A discussion on Precision OpAmps

[David Hess]

Would LT1008 also have been a good choice in the past, as the main difference (to LT1012) seems to be internal/external compensation?

The LT1008 is a fine choice and in some cases better because the external compensation allows its dynamic performance to be optimized.  Before I found the OPA140, I was going to use the LT1008 to replace the damaged input buffer on my Tektronix 4.5 digit DM501A multimeter.

Because of the flawed design of the ADC in the DM501A, the input buffer must have high CMRR so even the best older precision JFET operational amplifiers are barely good enough unless tested and selected for high CMRR.

I´ve looked at the input amplifiers of a few other DMM´s, and found a few interesting OP´s that could be used:
AD795 (from the HM8012, 1-2pA Ib but -100µV Vos and 1-3µVpp Noise)
AD706 (from the 34401A, 50-200pA Ib and +30µV Vos and 0.5µVpp Noise)

In the past if a low input bias current bipolar part was not used, then the highest precision JFET input part was.  The AD542 (25pA Ib, 250uV Vos, 1uV/C Vos Drift, <100uV/V CMRR, and 2uVpp Noise) was an early precision BIFET part recommended for this application.  I am sure Burr-Brown had a bunch.  Another recommended low input bias current bipolar part was the National LM11 (25pA Ib, 100uV Vos, 1 uV/C Drift, 0.3uV/V CMRR, >5uVpp Noise)  which is an improved 308.  In the past I used a lot of LM11s.

The limit to performance with JFET parts was usually their common mode rejection ratio.  I assume that high resolution meters which used them without an automatic zero implementation like the ICL7106 bootstrapped them?

An Idea that I had was to attach a Peltier Element to the Input Amps and alter their temperature to the point, that their offset voltage is nearly zero.
The primary problem with that, besides the thermal EMF caused by the leads, would be that the offset drift and the offset are rarely of the same polarity.

They used to make constant temperature oven "hats" for the metal TO-99 package.

Bipolar precision operational amplifiers since before the OP-07 are designed so that the offset null adjustment also minimizes  offset voltage drift.  I do not know of any JFET parts with this property but if the temperature is regulated, then the offset null adjustment can be used to trim the offset.

For an DMM input amplifier (unless using an old chip like ICL7106) one would not care about the absolute offset voltage, as this can be subtracted in software anyway. So critical parameters are more like the input bias current (usuallaly < 50 pA are aimed for, ideally < 10 pA) and the voltage dirft.

The ICL7106 has its buffer built in with offset corrected as part of its automatic zero process, but there were variations of that idea intended to be used with an external buffer.

except in low-cost applications it is never a good idea to use the input amplifier as the ADC driver. If one is already concerned about the current noise a zero-drift amplifier produces, the noise injected by the ADC is orders of manitude higher than that - considering the internal buffers are bypassed for accuracy reasons.
An opamp output appears as an inductor to a signal (or noise-) source attached to the output. That means the noise will be pushed into the signal source only slightly attanuated at best.
The only proper way to mitigate these issues is to have a dedicated ADC driver and appropriate filtering in between the amps.
This problem gets even worse if the input amplifier has a gain (much) greater than A=1.

I am not sure what you are saying here.  Some precision operational amplifiers will have problems driving against the charge injection of the ADC which requires fast settling time but that will not show up at their input.  It is easily solved by adding a low impedance output buffer *within the feedback loop* and capacitive decoupling of the signal at the ADC input.

Using a JFET amplifier (like OPA140 as a modern chip) and seprate switching the input is an option, but also not that easy. There are similar or possibly even worse switching spikes from doing the choper type offset correction under software control.

Potentially it is no worse than the ICL7106.  I have been working on this problem for a while and suitable input multiplexers are few.  Did any production multimeters use an input multiplexer made from discrete transistors like the 3N series small signal MOSFETs, or even JFETs?

[KT88]

I am not sure what you are saying here.  Some precision operational amplifiers will have problems driving against the charge injection of the ADC which requires fast settling time but that will not show up at their input.  It is easily solved by adding a low impedance output buffer *within the feedback loop* and capacitive decoupling of the signal at the ADC input.

I'm referring to attempts to use a single amplifier for both input and driving the ADC.

[Kleinstein]

Did any production multimeters use an input multiplexer made from discrete transistors like the 3N series small signal MOSFETs, or even JFETs?

The older DMMs (e.g. HP3456, 3458, Keithley 19x, 2000) use mainly JFETs for the input switching. It still looks like the DMM6500 still uses JFETs at least for some parts.  Discrete MOSFETs are rare, due to the parasitic diode. So it would need 2 in series and maybe an isolated drive - still an option for higher voltage.  One can also use CMOS switch chips (like DG508, ADG....) and some type have good specs. Some DMMs like Sigilent SDM3055, SDM3065, KS3446x seem to use such CMOS switches. At least for later stages where leakage and charge injection is less of an issue CMOS switches are very common.

When switching directly at the input one should use a kind of precharge: So there is an extra buffer to provide an auxiliary signal for guarding (and maybe gate drive for JFETs) and the sequency is so that before switching to the main input path the auxiliary buffered signal is read. This can reduce the switching spike quite a lot. This is really help when doing classic auto zero switching all the way to the input.

To get really good CMRR (and also more loop gain and higher input resistance) it is quite common to use a bootstrapped supply for the main input supply. With this one can also use an OP for a lower supply (like 5 V) for the input, espeically also 5 V AZ OPs like the AD8628 or max4238 as lower bias AZ OPs.
Compared to a bootstrapped OP the classical discrete JFET amplifier is also not that much more complicated.

[Ole]

So, first of all
I am not expecting any input impedances higher than about 1MOhm and the voltage noise level should be as low as possible

and secondly I do have to try and wrap my head around the concept of bootstrapping.

[Kleinstein]

The input stage of the Keithley 2000 is a very simple case of bootstrapping the supply. This is not bootstrapping in the stricter sense, but more in a feed forward sense. Bootstrapping in the stricter sense would derive the supply from the feedback network and not directly from the input. The extra delay makes this case a little more tricky to get stable and usually slower (slowing down the driven supply is one way to stabilize).

The idea is to have the supply of the critical OP to follow the input voltage, so that this OP allways sees an essentially constant input relatively to it's supply. So it only has to correct the errors of the circuit (e.g. OP) to drive the supply.
For just a buffer, like in the K2000 bootstrapping is still relatively easy. It gets more tricky with gain, as the output voltage range of the floating OP is limited. This needs an extra amplifier stage to get the full output range.
An amplifier with bootstrapped supply can indeed be mind bogling.  Getting it stable, well protected and latch-up free can also be a slight challange.

[RoGeorge]

I do have to try and wrap my head around the concept of bootstrapping.

Fig. 1 from AnalogDialogue Vol 53 No 3 August 2019 makes the main idea of bootstrapping easier to understand.  It is exactly like the tale that gave birth to the bootstrapping word: lifting yourself by pulling up from your own shoelaces. 



The whole opamp is floating relative to its own output voltage (which is the same as the input voltage). 

[KT88]

As Kleinstein mentioned, bootstrapping avoids errors due to CMRR. The LTC1050 has 125dB which is not bad. A more recent one like the ADA4522 is specified with at least 140dB at room temperature.
A non-AZ amp like the LMP7721 has only 83dB in comparison - boostrapping is a must with such an amp...
Another important reason for bootstrapping is the limited supply voltage of older AZ amps which mostly stays well below 20V thus prohibiting a true +/- 10V-range.
The ADA4522 can operate up to around 55V (+/-27V).

[RoGeorge]

It looks like a big PSRR (Power Supply Rejection Ratio) will also be desirable, but I have no hands-on experience and can not tell how much the PSRR influences the end results in practice.

[KT88]

PSRR refers to supply noise. A very clean supply voltage is of course mandatory.

[martinr33]

A discrete FET input switch in an early TR 6.5 digit multimeter, the predecessor of the Advantest 6581.

The transistors are 2N4117A devices, now about $10 each from Interfet.

In the 6581, this system is updated to a hybrid module.

[Ole]

I think I might have an idea how to get the Vo+5V and Vo-5V or in the case of the LMP7721 +/-2.5V:
Simple linear regulators, a metaphoric ton of capacitance and a overvolt protection zener diode and a minimum load resistor
The latter would only be added if required.

With a supply of +/- 15V, closely connected linear regulators and hopefully low dropout voltages, I could get +/-12V through the input.

[Kleinstein]

The simple schematics from AD for the bootstrapping is missing a point that is usually needed for stability: the drive for the floating OP supply needs to be somewhat slower than that OP.  Another point is that the output should not be loaded very much, especially capacitive load can be a problem.

Attached is an example for a relative simple bootstrapped buffer I use for a high impedance (500 K ohms at lower side) voltage divider. The 2 zener diodes are there to limit the swing of the supply, especially to avoid going too positive so that the supply for floating OP breaks down. The TL031 is not the best choice, but was what I had at hand.

Getting a full +-12 V range with a +-15 V supply is challanging, but not impossible.

A minimum load for the OP can be a good idea (e.g. to avoid the output stage cross over error, that can be at around 1 µV for a non Az precision OP).  A resistor to the floating OPs supply gives an essentially constant current.

[macaba]

Attached is an example of feedforward bootstrap to give you an idea. U1 and U2 do not need to be expensive OPA140, but perhaps some advantage to having everything be the same speed.

[3roomlab]

Attached is an example of feedforward bootstrap to give you an idea. U1 and U2 do not need to be expensive OPA140, but perhaps some advantage to having everything be the same speed.

i remember the k2000 version
i think opa141 is enough?
Qn : is there CMRR improvement?

[Gyro]

Attached is an example of feedforward bootstrap to give you an idea. U1 and U2 do not need to be expensive OPA140, but perhaps some advantage to having everything be the same speed.

U5 doesn't need to be expensive either if you connect it to the output of U6 rather than the input. When connected to U6 input, it will also adversely affect input bias and offset currents.


P.S. The Datron approach to input bootstrapping... https://www.eevblog.com/forum/projects/very-low-bias-current-op-amp-to-buffer-a-kelvin-varley-divider/msg694586/#msg694586

[macaba]

U5 doesn't need to be expensive either if you connect it to the output of U6 rather than the input.

This is then not feedforward bootstrapping...

[Gyro]

Sure, I realise that. I'm just not sure what it buys you (apart from maybe speed? U6 supplies will still be lagging anyway due to 2 opamp delays in the bootstrap path), when it compromises input characteristics by having a non-bootstrapped opamp connected to the input.

[Kleinstein]

Attached is an example of feedforward bootstrap to give you an idea. U1 and U2 do not need to be expensive OPA140, but perhaps some advantage to having everything be the same speed.

i remember the k2000 version
i think opa141 is enough?
Qn : is there CMRR improvement?

The bootstrapping is improving the CMRR and thus the linearity quite a lot. Ideally it adds the accuracy of the supply drive, so something like 40-100 dB to the CMRR. Even the OPA141 is already overkill to drive the supply, unless one needs a very fast response in the feed forward version. The TL031, OPA990 or TL072H would be reasonable candidates to drive the supply. The main point is low bias and a sufficient output range.

Classic bootstrapping from the output is a bit slow (it needs the slow down to avoid oscillation), but avoids extra bias.
Feedforward driven supply from the input side gives a little more bias current (thus the need for a low bias amplifier), but can be higher speed / faster settling.
So both versions have there justification.

[RoGeorge]

Another example, with a simpler implementation than the one from the AnalogDialog pdf:
https://www.pi.infn.it/~federico/Immagini/bootstrap.pdf

[Ole]

Getting a full +-12 V range with a +-15 V supply is challanging, but not impossible.

I noticed that the +/-12V are the differential voltage between the HI and LO inputs.
So the maximum Voltage on either input would be about 6V from GND.
macaba posted a concept which, I must admit, I do like it for the simplicity.
Although I´m not sure if the second Amp, the one that handles the bootstrapped voltages needs to be in parallel to the bootstrapped OP
Alternatively my Idea would be to put resistors in paralell to the constant curent J-FETs (say 50-100kOhm) so that the Bootstrapped Supply Voltages Vb+/- center around GND, if nothing is connected to the input.

Cheers
Ole

[Kleinstein]

Resistors are the JFET current sources would not effect the behaviour with an open input: this input voltage would just float and with the high impedance and low bias would slowly dirft up or down and pic up hum. This is OK for a DMM mit high impedance input.

With a +-15 V supply and moden OP with a near rail to rail output one can get +-12 V for the voltage at either side / input.

It is still a decision to make if the input low side terminal is fixed to the supply ground, which is the more conventional design, or the negative side terminal also moving to get a symetric signal. The second way can get a bit confusing but allows the larger voltage range. E.g. the Sigilent SDM3065 this way gets a +-20 V range without a divider at the input with only a +-15 V supply. With an ADC with a differential input the vesion with a driven low side would definitely make sense as it is a way to generate a differential signal for the ADC even in a mode with no extra gain.

This choice of amplifiers and the general DMM input setup are somewhat seprate topics. The available amplifiers effect what configuration is practical.

[David Hess]

Did any production multimeters use an input multiplexer made from discrete transistors like the 3N series small signal MOSFETs, or even JFETs?

The older DMMs (e.g. HP3456, 3458, Keithley 19x, 2000) use mainly JFETs for the input switching. It still looks like the DMM6500 still uses JFETs at least for some parts.  Discrete MOSFETs are rare, due to the parasitic diode. So it would need 2 in series and maybe an isolated drive - still an option for higher voltage.

Discrete MOSFETs like the 3N series that I mentioned are 4 wire parts with a separate substrate connection, so they have no body diode between the source and drain.  Signetics and Siliconix used to sell them but now they are available from companies like Linear Integrated Systems and Calogic, at least in theory.

One can also use CMOS switch chips (like DG508, ADG....) and some type have good specs. Some DMMs like Sigilent SDM3055, SDM3065, KS3446x seem to use such CMOS switches. At least for later stages where leakage and charge injection is less of an issue CMOS switches are very common.

I have had problems finding integrated CMOS switches with low enough guaranteed leakage specifications.  The DG419 is on my order list for my delayed DM501A repair project.  I have been assuming that precision applications for these parts are selecting them for low leakage.

When switching directly at the input one should use a kind of precharge: So there is an extra buffer to provide an auxiliary signal for guarding (and maybe gate drive for JFETs) and the sequency is so that before switching to the main input path the auxiliary buffered signal is read. This can reduce the switching spike quite a lot. This is really help when doing classic auto zero switching all the way to the input.

Precision sample and holds invert the gate drive and apply it to the output through a capacitance to cancel charge injection.  The same could be done to suppress charge injection at the input, and I suspect parts like the LTC1043 do this.

It looks like a big PSRR (Power Supply Rejection Ratio) will also be desirable, but I have no hands-on experience and can not tell how much the PSRR influences the end results in practice.

Power supply rejection applies to symmetrical voltage variation in the supplies, including noise.  It is only relevant if the supply voltage variation is large which will not be the case with regulated supplies.

The same applies to common mode rejection in the inverting configuration since then the common mode voltage does not vary.  If the operational amplifier is bootstrapped, then the common mode voltage also does not vary so common mode rejection is no longer important.

I believe in a discussion on the forums about the OPA140, it was stated that the OP140 series uses internal bootstrapping of its input stage and input protection network to achieve its outstanding specifications.

[Kleinstein]

For the charge spike when switching there are 2 parts: one is the charge injection in the chip itself, which is normally measured when both sides are at the same voltage and usually for the turn off case. The other part is the extra charge spike when a connection is made to a different votlage. This is from the capacitance of the switch part and also from the input capacitance of the amplifier and possibly extra input current when exceeding slow rate limits, so that amplifier internal parts can not follow and may increase the charge beyound what is expected from small signal input capacitance.

The 4 pin MOSFETs with separate substrate are quite often only low voltage. The Keithley2000 and similar meter use such FETs as switches in the ADC.

The guarantied leakage specs on the CMOS switches are usually not that low - testing to that level is quite expensive.
There are a few Maxim parts (e.g. max327), but with relatively high on resistance.
Typical leakage at least can be quite low with quite a few more afforable parts too (e.g. DG211B, TMUX6111). It still gets tricky when the meter gets hot.
Separate screening for low leakage is very likely in some cases. This also applies to JFETs used as switches - the guarantied specs are often not sufficient and selcted parts are quite often specified.

[RoGeorge]

LMC662 of former NS now TI, CMOS dual op amp, single supply 15V, RRO, 22nV/sqrtHz, bias 2fA typical, input R > 1Tera\$\Omega\$ and in stock at Mouser for about $2-3/1pcs.

Voltage offset is 3mV and drift 1.3uV/*C, but you'll probably chop it in software anyway.

[David Hess]

LMC662 of former NS now TI, CMOS dual op amp, single supply 15V, RRO, 22nV/sqrtHz, bias 2fA typical, input R > 1Tera\$\Omega\$ and in stock at Mouser for about $2-3/1pcs.

Voltage offset is 3mV and drift 1.3uV/*C, but you'll probably chop it in software anyway.

Or use the better LMC6081 or LMC6082, but these parts are limited in follower applications by low common mode rejection.  They are also pretty noisy.