Simple 4-quadrant PSU with remote sense - please recommend topology/schematic

[Marco]

A *difference* amplifier absolutely is the correct topology for this, as you obviously need to measure the difference between Sense + and -.

Well that's treading into semantics, you need to output/amplify the difference but generally difference amplifier is understood as a specific way of doing it. The one opamp with 4 resistors difference amplifier, a chopping differential amplifier like the MAX4208 is clearly superior in most ways and is internally something fundamentally different (different from a 3 opamp inamp topology too, even though it's still called an inamp, semantics strike again).

I'm not sure what you'd gain by converting the voltage to current and back to voltage in this case?

The same thing as with current sensing, common mode rejection.

Don't these current shunt monitor IC's rely on laser-trimmed matched resistors, too, or am I completely wrong here?

It minimizes the number of high precision resistors needed ... and their drift doesn't impact common mode rejection by design. Doesn't seem a big win here though in retrospect.

[timb]

A *difference* amplifier absolutely is the correct topology for this, as you obviously need to measure the difference between Sense + and -.

Well that's treading into semantics, you need to output/amplify the difference but generally difference amplifier is understood as a specific way of doing it. The one opamp with 4 resistors difference amplifier, a chopping differential amplifier like the MAX4208 is clearly superior in most ways and is internally something fundamentally different (different from an inamp too).

I'm not sure what you'd gain by converting the voltage to current and back to voltage in this case?

The same thing as with current sensing, common mode rejection.

Right. The OP wanted simple and cheap. Doesn't get much simpler than four resistors and an op-amp. (You don't even need the extra op-amp since you can feedback right to the power amp itself as I showed in another post.)

For kicks, how exactly do you propose to convert the voltage to current?

[Marco]

For kicks, how exactly do you propose to convert the voltage to current?

"something like INA170 for instance"

As I said before though, before you even commented ... "Considering it further though all you would really accomplish is make common mode rejection independent from the resistors drifting, offset and amplification would still be drifting along." So it doesn't really seem worth the effort.

[timb]

For kicks, how exactly do you propose to convert the voltage to current?

"something like INA170 for instance"

As I said before though, before you even commented ... "Considering it further though all you would really accomplish is make common mode rejection independent from the resistors drifting, offset and amplification would still be drifting along." So it doesn't really seem worth the effort.

Ah, gotcha. I didn't realize that had a current output. I was thinking of some of the other INA current shunt monitors with voltage outputs.

It would increase common mode rejection, but yeah not sure if it's worth the trouble in this particular case. An LT1990 difference amplifier has similar levels of CMMR and has a higher common mode range; there's also a couple of in-amps from TI and AD with better CMMR but at much lower voltages.

Though, I can think of a couple of niche applications where doing bipolar differential voltage sensing to current could be extremely useful. So thanks for thinking outside of the box! I might be able to use this in a future circuit.

[macboy]

I didn't see anyone mention using audio power amps as the power op-amp. On possible device is the LM3886 which can dissipate up to 125 W. Although marketed as an audio amplifier, it is fundamentally just a big power op-amp with 8 MHz of GBP and 115 dB open loop gain. The device is very well specified in the datasheet for things like SOA, pulse power limit, and many other parameters (no fewer than 45 graphs), and being an audio amp, it is designed to drive reactive and low resistance loads. Any comparable power op-amp specified (read: marketed) for industrial applications will cost 10x as much.

[Yansi]

Noone didn't mention, because this idea is wrong from a few perspectives of this PSU design:

1) You cant easily heatsink the package like that MULTIWATT thingy of LM3886. The only one M3 screw for such package is just hilarious and any other means of mounting is just stupid. Too complicated. You cannot beat two TO3 mammas on a chunk of aluminium.
2) Loop stability. In general, the audio power amps are mostly designed so they are stable only at certain level of gain. For example TDA2005 is the most know bastard, that just needs at least those 20dB to be really stable. I think the LM3886 won't be much different. And as long as I need only +-1V output range, 20+dB gain does not make much sense.
3) minimum supply voltage. I doubt the LM3886 will work at about +-3V supply.  Why would I use +-9V minimum supply for a +-1V output range? But this can be easily done with two TO3 mammas... no problem.

Conclusion: integrated audio poweramps are really not suitable for this low voltage high current application. Difficulty of heatsinking, too high minimum gain for stability and too high minimum operating voltage are those culprits.

NOTE:   LM3886 is stable above 20dB gain, I was right. ...just have looked into the datasheet.


Since last time, I have made some progress with breadboarding.  The discrete power stage design is I think closed, schematic below. Sorry for not filling in the resistor values, these are still  TBD. The breadbord version of the circuit is simplified and using little bit different scheme.
I have also breadboarded the difference amp section for remote sensing. The whole is connected as on the image below below.
Please don't be worried about those probably unknown to you semiconductor types used, these are some old beefy mammas made by our old national manufacturer, that has quit doing some decades ago.

Now only CK1 is used and it seems to be stable. CK2 and RK2 not used (bcs of lacking knowledge how to calculate them. Can somebody help please?).
However serious stability issues occur when capacitively loading the supply output. 100n or 3uF or whatever small cap on hand, wild oscillation starts. How to solve that issue, how should I tune the feedback loop so it will be stable with C loading on the output?

Thx, Yansi

[Yansi]

Huh? I've just might solved the stability issue.

By looking in the Keithley 228A schematic, the control opamps seems to have only about 5kHz bandwidth (47K x 680p). I have calculated what I have there just to compare... and ahaa moment! 670kHz bandwidth... So instead of 250pF x 500ohm I have increased the feedback capacitor to 3n3 (meaning the bandwidth is about 96kHz) and bingo!  It does now tolerate the 3uF MKP foil cap directly on the output.

So I am quite happy now. Let's proceed to the current sense and feedback design.

//EDIT: But still, after more slowing down the BW to only 14kHz it does not like capacitive loading of 1uF or less. :-/

Yansi

[Kalvin]

This application brief might give some help:

"Unlimited Capacitive Load Drive Op Amp Takes Guesswork Out Of Design"
http://www.ti.com/lit/an/snoa808/snoa808.pdf

[Yansi]

Might, thanks, but did not :-/

For me, a common undereducated hobbyist, such appnote looks like "buy this special opamp everything will suddenly work blah blah blah" 

But I know, it is not funny.   

[Kalvin]

Might, thanks, but did not :-/

For me, a common undereducated hobbyist, such appnote looks like "buy this special opamp everything will suddenly work blah blah blah" 

But I know, it is not funny.   

The LM8272 discussed in the application brief has capacitors placed across the output transistors' base and collector increasing the Miller capacitance thus reducing the bandwidth when driving capacitive loads. I haven't been simulating the concept but it might help. Simulating your whole circuit and evaluating the phase margin with the Miller capacitors in place would give some practical insight, too.

[Marco]

The output totem pole is not a pair of followers ... replicating this circuit semi-discretely seems a non starter.

In loop compensation seems the most natural fit.

[Yansi]

It not only seems, but it it really does fit! That is what I have exactly done, but based on stupid experimenting with the circuit. If you look in the circuit few posts above, I have added some small cap  accross the opamp and that helped it stabilize.

Now I have also verified by breadboarding the current limiter circuit from LM12 datasheet, with that transistor feedback thing. It really does work as advertised.

Now I only need to add current limiter active indicator (some comparator craziness like in the Keithely 228A) and hooray, we can make us a PCB. (A then do the whole capacitor tweaking process again) 

[Yansi]

Oh goodness... Nobody spotted my superfail in the schematic? The buffering stage before the power darlingtons will burn like shit, when the +-3V power supply goes out first, or when the opamp saturates. FAIL right there!

The buffer stage must be supplied from the 3V supply (I don't like that cause too much voltage drop accross 3 BE junctions) or I will throw em out, as those darlington mammas has enough h21e at 5A I can easily drive them with th opamp directly (as is also currently done on my breadboard).

So here's the result, after few days of thinking and one whole day of breadboarding. Ugly, but works. Stable (or stable enough), current limiter works fine and so does current limit indicator, providing nice open collector output for the stupid arduino they want to drive this beauty with 

Now it's time for PCB layout.  I'll try to have it on single sided board. Will throw in the full schematic for check or diy, if there will be interested people.

Thx, Yansi