Precision opamp power stage to increase V and I out

[RoGeorge]

Found a few NOS precision opamps K153UD5, which is the Russian version of the now discontinued uA725, and would like to add a power stage to it, as a design exercise.  Posting here to ask for guidance. 

The goal is to use the circuit as a power stage for a DDS, so to extend the max V and I out of the DDS (Rigol DG4202).  It's a hobby project, so a one off prototype for the home lab, no design for manufacturing required.

Reusing parts from the scrap box is one of the most important requests.  The ones I already have are:
- K153UD5 (uA725) precision opamp, +/-16.5V max supply, +/-15V unprotected inputs
- BD680/BD681 Darlington power transistors, 100V, 4A, 40W

Aimed specs:
- supply voltage range:  +/-25 ... +/-50V
- input voltage:  max. 20Vpp, 50\$\Omega\$
- output voltage:  Vpp 3-5V less than the supply range
- output current:  less than 3A
- frequency range:  power swing for DC to audio would be OK, 1MHz or more for unity gain would be nice to have but not required

Nice to have:
- overload protection
- zero offset manual adjust from the front panel


First idea was to use a bootstrapped power supply to extend the output range (flying power supply for the opamp):


Image from:  https://www.edn.com/bootstrapping-your-op-amp-yields-wide-voltage-swings/

That would still need a current amplifier, so not ideal for this application.  Second idea was to use a less usual topology, starting from a typical application schematic seen in a datasheet for A741:



The trick there is to connect a low impedance resistive load at the output of the opamp (47\$\Omega\$ in the above example), then to close the negative feedback loop from the output of the external power transistors.  The current needed to drive the 47\$\Omega\$ load will produce a current imbalance in the V+ and V- supply currents of the opamp, and those currents will drive the external power transistors.

The circuit is simple and very clever, looks more like a hack.    However, the voltage supply for the opamp is exactly 1.2V lower than the voltage supply for the power transistor, or else said this circuit does not extend the maximum output voltage swing of the 741, it only extends the max current.

To extend the output voltage swing, I've decided to power the final transistors at a higher voltage, then to lower it to +/-15V for the opamp only while still preserving the supply current imbalance in the V+ and V- supply, so to drive with it the power transistors.  Something like this:



Does this look OK?  Any obvious mistakes?  Compensation/stability was not considered yet, and I'm not very confident at analog design.  Should I choose some other topology?

Other observations or alternative ideas are welcome.

[Kleinstein]

In priciple the circuit could work, but there are a few weak points:
The power transistors provide quite a bit of the voltage gain and this limits the speed.
For the circuit with the power transistors at the supply pins the stability can be tricky, if at high frequency the supply current changes. So some OP-amps are more suited than others for this type of circuit.
It often work, but may fail with a few chips. The rather slow µA725 equivalent amplifier may also be a problem - more suitable to do DC correction and have a discrete amplifer for the rest.

The BD680 transistors have a rather limited SOA. In the Motorola data-sheet the permitted power already above some 25 V. At some 50 V the permissible current would be down to the 100 mA range.
These transistors are not really suiteable for a linear power amplifier, at least not with more than some +-20 V supply.

[Zoli]

If you're looking for AWG amplifier, read the attached app note(don't get confused by the title, is rated for 1MHz ) for inspiration. It still can use the K153UD5 instead of OP97.
BD680/BD681: f_T is ~3MHz IIRC(dig out your old ß catalogues, since the current data sheets don't have the parameter), which make them suitable maybe for audio frequencies; for alternatives, look for onsemi and Sanken darlingtons; MJE15034, MJE15035, 2SB1559, 2SD2389 are examples.

[strawberry]

additional components could potentially add THD
now days, benefit going discrete is for high speed/wideband or high power amplifiers or high voltage amplifiers

TTA/TTC004B ft=100MHz 10W
for example 33120A use ft=1GHz output transistors (it is about parasitic capacitance)

new stock IC https://www.ericasynths.lv/shop/ics/

[mawyatt]

If you just want to buffer V & I for the AWG, then the LM1875, LM3886, PA443, & OPA462 are some ICs that come to mind. We did a few AWG buffer amps awhile back if interested.

Best,

[RoGeorge]

Integrated high voltage power opamps are very tempting, indeed, they seem ideal as external amplifiers for a signal generator.

The "Alexander Current-Feedback Amplifier" topology seems to be quite famous in the audio amplifiers world (e.g. https://www.diyaudio.com/community/threads/kuroda-tl071-60w-fet-amp-1982.164476/#post-2145208).  The schematic looks very complicated, but maybe I should study it closer.  Same as "Kuroda" amplifier for which there is even a patent https://patentimages.storage.googleapis.com/d2/f5/7f/028ea37d2fda6f/US5097223.pdf


For a very first version of an AWG buffer I would like very much to use these parts that are now sitting in a paper envelope doing nothing for the past 20-30 years. 
- uA725 opamp https://media.digikey.com/pdf/Data%20Sheets/Texas%20Instruments%20PDFs/LM725.pdf
- BD680/681 Darlington https://www.onsemi.com/pdf/datasheet/bd681-d.pdf https://www.onsemi.com/pdf/datasheet/bd682-d.pdf

Another goal is to make this amplifier as an exercise in analog design and feedback loop stability.

Did a brief simulation with LTspice.  Found some BD681/682 models from Onsemi and cobbled some symbols for them (all needed files are included in the attached LTspice zip).  Couldn't find any model for uA725, so used the OP07 model from Analog Devices instead (OP07 was a followup of the uA725 design - https://www.analog.com/media/en/training-seminars/design-handbooks/Op-Amp-Applications/SectionH.pdf page 57).

The simulation worked from the first try, great success! 



The last trace, V(out) in yellow shows the output for a 100kHz square wave input of 1Vpp.

Well, it seems to be working OK, yet it's not very reliable.
- final stage current is very sensitive to R5 and R6 (the current to voltage R parallel with the Darlingtons' BE)
- tendency to oscillate under different loading conditions, e.g. with capacitive loads
- no short-circuit protection yet

The one that bothers me the most is the stability.

Naive approach would be to limit the speed of the opamp (uA725 has external compensation pins) so the power stage would be fast enough (for the slowed down opamp) that it wouldn't matter for the feedback loop.  Seen similar ideas in some LT application notes (i.e. APPENDIX C in the AN-47 page 86 "The Oscillation Problem – Frequency Compensation Without Tears" https://www.analog.com/media/en/technical-documentation/application-notes/an47fa.pdf )

I'm a complete beginner in stability design, so how do I start improving the AC performance for this circuit?

[DavidKo]

In AN86 and LT1055 datasheet there is and amplifier with +-100V output and 25mA (with current limiting). You can tune it to your requirement.

[Zoli]

...
The one that bothers me the most is the stability.

Naive approach would be to limit the speed of the opamp (uA725 has external compensation pins) so the power stage would be fast enough (for the slowed down opamp) that it wouldn't matter for the feedback loop.  Seen similar ideas in some LT application notes (i.e. APPENDIX C in the AN-47 page 86 "The Oscillation Problem – Frequency Compensation Without Tears" https://www.analog.com/media/en/technical-documentation/application-notes/an47fa.pdf )

I'm a complete beginner in stability design, so how do I start improving the AC performance for this circuit?

Add emitter resistors to the BD's; 0.22Ω or greater should do the job, and help implement the short circuit protection. Another stability option would be parallel RL in series with the output; 10µH||10Ω is a good start.
Alexander amplifier: I've presented as example for discrete CFB amplifier, which is typical as AWG output amp. Most of the original parts are unobtanium(well, except fleabay's NOS), as result, it needs redesign for the current reality(pun intended).
Mixed design(opamp+discrete): local feedback(like the BD's emitter resistor) has increased importance to keep the situation under control; general feedback doesn't always solves the problems.

[KRISTOFFER]

I spent a long time playing with bootstrapped op amps to yield a high voltage swing. Seems to be trick in selecting the overall gain. Some stand alone op amp circuits with a low gain will become unstable (often stated in the data sheet as stable with a gain of, for arguments sake, 10 or above). In bootstrapped circuits the same problem exists. You know when you have this problem when you power it up. The ouptut will immediately swing to the positive or negative rail.

As regards more current, a simple totem pole output, as long as you are happy with the cross over distortion.

My final approach was to build a discreet op amp : differential pair, class A amp and output stage. The first two sections using current sinks. With BCX53 and BCX56 transistors throughout it was happy working on +/- 40 volts.

[David Hess]

Naive approach would be to limit the speed of the opamp (uA725 has external compensation pins) so the power stage would be fast enough (for the slowed down opamp) that it wouldn't matter for the feedback loop.  Seen similar ideas in some LT application notes (i.e. APPENDIX C in the AN-47 page 86 "The Oscillation Problem – Frequency Compensation Without Tears" https://www.analog.com/media/en/technical-documentation/application-notes/an47fa.pdf )

I'm a complete beginner in stability design, so how do I start improving the AC performance for this circuit?

Fix the local feedback which goes to the output of the operational amplifier.

[RoGeorge]

Ha, ha, nailed it! 
I mean, cobbled together a schematic that is simple, yet very sturdy against loading conditions, and high speed, too. 

The main hint was this:

[Current-Feedback] ... CFB amplifier, which is typical as AWG output amp ...

That made me curious to learn more about Current-Feedback (CF), and digging through literature found that multiple references were pointing to a Sergio Franco's article in the EDN magazine from Jan 5, 1989:  "Current-feedback amplifiers benefit high-speed designs", which happens to be available at the archive.org, as a scan of the paper-printed magazine (see page 161 of 286):  https://archive.org/details/edn-1989_01_05/page/160/

Once knowing the basics, it was an exercise to convert this:



which uses voltage-feedback and increases only the max current, into a current-feedback amplifier.  Also wanted to increase both the max current and the max voltage of the opamp, so this is what I came with:



To understand why that works, let's remove the current feedback resistor, Rcf.

The idea was to use the opamp as a voltage repeater, so the output of the opamp to become a zero-impedance inverting-input for the CF network.

Since the positive-input of the opamp is tied to ground, and with the opamp as a voltage repeater, the output of the opamp will always be zero, so no need for a load resistor for the opamp.  The output of the opamp (which becomes the inverting input for the CF network) will appear as if it were acting on a zero ohms load (a zero ohm virtual load).

The zero ohms load is virtual, but the current has to go somewhere, so the current will close through the power supply pins of the opamp.  This means the current through the emiters of Q1/Q2 will mirror the virtual current through the zero-ohms opamp's virtual-load.

R1 and R2, together with the quiescent current of the opamp will set the minimum idle current through Q5/Q6 collectors, so to avoid the crossover distortions.

Notice the base of Q1/Q2 are kept fixed, at +/-15V respectively, and this will be the opamp's supply voltage (minus 0.6V drop on each BE junction).  Since the Ib of a transistor is much smaller than Ic, the BE current of the Darlingtons Q5/Q6 will see a current proportional with Ic of Q1/Q2, which is the current going through the supply pins of the opamp, which for this configuration is the same as the current going through the zero-ohms virtual load.  The Darlington pairs Q5/Q6 will amplify further, and they'll sum their collector currents at the OUT node.

Q3/Q4 are there to dial down the power transistors Q5/Q6 in case of short-circuits or over-current at the circuit's output OUT.  D1 and D2 are to protect the opamp against voltage inputs greater than +/-15V.

Finally, let's close the CF loop.  From OUT, the Rcf resistor together with Rin form the CF network.  Notice the Rin also acts as a terminator for the 50ohms cable, cable that is coming from the signal generator.  Notice that Rin is connected in series with the cable, not in parallel.



And that is how it all works as a Current-Feedback Buffered Opamp, and why all the advantages of a CF topology apply to this schematic, too (or at least that is how I assumed it works while designing it, 'cause this is my first current-feedback design  ).



The LTspice plots are for a 1MHz square wave input of 1Vpp, and the output produces about 100Vpp into a 50ohms load (+/-1A) parallel with a 1nF.

In comparison, the former voltage-feedback topology was struggling at only 100kHz, while the current CF has no problems at 1MHz, and seems to be going OK up to about 10MHz.  I'm very impressed how well it performs in simulation.  Will see how well it works in practice, or if works at all. 

[Zoli]

Good job ; glad to supply you the proper hints. Now, real word behaviour is a little different, I don't expect even 1MHz(limit: the Darlingtons f_T; if 3MHz, you need to feed the base around 400mA losses included around 1MHz signal).
If the OP07/K153UD5 can supply 30mA(stretch), I expect around 100kHz real life bandwith, which is an extremely respectable number.
Again: good job , and keep us posted with the results.

[David Hess]

which uses voltage-feedback and increases only the max current, into a current-feedback amplifier.

It uses both.  Feedback to the inverting input is voltage feedback.  Feedback into the output of the operational amplifier goes into the emitters of the output transistors operating in common base mode and is current feedback.

Another way to look at it is that the current feedback into the output of the operational amplifier is local feedback which controls the voltage gain of the booster stage, made up of the operational amplifier's output transistors and the external current mirrors, which is all enclosed within the voltage feedback loop.  The local feedback gives the booster a fixed voltage gain simplifying frequency compensation of the voltage feedback loop.

It helps to include the operational amplifier's output transistors as part of the schematic.  Internally the operational amplifier's output stage is equivalent to the diamond buffer used by current feedback operational amplifiers for exactly the same purpose.  This makes me wonder what happens if you try this circuit with a rail-to-rail output operational amplifier where the collectors of the operational amplifier's output transistors are exposed instead of the emitters; the local feedback would no longer work the same way.

Below are a couple of modern examples from Walter Jung.

[RoGeorge]

Thanks for the attached examples, though they puzzle me. 

At a first look, I don't see what would be the advantage of having two global feedback loops, one for voltage and one for current, like in the two schematics from Walter Jung.  Will have to brush my knowledge about multiple feedback loops first, then think about it again.

For now, anything less than a voltage follower opamp seems like a waste of bandwidth.  Av=1 (as in inverting input tied to opamp's output) will transform the opamp alone into a black-box working the same from DC to its GBW, because the noninverting input is tied to ground (in my schematic), which will keep the inverting input and the opamp's output also at zero (in an ideal opamp).  That will make the slew-rate of the opamp irrelevant, because the opamp's output pin will stay at zero volts all the time so no voltage swing there.

However, the output stage inside the opamp is in use, and it will push and pull current through the opamp's output pin, either into Rin (when the Rcf is disconnected) or in both the CF resistor Rcf and the input resistor Rin when the Rcf is connected in the CF loop.  The current through the opamp's output pin will be reflected into opamp's supply pins, one supply pin for each alternance.

If I allow the opamp to have any Av>1, then the output pin will have to swing it's voltage, the slew-rate comes into play, and the max usable frequency becomes smaller than GBW.

Av=1 (locally, for the opamp alone configured as a voltage follower) seems the best choice, so why this isn't the norm? 

About the rail to rail implications, in my schematic both the input and the output pins are always at 0V.  Only the currents are swinging, the voltage is flat zero (for an ideal opamp).  Does the opamp's RR specs matter, as long as I stay inside the limits of the maximum output current for the given opamp?

[David Hess]

At a first look, I don't see what would be the advantage of having two global feedback loops, one for voltage and one for current, like in the two schematics from Walter Jung.  Will have to brush my knowledge about multiple feedback loops first, then think about it again.

It is not two *global* feedback loops.  There is a local current feedback loop, which controls only the gain of the output stage, and there is a global feedback loop, which controls the overall gain.

For now, anything less than a voltage follower opamp seems like a waste of bandwidth.  Av=1 (as in inverting input tied to opamp's output) will transform the opamp alone into a black-box working the same from DC to its GBW, because the noninverting input is tied to ground (in my schematic), which will keep the inverting input and the opamp's output also at zero (in an ideal opamp).  That will make the slew-rate of the opamp irrelevant, because the opamp's output pin will stay at zero volts all the time so no voltage swing there.

The booster still has voltage and current gain no matter what the operational amplifier's voltage feedback loop does.  As a matter of fact, the booster output stage can use a gain of one buffer, like an LT1010, to drive it.  Current feedback amplifiers use a diamond stage buffer to drive the output transistors.  In both cases a gain of 1 buffer is providing both voltage and current gain.

So below for an example where an LT1010 is used to do this.  The LT1010 buffer by itself has no voltage gain, but combined with the external transistors, the two resistors providing current feedback create voltage gain.  The added transistors implement simple Vbe current limiting to protect the output transistors.

If I allow the opamp to have any Av>1, then the output pin will have to swing it's voltage, the slew-rate comes into play, and the max usable frequency becomes smaller than GBW.

The boosting output stage multiplies the operational amplifier's slew-rate by its own voltage gain, which is one reason this configuration was popular in the past.  It allows a 741 type operational amplifier to have enough slew rate to more easily support audio applications.

About the rail to rail implications, in my schematic both the input and the output pins are always at 0V.  Only the currents are swinging, the voltage is flat zero (for an ideal opamp).  Does the opamp's RR specs matter, as long as I stay inside the limits of the maximum output current for the given opamp?

The rail-to-rail specifications will not matter because very little output voltage swing is required to generate the current to drive the external transistors.

[magic]

Things can be rearranged a little. This is now clearly a CFA whose offset voltage and inverting input output impedance are being corrected (at DC and low frequencies) by the precision VFA.

This is of course mostly equivalent to the original, where the job of Q1,Q2 was done by the internal EF output stage. Mostly, because internal output current limit is now bypassed (at a potential risk of frying Q1,Q2) and perhaps the better PNP would bring benefits at high signal frequencies.

It's not very practical though, because quiescent current is poorly defined IRL and Q1,Q2 seem prone to thermal runaway which may subsequently blow up the output stage. Some emitter degeneration could help, opamps have it too.

This makes me wonder what happens if you try this circuit with a rail-to-rail output operational amplifier where the collectors of the operational amplifier's output transistors are exposed instead of the emitters; the local feedback would no longer work the same way.

Probably nothing interesting, because the output stage still has local feedback at AC and well beyond the bandwidth of the opamp by means of Miller compensation. The collectors are high impedance so you can shift the output voltage no problem, and then the bases shift by the same amount and conductance is modulated exactly the same as in an EF.

Okay, actually less than EF, because you are now working with the emitter-to-emitter combination of the internal output transistor and the external cascode. So output impedance is doubled.

That's what the theory looks like to me, at any rate. SPICE would probably agree.

[RoGeorge]

That one should work, too, but then it doesn't use "the trick" any more.  The one in which the current through the opamp's supply pins is used to drive the rest of the power stage.  I was liking that trick so much... 

Then I'll need to add 2 more transistors to make the Vcc_oa and Vee_oa starting from the +/-50V (Vcc and Vee).  In my CF buffer schematic it was not shown, but Q1 and Q2 are there to prepare the +/-15V supply for the opamp, with a Zener in their base, but I've drawn a Vcc_oa and Vee_oa labels instead, to make the schematic look cleaner.

In your version Q1, Q2 should be something with higher Vce (these are 45V), but that's irrelevant, what I think would be a real disadvantage is that the output pin of the opamp will see a swing of up to 1.4V (because of D1+D2), and that will make the slew-rate of the opamp to matter again, as if it were in a normal voltage-feedback with 1.2Vpp on the opamp output pin, so the opamp (which mainly dictates the bandwidth of the entire schematic) will lose some bandwidth.

I think the opamp configured as a standalone voltage follower is a key aspect, and it shouldn't be discarded.  A voltage follower will give a useful band equal with the opamp GBW.  Any voltage swing in the opamp output pin will make the max frequency to become smaller.

I like that we have the same color settings in LTspice. 


LATER EDIT:
Now I've noticed the Vpoz_oa and Vneg_oa voltage source definitions are not visible in your picture, which combined with the same color scheme makes me think your modified schematic might have been made by photo-editing (so no LTspice simulation posibile).  If so, the LTspice files were attached already together with the png waveforms, so the .asc should open in any current LTspice such that the schematic can be edited from LTspice, and simulated, too.

Unzip the original attachement "RoGeorge's CF buffered opamp.zip" in a new directory, then open the .asc with LTspice.  The color settings were not attached in the zip file, but they were posted once (I hope I didn't change them much since):  https://www.eevblog.com/forum/eda/where-does-ltspice-keep-its-settings/msg4150582/#msg4150582

[magic]

This makes me wonder what happens if you try this circuit with a rail-to-rail output operational amplifier where the collectors of the operational amplifier's output transistors are exposed instead of the emitters; the local feedback would no longer work the same way.

Probably nothing interesting, because the output stage still has local feedback at AC and well beyond the bandwidth of the opamp by means of Miller compensation. The collectors are high impedance so you can shift the output voltage no problem, and then the bases shift by the same amount and conductance is modulated exactly the same as in an EF.

On second thought, it's not that trivial, because the compensation caps need to carry the base or gate current of the output devices, which may be considerable, particularly at frequencies beyond the design bandwidth of the opamp. OTOH, in an EF output opamp, this current still needs to be generated by the VAS.

the output pin of the opamp will see a swing of up to 1.4V (because of D1+D2)

Not sure what you mean, I re-checked the schematic and I don't think I have made a mistake.

The diodes bias the BC5x7 buffer into class AB to avoid slewing 1.4V on transitions between sourcing and sinking, so this should be sorted.

Small voltage swing will occur to cancel the base-emitter voltage shift due to external load. But I made a point that this circuit is little different from using the EF output stage of OP07, and indeed, the internal stage's base voltage also must shift to accommodate loading of the output.

the opamp (which mainly dictates the bandwidth of the entire schematic) will lose some bandwidth

This output booster has a life of its own. You could remove the opamp and simply ground the junction of the diodes (on my schematic) and it still works, potentially with bandwidth in excess of what OP07 can do. Perhaps not a lot of gain, it's going to depend on output loading a lot.

Now I've noticed the Vpoz_oa and Vneg_oa voltage source definitions are not visible in your picture, which combined with the same color scheme makes me think your modified schematic might have been made by photo-editing

It might have been

[RoGeorge]

the output pin of the opamp will see a swing of up to 1.4V (because of D1+D2)

Not sure what you mean, I re-checked the schematic and I don't think I have made a mistake.

Doh, I was wrong with that one, sorry.

[Terry Bites]

You can use a readymade audio amplifer inside the loop of a more precise opamp. You need to apply compensation.

[RoGeorge]

Indeed, a power amp would be very good, just that I don't have any, except one or two A2030 that can be powered at max 36V only.

Meanwhile tinkered around scavenging radiators from former PC power supplies, for the power transistors, then got sidetracked from this proj, but it happened that I've just found minutes ago an entire application note exactly about buffering opamps, AN-18 Power Gain Stages for Monolithic Amplifiers, by Jim Williams. 

Will read it tomorrow hoping that that won't make me to redo all, fingers crossed.

[KRISTOFFER]

Try this.


Remove R1, D1 and D2, and use R4, D3 and D4 to bias Q3 and Q5.
With supply rails above +/- 24 Volts increase R6 to about 1K.
You need to select R3 to give 0 Volts at the output for symetry.

The last one I did was R6 = 1K, R3 = 3K9, all transistors BCX53 and 56.
Supply +/- 40 volts. Output swing = 60V Peak to Peak.

[magic]

ELI5 how this input stage turns 200µA into 2×50µA while only drawing 200nA bias current?

It's obviously imbalanced, which costs you over 20mV unnecessary offset voltage.