Current feedback questions

[Consul]

I've read lots of electronics books that supposedly explain these things, but for some reason, I still have trouble with some concepts.

Okay, here's another question I have a hard time wrapping my head around: What is the difference between voltage feedback and current feedback?

I know it sounds like it should be obvious, but I seem to be missing something. I don't understand how current feedback would work at all, since it's a varying voltage that carries your signal, right?

Actually, a discussion on feedback in general might be valuable to a lot of people here.

Thank you all for the help!

 

[bcarso]

Your difficulty is not that odd since there's some ambiguity in the nomenclature.

So-called "Current feedback amps" connected in the conventional fashion use voltage in and voltage out in the circuits they are part of, and look on paper very much like other op amps. But the difference is that the inverting input is very LOW impedance (ideally, zero), while the two inputs of a traditional op amp are both relatively high Z.

Amplifiers come in at least four different flavors, ideally. Any real amp can be a mixture of them, but is usually trying to be as close to one of the ideals as possible.

The most familiar ideal is the voltage in/voltage out amp. The ideal is infinite input Z, zero output Z, and some voltage gain, maybe a lot.

There are also the transconductance amps: voltage in, current out. The ideal is infinite input Z and infinite output Z. The "gain" is delta current out/delta voltage in.

Then we have current amps: current in and current out, with zero input Z and infinite output Z. The gain is in delta current out/delta current in.

Finally in the last quadrant is a transresistance amp: current in, voltage out, zero input Z, zero output Z. The gain is delta voltage out/delta current in.

We can choose to use current as our signal variable if we like, and sometimes there are advantages. Where, with voltage as the variable we can send the same signal to a number of inputs, with current as the variable we have to replicate the currnt somehow for multiple inputs, or divide it with the associated losses, or have a way of recycling it somehow. This is usually less convenient than feeding a voltage source to multiple inputs. We can use current as the feedback variable as well, and so have a "current-feedback" system.

So far all that is described applies to amps with a single input and a single output. So we haven't even gotten to an op amp yet.

Conventional op amps are a two-input affair with high Z at both inputs; the output is a voltage that is the highly amplified difference of the two input voltages. If this differential gain is high and falls off in a nice way with frequency then we can apply feedback in various ways and get a smaller but more precise gain, or any number of other overall circuit transfer functions and behaviors.

All of the other flavors of amps described can be extended to two-differential-input forms as well, and that generates a bunch more ideal cases. And then we can mix the types of input too: our "current-feedback amp" has a voltage-input noninverting input, a current-input inverting input, and a voltage output. So there are quite a number of possibilites!

If that weren't enough, there are also differential output amps as well.

It would be interesting to see a graphic of all the possible ideal types of amplifier. But then, I don't get out much.

 

[Svart]

"Current feedback" is used a lot in motor control. I am getting ready to start a design using a current feed back PWM controller for super high brightness white LEDs(yes, I understand this isn't exactly what you asked but it's a current based amplifier system with a lot of extra junk thrown in to make it look cool to the ladies...) which senses through a pair of pins through a low value resistor on the low side of a load. changes in voltage/current(ohms law) across the resistor vs. duty cycle can either cause the IC to bring up the duty cycle or bring it down depending on the *set* current thresholds. in fact most switchmode PSUs are this type.

ah screw it just trying to throw in a few cents. back to my :guinness:

:green:

 

[bcarso]

"with a lot of extra junk thrown in to make it look cool to the ladies...)"

It appears to please Darren's Avatar; YOU ARE WORTHY.

 

[PRR]

> "Current feedback" is used a lot in motor control.

That's different. In motor control, motor current is the important fact about the output.

"current feedback opamp" is partly marketing. Conventional opamps have high-Z inputs. The designers bust a gut getting a high input impedance. "current feedbag" amps may have a fairly low impedance at the inverting input where we apply feedback. Aside for reducing gut-bust, this has some performance advantage. As bcarso says, we want "gain is high and falls off in a nice way with frequency". The nice-way fall-off is set by an R and a C. In conventional opamps, both are hidden inside, which means they have to be selected for worst-case feedback (unity-gain). It is possible to use an external C that is selected when the feedback is designed (709, LM101); but cheap chip opamps are supposed to free us from such considerations. The real trick in current-feedback opamps is that the R is external and is the feedback network impedance seen at the feedback pin. For high gain, use a low-Z network; for low-gain use a high-Z network, and the nice-falloff shifts to suit the gain. In many-many applications, this reduces to using a semi-constant value (like 1K) for the series (output to "-" pin) resistor, and figuring the shunt ("-" to ground) resistor to set the gain. Now the nice-falloff or compensation changes with closed-loop gain in such a way as to make closed loop bandwidth constant and as good as the chip can do for any closed-loop gain.

 

[Samuel Groner]

Is there some good literature on current feedback? Something like a "Douglas Self for Current Feedback"? I found many sources which describe the theoretical background well, but not much that helps with successful implementation in practice by giving working examples.

Edit: I'm not asking on how to use CF ICs, rather how to design CF discrete (or IC) amps.

Samuel

 

[bcarso]

Samuel I think that stuff is still fairly proprietary. Most of the articles and material in books is still fairly abstract. Some manufacturers show fairly accurate schematics for what is inside their ICs and you can kind of go from there (AD, Linear Tech in particular).

Basically the input stage is often a diamond quad like Jung's buffer, with the "output" as the inverting input and the "input" the noninverting input. Then the collectors of the "output" devices go to current mirrors off the respective rails and the output of the mirrors tie together and are buffered by another diamond follower.

Comlinear tried to patent this whole approach years ago but I believe were rebuffed as someone pointed out that this kind of feedback had been used for years without it being made as explicit.

 

[Consul]

Wow, thanks for all of the information, everyone! And it's nice to know some folks here like Mai, too. :wink:

What I'm getting from all of this is that current feedback seems to be a more advanced subject that I don't really need to worry about too much just yet. Is that a fair assessment?

 

[bcarso]

"What I'm getting from all of this is that current feedback seems to be a more advanced subject that I don't really need to worry about too much just yet. Is that a fair assessment?"

Well, sorta. Master voltage amplifiers and their associated feedback schemes and you will be able to handle an awful lot of real-world stuff.

But do realize that current can be the signal in circuits as well as voltage. You can't really have one without the other, but many times design and analysis in terms of signal currents is far more straightforward.

 

[Consul]

Maybe I should just stick to building from known schematics for now. *sigh*

Oh well, I'll get the hang of this eventually. Thanks for the help!

 

[bcarso]

Play with them and think about them, and calculate as often as possible, and you will be ahead of a legion of people who don't.

 

[Samuel Groner]

Some manufacturers show fairly accurate schematics for what is inside their ICs and you can kind of go from there (AD, Linear Tech in particular).

I tried to work a bit from a schemo found here (bottom of the page), which seems to be of the topology you described.

Unfortunately, simulation is not that successful; using four ideal 1 mA current sources does not work at all (output sits at -19 V or so), replacing them with 1k resistors makes operating points better, but no usuable small signal response yet. Any hints?

Samuel

 

[bcarso]

That's the topology all right. It should simulate---what are you using for the resistor values?

The C's shown could be consolidated into a single one to ground, not that this should change things much in sim.

Also, often the current mirrors are three-Q Wilson ones for higher accuracy, but again that shouldn't change the sim too much. Bandwidth will be degraded if the R's in the two-Q mirrors get big (say >>100 ohms).

One thing to note: as such without a lot of massaging of the device geometries the PNP and NPN parts don't cancel their Vbe's very well, so there will be an appreciable d.c. offset. I don't know how the makers of the real ICs deal with this---probably with (a) little offset R('s) in the common-collector input device emitter(s).

That's a nice little ref btw. Not too many mistakes at first glance.

 

[Samuel Groner]

what are you using for the resistor values?

R-sub-M = 180
R-sub-T = 15
R-sub-F2 = a few kOhm
C-sub-D = 15 pF

Re-reading the text, I realize that R-sub-T needs to be higher; it just looks so unfamiliar! Any suggestion for a good value?

The C's shown could be consolidated into a single one to ground

So one C from the collectors of Q10/Q12 to ground?

Samuel

 

[bcarso]

Yeah one C to ground.

You might want to run a little more current through the output buffer section, and for wideband op the r's could be a hair smaller, but otherwise there isn't anything wrong with those values that I can see. Beware of sims with i sources that the simulator assumes have been on forever before the supply rails come up, but if both are assumed on before the sim starts things should be o.k.

I don't think R sub T need be higher. The output quiescent I will be the same as the current source I for the assumption of perfectly matched transistors. You will want some R there for thermal considerations---15 ohms sounds o.k. to me, depending on the transistors.

Don't go for a whole bunch of closed-loop gain as the offset voltage mentioned will throw you towards the rail. Gains of ~10 ought to be fine though. What transistors are you using?

Maybe I'll throw something like this into sim at some point time permitting.

 

[clintrubber]

from Samuel: Edit:

I'm not asking on how to use CF ICs, rather how to design CF discrete (or IC) amps.

There's a nice design started by Tamas just around the corner: the GainBloak
http://www.groupdiy.com/index.php?topic=5866


Cheers,

Peter

 

[clintrubber]

from Brad:

One thing to note: as such without a lot of massaging of the device geometries the PNP and NPN parts don't cancel their Vbe's very well, so there will be an appreciable d.c. offset. I don't know how the makers of the real ICs deal with this---probably with (a) little offset R('s) in the common-collector input device emitter(s).

FWIW: Obviously, that DC-shift can sometimes just be ignored. Say when an amp has (closed loop) gain above unity, the output doesn't mind the input lives one or a few Vbe's above the mid-supply point of the output.

Regards,

Peter

 

[bcarso]

Agreed---after all Tamas gets by with a whole Vbe---cap coupling solves a multitude of problems for audio apps.

OTOH, as they say, there's no capacitor like no capacitor.

I would still like to know how the comlinear-type topology gets rid of the offset though. I know AD actually just use diodes out of the same transistors to bias with, at least according to their simplified schematics, as opposed to e-followers. But that takes the n.i. input Z down a bunch.

 

[clintrubber]

Hi Brad,

Fully agreed, the DC-shift can be kept, but you'll need a cap. Luckily it can be a film-cap because of the relatively Hi-Z of the n.i. input.

FWIW (& still not answering your comlinear-type topology question ) is this:
how about a diff-in, diff-out version of the Gainbloak ? The circuit is simple, so easily duplicated for fully balanced.
(Simplest incarnation would be double single-ended iso of 'true' balanced, but it'll just work I think)

Using TXs for in and out, these can simply live at different voltages above ground
(the sec of the input-TX and the prim of the output-TX).
No cap needed anywhere, if I'm right.
But OK, there are TXs now.

Regards,

Peter

 

[bcarso]

I think what I would do would be to insert a diode-connected transistor or maybe a few in parallel in the feedback network to the input emitter. Drive a current from above and sink it out below. So you wind up with this tracking Vbe with low impedance---adjust the bias current for the same current density as the input device in the case of the multiple paralleled devices. Maybe even bypass with a small C across for the highest frequencies.

But this is significant complexity---two current sources and one or more diode-connected bias transistors. You could just preface the existing amp with a PNP e-follower and nearly get there, except for the same NPN PNP offset issue. This would add another series noise generator as well.