10V Reference (I Did It My Way) No1

[blackdog]

Hi,


I'am sorry...
That i used a good quality product, from the 1990's, i'am a bad boy 

The next schematic that I present here, will only have components that are not older than 6 months. 

Kind regarts,
Blackdog

[babysitter]

Preferrably Infineon, which have at least had enough short engineering-sample-to-obsoletification-cycles for me to realize, so if people are not fast they will immediately be punished

[branadic]

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

In addition, you can also use thin base material such as 0.1mm FR4 or Polyimid the reference is soldered to and solder this flexible pcb to your main pcb or use a semiflex pcb, with the flexible part for the reference. This ensures a much smaller stress relief because of less swelling of the pcb by humiditiy, less thermal mass, but causes maybe other problems. However, this solution is more process reliable.

branadic

[babysitter]

Instead of going for a quite expensive flex-rigid processed PCB, consider using cheap flat foil cable.
(Says someone who finally decided for a flex-rigid pcb at work after starting his design with ffc)

[JohnnyBerg]

is this a viable way to do the buffer/pre-buffer? (if assuming, i dont need 10v)

Hmm .. you certainly have me confused .. what is this supposed to do?
Is it some kind of noise generator?

Generally speaking when doing voltnut stuff: "less is more"

[JohnnyBerg]

is this a viable way to do the buffer/pre-buffer? (if assuming, i dont need 10v)
**edit (or as pointed out by johnnyberg, maybe without the quad opamps?)

Hmm .. a opamp as a voltage follower behind a reference will add offset, offset drift and noise to the reference voltage.

When amplifying the voltage, you get the TC of resistors added to the TC of the reference. You need very expensive resistors, or put the whole thing in a oven.

[T3sl4co1l]

Any time you see "three terminal model", it means the output is unhindered by conventional notions of things like "power supplies", or "breakdown voltage of the quantum vacuum".

Tim

[babysitter]

Hmm .. a opamp as a voltage follower behind a reference will add offset, offset drift and noise to the reference voltage.

Did right that with a second LT1013 as a x1 buffer for my LTZ1000A (see in the ultra-thread), thus preventing heater going berserk if I ever short it out. Influence is small. You even can find it in the datasheet how small you can expect it to be. Comparisons against a bunch of unbuffered LTZ's showed, if all, contributions to a very small overall impact.

[blackdog]

Hi Group.... 

I've done a number of tests on the LT1236-5V references.

This version => LT1236A-5V LS8


I placed them on a smal oven to heath them to 42,5C and back to room temp.
I have some pictures so you guys can see how the LT1236 behaves when cycled between these temperatures.

The first image is LT1236 number-1, and shows the drift when cooling down.
At the beginning of the chart is power of the oven is turned off.
At the end of the chart where i stopt measuring, the temperature is cooled down to 29 Celsius.
The temperature sensitivity from this LT1236,  is a little more than 2PPM / C



LT1236A LS8 Number-4
Sorry different style of the picture.
Look at the middle of the chart, there i started the oven.
In the middle to the end is about 90 seconds (fast oven!!!)
Afther 90 seconds the temperature is stable and you can see there is a hump in the graph.
This graph is totally different, the other graph was linear and this looks like a camel 
At the beginning of the chart, the power of the oven is switchd off.
At the end of the chart where i stopt measuring, the temperature is about 42.56 Celsius.
The temperature sensitivity from this LT1236,  far below 1PPM / C


Next time also some pictures of the "oven" i used.

Kind regarts,
Blackdog

[JohnnyBerg]

Are those LT1236's from the same batch?

[blackdog]

Hi,

Not all LT1236 references are measured, it cost a lot of time...
I try to find a good way to do these measurements.
If I know how I can do good measurements, which will be done on 10x LT1236-5V IC's, with half of them from one SMD plastic strip.
You guys will have to wait until I'm done. 

Kind regarts,
Blackdog

[Andreas]

LT1236A LS8 Number-4
Sorry different style of the picture.
Look at the middle of the chart, there i started the oven.
In the middle to the end is about 90 seconds (fast oven!!!)
Afther 90 seconds the temperature is stable and you can see there is a hump in the graph.
This graph is totally different, the other graph was linear and this looks like a camel 
At the beginning of the chart, the power of the oven is switchd off.
At the end of the chart where i stopt measuring, the temperature is about 42.56 Celsius.
The temperature sensitivity from this LT1236,  far below 1PPM / C

Hello blackdog,

the typical T.C. for the LT1236 is around 2-4 ppm/K (relatively linear but with a slight hysteresis).
From my measured 20 LT1236 I have only 2 which are deviating to this rule.
So when having extremely low T.C. I would consider a mounting anomaly of the chip within the housing.
(perhaps too much die attach or a edge of the chip which has contact to a sidewall).

The one of these low T.C. references (#2) I had large initial ageing drift. Now I am using this chip in one of my ADCs (ADC17).
(I really do not know if I can trust this candidate or if he will start larger ageing drift some day).
The other (#8) has still large 0.1-10 Hz noise. About a factor 2-3 more than the usual 2-3uVpp.
(see oscilloscope photo with 2 uV/DIV and 10 sec sweep time).

With best regards

Andreas

[blackdog]

Hi Andreas :-)

At the moment i have only 10x total LT1236, in a and b version.
The problem is that you are never sure of how a component behaves over time ...
Check, check, check over time, References are a Bitch 

Wat kind of preamp are you using to measure the noise, one of Jim Williams designs?
Gain 60, 80 100dB?

OK, Picture time!
Let me start first with the oven, it is built on 2x a BPR10 resistor (50 Ohm).
On the picture below you see 5 pieces and i use in this oven 2 pieces.
I remove the connection pins and glue on each site a thin double sided printed circuit board (0.25 mm thick board)
This gives some extra thermal mass and it is good solderable.
The copper of both board (top and bottom) helps distribute heat very well.


This is one of my designs for a oven controler, this one, is using a TI, INA125 Chip, little expensive,
but you have good control over characteristic's (compensation)


The LT1236 are powerd bij a 9V battery.
The breadbord holds the decoupling capacitors and a easy way to connect my testclips.


Close up...
There are three PT1000 sensors, the sensor in the center, goes to the TEK DMMM4050 multimeter to measure the temperature of the oven.
The two other sensors are connected in series and are for controlling the oven temperature.
The "big" red cap is for power supply decoupling.



This is the oven controler "Manhattan Style", on the right side of the sensitive components are buried under a layer of thermal adhesive (WLK30 Fischer)
At the bottom left, one can see a small power FET, which regulated the current through the resistors.



LT1236 Pictures, Power Up and Power Down
I measured six LT1236, as the oven heats up very quickly, this measurement takes only 1 minute.
In the first picture you can see that there is a hump in the graph.
This little anomaly is due to the oven current wich drops when reaching the selected oven temperature.
This has nothing to do with the LT1236, it is a measuring abberation.
In the first picture i placed the temperatures at two points, in the middle that is about 36C and at the little hump, thats about 41C.

Power Up LT1236-01


Power Down LT1236-01


This is the oven current, you can see, its nicely damped.
And the current drops at the same time the is a nanomy in the reference output.



Power Up LT1236-02


Power Down LT1236-02



Power Up LT1236-03


Power Down LT1236-03



Power Up LT1236-04


Power Down LT1236-04



Power Up LT1236-05


Power Down LT1236-05



Power Up LT1236-06


Power Down LT1236-06



It is clear, LT1236-6 is a BAD BOY!
Yes its whitin the specs, but not as nice as the other ones.
With the 34461A multimeter, it is not possible to measure hyteresis at the 20 degrees temperature variation,
these variations are too small at the six measured LT1236 references.

Shoot @ it! 

Kind regarts,
Blackdog

[Andreas]

Hi Andreas :-)

Wat kind of preamp are you using to measure the noise, one of Jim Williams designs?
Gain 60, 80 100dB?

Hello,

I was inspired by AN124 but did a much simpler design.
Just 2 simple OP-Amps with a 4rth order Bandpass. And selected standard 85 deg C capacitors.
80 dB together with 2mV/div lowest resolution on my analog oscilloscope give 0.2mV/div maximum sensitivity.
Noise floor is around 0.12uV rti.
So the noise of a LTZ1000 can be easily measured.

But you will have to add a series resistor (3k6) and a switch (closed after charging) to charge the capacitor in order not to kill the LTZ1000 by the inrush current. (heater setpoint will go to "infinity" when the output is loaded by more than about 2mA so the device ages very fast).

With best regards

Andreas

[Andreas]

With the 34461A multimeter, it is not possible to measure hyteresis at the 20 degrees temperature variation,
these variations are too small at the six measured LT1236 references.

Hello,

try harder !
Perhaps you have to use a additional temperature stabilized reference and measure the difference with a higher resolution. (in the 100mV range).

On the other side you would have to stay at minimum for half an hour or so on the hot temperature so that the die attach on the reference can begin to creep. Half an hour is about the time constant that I found at least on my #02 sample.

With best regards
Andreas

[blackdog]

Hi Andreas, :-)

Thanks for the schematics of the measuring amp!
I wil built one this weekend!

Tomorrow I'll show my adapted schematic of your preamp.
And I understand your comments on the load of the reference measured.

But, I have a question, I do not understand the filter at the output of the LT1012.
Strange corner frequenties...
Can you explain this for me?

Thanks!
Blackdog

[Andreas]

Can you explain this for me?

C6 + C10 are only for EMI supression (the oscilloscope has a mains line) no filtering function.
All other parts including the reference are battery supplied.
And do not forget the (metal) cookie box. (or a paint tin can if measuring TI devices).
I use a additional TEKO aluminium box for the amplifier.

with best regards

Andreas

[blackdog]

Hi Andreas, :-)

First of all, i will try harder for the hysteresis effect... 

OK noise amplifier question!
I'am talking about R12, C10 (1,6Khz) and after R12 there is a resistor 2k2 or 1k? in series with C8?
I know that C8 and R10||1Meg scoop resistor, makes a ~60mHz filter.
But why the two resistors in series?

At the moment i'am testing some capacitors, but its difficult to find some with low leakage currents.
What brand and type you've selected??
I tested already 10 pieces, the lowest  1000uF, was approximately 60nA...
Afther some time formatted at 15V.

Kind regarts,
Blackdog

[Andreas]

But why the two resistors in series?

At the moment i'am testing some capacitors, but its difficult to find some with low leakage currents.
What brand and type you've selected??
I tested already 10 pieces, the lowest  1000uF, was approximately 60nA...
Afther some time formatted at 15V.

Hello,

Try it out. Short the resistor comming out from OP-Amp to the 100nF capacitor.
You will need at least 2 days forming/formatting? at >= 10 V.
And do not heat the capacitor during soldering.
It´s also a good idea to reform the capacitor 2 days before measuring.
(otherwise there might be some popcorn noise).

I used just that what I had in the drawer. (2200uF Elna NOS and some 1000uF YAGEO?)
Higher voltage rating than needed could be better.
From 10 capacitors there should be 2-3 which meet the spec.

By the way: for decoupling the amplifiers I use 1000uF + 100nF for each supply.
And between pre-amplifier and end amplifier 100 Ohms for further filtering.

With best regards

Andreas

[blackdog]

Hi Andreas,

I'am aware about the load a opamp can't handle, R12 and C10 are clear to me.
But why is there a resistor between C10 and C8?
Is this resistor 2K2 or 1K?
RFI filtering was not needed?, chokes ect.

Kind regarts,
Blackdog

[Andreas]

Hello,

Resistor (name = 2K2) with value 1K (as can be seen on the photo) prevents together with C10 noise comming from the mains line to be demodulated on the OP-Amp output.

I rather use resistors than chokes if possible. They have lower parasitic capacitances.
The whole cirquit is double shielded (cookie box and TEKO A3 enclosure).

With best regards

Andreas

[blackdog]

Hi All 

I've been a little busy...
Now an update on the changes of the Quad LT1021 reference.

First update of the schematic.


Link to the big schematic
www.bramcam.nl/Diversen/Quad-LT1021-10Vref-03.png

If anyone is interested in more than 50 hours of data, below are two files for the voltage and temperature in the OLD situation.
10V data, there is one big spike, sorry ;-)
www.bramcam.nl/Diversen/Quad-LT1021-51Hour-10V-2015-03-07.csv

Temperature data, the sensor is between two of the four LT1021 references, see picture below.
www.bramcam.nl/Diversen/Quad-LT1021-51Hour-Temp-2015-03-07.csv

This is the old situation of the reference circuit board.


This is the new setup, the 2x opamp is now a LTC2057, you can just see him, in the blue circle.
The temperature sensor is mounted between the middle two LT21021 references in the red circle.



I've tried various circuits for the oven controler circuit, but I went back to the LT1006 circuit.
The problem is not the LT1006 controler, but the mechanical layout of the Power FET and the two heather resistors and the sensor place.
If the oven is cold, there is about 1.2Amp of current through the resistors and there is almost no voltage over the Drain/Source of the Power FET.
The sensor is glued between the two resistors and that is OK but...
When the oven is reaches its equilibrium, most of the voltage is across the FET and de place of de sensor is now not optimal.
The distance is about 3cm and the thickness of the aluminum is not enough for good thermal coupling.
When the oven is powerd up, the resistors dissipate about 0.1 Watt and the FET about 1.2 Watt!


Between the two Rifa 6,8uF and the blue 150uF capacitor is the Power Fet.
On the right side you can see the two Vishay 5 ohm power resistors, and in between, the sensor LM335.
The yellow ntc on the right side, was only for testing, its not there anymore.



And what can I do now to improve the oven...
Make the bottom of the oven thicken, so give a lower thermal impedance.
I orderd some sheet copper, 0.1, 0.8 and 2mm.
And have a piece of copper cut out the size of the bottom of the oven.
Here you can she the piece of copper, and the polystyrene box, fore size reference i placed a 9V battery on the oven.



Yes, 2mm of "Low Impedance" 



This is a second use of isolation transformers 





Is it now a good oven, it is OK but i will not use this setup anymore.
Later more about ovens and the problems.

Some more pictures, this one is about 150 seconds of measuring time.
1uV/Div.


Almost three hours, After 30 minutes, the sun began to shine in my LAB, the voltage went up.
This can be the reference drifting, but i think, a big part is the 34461A DMM :-)
5uV/Div



Shoot at it!

Kind regarts,
Blackdog


PS
Temp and Voltage a little more then 3 hours.

[JohnnyBerg]

Nice pictures, again!

I have done some experiments with ovens lately. Of course, they oscillate like hell 

From your descriptions I read that the location of the sensor is critical. But what if you have no freedom on the place of the sensor.
It is on the die, for example?
A lot of bandgap references give a PTAT signal for free, something we Dutch like a lot 

I noticed that your temperature regulation is a simple P regulator. I guess, by using a PI or even a PID regulator, sensor location will be less critical and the regulation more stable? I remember the professor saying that for a P regulator to work, there must always be a error signal, eg a difference between set and feedback.

Furthermore, you test a lot to see if you got a long term stable situation. My experience is that when cycling the temperature, thermal hysteresis is a very important factor. 10ppm difference after a cycle is not unusual. Unless your reference is 24/7 on, how do you solve this?

[Andreas]

Hello Blackdog,

for ovens I usually use 2 temperature sensors.
The first immediately at the heater.
(fast response for the P-controller).

The 2nd is at the point of interest.
(slow I-Part for the I-controller).

Usually I use the I-part to tweak the setpoint of
the P-controller so that the final temperature is reached
at the point of interest.

with best regards

Andreas

[blackdog]

Hi,

Thanks for the remarks.

Most of the ovens i make, are small like the other topic on this forum.
https://www.eevblog.com/forum/projects/small-oven-for-2x-lt1236-%28python-reference%29-no1/

The small ovens almost never need capacitief compensation in the loop, this is because the good coupling between the power source and the sensor.
Also becouse there is only a little thermal mass.

JohnnyBerg
When the TI REF5050 came out, i played with the temp output, but you have to use a smal thermal mass and good coupling between the REF50550/REF5010 and the heather.
Someting like a TO220 transistor.
Cut of the thick wires of the TO220 housing, and use thin wires.
Do not mount the TO220 connections on a print, you wil lose so mutch energie and it wil be inposible to get is stable :-)

This is one of the controlers i designed, and i like the most, simple and good controle over the setting.
R1 in the schematic controles the temperature and R3 sets the gain, the do not interact like most of the schematics you see.
Use max. gain as posible, without oscilation.
R4 and R5 keeps the output relatively in the middle of the commonmode range of the INA125.
If you want to use a transistor for the FET, no problemo, just change R4 and R5.
And if you dont like the BPR10 resistor, kick it out, and use only the transistor as the heath source.
I use about 10x the power as Max to setup the current limit at normal LAB temeprature.
I am not designing for military use, in my LAB its between 18C in the winter and 30C in the summer.



Some pictures of oven design...
This is a piece of ALU tube, the wall is 5MM and you can see a small hole wherethe NTC fitted in, not tested yet.



This is a bad way to do it, a small sheetmetal box...



This is a project i stil have to finnish, a 1Khz AC reference, yes, in the middle a LT1021 as a DC reference.
On the right the LM10 oven controller and on the left you can se a small ceramic TCXO for the 1Khz reference generator.
On the left en right site of the LT1021 you can see a PT1000 sensor for controling the temperature of the oven.
It is al mounted on a big sister of the BPR10 resistor.
On the other site is the control circuit for making 1Khz 1V 0.01% and the distortion in below 0.001%, the 1 Khz wil be within 1PPM.


Maybe this week i finnish this setup of a oven, i uses 4 powerfets to distribute the heat for the oven.
I hope that is wil not be a 0.1Hz oscilator 
The black and the white twisted wires are both 5K NTC resistors for controling the oven.



This alu tube is 4mm thick and on top is a thin layer of FR4 mounted.


This are the last ones i show today, smal thick wall alu box.
The design is such that it there is no setting for the temperature.
It wil be whitin 41 tot 44C by design, sweetspot around 42C.
Im made a tester, to test the temp sersor before mounting it in the box.
You can see the sensor in the middle bonded to the aluminum.
This is stil a design in progres, and maybe i wil sell is as a product.



Hysteresis
I know, for my own use, i select the parts.
This week i received more LT1236A references and i am testing these.
It is difficult to see the hysteresis on a 34461a and i want to ask Andreas how he is using the 3458A DMM.
I have a measuring computer W7 USB-GPIB of HP and a PCI HP GPIB card a modern one.
In the Agilent IO, i can talk to the 3458a, simple commands...
I also installed IZGPIB on this system, but i have yet to learn the command structure.

So Andreas can you kick me in the wright direction so I can make pretty graphs with the 3458a?
If the output is a .cvs, then i am happy Excel wil do the rest.

Kind regarts and thanks,
Bram

sorry for the many grammatical mistakes, strong headache today...