CX Reference

[CalMachine]

I know what you're probably thinking...  Another damn X"X" reference?!

But, after many of poking and prodding from fellow community volt-nutters, I come here with the beginning stages voltage reference project! 

CalMachine EXperimental
Main design choices I wanted to include:
~Implement Main LTZ Reference Circuit with few tweaks.
~Use mostly THT/DIP components
~LTZ1000 non-A ~LTZ1000A
~2x2 Resistor arrays (PWW technology)

End Goals:
~A more stable reference than the original circuit.
~Eventually design PWM DAC implementing this as the primary reference

Current schematic iteration and initial PCB component placement tweaking is attached. 
I am new to, and very inexperienced, with KiCad..  There will be many amateur mistakes made here, so bear with me

This is just 1 of the many VRef ideas I want to pursue.  The thought behind using 2x2 resistor arrays is to cut down on self heating and power through the resistors, reducing their lifetime stress and aging.  I know the drift of resistors are attenuated by many factors... but it is still something I want to try out.  The main reasoning behind using a non-A LTZ is so I can run at a lower setpoint, also decreasing long-term aging.  And also as a side-thought, lower temp setpoint will induce lower thermal gradient on the PCB and in the enclosure.

Comments/questions are welcome as this project is in its infancy.

[ManateeMafia]

Thanks for posting. There's always something to learn from doing it differently.

Have you considered interleaving R4 and R5?

[mimmus78]

With four resistors you are starting playing with big numbers ... so you will get less melons but also less outsiders.


Inviato dal mio Nexus 6P utilizzando Tapatalk

[chuckb]

Good luck with the design!
I would make D3 a 1N400X series diode to handle the surge current for the electrolytic caps.
U1 needs a 0.1ufd cap directly across it's power pins.
As I mentioned in the main LTZ thread a few days ago, installing a zener in the collector of Q1 can help to prevent excess stress on the chip and to prevent damage.
You will want to minimize the distance (copper resistance) between the M2 and the M3 nodes.

Will this be for you lab use only or will it be transported?

[Kleinstein]

I don't think it is worth using 4 resistors each in stead of a single one. This drives up the costs, but with very little advantage. There is not much to improve one the standard LTZ1000 circuit. The important point is to get the layout right so that not too much copper resistance is included and to keep RF noise out. A suitable stable thermal layout can also be important. If at all one might consider the modification that allows more capacitance at the output and thus more filtering there.

Doubling might be an option for the heater transistor: having a symmetric layout with 2 transistors could be used to reduce the possible temperature gradients due to the heater transistor(s). If well distributed, there heat could even be considered a partial temperature stabilization for the whole circuit. Another point is a limitation of the heater power to prevent damage - just in case.

[CalMachine]

Thanks for posting. There's always something to learn from doing it differently.

Have you considered interleaving R4 and R5?

I have not!  I could possibly see that being beneficial..  To help with thermal coupling of the resistors?

Good luck with the design!
I would make D3 a 1N400X series diode to handle the surge current for the electrolytic caps.
U1 needs a 0.1ufd cap directly across it's power pins.
As I mentioned in the main LTZ thread a few days ago, installing a zener in the collector of Q1 can help to prevent excess stress on the chip and to prevent damage.
You will want to minimize the distance (copper resistance) between the M2 and the M3 nodes.

Will this be for you lab use only or will it be transported?

Thank you for the advice!   These changes should make their way into the next revision.

This will mostly sit in the standards lab on the DP scanner and not move.  For transporting, I've got 2 732Bs in the lab.  Once of which is my personal unit (which I got fairly recently) and the other is the lab's main transfer reference that we've had for many years now.

[Edwin G. Pettis]

There will be little to gain from quadrupling the resistor count, particularly with PWW, the power levels are very low and will have no significant effect on them as far as aging goes.  In the case of film/foil, having a very small mass for a resistor element, the effects, however small, would be multiplied due to the very small mass.  In either case, once the unit has achieved equilibrium and stays at a relative constant temperature, everything will settle down to a minimum drift with time, all it takes is patience, it will not happen in months, it takes years for it to happen.  Even accelerated aging has its drawbacks.

[Kleinstein]

The possible lower temperature in the non A version is a two sided thing: Aging is slower, but this also means it takes longer to get a "stable" state. The difference is not that large anyway. With an additional about 300 K/W and a typical power dissipation of the LTZ1000 (without the heater) of some 30 mW it's about a 10 K higher temperature that is needed for the A version. An extra 10 K will speed up aging by something like a factor of 2 to 4. With a relatively high temperature there is a chance to get the fast processes settled after 1 year or so, so that after that mainly a single slow process will remain active and thus drift gets predictable and allows for extrapolation.

The A version allows a lower power consumption, despite of the higher temperature set-point, unless very good thermal isolation around the non A version is used - but this tends to reduce the performance of the temperature regulation.

The temperature gradients on the board are more due to the power level. With the A version more of the temperature gradient is inside the case and less outside. So the A version can help to build a more compact version.
The logical way to fight temperature gradients (especially those than can change) would be a metal case all around the board and a rather compact design. The natural place for the heater controlling transistor would than be mounted to the case.  An always slightly elevated temperature can also help with humidity effects, especially if the unit is always on.

[CalMachine]

The possible lower temperature in the non A version is a two sided thing: Aging is slower, but this also means it takes longer to get a "stable" state. The difference is not that large anyway. With an additional about 300 K/W and a typical power dissipation of the LTZ1000 (without the heater) of some 30 mW it's about a 10 K higher temperature that is needed for the A version. An extra 10 K will speed up aging by something like a factor of 2 to 4. With a relatively high temperature there is a chance to get the fast processes settled after 1 year or so, so that after that mainly a single slow process will remain active and thus drift gets predictable and allows for extrapolation.

The A version allows a lower power consumption, despite of the higher temperature set-point, unless very good thermal isolation around the non A version is used - but this tends to reduce the performance of the temperature regulation.

The temperature gradients on the board are more due to the power level. With the A version more of the temperature gradient is inside the case and less outside. So the A version can help to build a more compact version.
The logical way to fight temperature gradients (especially those than can change) would be a metal case all around the board and a rather compact design. The natural place for the heater controlling transistor would than be mounted to the case.  An always slightly elevated temperature can also help with humidity effects, especially if the unit is always on.

Thanks for the valuable information    Always a pleasure reading your posts.  I kind of like the Q1 mounted to case idea.  How compact would you say 'compact' is?  The current board I have is 150x100mm and it's the nearly the exact size of one of the aluminum enclosures I have.   Thermal gradients would better be tackled with a smaller enclosure?

I don't think it is worth using 4 resistors each in stead of a single one. This drives up the costs, but with very little advantage. There is not much to improve one the standard LTZ1000 circuit. The important point is to get the layout right so that not too much copper resistance is included and to keep RF noise out. A suitable stable thermal layout can also be important. If at all one might consider the modification that allows more capacitance at the output and thus more filtering there.

Doubling might be an option for the heater transistor: having a symmetric layout with 2 transistors could be used to reduce the possible temperature gradients due to the heater transistor(s). If well distributed, there heat could even be considered a partial temperature stabilization for the whole circuit. Another point is a limitation of the heater power to prevent damage - just in case.

I realize a design with 2x2 resistor arrays is going to drive up cost.  I don't care.  I know there is not much to be improved.  Hence, why I'm mostly sticking with the basic reference circuit and very minimal changes. 

I think where the added difficulty from these tweaks come when I delve more into the board layout.  I'm intrigued by the double transistor idea.  I'll look into that more. 

LTZ1000 non-A
... The main reasoning behind using a non-A LTZ is so I can run at a lower setpoint, also decreasing long-term aging.  And also as a side-thought, lower temp setpoint will induce lower thermal gradient on the PCB and in the enclosure.
Comments/questions are welcome as this project is in its infancy.

Hello CalMachine - the LTZ1000ACH has a much better thermal insulation - 400K/W vs. 80 K/W for old LTZ1000CH part - this means you need to pump in less energy in system for -A part vs. non -A to reach same oven setpoint ... ? Are you sure about your above statement in bolt ?

cheers
Butterfly

I'm fairly certain I'm right in that train of thought.  I could always be wrong though. 

Here's what the datasheet says.

Because higher temperatures accelerate aging and decrease
long-termstability, the lowesttemperature consistentwith
the operating environment should be used. The LTZ1000A
should be set about 10°C higher than the LTZ1000. This
is because normal operating power dissipation in the
LTZ1000A causes a temperature rise of about 10°C. Of
course both types of devices should be insulated from
ambient. Several minutes of warm-up is usual.

There will be little to gain from quadrupling the resistor count, particularly with PWW, the power levels are very low and will have no significant effect on them as far as aging goes.  In the case of film/foil, having a very small mass for a resistor element, the effects, however small, would be multiplied due to the very small mass.  In either case, once the unit has achieved equilibrium and stays at a relative constant temperature, everything will settle down to a minimum drift with time, all it takes is patience, it will not happen in months, it takes years for it to happen.  Even accelerated aging has its drawbacks.

You can just call me the Ronco Showtime Rotisserie Oven..  I'm just going to set it and forget it 

[mimmus78]

You can just call me the Ronco Showtime Rotisserie Oven..  I'm just going to set it and forget it 

You know it never ends this way, aren't you? :-)

Inviato dal mio Nexus 6P utilizzando Tapatalk

[Kleinstein]

With the A version the extra temperature gradient will be inside the case. Despite the higher set point the case of the ltz1000A can already be at a lower temperature than with the non A version. The higher internal temperature gradient seems to be no problem at all. The A version seem to have an even lower TC and thus does not need the 400 K external compensation resistor.

I would consider something like the HP ref in the 3458 or the Datron LTZ module (kind of cheating with a special resistor hybrid) relatively compact. If not so large resistors (like the Vishey foil types) and cap form factors are used something like 40x80 mm should be possible. With a cover over the LTZ there should be no need for a really large distance. Not sure if you even need extra slots of holes for the thermal design, but they might help. I would be more afraid of large pockets of air, as this is where convection currents can develop.

One of the few point's where it standard LTZ circuit might be improved would be heater control at low power. With linearization of the power it should be possible to get higher gain and thus possibly better regulation at the low heater power end. This can help a little with performance when the heater is nearly off and thus could extend the working range a little to higher temperatures. I am not sure if it is worth it.  To see how big the possible effect is, one could do a test with a standard circuit: close the upper environment temperature limit, the low frequency noise and TC is expected to get worse. One could check how large that range is where the temperature control is not working well enough. A minimal version could be just limiting the heater power (limited voltage) and based on the limit and maybe the individual heater resistance increase the loop gain (e.g. increase the 1 M resistor and reduce the series cap). The next step would be something like resistor + LED(s) from the heater transistors base to ground, as a very crude approximation of a square root law.

[CalMachine]

Thank you everyone for your valuable feedback thus far!

I've taken into account a lot of the feedback I have received.  As something that can be added relatively easy and doesn't alter the main reference circuit terribly... I've decided to throw in a series zener on the heater side, in an effort to reduce excess stress.  In my next reference design, I would like to tackle the efficiency of the heater side of the circuit, in more depth.

I've also decided to go with the 'A' verson of the LTZ, hoping that it will radiate out less heat (outside of the TO-5 can) and cause a smaller delta from ambient inside the enclosure.  I'm hoping this will create a more stable, more uniform temp inside the enclosure/oven.  I left the temp compensating resistor R9, in case I wanted to throw a non-A reference in.

I've attached my most recent schematic and PCB revision.  Further comments/questions/advice is welcome

[Andreas]

I've attached my most recent schematic and PCB revision.  Further comments/questions/advice is welcome

Hello,

if you really use the BC639 (and not the 2N3904) be aware that the wiring on your PCB is wrong.
(the BASE is not the middle pin on BC639)

with best regards

Andreas

[CalMachine]

I've attached my most recent schematic and PCB revision.  Further comments/questions/advice is welcome

Hello,

if you really use the BC639 (and not the 2N3904) be aware that the wiring on your PCB is wrong.
(the BASE is not the middle pin on BC639)

with best regards

Andreas

Thank you for catching that!  I'll go back to a 2N3904 for now.

[Kleinstein]

It is still odd using 4 resistors each for the precision ones. It usually does not help much in this application. It might if you need higher power rating to lower self heating, but this is not the case here as the critical resistors all see an essentially constant power.

The PCB lines are quite thin in areas where they don't need to, this can add to trace resistance, especially around the 120 Ohms. The sensing lines to the reference usually go right up to the chip. With the 4 mA reference current it's mOhms that can matter.

The temperature sensor should be placed where it matters, so more like at the 13K/1 K divider, but not so close to the LT1013.

The 13 K / 1 K divider is rather close to the heater controlling transistors - this might not be such a good idea, as the transistor is a variable heat source. One can not use the heat from the transistor to compensate for environmental temperature change, as the heat from the transistor and the zener is approximately  proportional to the current and the current is more like changing like a square root law with temperature. So for the transistors there are two good options: one is far away from anything critical - that is what most circuits use. The other good option would be mounted to a metal case that surrounds the whole circuit or at least the LTZ1000 part - in this case the transistor would be a secondary heater for the outer shell. However the contribution of the transistor would not be very large, so not sure it is really worth the effort.

Similar, if no extra driver transistor is used the heat from the LT1013 can be more of a possible problem.

 

[zhtoor]

It is still odd using 4 resistors each for the precision ones. It usually does not help much in this application. It might if you need higher power rating to lower self heating, but this is not the case here as the critical resistors all see an essentially constant power.

series parallel combination of resistor enhances the accuracy spec.
if you connect 100 0.1% resistors in parallel, you get a resistor which accurate to .01%

moreover, series-parallel combination can also help in tempco adjustment of the composite.
ie; one leg having a +ve tempco and the other having -ve tempco balances out.

regards.

[The Soulman]

It is still odd using 4 resistors each for the precision ones. It usually does not help much in this application. It might if you need higher power rating to lower self heating, but this is not the case here as the critical resistors all see an essentially constant power.

series parallel combination of resistor enhances the accuracy spec.
if you connect 100 0.1% resistors in parallel, you get a resistor which accurate to .01%

moreover, series-parallel combination can also help in tempco adjustment of the composite.
ie; one leg having a +ve tempco and the other having -ve tempco balances out.

regards.

Accuracy won't be enhanced by default, only after characterization of each resistor and matching those with opposite properties.

[zhtoor]

It is still odd using 4 resistors each for the precision ones. It usually does not help much in this application. It might if you need higher power rating to lower self heating, but this is not the case here as the critical resistors all see an essentially constant power.

series parallel combination of resistor enhances the accuracy spec.
if you connect 100 0.1% resistors in parallel, you get a resistor which accurate to .01%

moreover, series-parallel combination can also help in tempco adjustment of the composite.
ie; one leg having a +ve tempco and the other having -ve tempco balances out.

regards.

Accuracy won't be enhanced by default, only after characterization of each resistor and matching those with opposite properties.

assuming a usual gaussian curve distribution of resistor tolerances, say 100 of 0.1% each, i would say
the composite (parelelled) would be pretty near 0.01%

and of course, tempco balancing would need characterization, at least 2 bins +ve and -ve tempco.

regards.

[Echo88]

Resistor-accuracy doesnt matter in this application, only long-term-drift, absolute tempco and matching-tempco of some resistors. Noise is sufficiently low due to resistor technology (pww or foil).

[Kleinstein]

The resistor noise also does not matter in this application. It would need rather tiny (e.g. 0201 SMD maybe)  or carbon resistors to get enough noise to only have chance to see the noise. A relative resistor change is attenuated by something like a factor of 100 (can be even more) at the output. The reason for using PWW, foil or good thin resistors is mainly long term stability and to small part the low TC.

[CalMachine]

Resistor-accuracy doesnt matter in this application, only long-term-drift, absolute tempco and matching-tempco of some resistors. Noise is sufficiently low due to resistor technology (pww or foil).

The 2x2 array is not being implemented to achieve a more accurate nominal value of each resistor. 

The first and biggest reason behind the 2x2 arrays is to average the drift of the critical resistors across 4 resistors.  In turn lowering the drift of the equivalent single resistor.  The lower power draw and less self heating result, was more of an after thought.  I understand now that reducing power consumption for each resistor won't make any improvements really.

It is still odd using 4 resistors each for the precision ones. It usually does not help much in this application. It might if you need higher power rating to lower self heating, but this is not the case here as the critical resistors all see an essentially constant power.

The PCB lines are quite thin in areas where they don't need to, this can add to trace resistance, especially around the 120 Ohms. The sensing lines to the reference usually go right up to the chip. With the 4 mA reference current it's mOhms that can matter.

The temperature sensor should be placed where it matters, so more like at the 13K/1 K divider, but not so close to the LT1013.

The 13 K / 1 K divider is rather close to the heater controlling transistors - this might not be such a good idea, as the transistor is a variable heat source. One can not use the heat from the transistor to compensate for environmental temperature change, as the heat from the transistor and the zener is approximately  proportional to the current and the current is more like changing like a square root law with temperature. So for the transistors there are two good options: one is far away from anything critical - that is what most circuits use. The other good option would be mounted to a metal case that surrounds the whole circuit or at least the LTZ1000 part - in this case the transistor would be a secondary heater for the outer shell. However the contribution of the transistor would not be very large, so not sure it is really worth the effort.

Similar, if no extra driver transistor is used the heat from the LT1013 can be more of a possible problem.

 

Thank you for the pointers!  Keep them coming, I appreciate it.  I can see that using the 2x2 arrays is throwing a big wrench in the layout design, lol.  I'm enjoying the process though

How do you feel about the idea of making a cap, like for the LTZ1000, but for the heater transistor?

 

[The Soulman]

It is still odd using 4 resistors each for the precision ones. It usually does not help much in this application. It might if you need higher power rating to lower self heating, but this is not the case here as the critical resistors all see an essentially constant power.

series parallel combination of resistor enhances the accuracy spec.
if you connect 100 0.1% resistors in parallel, you get a resistor which accurate to .01%

moreover, series-parallel combination can also help in tempco adjustment of the composite.
ie; one leg having a +ve tempco and the other having -ve tempco balances out.

regards.

Accuracy won't be enhanced by default, only after characterization of each resistor and matching those with opposite properties.

assuming a usual gaussian curve distribution of resistor tolerances, say 100 of 0.1% each, i would say
the composite (parelelled) would be pretty near 0.01%

and of course, tempco balancing would need characterization, at least 2 bins +ve and -ve tempco.

regards.

The center of the Gaussian bell shape doesn't have to be "on target".
I.e. if the resistors on average are 0,05% low then also the bell is 0,05% shifted low.

[Kleinstein]

....
How do you feel about the idea of making a cap, like for the LTZ1000, but for the heater transistor?

An isolating cap just for the transistor would not help - the power is fixed, so all it would do is to make the transistor hotter.

What would make sense it to add a heat sink to distribute the power a little. The zener in series is a little like this, as some power will be from the zener too. A more distributed heat source would cause less gradients. Still the best place would be away from the resistors. HP put the transistor right at the far edge, with an extra isolation toward the rest of the circuit.

The other extreme would be a metal cap around the LTZ  (or the whole circuit) and mount the transistor (e.g. BD135 in this case) on the metal cap.

[MisterDiodes]

Just some general head's up - and nothing is cast in stone here, just some friendly experienced design tips keep in mind.  They are worth at least what you paid for them.

Hmmm... The quad resistors I really don't think are going to get you anything - as been shown over and over, and in the LTZ datasheet, the resistor drift is -not- driving force behind overall output drift.  You don't want absolute cheap-ass noise-generator SMT thick films for resistors (like a 34470a Vref module), but they really don't have to be ultra-precision as long as you get some reasonable quality TC.  That's the whole point and beauty of the circuit - the resistor values are forgiving, and absolute ohmic value is not too important!!  The most important is the heater resistor -ratio TC- but again absolute value is not too important; you do want to see some good stability and repeatably on those two, any problems on the heater ratio TC will certainly show up on the Vref output.

Everything that is critical and the biggest contributors to over drift are already hermetically sealed in the LTZ can.  The main contributors of overall noise and drift is your luck of the draw on each LTZ die, and it's inherent stress in the crustal lattice at the moment it was born as a single die.  That is beyond your control.

But it's your experiment and path to discovery to enjoy.  You might find out something interesting!  It all depends on your expected output noise and drift rate requirement vs. the practical limits of how every unique LTZ die is going to stabilize over time.

What you do have to watch out for (and I can guarantee can be a problem based on my own bad discovery experience) is making the board bigger than required  and adding unnecessary noise injection antennas.   PWW resistors are excellent low-noise components (And the preferred LTZ datasheet recommended choice of course) but realize the even the PWW reverse-wound winding techniques don't cancel out their self inductance 100%, so you still do want a shielded enclosure (you'll do that anyway for serious work, shielding is another key to low-noise success).

For best chance of success I would go with something similar to the proven, compact board design around the size and general layout like the 3458a module - remember you're after small temperature gradients across the board, and to give that the best chance you want the board smaller, not larger.  I wouldn't make the board so small that you can't lay the PWW down on the board - Trying to install an axial PWW vertically with that one lead waving around loose in the breeze is not the best construction, but it an work in a pinch if no other way.

Pay close attention to keeping the every current loop enclosed area as small as possible - because every current loop is also an antenna for whatever interference comes your way, and every PN junction on a current loop is considered a potential demodulator than can result in Vref noise.

Of course keep in mind Star Ref points, etc.

For all those reasons:  I wouldn't make the board any bigger than it needs to be - you want to go for an optimized, compact design.  At least that's how we do it, and that's what leads to proven success for what we need.  Your needs might be different.

[kj7e]

... The most important is the heater resistor -ratio TC- but again absolute value is not too important; you do want to see some good stability and repeatably on those two, any problems on the heater ratio TC will certainly show up on the Vref output. ...

In an attempt to identify the small TC drift I see in my two LTZ1000A based references, I was using precision heating and cooling of individual components.  I found the heater ratio resistors to be the most sensitive components leading to TC drift.  Seriously considering ordering a VHP200 1K/13K .001% match network.

[Edwin G. Pettis]

The point being, you're actually looking for a stable equilibrium in the environment around your circuit board, no air drafts in the enclosure.  The main reason that heater ratio is so sensitive is that there is a big power dissipation difference between the two resistors, a slight draft is going to wreck havoc.  The cheapest way to get stability in the heater divider is provide good equilibrium with no air drafts, hermetic may slow down the drift under the SAME conditions but eventually whatever is causing drift with regular resistors is going to get to the hermetic too, they are not immune to temperature variations.  The 3458A's Vref board doesn't use hermetic and you don't need to either unless you like to waste money on band aids.

[CalMachine]

After digesting all of the information put forth, going back over ap notes, and looking at already proven layouts, I think I am finally ready to get some boards made!  Any last minute quick changes that would behoove me to make, before boards are made, due to a mistake I've made?   Thanks for all of the advice thus far.

Key design changes/additions
~Finalized board size and first enclosure.  (PCB = 100mm x 50mm.  Enclusure = Hammond 1590B)
~Decided to hold off on 2x2 array until I can get a single resistor design down
~Added Ground Plane and +15V Plane
~Removed Ref_Drive connections (not driving anything)
~Added alternate footprints for Vishay Resistors
~Postive/Negative net star points, both, on front side of PCB


Notes
~Paid close attention to balance and equal out the trace lengths of the most important connections, coming to/from LTZ and star points
~Streamlined component placement
~Attempted to get most efficient layout of resistors around LTZ without copying, verbatim, off another reference layout.


I'm excited to have this project come to fruition and join the LTZ reference club.

[Andreas]

Hello,

some of my thoughts:
- where is the connection between GND and the negative star point?
- The usual stocked vishay resistors have 150 mils pin distance. The 200 mils are not so common.
- how do you mount the PCB within the housing?
- from where do you get the plastic cap? / will it also shield the soldering side of the PCB? Is there enough height in the housing?
- wouldnt it be better to place R4/R5 in direct thermal contact?
- I would put the voltage regulator within the housing (can be on a separate PCB) but D3 should be moved to the input of the voltage regulator.
  (I use LTC1763 as voltage regulators which have a built in reverse polarity protection but are for maximum 20V input voltage.)
- after 3 accidents on unbuffered references (each time loosing 2-5 ppm + a horrible drift the next 6 months),
   I appreciate the output buffer on my newer design.
- no EMI measures? (except the housing, but without line filtering)

with best regards

Andreas

[CalMachine]

Hello,

some of my thoughts:
- where is the connection between GND and the negative star point?
- The usual stocked vishay resistors have 150 mils pin distance. The 200 mils are not so common.
- how do you mount the PCB within the housing?
- from where do you get the plastic cap? / will it also shield the soldering side of the PCB? Is there enough height in the housing?
- wouldnt it be better to place R4/R5 in direct thermal contact?
- I would put the voltage regulator within the housing (can be on a separate PCB) but D3 should be moved to the input of the voltage regulator.
  (I use LTC1763 as voltage regulators which have a built in reverse polarity protection but are for maximum 20V input voltage.)
- after 3 accidents on unbuffered references (each time loosing 2-5 ppm + a horrible drift the next 6 months),
   I appreciate the output buffer on my newer design.
- no EMI measures? (except the housing, but without line filtering)

with best regards

Andreas

Thank you, Andreas!  Catching many of my mistakes

-The PCB will not be explicitly mounted.  It will sit snug (although freely) in foam/padding.
-Plastic cap will be 3D printed by fellow EEVBlog member and friend, MM.  The cap should have a little clearance within the enclosure.
-What advantage would using a voltage regulator over my current lay, bring?
-R4/R5 I was contemplating reworking to get them both vertical next to each other.

Can you elaborate on your output buffer setup?

What would you suggest for EMI measures?  I ran traces in tandem and tried to reduce loop area, atleast the ones I thought were necessary.


Fixed grounding net / start point issue and a few other minor tweaks.

[nikonoid]

Cool design, CM. Not that know a lot about the stuff, but let me through few ideas in anyways.

You are trying to thermally insulate reference from the rest of environment. Out of all materials we are dealing with the air is actually the best thermal insulator. With that in mind you probably want to make sure the foam is not touching the cap. Additionally with cap in place the most of thermal exchange between reference and outside world will be the PCB material. I would route a channel around the reference just inside of the cap or just outside of it. Having less of PCB material there should minimize thermal exchange.
Let me know if this makes sense or if I am completely off base.

Additionally have anyone considered making a dual zone temperature controlled unit? In a other words this board that you are designing would have its own heater/regulator to keep it at 40-50C then foam would go on the outside of it (not inside), to be enclosed by another metal box. This also could allow the use of resistors that have slightly inferior tempco, but possibly other good qualities, like long term stability, noise, price, etc.

[Andreas]

-What advantage would using a voltage regulator over my current lay, bring?
Can you elaborate on your output buffer setup?

What would you suggest for EMI measures?  I ran traces in tandem and tried to reduce loop area, atleast the ones I thought were necessary.

Hello,

PSRR measured on 2 of my units is -0.3ppm/V without voltage regulator.

Possible output buffer attached. (other possibility see DATRON reference cirquit).
R23 is not populated in a 7V reference.
R25,C24 will be added in my next revision to improve behaviour in case of a short cirquit on the output of the buffer.
(be aware that many precision OPs have diodes between their inputs: so above 0.7-1.4V they are shorted against each other).

I use capacitors to do EMI filtering: see LTZ1000 thread
 (but you would also need slight changes against the datasheet cirquit to maintain stability).

Dr. Pyta uses ferrite beads.  (I would add at least one additional into the negative power supply).
https://www.eevblog.com/forum/metrology/ultra-precision-reference-ltz1000/msg941288/#msg941288

Reducing loop area helps against magnetic pick up. But not against common mode noise from the mains line.

with best regards

Andreas

[Kleinstein]

Air is a good isolator, but if the air space is too large, there tend to be thermally driven convection. Convection will depend on the orientation, air pressure and will make a kind thermal noise from fluctuations in the temperature if the air current starts to get turbulent. So it is a good idea to a thermal setup where convection currents are not significant and at least not turbulent. As a rule of thump this mean avoiding air spaces with significant temperature differences that are larger than a few millimeters.

For the board usually the copper traces are more important than the pure FR4. The purpose of cuts is more to reduce mechanical stress. The TO99 case is not very sensitive to board stress though. So one can likely get away without cuts.

A very good insulation the the reference is not needed and not even wanted - the regulator circuit needs some heat flow to work well. The main task is to avoid convective air currents and drafts in a system with a fan.

EMI can be an important point, especially with mobile phones everywhere. However in the GHz range EMI tolerant design can be tricky - down to the black magic level for those of us not used the RF signals. Ferrite beads and feed-through caps are likely a good idea. However the output signal is kind of sensitive to capacitive load, which makes EMI filtering even more tricky.

[CalMachine]

~* Update *~

I sent OSH park my latest revision and received it back a few weeks ago.  I've been waiting to receive resistors from Hifi.  Hooray snail mail from China!   Without realizing the PCB thickness would be reduced, I sprung for OSH park's extra copper fill (2 oz instead of 1 oz).
These boards are very similar to the boards Datron used for their reference modules.

Attached are some pictures of the unpopulated board, populated board in it's housing, and a 1 hour graph monitoring output and ambient temperature.  Blue line is Voltage measurement, Yellow line is temperature measurement via BME280 within 15 cm of the CX box

I don't currently have a cap on the LTZ, so I think the ~0.2ppm/ºC I'm observing is not too bad.

[CalMachine]

I was able to snap a few thermal images with a Fluke Ti9 that is currently here. 

I know not all of the materials here have the same emissivity, so there isn't a whole lot of concrete data that can be drawn.  But, you can still locate hotter/colder spots on the board.

[cellularmitosis]

Possible output buffer attached. (other possibility see DATRON reference cirquit).
R23 is not populated in a 7V reference.
R25,C24 will be added in my next revision to improve behaviour in case of a short cirquit on the output of the buffer.
(be aware that many precision OPs have diodes between their inputs: so above 0.7-1.4V they are shorted against each other).

I use capacitors to do EMI filtering: see LTZ1000 thread

Andreas, I'm looking to add a similar output buffer to my design and wanted to understand your design.

I started with this (first attachment).

The 1k resistors were added because of https://electronics.stackexchange.com/a/56736

Considering your comment about op amp input diodes, the 1k isn't ideal (7V through 1k into a short is 7mA, and we probably don't want to load the LTZ that heavily) -- is that why you went with 10k?

Can you talk about C24?  I see it forms a low-pass filter with R25, but is it a concern that its energy would be dumped into the op amps input diodes if the op amp output were shorted?

Is R21 there because op amps don't like capacitive outputs (C23)?  And is C23 there to absorb EMI in the measurement leads?

What is C22 for?  To slow down the loop?  Or is that for EMI as well?

Thanks!

[Kleinstein]

An important function of the resistor is some additional filtering. This is higher frequency noise of the reference circuit and those high speed current spikes from the chopper amplifier. So an additional resistor to ground at the OP is likely a good idea.

With the LTZ reference circuit load current comes from the OP, not the reference itself. The LT1013 should be OK with 7 mA worst case. The trouble with too much load is that if the voltage drops, the temperature regulation will go to a higher temperature, possibly too high. So it might be a good idea to have a power limit for the heater, just in case and startup.

[Andreas]

Hello,

I would not load the LTZ output with more than 2 mA.
So the sum of the feedback and +Input resistor should be around 4K in sum at least.
1K for both is lower than that.
The 10K value is mainly for having a possibilty of a 7->10V transfer together with the 25K Z201 Resistors that I already have.
And both resistors should be equal to equalize the bias current errors.
C24 is intended as low pass and EMI reduction. Its energy into the diodes is limited by R22.

Yes R21 is the standard measure together with C22 to isolate capacitive loads. (see any old analog data book).
(Adapt the values when you want to use different amplifiers).

C23 is against EMI. Without it I had several tens of uV difference between unbuffered and buffered output.
(the offset of the LT2057 is much lower).

C22: see above.

@Kleinstein: try the 7mA with your own reference before writing (and do not tempt others to do silly things).

with best regards

Andreas

[CalMachine]

I think it's about time I've posted some measurement results of the first spin of CX Reference.  Finally have the equipment to allow me to properly test CX at home! 

There was a few weeks there of headache due to wild TC's and instability because of the series zener I attempted to try to limit the power the heater would see at turn-on.  That zener was removed and the circuit fell back more in line with how it should perform.  In hindsight, I think I chose the wrong zener and the heater was being starved of the much needed current to properly ovenize the buried zener.

I'm observing ~0.15 ppm/ºC.  You can also see the TC of my measurement setup rearing its ugly head when my heat kicks on!  Interesting stuff     Next will be TC adjustment attempts

Plot 1: ~8hr run of Vref vs. Time

Plot 2: Same run Vref vs. Box Temp

[cellularmitosis]

There was a few weeks there of headache due to wild TC's and instability because of the series zener I attempted to try to limit the power the heater would see at turn-on.  That zener was removed and the circuit fell back more in line with how it should perform.  In hindsight, I think I chose the wrong zener and the heater was being starved of the much needed current to properly ovenize the buried zener.

Since I've implemented the heater zener as an option on my board as well, I'm interested in which zener you saw problems with.  I initially went with a 3.3V zener on a 15V VCC, which is pretty conservative.

[CalMachine]

There was a few weeks there of headache due to wild TC's and instability because of the series zener I attempted to try to limit the power the heater would see at turn-on.  That zener was removed and the circuit fell back more in line with how it should perform.  In hindsight, I think I chose the wrong zener and the heater was being starved of the much needed current to properly ovenize the buried zener.

Since I've implemented the heater zener as an option on my board as well, I'm interested in which zener you saw problems with.  I initially went with a 3.3V zener on a 15V VCC, which is pretty conservative.

I had it populated with an NZX2V4.  Don't ask me why I chose that one back when I was putting the circuit together... 

https://assets.nexperia.com/documents/data-sheet/NZX_SER.pdf

[cellularmitosis]

That's only a 2.4V zener, so that shouldn't be a problem unless you are trying to run the circuit on a fairly low VCC.

You can measure the voltage drop across the 2N3904 transistor, and if it is greater than 2.4 + 0.6, then the zener isn't the problem.

Edit: oops, I meant to say that if you short the zener, and can then measure more than 2.4 + 0.6 across the transistor, then the zener isn't a problem.  Alternatively, if you measure more than 0.6V across the transistor with the zener still in-circuit, then the zener isn't the problem.

[CalMachine]

I'm sorry by the time you suggested the measurements, the diode had been removed. 

As of now, I've tweaked the TC of the board to <0.1 PPM/ºC.  I was observing a few orders of magnitude greater TC with the diode in place.  I had played with input supply voltage up to 15V while testing the board in ambient.  I, however, did not test input voltage levels during TC tests.

Data to come soon!

[CalMachine]

I've done some TC tweaking to the first revision of the CX layout.   This board has a normal LTZ1000 populated and started off with 402k for my R9.  Calculator value in screen caps is a ~ppm/ºC estimate.   As I increased the R9 value, my TC slowly dropped.  I ended up stopping at ~ 0.08 ppm/ºC.


For all of the graphs.  Ambient Temp, RH, Box Temp and 3458A temp are on the right secondary Y-axis.  Voltage measurements on the main Y-axis

Nominal R9 @ 402k


@ 470k


@ 560 k


@ 1 M


@ 2 M


@ 3.3 M




Rev 2 of CX reference is currently undergoing TC tweaking.  Here is a screen cap of TC data  with R9 value @ 150k.  Observing some funky jumps and hard to determine cause due to little apparent TC.




Still have Cells 1 and 2 of rev2 to tweak the TC on while I'm currently working on the layout of rev3 (which includes LDO and 10V booster)

[cellularmitosis]

Fantastic!  I think this is the first post I've come across with this much detail on fiddling with the 400k resistor.

Any chance I could trouble you to render those graphs using the same Y scale?

[kj7e]

At 3.3M you had the best results, did you test it with out the resistor populated?

[CalMachine]

Fantastic!  I think this is the first post I've come across with this much detail on fiddling with the 400k resistor.

Any chance I could trouble you to render those graphs using the same Y scale?

How about I throw you the CSVs?  It takes awhile to do these large data sets on my  laptop, here.

At 3.3M you had the best results, did you test it with out the resistor populated?

I did not.  I figured I was getting pretty close to open that ~0.08 ppm for a first rev board was pretty good..  and I wanted to get testing on the rev2 boards

[cellularmitosis]

Oh, nice, I see the CSV's here: https://xdevs.com/cm/teckit_test/

[CalMachine]

Oh, nice, I see the CSV's here: https://xdevs.com/cm/teckit_test/

You found my little stash

Here are pics of my tecbox setup, as well!  For those that are interested.

https://xdevs.com/cm/TECBOX/

[Pipelie]

the funky jumps look like have something to do with the room temperature and the TC of 3458.
did you turn on your AC?

[CalMachine]

the funky jumps look like have something to do with the room temperature and the TC of 3458.
did you turn on your AC?

The small oscillations are definitely coming from the TC of the 3458 in the test setup.  It's been in the single digit ºC here, so I've got my heat on.  Some of the abnormal offset jumps could be from my dishwasher or clothes dryer turning being used during the test.  I've not had the setup going long enough to really weed out my environmental disturbances.  Also, the RPi time stamp is off by a few hours.

[Andreas]

Fantastic!  I think this is the first post I've come across with this much detail on fiddling with the 400k resistor.

Any chance I could trouble you to render those graphs using the same Y scale?

Hello,

I fear you should read the whole LTZ1000 thread.

A summary about T.C. trimming (for the A-Device) is here:
https://www.eevblog.com/forum/metrology/ultra-precision-reference-ltz1000/msg999857/#msg999857

with best regards

Andreas

[CalMachine]

Happy new year!

This project was stalled for a majority of the summer and fall, but I've been slowly working on it on occasion.  Currently up to rev4 and it includes

-Power Regulator Section LT3042
-ADA4522 OP for 10V booster

Currently undergoing TC trimming and stability tests!  More to come 

[vindoline]

Looking good!