Thoughts on voltage reference board?

[c4757p]

I've been working on building replacements for some of my old, bulky equipment (mostly just for fun), and the current contestant on Who Wants to Be Replaced is my Power Designs 2005 precision power supply. It's specified at 0.1% ± 1mV with five digit precision to 20.000V and 10ppm/°C; I'd like to at least match that, adding computer control, multiple outputs (one high precision and two lower precision, all operating off the same reference, part of my recent efforts related to automated testing) and better constant-current accuracy.

Here's what I have for the voltage reference board. Does anybody have any suggestions or corrections before I make it? Some of the parts are pretty expensive (free samples, hell yes - but I've only got one of each), so I don't want to screw it up. I'd especially rather not have to desolder the reference chip once it's been soldered on.

The following parts (at least - depending on what I can fit, I haven't done the PCB yet) will be under a small metal can with a heater, which is then surrounded by insulation, for thermal stability. (Gilding the lily a bit, most of this has good specs over a full typical ambient temperature range.) It will be kept at 50°C by an external PIC running a P-I algorithm.

- U2 (LM317, powers the reference) and its feedback resistors
- TL431 (powers U3 - missing a ref des... oh well)
- U1 (ADR4550 voltage reference, 2ppm/°C).
- Output voltage divider and trimmer.
- U3 (MCP6V06 output chopper amp).
- Of course, U4 (LM35 temperature sensor), and Q3 and Q4, the heaters.

Keep in mind, I've chosen some of the parts simply because I have them, so I know some of it isn't the most ideal design. Still, if my calculations are correct, I should be able to very easily achieve 1-3ppm/°C and hopefully better than that.

Edit: Removed the old schematic to not waste too much of Dave's server space. See below for updated.

[c4757p]

I may toss the temperature controller PIC on the board too, outside the can. I didn't want anything digital on it, but with the grounded metal can surrounding the oven area, it's pretty much a separate PCB-within-a-PCB, and that would make testing easier since I can get it up to fully regulated temperature by itself without having the main board attached.

[GK]

I'm pretty sure the TL431 is intollerant of capacitive loads (oscillates with a parallel capacitor).

[c4757p]

It's intolerant of capacitive loads between about 10nF and 5uF (depending on voltage and current), but loads much larger than that are OK.

Updated to include PIC.

Of particular interest to me is the trimmer network for the reference. I borrowed that arrangement from the TI datasheet for LM399. They don't address why it's like that, but as far as I can tell, it's to compensate for the shit tempco of the trim pot. While the total resistance of the pot may drift, the proportion should not, and the dependence here on the total pot resistance is small. (I may decrease it further to 10k to decrease the dependence even more, but I want to keep the output current from the ADR4550 low.) Is this correct?

Edit: Removed the old schematic to not waste too much of Dave's server space. See below for updated.

[quantumvolt]

...
Of particular interest to me is the trimmer network for the reference. I borrowed that arrangement from the TI datasheet for LM399. They don't address why it's like that, but as far as I can tell, it's to compensate for the shit tempco of the trim pot. While the total resistance of the pot may drift, the proportion should not, and the dependence here on the total pot resistance is small. (I may decrease it further to 10k to decrease the dependence even more, but I want to keep the output current from the ADR4550 low.) Is this correct?

Yes. The op amp sees 5 k from the two 10 k resistors, but more than 68 k from the trim. So the effect is small. If you don't like the full swing of the trim, you can change it for two (possibly different low tempco) resistors and a smaller trim.

In general, imo your design cannot be flawed. All you do is hooking up the parts in ways that have been done over and over again. No fancy analog regulating or feedback circuitry. You might have to change/add/remove small bits if you want to optimize the circuit. But that is normal.

If the 7.5 V is solely an internal supply I don't see the point in protecting it unless you will plug in modules with possibly charged capacitors. Also the 10 Ohm low pass filter with the small 100 uF looks special imo.

[c4757p]

Thanks for looking at this!

If the 7.5 V is solely an internal supply I don't see the point in protecting it unless you will plug in modules with possibly charged capacitors. Also the 10 Ohm low pass filter with the small 100 uF looks special imo.

Diodes are cheap, and I don't want to try to guarantee that the 12V coming in won't be shorted, since I'm not 100% sure what will be on the main board.

"Special"? You mean excessive? Probably.

[blackdog]

Hi c4757b :-)


Some remarks about your schematics.

1e
Keep the power of the heater circuit/Pic separate from the reference!

2e
Make R6 as high as posible, just a smal range so yo can Cal. your reference.

3e
There is no protection on the output of your Reference, you WILL kill it with static and or power from another circuit 

4e
To mutch decoupling (this is the first time i ever saying this) 
Lose these components, R10-C18, C1-L1, R3-C12, make R5 330 Ohm.

5e
Keep the heater controler (Pic) as far as posible away from the reference circuit.

6e
Generate the heat as close as you can to U1.

7e
Mount the LM35 as close as you can to U1 and the heat source.

My 2 cents.

Kind regarts,
Blackdog

[c4757p]

1. The PIC power is separate - the PIC, the temperature sensor and the LMV358 are the only things on the +5V rail. Do you really think the heater needs to be separate? It's not pulsed, it's run in a steady constant-current mode, so once the controller settles it shouldn't be putting noise on that rail. I'm not sure how separate I can really make it, anyway, the power has to come from somewhere on the main board, after all.

2. I changed both R6 and the trimmer to 220k and 10k, giving an adjustment range of about ±50mV.

3. Noted.

4. Yeah, there's a lot. I've since removed the R-C low pass on the 7.5V output. I think I will remove L1 and merge C1/C3 into a single 22uF. The R3/C12 setup was actually straight from the MCP6V06 datasheet, left over from earlier when I was running it off the main +5V rail and wanted a lot of filtering. I think I will remove them.

5. I'm almost done with the PCB, so I'll post it when I have it done. So far, the PIC is fully outside the metal can, electrically separated fairly well too.

6. I instinctively put the heaters at opposite corners of the can. After the whole thing has been running for a while, it should all be very close to the same temperature. I figure that if the heaters are farther away, the distance and thermal mass in between might help to absorb a bit of temperature noise.

7. U1, U3 and the LM35 are all close together, near the middle of the can.

Thank you for your 2 cents!

[c4757p]

Here's what I have so far for the PCB, as well as an updated schematic.

(Updated.)

[Andreas]

Hello,

there is 1 thing which I do not understand:
If you need 2.5V as output voltage why do you not use a 2.5V reference like ADR4525. With the external 25ppm/K voltage divider you will never reach the long term stability (humidity) like with the 2ppm/K internal voltage divider.

And there are other considerations to do.
If you want to use a capacitive load I would add a small resistor 10-100 Ohms between C16 and J2. C16 should be increased.
Use film capacitors for precision circuits. With ceramic types you will get piezoelectric effects picking up mechanical noise to electrical output voltage.
The ADR4550 is in my opinion not a device with good long term stability. My ADR4550 has drifted around 65 ppm within 60 days after a 14 day run in phase. Other (buried zener) devices drift usually below 10-20 ppm per year after some run in phase. Usually these devices have a dedicated trim pin which can be used for a overall system trimming without influencing tempco or long term stability.
A good thermal design does not use punctual heaters but heater foils. Any temperature gradient on the pcb will give thermoelectric voltages.

With best regards

Andreas

[c4757p]

If you need 2.5V as output voltage why do you not use a 2.5V reference like ADR4525. With the external 25ppm/K voltage divider you will never reach the long term stability (humidity) like with the 2ppm/K internal voltage divider.

Because as a one-off made from AD and TI free samples, I did not know which DAC I would have and what it wanted for a reference. Also, I want to be able to trim the reference; none of them have both the initial accuracy and thermal performance I want without being ridiculously expensive.

As for humidity - that is a concern. It varies widely in my house. I will consider potting the circuit in something thermally conductive, hopefully that will help.

And there are other considerations to do.
If you want to use a capacitive load I would add a small resistor 10-100 Ohms between C16 and J2. C16 should be increased.
Use film capacitors for precision circuits. With ceramic types you will get piezoelectric effects picking up mechanical noise to electrical output voltage.

I must be misunderstanding you about the resistor - wouldn't that be R29?

Noted about the film caps, I will switch.

I don't really plan on driving a major capacitive load, I just wanted to make sure there was a bit of filtering so that the transmission line effects on the output and sense lines don't become an issue. I'll try to keep them as short as possible on the main board, but I haven't routed it yet so I can't guarantee anything.

The ADR4550 is in my opinion not a device with good long term stability. My ADR4550 has drifted around 65 ppm within 60 days after a 14 day run in phase. Other (buried zener) devices drift usually below 10-20 ppm per year after some run in phase. Usually these devices have a dedicated trim pin which can be used for a overall system trimming without influencing tempco or long term stability.

That seems to match the long-term drift plot in the datasheet, though it is a bit on the high side. I also considered using an LM399; I now think I'd rather use one, but I already have the ADR4550 here, so I am going to continue with it. I'll make sure to leave this running continuously for a very long time and re-trim it periodically until I see the drift start to settle.

A good thermal design does not use punctual heaters but heater foils. Any temperature gradient on the pcb will give thermoelectric voltages.

What do you mean by "punctual heater"? Do you mean heaters run digitally, switching on and off? These are not. They are driven in constant-current mode and the current is varied. With a properly tuned P-I controller, I expect a maximum of 200uA jitter. I'm sure there will be a bit of thermoelectric voltage, but hopefully the temperature will be stable enough for this to be calibrated out.

[quantumvolt]

There are pictures of resistive heater foils in the "732A Teardown" tread. (Without interfering I think what Andreas is saying is that your transistor heaters are point heaters, not surfaces. You can mend this to a certain degree by mounting the transistors on a separate copper clad PCB shield on top of the reference board or directly at the metal can which will contain it.)

[c4757p]

Ah, that makes sense.

I'll consider moving them. Heater foils are a bit excessive IMHO - I'm only trying to beat the accuracy of a couple diodes and transistors in a big can with a mains-voltage heater filament and hysteretic control, it can't be that hard...

Seems like it would still be very difficult to eliminate temperature gradients. Even if I put the heaters on the can, there will be a radial gradient from the edge of the board inward, starting where the can is soldered to the board. I am hoping that it will be enough to make the gradient repeatable so it can be trimmed out. Outside the metal can, there will be a rather large amount of insulation as well, so hopefully the heaters will not have to dissipate much power once everything is up to temperature.

[Andreas]

As for humidity - that is a concern. It varies widely in my house. I will consider potting the circuit in something thermally conductive, hopefully that will help.

I must be misunderstanding you about the resistor - wouldn't that be R29?

That seems to match the long-term drift plot in the datasheet, though it is a bit on the high side.

What do you mean by "punctual heater"?
but hopefully the temperature will be stable enough for this to be calibrated out.

Potting will only slow down the time constant for humidity but not really help since all plastic materials are swelling with rising humidity. I use ceramic packages for the references when having high demands.

To R29 you are right - I simply had the old schematics. But 1k might be too high depending on load: usually a DAC has 15-5K input impedance at the reference pin.

The datasheet values are under idealized conditions: absolute stable temperature and no humidity drift.
And the salesmen (not the development engineers) who write datasheets will not tell you the bad features.
Mhm: why do they measure the long term drift at 60 degrees? Other references (ADR435) are usually measured at 25 degrees. Do they want to minimize the ageing drift due to humidity?

You have 2 punctual heaters Q3 + Q4.
I have a NTC directly at the GND-Pin of my references. I usually measure 2-3 degrees celsius more at the NTC compared to the environment temperature. And this only through the self heating of the 2mA * 10V = 20mW of the reference.

with best regards

Andreas

[C]

First look at C8 & C9.
Are they there to keep Vout constant, if so then where you show ground connections to these should be a direct connection to Pin 4 and not to ground noise.
If you think that these two are to keep a constant voltage across R1 & R2 then a direct connection to the wire connecting R2, R4 & C13.

Now to see what I am saying put some boxes in side boxes. The inside box contains U1, C5 & your thinking above for C8 & C9. The next box out contains the first box and R1, R2 E4, R6, C13. The next box out contains these two boxes and U3.

Now anytime you have a wire crossing one of the box walls, think of what is going in or out.
For example, Any current change to the wire from U1 gnd to J2 pin 3 causes a voltage change in your reference output. Any change in U3 output also causes a change in the power pins of U3 changing the current ratios on the power pins. Any change to U3's output also causes a change in U3's + Input. Any power or ground connections also couples in the noise on these power connections.
Keeping R1 close to R2 means that a temp change should effect both more equal. Matched resistors are more important here then the actual value.

Even with a static design, the more you can make a possible change have a equal but opposite effect also the more stable the output will be. 

With single ended design this is harder but can still be done some what.  Add three or four matched resistors across U1 Vout to U1 ground.  One resistor away from U1 gnd pin on this new divider is a signal that changes more with a ground side change then Vout -  GND change. If you use this signal as part of the inputs to U3, you will be countering any changes that happens on the ground side.     

C

[c4757p]

C8 and C9 would be for Vout, not the divider; they're required for stability. I should have made this more clear in the schematic.

Good point about the "boxes", and really, I should have known to do that. Since this is a component of a system, I've neglected in my design mentality to recognize that it also has subcomponents and should be designed accordingly.

Okay, "boxes", as well as moving the heaters. (Not just a simple move - I want to solder the heaters to the can, which should be grounded, so I've got to rewire it to use PNP so the collector is grounded.) I'll come back in a while with a somewhat redesigned circuit.

[c4757p]

What a surprise. It's ten times easier to lay out the board when you do it right.

[c4757p]

OK, try again.

The heater transistors are the same, they've just been moved.

Re-uploaded after adding some finishing touches.

(Removed again, see below)

[senso]

I dont understand a lot of precision things, but isn't this a ground loop(marked in red)?

[c4757p]

Possibly, in the same way that the same "loop" would still exist in the copper if I just went and filled the whole thing with a copper pour. I don't think it will be a problem because of the paths the current should take.

Does anybody see a decent way to remove that loop, though? I don't want power supply currents flowing up the signal ground. I considered having the output amplifier sense the reference voltage differentially, but then I would be subject to the matching of yet another set of resistors.

[kt315]

Hey, sorry for the offtopic, but I am quite curious about the specs you are going after and some high level design overview.

With all the PSU discussions going around lately I can't help but think about building one as well. For me it is primarily a learning exercise, so sorry if the questions are rather ... dumb. 

Can you shed some light on the following:

• What power are you going after for each output?
• Tracking or independent?
• How do you plan to power each stage? I would expect you will need some additional split supply for each power phase to provide some juice to opamps
• Do you plan to use switching pre-regulator?
• How do you plan to provide voltage reference to the power stage?
• What about MCU control?

My plan was to "borrow" voltage reference optocoupling from HP PSU schematics (use linear optocouplers with the feedback), and implement a switching pre-regulator. What scared me quite a bit in this plan was the amount of independent supply voltages that would have to provided for the power stage, opapms, mcu, etc.

[c4757p]

I've dragged so many threads off topic, I deserve it...

My PD supply that I'm replacing only does 500mA. I'm either going to go for 500mA, or double it to 1A - not sure yet.

Independent. The main output will be the only "power" output, actually, the two others are just signals (probably up to 100mA), really just to take advantage of the voltage reference. I've been playing around a bit with automated testing lately, and I'd like to have a few computer-controlled signal outputs.

The main supply section will most likely have +25V, -5V for a few negative biases here and there, and either 5V or 3.3V, most likely generated from an 18-20VAC transformer and a couple switching regulators. This will be isolated from a small USB-connected, ground referenced section powered either by a separate mains transformer, or an inverter running off the main section.

No preregulator, the output power just isn't high enough to bother.

The voltage reference will feed into a DAC1220 (20-bit) to get the set point (I'll probably use the extra bits beyond the five-digit precision I need for a bit of software calibration wiggle room), and the output of that into a fairly standard power supply output section (but designed to higher precision). I can't remember which DAC chip I've got here for the lower precision auxiliary outputs off the top of my head, but it's a four-channel one from AD that will also handle the current limits.

Full MCU control. The PD 2005 uses some sort of precision voltage divider to set the output level, which is a bit silly in 2013. The MCU will be able to set all levels, including the current limit for the main supply and the shared current limit for the aux outputs, as well as read off the output current with at least four-digit precision.

What scared me quite a bit in this plan was the amount of independent supply voltages that would have to provided for the power stage, opapms, mcu, etc.

Try to find different parts with power requirements that overlap somewhat! You'd be surprised how much you can fit under a 5V rail with modern parts.

[C]

First note that you dropped the connection from U3 + to junction of R1 & R2


I may not got across what I was trying to say good.
In the box with U1, you have two grounds and a +7.5 to outside that box. These three connections all add some noise in to this box. C5 will have a better chance of removing this noise with a point connection from U1 pin 2 to +7.5. Like wise a point connection from U1 pin 4 to ground would work better on that side.  What I was trying to suggest is moving +7.5 and ground outside this box on the drawing and connect them  with point connections to U1.
With the current circuit I think you still want a point connection from J2 pin 1 to U1 pin 4. This would keep any current induced voltage changes on existing line from effecting J2 pin 1 as much.

A point connection for the power leads of U3 should allow C17 to remove more noise. Granted this may be small, but a lot of small things add.   

Not sure of your reasons for J2 pin 3. I see this as a way to mess up everything with out a way to tell how bad it is getting messed up.

[c4757p]

Nah, that connection is on the back side. Edit: You're right, I see it. It got blown right out of the schematic.

I'm still not sure exactly what you're describing about a "point connection from U1 pin 2 to +7.5". Maybe I'm just being dense - is there any way you'd take the time to sketch it over the picture?

I do see what you're saying about J2 pin 1 and I agree. I'll change that.

J2 pin 3 is the sense line return to the op amp, to allow it to compensate for losses caused by any current drawn from the output.

[kt315]

Thanks for the reply. Good luck with your build. Looking forward for the future posts on your progress.