DIY high resolution multi-slope converter

[Rerouter]

Have the reference part updated, Guessing the Jfet is just acting as a current limit on the heater to make things more predictable at the cost of a longer time to steady state?

Also curious why you changed the zener current down to 0.7mA, the curves from the datasheets tend to have 0.9-1.4mA as the region with the least temperature variation?

Have the ADC almost how I like it, the input buffer is going to take some time to unwrap in a nice way.

Yes the guard traces are overkill, but for now trying to work out where some of the INL error sources could be. and having them makes the math easier.

[Kleinstein]

The JFET at the reference heater is there to limit the current on start up. It may be enough to have a resistor there. Without a current limit I had a problem with start up with a relatively low power DCDC converter. It also helped to adjust the brown-out detector at the µC.
I have not really checked the reference current very much. AFAIK 0.7 mA should be just OK, at least in most cases, but 1 mA is the more normal current. The extra power is part of the heating anyway, so not much need to keep the current at the low edge.

I rechecked the operation with 10 K resistors at the integrator, this time with more stable PTF56 resistors.  For the turn over error the resistor TC is not a significant contribution, but stable resistors help to do the test.  In a quick check without much warm up, the turn over error was relatively large ( some 25 µV at 9.3 V and ~5 µV at 5 V). So it looks like INL gets worse with smaller resistors and 10 K seem to be well on the low side.
So I am afraid using 20 K as 2x LT5400-10K in series is not a good solution. 2x100 K in parallel would be the preferred solution, despite of the slightly more complicated layout.

Some overkill with respect to INL is understandable. It is hard to measure and every source avoided helps, even if likely small.
The current mirror is a little in this direction.

[Rerouter]

Already have parallel covered, there is the same footprint on the back of the pcb, as well as the 0805 option... First revision option that I left in, if you want parallel on the same side It will take some thinking.

For the integrator output, that current source is my main bit of grief at the moment, It appears to only cater to an integrator voltage up to 550mV positive (so only effecting negative input signals),

Not clear how fast your modulation is now, but I recall it being about 1uS per pattern step with a 12 step pattern, for 50K input and 2nF cap we would need to cater to about 2.7V to work for your full input range.lets say +-3V to cover for any weird points.

Simplest to fix it at present would be to swap out R30 with 68K, would ensure it remains constant, downside is it is a lot of current to sink, about 1.14mA, and around 15mW of heat extra, as well as heat loads that change with input voltage.

Harder fix is probably going to be making a constant current source / sink track a bit nicer, as right now it just shoots off towards 1.4mA with a +10V input when in reality it should be almost turned off. the only time the additional current is needed is for negative input voltages as best I can tell, and for a +-10V range about 2.45V is needed for integrator voltage

If your modulation has changed I will update the math accordingly, but right now that current source is not working for negative voltages over about -4V

[Kleinstein]

For the modulation in the run-up phase I have several versions to choose from the short times are from some 0.66 to 1.5 µs, normally 1 µs and the signal dependent part is 6.5 µs  in the normal version and up to 15 µs in the slow one.
For the larger current with small integrator resistor (10K) I have doubled the integration capacitor.

The current mirror part has some gain, so getting a higher current out than in. So the current out from T6  is about 4.5 times the current going into T5. So R30+R19 should be about 4.5 times the resistance at the integrator. So R30 should be adjusted with the resistors at the integrator.  The current trough R76 gives an addition bias - ideally to avoid the output cross over region of the integrator all together. A smaller value (e.g. 100-150 K) for R76 may be a good idea.
For the analysis it is important that the current mirror input voltage is at the positive supply.
With a high positive input the current mirror will give less current.

I get some 0.6 mA with zero input and 0.4 mA / 0.8 mA with a +10 V / -10 V input.

[Rerouter]

I was modelling the whole chunk, so the current through the integrating capacitor + current in or out of the slope amp, this is where it swaps. pretty much what direction is current flowing at the output of the op amp U2

At the cost of some extra heat, swapping R30 for 68K would be something easy to test, assuming one of your current test boards has it fitted.

[Andreas]

The JFET at the reference heater is there to limit the current on start up. It may be enough to have a resistor there.

are you really using a SST5485 ?
with only 4-10 mA constant current you may get a instable heater which needs up to 300 mW in steady state.
So you might want a current source which only limits above 20-30 mA.

According to (old NS) data sheet you will need a small electrolytic (> 2uF) in parallel to the heater when using a series resistor (or FET) for heater stability.

My observation is that the LM399 also has a measurable PSRR on the heater voltage.
Especially with heater voltages below 12V.
So I want a stable heater voltage.

with best regards

Andreas

[Kleinstein]

I currently don't have the JFET at the heater. The 1st board has a direct link, the 2 nd board a small resistor. The SST5485 was more like one of the very few JFETs in SOT23 case I have in the old eagle libraries. The more obvious choice is a MMBF4392 or MMBFJ111.

Thank you for mentioning the capacitor at the heater. This makes sense.

@Rerouter:
The current to the slope amplifier could indeed be a problem. With the DG419 part used the resistor R12 should become larger (e.g. more like 22 K, maybe more), so that less current will flow there. During run-up the DG419 is turned off - it would be on only during run-down.

R30 should be adjusted to directly compensate the current from the input. R76 gives an extra constant part - this has no obvious ideal value and would be a point to tweak.  The current to the slope amplifier changes with integrator output voltage and can thus not be compensated by the current mirror. A larger values for R12 can help here. However without the DG419 switch this would add to the noise for the run-down. The 5 K already give more noise than the OPA1641 or NE5534. So the 5 K value is kind of a compromise.

[iMo]

With 30-50mA current limit it may take up to 2minutes until the 399 "starts" itself..
PS: I started my 399 from cold, my "foldback" current limit is around 40mA, it took 101 seconds (with 15V at the heater)..

[Kleinstein]

The heater gets +-15V minus the drop on the JFET, so a total of near 30 V. This would be some 10 mA steady state.
It depends on the supply whether the extra limit is needed. I currently get away without with a 3.2 VA 2x18V transformer as well as a DCDC converter limited to some 50 mA. It currently takes some 15 seconds until the supply is reasonably stable. The version with a sereis resistor takes a little longer.

[Rerouter]

Suggest moving the current source to a fixed 1.1mA one as the thermal variation is about the same and removes another unknown from the mix. including not having to route the buffer output across half the PCB. Also want to shift it on to the analog supply rails as it shares a current loop with these devices. from some simulation sweeps this covers as best I can tell all input voltages between +-10V,

This is based on 50K input resistance, 2/2.2nF capacitor and a pattern length of up to 12uS, longer patterns and lower input resistances will need to be wound up a little more to keep that +-10V input, smaller inputs can still be quite longer.

For the average integration voltage op amp I feel needs to be tweaked to work how you described. right now any positive average integrator voltage will just be stuck at 5V to the ADC, removing R67 will have it reflect an average of +-5V on the integrator to the ADC, I can keep that as a footprint for scaling options, just right now I feel it should be non-populated by default.

Edit: now that the slope amplifier is using Q2 for a non-linear element, I presume there is some reason why D2/D10 did not get the same treatment?

Edit2: Had to tweak the current up to 1.33mA, to really cover every possible case up to 12uS patterns, so a 180K resistor

[Kleinstein]

Using Q2 instead of the 1N4148 is because I saw some temperature dependence of the gain, even at around zero. The transistor instead of the diode helped by having less leakage and hopefully still enough speed. At least it now seems to work better.  For D2/D10 this is not needed, as there is the extra resistor in between, so that leakage current is not as critical.

The point with R67 is very valid. I had the extra gain as an option but so far not populated.

I know it is a kind of pain to route the buffer output (or current mirror input node) all across the circuit. However the main idea was to compensate for the changing current level with a change in input voltage. So I would at least keep this as am option, just in case. Worst case one can always add a resistor and bodge wire.
Having just a constant current source that is strong enough to avoid cross over may also work - kind of a different way to fight a possible weakness. As there is the current through R12 this may even be the better way. For a constant current source it does not really matter where it gets its supply from.
 

[Rerouter]

ADC and buffer are mostly fitted together, not entirely happy about the buffer and will space it out more so the reference designation can be better placed soon.
Thermals also included, so the input resistor should avoid a lot of the variation, If something does not have a value, it means it has less than 0.08mW of heat output on average while operating.

Will include the option for a jumper wire for the current source, on board trace would be difficult due to how noisy the area it would be passing is.

While I will likely keep the double input array, right now both resistors would be sitting at the RN202 spot. one on the front, one on the back, for 50K equivilent.

Current source thermal values will be a little different from the schematic, as I have been tweaking a little based on simulations and datasheet values, but the transistor heat output is within a factor of 2 of the latest schematic.

[Rerouter]

Moving through it all, getting back to roughly where I was on Rev1, just lacking some polish,

Things can still shift about but its leaving even more room than rev1 free for the micro / power supply / possible multi meter input respin.

The micro and similar I will be aiming to crowd up closer to the lower left, and may try to shuffle the power supplies a bit further away from the references as well. to keep as much free board area as possible around the reference and input area as this is what I suspect will be the most likely to change.

Any thoughts and I can add / change it.

[Kleinstein]

The electrolytic cap between the integrator and input buffer it quite large. I think the Food-print can be a little smaller.
The same is true for halve the caps at the supply. The ones at the output side can be smaller and they may be better placed closer to the demand side. The capacitors on the input side may need the size, especially if powered from a 50 Hz transformer.

For the integration cap there is no more real need for a THT footprint. The capacitor of choice is definitely TDK C0G in SMD (e.g. 0805 size), possibly 2 in parallel to get high capacitance if 20 K are used at the integrator or for other low loss caps that may only be available to some 1 nF. 

For additional points to add,  I would consider a connector for an external reference, e.g.  like the HP3458-A9: two rows with 5 pins.
With the slightly higher voltage of an LTZ1000 one may need JFETs instead of zener + BJT at the reference amplification so one can get closer to the +-15 V limit.

A true input stage would be quite a bit of circuit - probably a little too much for the board.  A minimal one would still need an AZ OP, some gain switching (e.g. 2/4 or 3/4 of DG201B), some switching at the input (e.g. DG408, DG201B) and input filtering and protection (incl. some guard / BS drive). The control would likely need some shift register (e.g. the usual 4094) connected to the SPI port.

[iMo]

I would add 1-1.5mm slots - like at the places indicated with the red below..

[Kleinstein]

There is not that much space left to really add a lot.

I would like a little more space between the input buffer and the resistor arrays.

With the buffer U16 one could/should have some extra filtering (Pi type) and protection (Zeners+2 diodes with bootstrapping). This would make it more like a real DVM input, though still without gain.
Without gain one could also add a simple slow inverter so one could have a +-20 V range. The inverter output would be the other side terminal, and should also get an input of the MUX. With a gain of 1 only, there is no need to turn the inverter part off, as one does not need a gain check all the way to the input. The zero check /cal would need user help to short the input.  So one would have the choice to use 4 channels +-10 V (one with protection) or 1 channel with +-20V.
 
I see very limited need for the guard traces around U4 / U20. The NE5534 is not high impedance anyway.

[jbb]

I had a look at the schematics for our Agilent DAQ DMM card ... and there’s a lot of stuff in there!

I suspect that trying to fit a ‘proper’ DMM input on the same board will open up a heap of issues, so it might  be better to design that separately.

That being said, an input protection network would be useful for system integration.

The 20V range trick is interesting, but seems to require some floating ground tricks, which could complicate system integration (eg might end up with floating supplies and AC CMRR issues when ADC ground isn’t the same as front terminal ground).

[Rerouter]

Are you willing to share those schematics?

The 100x100 was just a cheap board size. In reality it could be anything. I agree on a seperate sub board might be a valid answer.

[jbb]

It’s the Agilent / Keysight 34972A DAQ. You can download the service manual from the Keysight website, and the schematics are near the back.

Of note:
- lots of custom resistor arrays for things like V scaling, reference gain, ADC core
- some relays for mode switching (eg high V / low V / R / I)
- selected LM399AH reference

Edit:
- current input is protected by 4 diode bridge. An op amp is used to bootstrap the ‘live’ diodes (presumably reduces leakage)
- 0.1A current shunt for 1A range, schematic doesn’t indicate 4 wire
- extra 5R shunt stacked on for lower ranges
- resistance range current supplies by a 2 range low side current sink connected to a 4 range current mirror (to make a current source). Quite elaborate here.

[MegaVolt]

It’s the Agilent / Keysight 34972A DAQ. You can download the service manual from the Keysight website, and the schematics are near the back.

I could not find the circuit Tell me where can I see the circuit?

[Kleinstein]

I know the commercial ADCs usually use custom resistor arrays, but these are not available so easy. So this ADC is made to work with standard resistor arrays (e.g. 4 equal values) for most of the part. This is not that difficult. Going from 7 V to +-14 V range is not that inconvenient. It even has some advantage over the +-10 or +-12 V references used in the 34401 or 3458.

A full DMM input would be really more like a separate board, about the size of the ADC board, or even more with amps and ohms ranges. There can be enough issues with the input part and ADC - so separate boards make changes easier. I would even consider to have a voltage input stage only for the start and have amps and Ohms as a 3 rd part.

Attached is the idea for a little protection  filtering for the U16 (from the ADC schematics) input.
The exact resistor / cap values are not critical, more like order of magnitude. The 2 lines to the right would go to the MUX.

Something like CMRR or mains hum may be an issue - this is a reason to include the simple inverter part to test it. There can also be a positive side, like reduced turn over error or the possibility to get an auto zero mode for 50 SPS. So if it works it can be attractive.
There are issues to combine the driven common terminal with multiple inputs. It also makes the current modes a little more complicated, but not too bad if relays are used for range switching.

[Rerouter]

Shifted about as requested, not yet looked into slots around the reference,
Now both the resistor array and input mux are far enough away from variable heat sources that the thermal difference is too small for me to worry about further, but I am open to mixing things up.
Capacitors where mis-sized for a higher voltage, picked the most common footprint for 100uF >35V and ran with that for the lot of them, half of them should be 47uF which could get away with smaller, but I don't really see a need for it.

the +-15V caps being nearer to point of load is already mostly done, the biggest load is the reference, then the input buffer, everything else of significance is on the analog +-15V rail, where the largest loads are the second integrator op amp and the slope amp.
The main power traces right now are on the back of the PCB in 0.8mm thick traces, at the current levels involved the varience over the largest load difference should account to about 300uV of DC shift on the supply rails, everything is locally decoupled,
Its the AC influences I cannot fully control for,

For the external reference, you say 2 rows of 5 pins, I assume something like below?
Or skip the sense thought and instead its just for contact reliability

+15V  +15V
-15V   -15V
GND    GND
-Ref     -Ref_sense
+Ref   +Ref_sense

If there is a popular reference module e.g. the KX LTZ1000, and someone can give me its pinout and connector dimensions I could just space it out to make it plug and play.

the slope input trace is very overkill, but having it there makes the math easier, so unless it hurts something, I will leave it be, as the ground would be there either way to soak up some of the noise from where it passes through. There is a fair bit of energy every time the slope output moves, and that would couple in too easily otherwise.

U16 confuses me a little, where R10 and R7 are meant to connect, and assuming that only U16's reference is the same as the main PCB
Edit: Ok, just an inverter, so to 2 different inputs, What is pad 2 for?

[Kleinstein]

AFAIK there is no common interface for the DIY LTZ1000 boards.  The closest to a kind of standard in the this case would be the reference board in the HP3458.
Pins an A9 board
J401 1) Heater (base of transistor) - likely used for burn in only
J401 2) +18 V  (OP + startup resistor)  (15 V should be sufficient )
J401 3) Heater GND
J401 4) -15   for compensation of the ground current via a resistor
J401 5) +10 V heater
J400 1) divider for set temperature  (should be left open) - likely used for burn in only
J400 2) Ref GND  (0V)
J400 3) Ref. 7 V
J400 4) GND A# , ground for the OPs
One pin of J400 is unused / blocked as key.
The connections are not ideal, but also not so bad.  Ideally I would like one more GND pin and I see not much need for the heater base connection.
Still the original HP A9 board uses an odd connector so the board sits very close to the carrier. 

[branadic]

What I would like to see on a reference meter is an external input, for a well known 10.xxxxxxV reference.

-branadic-

[Kleinstein]

An external could be nice, but it's also not easy.
I would guess the more normal use of an external reference would be to drive some divider network from the reference and than sense a voltage as a differential input. So both voltage are liked but generally don't share a common ground. So either the reference input or the voltage input would need to be differential.
Also reference scaling from 10 V (likely x 1.5)  is more complicated than scaling form 7 V (x 2).

For some tests it could help if one would have a extra output from the ADC reference - this may be easier to realize than a true ratio mode. It still gets tricky with extra drive and sense lines and especially if the +- mode for 20 V input is used.