using an ICL7650S to buffer the output of an LM399

[robert.rozee]

hi,
    i'm wanting to set up an LM399 as an 'artifact' reference, ie, where the output is buffered only, and not scaled up to 10v. to do this i have a couple or three choices:

1. a regular precision opamp (such as an LT1001) with the output trimmed using pins 1 and 8, to achieve 1uV offset at 23 degrees C, and hope for minimal drift with temperature changes. how well does this work out in practice?

2. an ICL7650S chopper opamp, in a 14-pin DIP package. the 14-pin package comes with a 250Hz output on pin 12, which can be checked to ensure the part is genuine - the parts i've ordered ar us$7 for 10 pieces, so just a tad suspect! i already have some 8-pin DIP versions, but these have no 250Hz output to check.

3. an LTC1050 or similar chopper opamp, and hope that the offset falls within the typical range (+/-0.5uV). but these are relatively expensive, and i don't have any to hand.


the current for the LM399 zener will be derived using a second cheap op-amp.


how good/bad does the ICL7650S choice sound? how dodgy does the idea of just trimming an LT1001 sound, and will the drift scuttle the idea?


cheers,
rob   :-)

[Andreas]

Hello,

ICL7650 and LTC1050 (which is rather similar) can only deliver 1-2 mA.
For buffering references I would use LTC2057 or ADA4522 with some filtering against EMI.
See for example here:

https://www.eevblog.com/forum/metrology/ad587lw-10v-precision-travel-standard/msg1449488/#msg1449488
(later I added also a transient zener to the output)

but of course you can also use a LT1001 as Op-Amp if you regard the offset as part of the zener voltage.
Especially if you have not so much temperature change in your lab as I have in mine (18-34 deg C over the year)

with best regards

Andreas

[robert.rozee]

hi Andreas,
    i have read your thread on the AD587 travel standard a few times over the years, in many regards it is pretty close to what i have in mind.

regarding ICL7650S/LTC1050, even if only able to source a couple of mA, this should still be sufficient to drive into a 4k ohm or higher load; one of the reasons for using a buffer is so that whatever you connect the artifact's output to can't divert away any of the LM399's zener current. opinions seem to vary upon how important a precise current going into pin 1 of the LM399 actually is, but i am inclined to err on the safe side.

would adding a JFET to the output of the opamp (but within the feedback loop) offer any disadvantage? this would also eliminate the potential of any internal heating of the opamp die caused by the output load.




cheers,
rob   :-)

[iMo]

The dynamic impedance of the 399 is around 1ohm, of the 1399 0.011ohm to 0.04ohm.
To divert a "significant current" off the 399 the input impedance of the opamp has to be "comparable" with that of 399..
But usually it is at least 10^7 times higher for any type, imho.

[robert.rozee]

The dynamic impedance of the 399 is around 1ohm, of the 1399 0.011ohm to 0.04ohm.
To divert a "significant current" off the 399 the input impedance of [...]

oops, my mistake. when raising the issue of diverting current away from the LM399, i was talking about the case of an unbuffered LM399, with a load consuming a significant fraction of 1mA then applied to the output (between pins 1 and 2):




cheers,
rob   :-)

[dietert1]

At 1 mA output current, what is the required precision? If you use wiring with e.g. 0.1 Ohm, you already get an error of about -14 ppm = 1 mA * 0.1 Ohm / 7 V). You probably need a 4-wire system with two buffer amplifiers, one for plus and one for minus output.

Regards, DIeter

[robert.rozee]

At 1 mA output current, what is the required precision? If you use wiring with e.g. 0.1 Ohm, you already get an error of about -14 ppm = 1 mA * 0.1 Ohm / 7 V). You probably need a 4-wire system with two buffer amplifiers, one for plus and one for minus output.

hi Dieter,
    you raise a good point that i'd not thought about - given just 1mA of current flow through a set of leads having 0.1 ohm of resistance will cause a drop of 100uV across said leads. this is a good enough reason to always keep the current drawn from any voltage reference to an absolute minimum.

but the BF245C FET can still serve to protect the output of the ICL7650S. is this useful? or would it be preferable to use the opamp output pin directly as the reference voltage output.


cheers,
rob   :-)

[MK]

At 1 mA output current, what is the required precision? If you use wiring with e.g. 0.1 Ohm, you already get an error of about -14 ppm = 1 mA * 0.1 Ohm / 7 V). You probably need a 4-wire system with two buffer amplifiers, one for plus and one for minus output.

hi Dieter,
    you raise a good point that i'd not thought about - given just 1mA of current flow through a set of leads having 0.1 ohm of resistance will cause a drop of 100uV across said leads. this is a good enough reason to always keep the current drawn from any voltage reference to an absolute minimum.

but the BF245C FET can still serve to protect the output of the ICL7650S. is this useful? or would it be preferable to use the opamp output pin directly as the reference voltage output.


cheers,
rob   :-)

Its mixed, the fet reduces the change in heat of the op-amp, but reduces the cmrr and psrr.

[dietert1]

The JFET doesn't really solve the problem of an output short. Its gate diode will still route the opamp output to the short, possibly damaging the opamp and the fet. There needs to be a resistor between opamp output and gate.
Also the output should get some protection against overvoltage or reverse voltage (from external sources like ESD).

Regards, Dieter

[robert.rozee]

There needs to be a resistor between opamp output and gate. Also the output should get some protection against overvoltage or reverse voltage (from external sources like ESD).

like this?



MK: is cmrr and psrr likely to be a major issue? when in actual use i'm thinking of having everything powered by a set of 6x Li-Ion cells (giving a tad over 24v when fully charged, and a minimum of 18v at end of discharge). at other times an external supply will be used to keep things 'warmed up'.


cheers,
rob   :-)

[Kleinstein]

The ICL7650 is likely not so bad, but it is usually a good idea to have some filter (e.g. 1 K and 100 nF) between the reference and an auto zero OP-amp to keep away switching spikes of the amplifier from the reference.

[David Hess]

but of course you can also use a LT1001 as Op-Amp if you regard the offset as part of the zener voltage.
Especially if you have not so much temperature change in your lab as I have in mine (18-34 deg C over the year)

Nulling the input offset of an LT1001 type of operational amplifier also nulls its input offset drift with temperature.

For an LM399, 0.3ppm/C at 6.95V is 2.1uV/C, so the offset drift of an LT1001 type of operational amplifier can be ignored.  The noise from the LT1001 can also be ignored because the noise from the LM399 is so much higher.

In theory the LT1001 offset drift, which is proportional to input offset voltage, could be tuned to cancel the temperature drift of the LM399 over a narrow range of temperature.

would adding a JFET to the output of the opamp (but within the feedback loop) offer any disadvantage? this would also eliminate the potential of any internal heating of the opamp die caused by the output load.

I have seen JFETs used that way, but I think a bipolar transistor works about as well.  A MOSFET could also be used.

An integrated regulator with low adjustment pin current could also be used, like an LM317, which would have the advantage of providing built in current limiting and thermal protection.

At 1 mA output current, what is the required precision? If you use wiring with e.g. 0.1 Ohm, you already get an error of about -14 ppm = 1 mA * 0.1 Ohm / 7 V). You probably need a 4-wire system with two buffer amplifiers, one for plus and one for minus output.

hi Dieter,
    you raise a good point that i'd not thought about - given just 1mA of current flow through a set of leads having 0.1 ohm of resistance will cause a drop of 100uV across said leads. this is a good enough reason to always keep the current drawn from any voltage reference to an absolute minimum.

I have made 4-wire "power" references before using the LM329 where the output voltage changed less than 1 microvolt from 0 to 1 amp, so the 4-wire connection can make a huge difference.

[robert.rozee]

hi Kleinstein,
    i have added a 1k/0.1uF filter as you suggested:



David: so you believe that using a nulled second LT1001 will provide similar performance to an ICL7650S+JFET? i had shied away from using an LT1001 because of the untrimmed maximum input offset voltage of 60uV; if this LT1001 is considered a 'sacrificial buffer' then changing it out for a new one could introduce a significant shift. but if this offset can be reliably nulled out that is not an issue.

i have noticed in other's reference designs that offset nulling is generally not used, but then this could be because most folks are targeting a 10v output rather than just a zero-gain buffer, and that with a gain stage the offset is just a constant that can be ignored.

note that i'm only really interested in performance being maintained over a limited temperature range. lets say +10 degrees C to +35 degrees C, and over this range it would be nice to be able to keep the buffered output within 1ppm (7uV)  of the LM399's output; my 'best' meter is a 34401a, and i don't imagine myself ever getting anything fancier!


cheers,
rob   :-)

[Kleinstein]

The LM399 is anyway limited by the popcorn noise. For the LM399 this are random jumps between 2 levls some 4 µV apart on a time scale of usually some 1-10 minutes. So there is a limit on how much effort is justified for the amplifiers and so on. It may make sense to have the option to upgrade to a ADR1399, so add the RC series element parallel to the reference.

Adding offset adjustment is a 2 sided thing: one can trim the zero (though this is not easy in this circuit (e.g. with a resistor switched in/out between the inputs)). With quite some time for the test alternatively adjust the thermal drift. However the offest trim is often also susceptible to drift over time, as trimmers are not the most stable parts. The trim would only effect the 10 V output, not the 7 V.

A weak point with the circuit as shown may be the reaction to capacitive loading of the output (e.g. a cable or DMM internal capacitance and the 11 V zener). With too much capacitance (no so clear where the limit is, maybe 1 nF or so) the buffer could oscillate. With less capacitance there could still be a voltage shift as it changes the impedance need by the AZ amplifier input. I would add the local feedback capacitor (~ 1 nF)  and resistor (e.g. 1 K) in the feedbakc path to make the buffer more resitant to to capacitive loading.

[robert.rozee]

A weak point with the circuit as shown may be the reaction to capacitive loading of the output (e.g. a cable or DMM internal capacitance and the 11 V zener). With too much capacitance (no so clear where the limit is, maybe 1 nF or so) the buffer could oscillate. With less capacitance there could still be a voltage shift as it changes the impedance need by the AZ amplifier input. I would add the local feedback capacitor (~ 1 nF)  and resistor (e.g. 1 K) in the feedback path to make the buffer more resitant to to capacitive loading.

R+C added to feedback path:




cheers,
rob   :-)

[David Hess]

David: so you believe that using a nulled second LT1001 will provide similar performance to an ICL7650S+JFET? i had shied away from using an LT1001 because of the untrimmed maximum input offset voltage of 60uV; if this LT1001 is considered a 'sacrificial buffer' then changing it out for a new one could introduce a significant shift. but if this offset can be reliably nulled out that is not an issue.

As I explained, the untrimmed input offset voltage drift of the LT1001 is lower than the temperature drift of the LM339, so using it as an alternative will provide similar performance because performance is limited by the LM339, and then the input offset voltage and input offset voltage drift can be trimmed anyway.  Precision parts like the LT1001 and ICL7650S are limited by external factors like thermoelectric effects.

To put it another way, the errors from the LM339 dominate so there is little advantage to using the ICL7650S or any other chopper stabilized operational amplifier, at least compared to a precision part like the LT1001.

I have noticed in other's reference designs that offset nulling is generally not used, but then this could be because most folks are targeting a 10v output rather than just a zero-gain buffer, and that with a gain stage the offset is just a constant that can be ignored.

That is right, although the input offset nulling could be used to adjust the temperature coefficient of the LM339, but it would take some effort to get it adjusted properly.

[Andreas]

R+C added to feedback path:

but wrong.

The capacitor should be from negative input to directly output of the OP-Amp.
Otherwise you pick up too much EMI from the output line.
I would also add a 100nF (foil) capacitor in parallel to the output against EMI.

I hope the 24V for the heater are stabilized.
(the LM399 is sensitive to heater voltage variations).

with best regards

Andreas

[iMo]

BTW., why do you mess with two opamps now? You went from the 7V version towards the 10V version and that is a standard version people use here since ever, with a single opamp and the transistor(s)/fet buffer/limiter, its schematics depicted almost once a week here.. 

https://www.eevblog.com/forum/metrology/lm399-based-10-v-reference/msg5435552/#msg5435552

PS: couple of weeks back I started a separate thread here on nulling the opamp's offset and its TC. Still not sure how to exactly do it in this wiring, moreover I would be happy to read more on why the TC of the input offset gets zero when its offset is nulled.. On the other hand doing the nulling with a crappy 10T trimmer is not a good idea either (so you have to replace the trimmer with some quality resistors at the end). What had been told there - there is a chance to compensate the TC of the 399 - and that sounds pretty interesting..

https://www.eevblog.com/forum/metrology/nulling-the-opamps-input-voltage-offset-and-its-tc/msg5401565/#msg5401565

[robert.rozee]

The capacitor should be from negative input to directly output of the OP-Amp. Otherwise you pick up too much EMI from the output line. I would also add a 100nF (foil) capacitor in parallel to the output against EMI.

I hope the 24V for the heater are stabilized (the LM399 is sensitive to heater voltage variations).

Andreas, is the below what you had in mind?




i was hoping to get away with not needing to closely regulate the heater voltage, the LM399 datasheet contains examples with 15v-20v and 12v-18v heater supplies. to what extent does a change in heater voltage affect the reference voltage? i'm looking at a range of 18v-24v, but there is no reason why the heater could not be switched over to use the 15v regulated supply while in actual use (with switching back to the raw battery voltage when not in use).


cheers,
rob   :-)

[David Hess]

PS: couple of weeks back I started a separate thread here on nulling the opamp's offset and its TC. Still not sure how to exactly do it in this wiring, moreover I would be happy to read more on why the TC of the input offset gets zero when its offset is nulled.. On the other hand doing the nulling with a crappy 10T trimmer is not a good idea either (so you have to replace the trimmer with some quality resistors at the end). What had been told there - there is a chance to compensate the TC of the 399 - and that sounds pretty interesting..

The subject is covered in old application notes from about the time that the OP-07 was released.  Linear Technology may have published something about it also.

I know there was a more detailed publication, but check out PMI (Precision Monolithics Inc.) application note AN-12 which covers the math of what is going on with a matched differential pair.  It comes down to:

delta Vbe = (kT/q) * log(Ic1/Ic2)

If the currents in the two identical junctions are identical, and that is what is being trimmed by the offset null pins, then log(Ic1/Ic2) = 0 and the offset voltage drift also becomes 0.  Otherwise the input offset voltage drift is proportional to the log of the ratio of the currents, which is how the delta Vbe method of absolute temperature measurement works.

Of course the junctions are not quite identical, so we use the offset null connections to trim the currents to produce equal current density, which has practically the same effect.

Analog Devices published an application note where a precision operational amplifier was used as an accurate temperature sensor (and cable driver) by deliberately creating input offset using the null connections, but good luck finding it.

[David Hess]

Why not unload the first LT1001 with an NPN transistor?

[robert.rozee]

BTW., why do you mess with two opamps now? You went from the 7V version towards the 10V version and that is a standard version people use here since [for]ever, with a single opamp and the transistor(s)/fet buffer/limiter, its schematics depicted almost once a week here.. 

hi iMo,
    the design criteria i'm aiming to meet are:

1. no scaling employed, with just the raw reference output presented (or as close to as possible);
2. 1ppm performance over a limited temperature range;
3. preferably no adjustments required;
4. some degree of output protection/buffering;
5. relatively low cost.

these design criteria are inspired by a comment by Conrad Hoffman i spotted a year or two back about a move towards minimal voltage references without any scaling of the output. alas, i can't find his exact posting at the moment.

in terms of meeting these criteria: the LM399 is relatively low cost now, and seems to offer better than 1ppm performance. the LT1001 is also relatively cheap, and the accompanying resistors used (to derive the 1mA zener current) are nothing special. this is the place i started at - a good zener and a well regulated zener current, using the elegant schematic layout from 'cellularmitosiscircuit':
https://www.eevblog.com/forum/metrology/influence-of-resistors-in-lm399-reference-circuit/msg1475556/#msg1475556

the idea of buffering the zener voltage arose firstly to offer some protection to the LM399, and later to ensure that 'misuse' of the output wouldn't adversely affect the reference voltage (by diverting away zener current). again, the ICL7650S is relatively cheap, and there may be scope for even cheaper substitutes.

i've actually had a dual opamp solution cooking away for a few years (2x LT1001), but never done much with it due to the offset voltage that i noticed fairly early on. i can't see any way to easily have both an opamp generated zener current and a buffered output equal to the zener voltage without using two opamps. the discussion in this thread has offered a means of getting my original dual opamp setup into a usable state.

i also wanted the assembled device to be something that you'd be happy to post away to a friendly forum member or three to check the output voltage, and not be too concerned about it getting lost in the postal service. this was inspired by the recent forum threads about his Fluke 45 started by 'Fried Chicken':
https://www.eevblog.com/forum/testgear/easy-way-to-test-the-calibration-of-a-dmm-(fluke-45)/?all
and
https://www.eevblog.com/forum/testgear/looking-for-my-firstlast-bench-multimeter/?all


cheers,
rob   :-)

[iMo]

..
The subject is covered in old application notes from about the time that the OP-07 was released.  Linear Technology may have published something about it also.

I know there was a more detailed publication, but check out PMI (Precision Monolithics Inc.) application note AN-12 which covers the math of what is going on with a matched differential pair..

Yep, my ceramic PMI OP07AY wait desperately to be nulled out

.. and the "PMI Linear and Conversion Application Handbook 1986" with AN-12 in it..

http://www.bitsavers.org/components/pmi/_dataBooks/1986_PMI_Linear_and_Conversion_Applications_Handbook.pdf

And from an PMI 1976 opamp handbook - untrimmed and trimmed TC for OP07 A..E

[Kleinstein]

Those graphs look odd. The curves are not actual offsets for some sample parts, but still upper limits for the specs.

The LM399 has some drift in the early days. Settling in the frist 1000 hours may give some 20 ppm of drift. It may take long time operation (maybe more than 1 year) to get the LM399 stable to the 1 ppm level for a longer time. It could still be OK short time and when not powered on most of the time. Drift mainly happens when hot and the LM399 has rather little hysteresis from cooling to room temperature. So it is OK to power it down between uses.

I would consider the ICL7650 relatively expensive by now. The cheaper AZ amplifiers for a 15 V or higher supply are the OPA180 or MCP6V51. They can have a little more offset, but that offset is usually stable. With some case variants they can be exchangible, while the ICL7650 is more like an oddity from the old days with external capacitors.

If the 10V output has low priority the LT1001 is already overkill and a simpler OP07 or OPA202 could also do the job. The LM399 is not that sensitive to small changes in the current.

[Andreas]

Andreas, is the below what you had in mind?

i was hoping to get away with not needing to closely regulate the heater voltage,

a) yes
b) its time that you read the LM399 thread most of your questions are handled there:
https://www.eevblog.com/forum/metrology/lm399-based-10-v-reference/msg441913/#msg441913

[robert.rozee]

b) its time that you read the LM399 thread most of your questions are handled there:
https://www.eevblog.com/forum/metrology/lm399-based-10-v-reference/msg441913/#msg441913

i have read through the "LM399 based 10 V reference" thread a couple of times in the past, but the problem is that it contains too much information; with nearly 1400 postings many dozens of different approaches have been explored, analyzed, and numerous conclusions drawn in an attempt to wring every last ounce of performance out of the LM399. there have also been many paths that, while promising, have ultimately lead to no useful improvements.

whereas i'm just after something that is 'good enough' for a specific task, while fitting within a few non-onerous design criteria. and to this end, almost everyone who has contributed to the couple of dozen postings to my thread here has added valuable information in a form that i can make use of   


cheers,
rob   :-)

[robert.rozee]

I hope the 24V for the heater are stabilized. the LM399 is sensitive to heater voltage variations.

hi Andreas,
    i've been thinking about this one for the last few days. over in the LM399 thread, back in 2014, you made some measurements of this effect with the following results:



(copied from: https://www.eevblog.com/forum/metrology/lm399-based-10-v-reference/msg441913/#msg441913)

i've added in blue lines to mark off the region between 18v and 25v heater voltage (corresponding to 6x Li-Ion cells going from fully charged at 4.2v each, down to a 3.0v cutoff when fully discharged), and the corresponding change in zener voltage. in the setup you used at the time, this works out at about a 6uV drop after the 2:1 scaling, or 12uV directly at pins 1 and 2 of the LM399.

i presume that each run was conducted over a relatively short time interval? has anyone else repeated and verified this experiment since? i'm also wondering if there could be mechanisms other than the change in heater voltage coming into play; the 12uV shift in reference voltage corresponds to about 1.7ppm and thus is tantalizingly close to the 1ppm level that i'm trying to keep below.


cheers,
rob   :-)

[Andreas]

i presume that each run was conducted over a relatively short time interval?

Hello,

each measurement point is usually 1 minute averaged reading in my measurements.
And it looks like 250 mV steps between each measurement.

And of course both directions (rising and falling heater voltage) to make thermal effects visible (if any).
Have done the same for LM399#14 and for the first four ADR1399 in LS8 package

Edit: why do you think there is footnote (Note 3) below the ELECTRICAL Characteristics table in the LM399 data sheet?
And another measurement (from macaba) :
https://www.eevblog.com/forum/metrology/adr1399-reference/msg4114465/#msg4114465

with best regards

Andreas