Precision opamp + Power opamp = good?, bad? or ugly?

[Sleep Deprived Tommy]

Hello everyone, i happened to be in need of a precision power amplifier for a project of mine, at first i went with the classic push pull class AB amplifier driven by a precision opamp, but i quickly realized this isn't really ideal, specially for DC precision applications, so i decided to go with power opamps. maximum current of around 2A is required and for safety margin, i went with 3A, the desired offset value is lower than 1uV in a ±15V range (please cope with my delusion), power opamps with already low(ish) offset voltages exist, but if that was enough you wouldn't be reading this, nevertheless my options are very limited: LT1970A, OPA593. 6X LT1970A or 12X OPA593 in parallel would be the final spec, I'm a bit more leaned towards 12X OPA593s for improved noise performance and better overall performance. I'm relatively new to analog stuff but I'm getting better at it. as far is i know, paralleling opamps is a great way of increasing current output and reducing noise without any drawbacks, the drawbacks that i suspect might exist are the increase of input bias current and uneven(<3%) output current distribution between opamps, so rather than driving 12X parallel opamps directly, i want to drive them with output of a ADA4523 with the feedback to its inverting input directly from the output of the parallel power opamps. in theory there shouldn't be a problem since the ADA4523 will compensate for the offset of the power opamps. the fact that ADA4523 is a chopper worries my a little bit, i have seen choppers behave weird but that was a different scenario. I'm fairly sure that it isn't a problem here but better safe than sorry. is there something I'm missing on?

The answer to the question above might be obvious but I'm too elbows deep in another rabbit hole and I really don't want to risk it and i don't want to branch off from one rabbit hole to another, this project has already been too long. at first glance, It's a no brainer but I'm afraid there might be something that I just forgot to account for or didn't know at all that comes to bite my leg off afterwards.

Sincerely, Tommy

[mawyatt]

What about a high current power op-amp/buffer with a precision op-amp front end inside the closed loop feedback? Paralleling op-amps can get messy if done correctly (balancing the load current).

Best

[GigaJoe]

i'm not sure of possibility to reach about 2 amps output and 1 microvolt precision ... it required quite precision board layer and kelvin connection
it always possible to amplify current by power transistors   , depend on your corner frequency required .. 
composite opamp can be considered and transistors output ..

[Sleep Deprived Tommy]

What about a high current power op-amp/buffer with a precision op-amp front end inside the closed loop feedback? Paralleling op-amps can get messy if done correctly (balancing the load current).

interesting idea, the parallel ones are individually precision parts themselves, also the parallel ones would spread heat around the board more evenly and 1 part doesn't need to do the heavy lifting, also the improved noise figures makes me really wanna do it
a 1% shunt resistor (≤25mΩ) as a ballast resistor at output of each opamp would make the current imbalance negligible, even something as simple as a calculated pcb trace acting as a ballast resistor would reduce the imbalance to lower than 20% which in itself isn't that bad considering there is a safety headroom of 1A. but I'm going to go with a 100mΩ resistor, either 10 0.1% 1ohm resistors in parallel or for a bit more price, a 0.1% Vishay-dale shunt resistor.

i'm not sure of possibility to reach about 2 amps output and 1 microvolt precision ... it required quite precision board layer and kelvin connection
it always possible to amplify current by power transistors   , depend on your corner frequency required .. 
composite opamp can be considered and transistors output ..

kelvin connection is no problem in a 6 layer bard, the problem with power transistors is linearity and the "fiddly" nature of it in low voltages(<Vbe) and the additional biasing circuitry required, also there is a size consideration, for my application through hole components are eliminated due to height restraint and heat dissipation becomes a problem with smd ones because they dunk it directly in the pcb and in a situation where 1uV matter, you don't want that. also it would require a cascode MOSFET array before the collector and multiple rails to minimize heat dissipation; power opamps reduce component count(not in this case) and take care of all of the problems themselfs and do so magnificently in a small package.

[GigaJoe]

ok ... not in my league  ..
something as composite like precision and OPA2544 -  as parallel output ..

[Sleep Deprived Tommy]

ok ... not in my league  ..
something as composite like precision and OPA2544 -  as parallel output ..

5mV of input offset ..... i know it will be compensated for but i can't bring myself to do it, the OPA2544 also has this figure

can you point out what can go wrong with 12x OPA593s in parallel, with each one having it's own ballast resistor wouldn't it make a great solution for both increasing output current and improving noise figure? what am i missing here?

P.S. "not in my league" seems too humble, if it's not in your league i most certainly can't claim it's in mine, it's my obsession over stupid stuff that this is coming from

[mawyatt]

What about a high current power op-amp/buffer with a precision op-amp front end inside the closed loop feedback? Paralleling op-amps can get messy if done correctly (balancing the load current).

a 1% shunt resistor (≤25mΩ) as a ballast resistor at output of each opamp would make the current imbalance negligible, even something as simple as a calculated pcb trace acting as a ballast resistor would reduce the imbalance to lower than 20% which in itself isn't that bad considering there is a safety headroom of 1A. but I'm going to go with a 100mΩ resistor, either 10 0.1% 1ohm resistors in parallel or for a bit more price, a 0.1% Vishay-dale shunt resistor.

How do you plan on dealing with the different offsets of each op-amp, and the thermal induced offset drift, and the offset aging?  Unless you add local feedback to each op-amp individually inside the overall feedback, thus making each op-amp a fixed gain summation rather than a group of open-loop op-amps connected together in parallel with series isolation output resistors.

If you look at how the paralleled low-noise op-amps are used to reduce voltage noise at the expense of current noise, each op-amp is in a fixed gain arrangement (but not open loop) and summed output with a series "isolation" resistor.

As Joe mentioned, getting 1uV precision with 2 amps at +-15V output is going to be a challenge as that's the equivalent of 500nΩ. The TC of Cu traces, solder connections, thermal gradient induced EMF, thru hole Z and so on are going to require extreme care with Kelvin balancing sensing and the power supplies need to be stable under load and temperature as a 1V PS variation with an op-amp aggregate 120dB PSRR will cause a 1uV offset change at unity gain!

Anyway, seems a difficult challenge but might be possible with careful design and implementation. Good luck and keep us posted on your progress

   
Best

[mawyatt]

ok ... not in my league  ..
something as composite like precision and OPA2544 -  as parallel output ..

5mV of input offset ..... i know it will be compensated for but i can't bring myself to do it, the OPA2544 also has this figure

can you point out what can go wrong with 12x OPA593s in parallel, with each one having it's own ballast resistor wouldn't it make a great solution for both increasing output current and improving noise figure? what am i missing here?

P.S. "not in my league" seems too humble, if it's not in your league i most certainly can't claim it's in mine, it's my obsession over stupid stuff that this is coming from

That configuration shown doesn't improve the aggregate amplifier voltage noise. The voltage noise is set by the single amplifier, the unity gain "buffer" doesn't help the overall input referred voltage noise.

Best

[GigaJoe]

with 12 , oscillation .. most likely they will not able to be stable ...   my experiment with a 2 amps many in parallel give quite a negative result,  or you doing high resistors to balance, and loosing efficiency or low resistors and oscillation ...
as THD was kinda indifferent .. i finish as simple one opamp and 2 power transistors for output ...  ;  ( and of course precision indifferent as well )

again it not clear your frequency border ... or it DC .   for DC guessing .. DC servo would work.   

5 mv offset it usual for power opamps as tradeoff ... so composite where first has to be precision and control power opamp ...  and usually first opamp should be much faster then second , the power one ( i think im right ...  don't remember ),  there a lot of theory to calculate composite opamp ..

[Sleep Deprived Tommy]

ok, i think I'm starting to make sense of what might be the issue, specially with a chopper stabilized opamp in the first stage of the composite opamp, the oscillation is a bit concerning, if what i understood so far is correct, OPA541 should be even a all-rounder, the only remaining problem is heat, if there is any recommendations regarding what power opamp i should use, i would really appreciate it

[PCB.Wiz]

Hello everyone, i happened to be in need of a precision power amplifier for a project of mine, at first i went with the classic push pull class AB amplifier driven by a precision opamp, but i quickly realized this isn't really ideal, specially for DC precision applications

Why did you decide that?

maximum current of around 2A is required and for safety margin, i went with 3A, the desired offset value is lower than 1uV in a ±15V range

To sanity check that target, the best OpAmps TI make offer ±3µV max, so you will need to select or trim (or just cross your fingers )
The CMRR also means you need to run grounded in, not in follower mode.  (see the Offset Voltage vs Common-Mode Voltage graphs)

If you expect it to hold that 1uV from 0-2A, you will need some seriously low resistance leads!  (let's assume the AOL can regulate well enough)
Google says 10 Gauge (AWG) solid copper wire has a resistance of approximately 0.00328 Ohms per meter (Ω/m)
ie that's 3.2mV/A at 1m, or 320uV/A at 100mm

Or google says The DC resistance of a 10 mm x 5 mm copper busbar is approximately 0.532 mΩ/m (milliohms per meter) at 20°C.
- that's better, you are now down to 106uV over 2A load, for 100mm total busbar connection.

[Sleep Deprived Tommy]

well... this is embarrassing

apparently I'm not the first one to attempt doing something like this, there is an entire TI app note dedicated not only for paralleling power opamps but the exact one I was looking to parallel  

this clarifies most of the stuff i was worried about

[Sleep Deprived Tommy]

also my assumptions on reduced noise was on these two articles from AD
https://www.analog.com/en/resources/technical-articles/paralleling-amplifiers-improves-signal-to-noise-performance.html
https://www.analog.com/en/resources/technical-articles/paralleling-amplifiers-increases-output-drive.html

one important thing i forgot to mention was my target is a unity gain power amplifier, important thing to keep in mind, no resistor shenanigans.
this is what i meant at first

now that I'm looking back, my initial thing wasn't that bad, the app note mentions a suggestion for ballast resistor value for a 5A drive capability, this suggests 20 OPA593s in parallel, sounds crazy because it probably is. nevertheless based on the app note and AD articles, I'm sticking with my initial design, the worst that can happen is oscillation which if there are compensation capacitors wouldn't be a problem, and an active filter if need arises.

one thing it mentions is the Ladder-Follower configuration, unlike the mentioned circuit mine is closed around a precision opamp, it's similar in a sense that the leader/master is the ADA4523 but it's output is not connected directly to output circuit, but rather to input of the follower/slave power opamps. the Vout directly feeds to the leader/master so it will compensate for everything. in the app note, in both mentioned configurations the input(s) are directly connected to the signal, so i'm assuming this configuration is the one I'm going to go with. ballast resistor of 0.1% 25ppm/C 500mΩ(plenty enough), would result in a less than 0.2% mismatch which is in the order of micro amps and is negligible. the only thing left to do is determine a compensation capacitor and an active filter so it can handle capacitive loads (if need arises) and not oscillate

[Sleep Deprived Tommy]

To sanity check that target, the best OpAmps TI make offer ±3µV max, so you will need to select or trim (or just cross your fingers )

I get what you mean, it's an attempt nonetheless, as i said, I'm using ADA4523 which has a typical offset of 500nV and maximum offset of 5uV, trimming is the next thing, I'm going to do the trimming on the input side with external calibration and it will be compensated for it with a resolution of at least 1 order of magnitude lower lower (so 100nV)

The CMRR also means you need to run grounded in, not in follower mode.  (see the Offset Voltage vs Common-Mode Voltage graphs)

It has 3x lt3081 + 3x lt3091 for it's regulator feeding from isolated transformer output that has electrostatic shielding between windings going to the protective earth, so the common mode rejection isn't a problem, especially because the CMRR is 160dB for ADA4523!!! minimum 140
also Offset Voltage vs Common-Mode Voltage for the ADA4523 is attached, it's lower than 1uV

If you expect it to hold that 1uV from 0-2A, you will need some seriously low resistance leads!  (let's assume the AOL can regulate well enough)

isn't kelvin sensing's sole purpose to mitigate that?

also something everyone assumes is I'm trying to hold 1uv from 0 to ±15V !!! thats insane, 1uV is for the lowest range of the device which is ±100mV full scale

[MariuszD]

Achieving accuracy at the uV level is difficult because thermoelectric voltages are generated at the junction of different metals.

[David Hess]

Hello everyone, i happened to be in need of a precision power amplifier for a project of mine, at first i went with the classic push pull class AB amplifier driven by a precision opamp, but i quickly realized this isn't really ideal, specially for DC precision applications, so i decided to go with power opamps.

What was not ideal about controlling a class-ab power amplifier with a precision operational amplifier?

i'm not sure of possibility to reach about 2 amps output and 1 microvolt precision ... it required quite precision board layer and kelvin connection

I have designed and made power references that achieved that sort of performance.  It does require Kelvin connections which adds to circuit complexity.  For a power precision amplifier, this is handled by the feedback loop connection and remote sensing.

also my assumptions on reduced noise was on these two articles from AD
https://www.analog.com/en/resources/technical-articles/paralleling-amplifiers-improves-signal-to-noise-performance.html
https://www.analog.com/en/resources/technical-articles/paralleling-amplifiers-increases-output-drive.html

Noise in the power output stage is less important because it is reduced by excess gain of the precision amplifier's feedback loop.  In addition, the power output stage can be low impedance, making low noise performance easy.

[mtwieg]

Lots of discussion about specific components and circuits, but there's basically no description of what this amplifier is actually supposed to do.

Does the amplifier need to have voltage gain, or is it a unity gain buffer? What is the load? What sort of bandwidth does it need?

The 1uV input offset spec is particularly suspicious. Do you think whatever source is driving the amplifier input is going to have accuracy/precision on that order? If the offset was actually much larger (i.e. 10uV), would you actually be able to tell?

[Geoff-AU]

also something everyone assumes is I'm trying to hold 1uv from 0 to ±15V !!! thats insane, 1uV is for the lowest range of the device which is ±100mV full scale

We're not assuming it, you told us it's a requirement

the desired offset value is lower than 1uV in a ±15V range (please cope with my delusion)

[ejeffrey]

ok ... not in my league  ..
something as composite like precision and OPA2544 -  as parallel output ..

5mV of input offset ..... i know it will be compensated for but i can't bring myself to do it, the OPA2544 also has this figure

5 mV is nothing.  For applications that only need to operate in a single quadrant, it's quite common to use an opamp + single transistor with the transistor in the loop.  That's a full 600 mV offset voltage, plus a large tempco.  I've done this in precision reference applications and it works great.

[Koifarm]

Opa1611 with 8 lme49600 best I have build. Not designed for low dc offset.
But I could not measure DC offset.( < 10uV).

It was designed for low THD and noise.

[GigaJoe]

Oh .. my mistake ... I wrongly assume it AC output with microvolt precision ...

[Terry Bites]

I suggest you write down the spec you need. Why do you need it?
Gain, offset, bandwith etc and repost your query in the voltnuts arena https://www.eevblog.com/forum/metrology/

1uV is not a rational goal. The drift in external components, pcb leakage currents, grounding problems and thermal emfs will make this almost impossible to achieve.


Adding a power stage to a zero drift amplifier is fairly simple. I will null out offsets in the clock LPF and power stages. Look, a power stage> https://www.audiosciencereview.com/forum/index.php?threads/amp-too-good-for-diyaudio.67448/

I think you might want to read this AD notes on composite amplifiers.
Many examples of combining opamps to get the specs you want.

[Sleep Deprived Tommy]

What was not ideal about controlling a class-ab power amplifier with a precision operational amplifier?

V_BE compensation for different voltage and current ranges, mismatch between NPN and PNP pairs, terrible tempco of power transistors, hassle of paralleling multiple stages and so on. power opamps deal with these problems in a single DFN package and do so magnificently. the only exception is paralleling but it's far easier than paralleling discrete power amplifiers.

Lots of discussion about specific components and circuits, but there's basically no description of what this amplifier is actually supposed to do.

Does the amplifier need to have voltage gain, or is it a unity gain buffer? What is the load? What sort of bandwidth does it need?

The 1uV input offset spec is particularly suspicious. Do you think whatever source is driving the amplifier input is going to have accuracy/precision on that order? If the offset was actually much larger (i.e. 10uV), would you actually be able to tell?

well, at this point there is no point of holding back anymore, initially i wanted to make a DIY open source SMU and i was planning on posting it's "Initial Release" in a dedicated post. the amplifier in question is supposed to be the high current output stage. Resistive loads are the primary target over reactive loads since it's going to be a DC source meter, however sub 100KHz is the target bandwidth for programming set voltages and set currents, transient response (rise/fall) frequency of sub 10MHz would be nice to have but i haven't put much effort into it.
two AD5791s configured in parallel is the set DAC mainly because of results from a test done by AD themselves; other than those, usual precision buffering along with an LTZ1000 or an ADR1399(it has its own blog post) as a reference element are the big guns here doing the heavy lifting. this is almost the whole the drive circuitry in a nutshell.

We're not assuming it, you told us it's a requirement

yep, guilty. sorry mate, i was in a bit of rush and my lazy ass didn't bother to read it thoroughly before posting.

1uV is not a rational goal. The drift in external components, pcb leakage currents, grounding problems and thermal emfs will make this almost impossible to achieve.

Thermal EMFs are the worst to deal with simply because it's you vs nature, there are few methods to deal with it mainly coming from NIST, but it's hella expensive and complex to deal with, the effective measure in this and many other scenarios is compensation

[mtwieg]

well, at this point there is no point of holding back anymore, initially i wanted to make a DIY open source SMU

Okay, so why do you think a 1uV offset is relevant for the source of a SMU?

Have you ever used a SMU or looked at specs for one? AFAIK none of them claim a source accuracy anywhere near 1uV. The measure part of the instrument is really where accuracy is critical, but that's an entirely different circuit. And even then I've never seen an accuracy spec near that, +/-100uV worst case error would be considered excellent.