within the VDD+0.3V for VREF so that the input clamping diodes do not get forward biased.
with best regards
Andreas
-
As pigrew had already shown me, if you use the DAC reference input, it may not differ by more than +-40mV from the 5V power supply you are using for the DAC logic.
By supplying the DAC with a good 5V power supply and connecting the reference input to the Vdd,
I think I have overcome this problem and I have a stable DAC output voltage.
But if i see it the wrong way, please tel me.
I read the data sheet different than you both:
If (and only if) you use the internal (nearly rail to rail) voltage reference buffer of the DAC
then:
this buffer is only able to output from
40 mV to VDD - 40 mV
due to lacking true rail to rail features.
otherwise your external buffer sees a 210k resistor string in unbuffered mode.
Of course you should stay within the VDD+0.3V for VREF so that the input clamping diodes do not get forward biased.
This is more or less what I had initially said. However, the "VDD+0.3V" statement is under the absolute limit section, so that only guarantees that the DAC won't die. We should stay within the recommended conditions where in unbuffered mode V_REF <= V_DD. I embedded the table as an attachment to this post.The MCP4706 seems to be like many other Microchip DACs: a family of three, with a very good 8 bit, an okay 10 bit, and a poor 12 bit in the same 'family'. They add resolution, but keep the same accuracy throughout. Same story (but rather worse) on full-scale error.
Have you considered running the 5V logic at 5.5V? You would then have no issues with a 5.00V reference. I'm doing that in an ongoing design of mine (though I have more freedom to do so, perhaps, as the MPU is 3V3 so I need a voltage translator anyway, making it trivial to run the DACs at 5V5).
I like the idea of operating the DAC at 5.5V. The plots show the device is fairly temperature-stable and monotonic. If the goal is to do temperature corrections via the trim, life will be quite hard due to the INL pattern. But, if it's a set and forget type thing which won't be automatically adjusted, this DAC seems acceptable (though not great).When the outputs of the references are averaged (parallel), the output voltage remains the same but the noise gets divided by the square root of the number of references. When the outputs of the references are added (series), the output voltage is multiplied by the number of references but the noise is multiplied by the square root of the number of references.
In either case, the signal to noise ratio remains the same. In the parallel case, the signal remains the same but the reference noise is decreased and in the series case, the reference signal is increased faster than the noise is increased.
Very good point, though I think in this case the LT1021-10 should be used because it has a better specified 10 Hz to 1 kHz SNR for the 10V version versus the 5V version.
-
All 8 LT1021 references are now in parallel, and again the 0.4Hz low pass filter behind it.
Hello,
how did you came up with 0.4Hz value for the filter? when i try to simulate that i get different frequency (see attachment).
Am i doing it wrong? Also shouldnt be the bottom resistor value alot higher for lower leakage?
-
Hi
zhtoor
Aging of the LT1021-5V was verry simple, jus put it onder power for more then 3 years and sometimes measure the output of the references and write it down.
About the DAC
I dit not used this part before, only some of his brothers and sisters.
When I am back in my LAB I will test the MCP4726 extensively.
I chose the MCP4726 from the list because it has a memory function and if there is a "Power Down" it still automatically uses the last calibration value.
Afther testing i will show here how i wil use the component. (Power supply, reference mode, ect)
RR
Thank you for spicing the filter, ik wil check it, i scaled it down from a different application, it is posible i made a mistake.
LT1021-5V
About de LT1021-5V, yes! 5V version, i explaned dat now many times, there wil be no 10V version, also no 7V version, also no LTZ1000 version is this schematic.
I have many LT1021-5V, who aged more then 3 years and i am showing a way to use this references in one of my 42C component ovens.
Calibration function
I wil use a key switch and a extra button to calibrate this reference against my 3458A DMM.
And hopefully i dont have to use this many times.
Thre wil be no automatic calibration say against te temperature, because the hole reference is in a oven...
Only the drift wil hopefully need sometimes to be adjusted.
Logging
I'm going to try to log as much as possible, that is the temperature in the oven with at least two sensors, air pressure in the oven, humidity in the oven.
And also the air pressure, humidity and temperature outside the oven.
Kind regards,
Bram
-
Hi blackdog,
IMO the way Q4 is used for current limiting is not suitable for Q5 JFET. -
Hi,
zhtoor
My log book is somewhere else, so you'll have to wait a bit longer.
That also applies to the tests I still want to do, there is no time now for that...
doktor pyta
You are right, this is far from optimal(or just bad ), I have already been busy designing something else.
It wil be in the next version of the schematic, but I will probably have to do some consessions to the minimum supply voltage of this schematic which is now 12V.
rr
My Spice simulation also show that it is wrong, thank you!
I now have 0.6Hz at -3dB if the upper resistance is 4K and the lower 1K5.
But it all depends on the Nichicon capacitors I want to use, which are from the UKL series. (Low Leakage)
I will test some values in a 42C oven, and because I only need to filter at 5V DC, I can also try some larger values, so the resistor value can be lower.
I still have a lot of decisions to make...
Thanks for all the input!
Kind regards,
Bram -
I have been testing this capacitor series on my bench for the last week.
Nichicon UKL Series capacitor leakage -
At 23 deg C, a 3,300 ufd, 25V part with a 6V bias has a leakage current of 1.2 na and it's still dropping.
This is after a week at 25V to fully form the dielectric. I don't have an easy way to check these at a higher temperature.
Good luck! -
It wil be in the next version of the schematic, but I will probably have to do some consessions to the minimum supply voltage of this schematic which is now 12V.
I would rather keep the additional uncertainties down. You don't really need the x2 stage, as the references can be bootstrapped to 10V output easily (see the attachment). +15V supply can be made by bootstrapping another LT1021-5 from 10V output.
Cheers
Alex