Operational Amplifier Design

[Kevin Aylward]

IIRC this is one of Widlar's jewels. He was fond of leaving critical
components off of schematics, or using diffused structures that just
happened to create additional, invisible components.

I assume the reason for this is *ONLY* to keep things secret from the
competitor. Any other reason for using obtuse, hard to uanderstand
designs should result in the designer being burned alive in oil.

Yes, but to no avail for the gm/c GBW/slew limit.

and allowing the use of a

Ahmmm. Its a bit more subtle that that. The high current and then thrown
away again is used because the *Ft* is higher at higher current. That
is, FT is a non-linear function of current. You don't get any
*effective* gm/C advantage by throwing away the current. If you throw
away some of the current it cant go into the capacitor. Dah!!! i.e. both
current and gm are thrown away, which by itself, is useless. Its only
because the Ft gets better at higher current that this can work, i.e.
the whole amplifier gets faster so the parasitic poles are less bother
allowing for a smaller Ccomp.

Yes, that always surprises everyone... low gm is desirable in the
frontend for best GBW product _and_ slew rate.

Yes, within reason!

Emitter resisters of 100k aint gonnna get one much BW.

Kevin Aylward
[email protected]
http://www.anasoft.co.uk
SuperSpice, a very affordable Mixed-Mode
Windows Simulator with Schematic Capture,
Waveform Display, FFT's and Filter Design.

 

[Ken Smith]

Jim Thompson wrote...

Doesn't adding a modest series resistor (along with an ESD network
after the resistor) give one a chance to both stabilize the loop,
and avoid ESD failure?

Yes, adding a resistor in series with the compensation cap can add a zero
to the transfer function. If this zero lines up with the pole from, lets
say, the input pair, then stability can be improved.

 

[Kevin Aylward]

The "simplified schematic" is flawed... it can't be as-drawn... there
is a fundamental error. I tried to tease Win into _stepping_in_it_,
but he won't take the bait ;-)

I had assumed that the schematic was simplified but that the 3 diode
cross over issue was real because of what I had seen in in actual use.

And what's this statement, "...compensation does not enclose that
distortion..."?

Consider these oversimplifications:

ASCII art:

Not enclosed:
-----------
! !
-------+--------! Distorty !------ output
! ! ! stuff !
--- ! ! !
---Cc ! -----------
! !/
---+----!
!\ Q1
!
GND



Enclosed:

-------------------------------
! !
! ----------- !
! ! ! !
! --------! Distorty !---+-- output
! ! ! stuff !
--- ! ! !
---Cc ! -----------
! !/
---+----!
!\ Q1
!
GND


When the compensation capacitor takes the feedback from the far side
of the distortion, the gain of Q1 will be acting to make the whole
section act more linear. ie: its collector will slew more quickly
through the cross over distortion. This greatly lowers the cross
overdistortion of the circuit.[/QUOTE]

Not in my opinion. I first evaluated this 20 years ago when deigning a
0.002% amp, i.e. I actually measured it. I have also been re-evaluating
this recently in spice.

"There is no such thing as a free lunch".

If you move the cap to the outside of the follower things start
collapsing. When the cap is connected as usual, you get a low impedance
voltage driving the Distorty stuff. Without the cap the impedance is all
over the place, meaning that any variation in input impedance of the
following stage creates distortion.

Secondly, you end up with a response that doesn't roll off cleanly at
HF. Again, try it in spice.

In summary here, the "no free lunch" is something to take to heart. Many
configurations that look, different are essentially, topologically
equivalent, such that there is often not much to chose them apart.

Designs undergo Darwinian selection. Most designs that survive are there
because the obvious alternatives don't work.

This method only works for about 1 transistor followed by something
like a complimentry emitter follower.

It doesn't work at all. Try doing some SuperSpice on it, noting that SS
does small signal distortion and LTSpice dosnt

If you try to enclose too many
devices, the local feedback system can become unstable for some loops
gains.
Yes.

Transistor stages running right near the cut off point have a
much lower frequency responce. As the output switches from the NPN
to the PNP, the frequency responce of that part sweeps down and then
back up.
Yes.

The local feedback loop must be made stable for all of the
gain/pahse values the circuit sweeps through. I assume that this
problem is why the LM324 does not enclose the non-linearities in the

Yes, and other reasons, some of which are noted above.

Kevin Aylward
[email protected]
http://www.anasoft.co.uk
SuperSpice, a very affordable Mixed-Mode
Windows Simulator with Schematic Capture,
Waveform Display, FFT's and Filter Design.

 

[Ken Smith]

Yes. But I don't know of a single commercial amplifier that isn't set
up that way.

LT1364 and LT1814 do that.

 

[Kevin Aylward]

Yes, adding a resistor in series with the compensation cap can add a
zero to the transfer function. If this zero lines up with the pole
from, lets say, the input pair, then stability can be improved.

Ho humm.. never read my posts Ken?

Actually try (i.e Spice it) this. The results are not promising at all.
If it actually worked in practise (op-amp design), you would see it
used.

Look, heres the deal.

Consider a diff pair driving the high gain Miller stage. If the Miller
stage is well and trully clobering everthing else, all is well. If it
aint, like because one has included a resister, darkness awaits. This is
what happens.

1) The diff pair no longer has the same low value of impedance on its
collecter, becuse the input to the millor stage is no longer low
impedance. This means that there is HF gain at the first stage, thefore
creating a higher effective capacitance at the op-amp input. This causus
a roll off in the feedback loop.

2) What was a low impedance at HF driving the load capacitance of the
Miller stage, is now a much higher impedance. This means that this
capacitive load will roll of the gain with a pole much closer than what
it was. i.e. it will bring what was a good pole split, i.e. poles widely
separated, much closer together.

This is why the Millar stage is so universally used. It turns 3
potential poles into one dominant one. This extra zero idea only works
in practice if one is using a gain stage that is much nearer the ideal,
i.e. one that has a response that is not effected by the components
strung all around it.

The way to stabilise a typical amp, is summerised by
(http://www.anasoft.co.uk/EE/feedbackstability/feedbackstability.html)
the following:

"The only guaranteed way to make a small fortune, is to start with a
large one, and lose some of it."

i.e., throw away the gain, then allow some of it back in again.

Kevin Aylward
[email protected]
http://www.anasoft.co.uk
SuperSpice, a very affordable Mixed-Mode
Windows Simulator with Schematic Capture,
Waveform Display, FFT's and Filter Design.

 

[Terry Given]

Jim Thompson wrote:




I assume the reason for this is *ONLY* to keep things secret from the
competitor. Any other reason for using obtuse, hard to uanderstand
designs should result in the designer being burned alive in oil.

or at least being burned in olive oil

 

[Ken Smith]

Ho humm.. never read my posts Ken?

Maybe I should stop reading them but for now I'll continue to read some of
them :>

Actually try (i.e Spice it) this. The results are not promising at all.
If it actually worked in practise (op-amp design), you would see it
used.

The folks at Linear seem to disagree with you in that they include the
resistor in series with the compensation cap in several of their designes.

Look, heres the deal.

Consider a diff pair driving the high gain Miller stage. If the Miller
stage is well and trully clobering everthing else, all is well. If it
aint, like because one has included a resister, darkness awaits. This is
what happens.

1) The diff pair no longer has the same low value of impedance on its
collecter, becuse the input to the millor stage is no longer low
impedance. This means that there is HF gain at the first stage, thefore
creating a higher effective capacitance at the op-amp input. This causus
a roll off in the feedback loop.

Take a look at the LT1001 and you will see one of the simpler ways this
sort of thing is avoided. Linear used a cascode stage in that design.

I think even the LM101, (IIRC) used a funny sort of cascode that allowed
the common mode range to include the plus rail and some sort of a zero in
the compensation section.

Several of the fast op-amps from Linear contain the resistor you claim to
dislike so much.

2) What was a low impedance at HF driving the load capacitance of the
Miller stage, is now a much higher impedance. This means that this
capacitive load will roll of the gain with a pole much closer than what
it was. i.e. it will bring what was a good pole split, i.e. poles widely
separated, much closer together.

This is why the Millar stage is so universally used.

Miller effect increases the effectively size of the capacitor. I believe
that without it, the capacitor would have to be physically larger that the
chip makers would like it to be. This explains why in the op-amps that
contain the cascode stage or other interstages the compensation is still
done with a Miller effect

It turns 3
potential poles into one dominant one.
Yes
This extra zero idea only works
in practice if one is using a gain stage that is much nearer the ideal,

Or where the low impedance to the input section is otherwise assured.

 

[Jim Thompson]

Or where the low impedance to the input section is otherwise assured.

Ken, Just plonk Kevin... keeps the heartburn down ;-)

...Jim Thompson

 

[Ken Smith]

Ken, Just plonk Kevin... keeps the heartburn down ;-)

I had just unplonked[1] him. You may be right about the heartburn though.

[1] is that a word?

 

[Rich The Philosophizer]

Ken, Just plonk Kevin... keeps the heartburn down ;-)

I had just unplonked[1] him. You may be right about the heartburn though.

[1] is that a word?

It is now. ;-)

;^j
Rich

 

[Kevin Aylward]

Ken, Just plonk Kevin... keeps the heartburn down ;-)

What your problem Jim? My comments were technically accurate. If you
disagree with the technical content, lets hear your arguments. Someone
isn't a baddie just because he points out someone's misconceptions.

Again, with all due respect to Ken, its clear that he only has a basic
understanding of these issues. We all have to start somewhere, and I
started at 11.


Kevin Aylward
[email protected]
http://www.anasoft.co.uk
SuperSpice, a very affordable Mixed-Mode
Windows Simulator with Schematic Capture,
Waveform Display, FFT's and Filter Design.

 

[Kevin Aylward]

Maybe I should stop reading them

Why?

I might be an arrogant know it all, but out of 10,000 posts, you would
be hard pressed to find more than a couple of technically
er...misleading posts, of mine on technical maters.

but for now I'll continue to read
some of them :>



The folks at Linear seem to disagree with you in that they include the
resistor in series with the compensation cap in several of their
designes.

Only several?

How about counting those that don't?

Take a look at the LT1001 and you will see one of the simpler ways
this sort of thing is avoided. Linear used a cascode stage in that
design.

Yes, of course a cascode helps the input roll off, but it is not
completely effective. This cascode itself has Miller gain from its
emitter to collector. What saves one is that its the low impedance re
that drives this capacitor.

I think even the LM101, (IIRC) used a funny sort of cascode that
allowed the common mode range to include the plus rail and some sort
of a zero in the compensation section.

Several of the fast op-amps from Linear contain the resistor you
claim to dislike so much.

So what? So a in a few cases the numbers work out such that it *may* be
effective in some special cases. However, I would have to confirm the
sums myself. In many, many cases circuits have special topologies that
don't actually achieve anything with that topology.

As I said, go and do some typically spice simulations and it will be
very clear that this resister can only be used in special circumstances.

Miller effect increases the effectively size of the capacitor. I
believe ,
Ahmmmm...

that without it the capacitor would have to be physically
larger that the chip makers would like it to be.

Oh dear...with all due respect, your phrasing here indicates that you
only have a basic understanding of all of this. *Of course* the Miller
effect is a capacitor multiplier, this is 101 electronics, but that is
not relevant to the idea of why Miller compensation is used. Even if one
had an enormous capacitor easily available one would still not simply
plonk it down at the input of the stage. The performance would be
dreadfull. So no, that is not why it is used. Unfortunately, you have
been reading too many layman descriptions.

The gain of the Miller stage is gm.Xc. This is what is important. No
mention of its Xc/Av is required for this fundamental feature.

However, it can't be greater than gm.ZL. If gm.ZL is limiting the gain
at HF, then an added resister in Ccomp might be trying to command a gain
that is not actually available open loop at that stage. The gain stage
is not an ideal amplifier. It is a current source with only a finite
gain. This is one reason why adding in the resister doesn't work in
general. The gain is simple not available.

This explains why in the op-amps that contain the cascode stage or
other interstages
the compensation is still done with a Miller effect

No it doesn't. Not in the slightest. You simply don't understand what's
going on here. This is why:

POLE SPLIT LESSON 101:

The Miller capacitor is a *pole-splitter*. This is its *fundamental*
use. An amplifier usually has many poles. To ensure *stability* an
amplifier is usually designed such that there is only one main pole
dominating the loop gain. Without the Miller capactor, the main gain
stage has *two* fundermental poles, one at its input (Cin), on at its
output (CL). If the output of the gain stage has a low impedance, then
the effect of Cload will be minimised. For example:

Consider the Miller stage has a value of Ccomp. This negative feedback
results in an output impedance driving Cload. This results in the Miller
roll off and the Cload roll off. Now consider making Cmiller much
larger. This moves the Miller pole down in frequency, however, this also
results in more feedback reducing the output impedance. This means that
CL is going to have less effect on the output rolloff, i.e. it moves the
output pole further out.

FUNDAMENTAL POINT:

Therefore, the Miller stage converts what is two interacting poles into
two widely separated poles, where, ideally, the higher pole is well
above the unity gain frequency.

Again, the idea that the Miller cap makes for "a big capacitor" is just
layman's fluff.

Or where the low impedance to the input section is otherwise assured.

No.

The diff pair input rolloff is not the main issue, it is just one of the
side issues. The main issue is that the main gain stage has *two*
inherent poles. One at is input (Cin), one at its output (CL). Using
Miller compensation converts them effectively, into one. End of story.

Kevin Aylward
[email protected]
http://www.anasoft.co.uk
SuperSpice, a very affordable Mixed-Mode
Windows Simulator with Schematic Capture,
Waveform Display, FFT's and Filter Design.

 

[Winfield Hill]

Jim Thompson wrote...

I'll look thru my archives. I think that's one of the problem areas
for which I wrote an ICE report.

Hi Jim, we sure hope you find that old report, sounds like a fascinating
bit of detail we'd like to learn.

 

[Jim Thompson]

On 10 Nov 2004 02:32:23 -0800, Winfield Hill
<hill_a@

 

[Ken Smith]

[.. about ponk..]

Why?
I might be an arrogant know it all, but out of 10,000 posts, you would
be hard pressed to find more than a couple of technically
er...misleading posts, of mine on technical maters.

Consider this very thread. You said that placing a resistor in series
with the compensation capacitor of an op-amp was "bad design". You didn't
qualify it with "usually" or "almost always".

Do I get to call that 0.75 mistakes right here?

Yes, of course a cascode helps the input roll off, but it is not
completely effective. This cascode itself has Miller gain from its
emitter to collector. What saves one is that its the low impedance re
that drives this capacitor.

Please explain the term "Miller gain[1]" as you use it here. The
transistor has a capacitance from collector to emitter, but this does not
go via a gain to the emitter. There is also the collector-base
capacitance which does unless the base sees a very low impedance to
ground. For the collector emitter one, the low "re" does matter, for the
other it drops out of the issue.

[1] I usually take this to mean the increase in the effective capacitance
due to the far end of it being connected to an inverted signal.

So what? So a in a few cases the numbers work out such that it *may* be
effective in some special cases.

So, you were wrong, wrong, ever so wrong that's what. :> You overlooked
something, it happens.

As I said, go and do some typically spice simulations and it will be
very clear that this resister can only be used in special circumstances.

But, I don't need to do "some typically spice" simulations. I just need a
few weird examples to show that it works for those cases.

Oh dear...with all due respect, your phrasing here indicates that you
only have a basic understanding of all of this.

.... and then there are lines like this. No way to win friends there
buddy. My phrasing is a bit odd in many situations for reasons I do not
wich to go into here.

*Of course* the Miller
effect is a capacitor multiplier, this is 101 electronics, but that is
not relevant to the idea of why Miller compensation is used. Even if one
had an enormous capacitor easily available one would still not simply
plonk it down at the input of the stage. The performance would be
dreadfull.

Yes that would be dreadful and I don't think anyone has suggested that.

There was more but I just deleted it. Your tone makes it not worth it,
but go back and check it and you will find you made an "oversimplifcation"
(error).

 

[Winfield Hill]

Jim Thompson wrote...

I couldn't locate the uA124QB. Could you send a copy? Thanks!

Jim, did I send it this morning - Sheesh, I don't remember!

 

[Jim Thompson]

Jim Thompson wrote...

Jim, did I send it this morning - Sheesh, I don't remember!

Win, What do you sip on there in Massachusetts? We E-mailed back and
forth in regard to some typos ;-)

...Jim Thompson

 

[Terry Given]

[.. about ponk..]

Why?
I might be an arrogant know it all, but out of 10,000 posts, you would
be hard pressed to find more than a couple of technically
er...misleading posts, of mine on technical maters.

Consider this very thread. You said that placing a resistor in series
with the compensation capacitor of an op-amp was "bad design". You didn't
qualify it with "usually" or "almost always".

Do I get to call that 0.75 mistakes right here?

Yes, of course a cascode helps the input roll off, but it is not
completely effective. This cascode itself has Miller gain from its
emitter to collector. What saves one is that its the low impedance re
that drives this capacitor.

Please explain the term "Miller gain[1]" as you use it here. The
transistor has a capacitance from collector to emitter, but this does not
go via a gain to the emitter. There is also the collector-base
capacitance which does unless the base sees a very low impedance to
ground. For the collector emitter one, the low "re" does matter, for the
other it drops out of the issue.

[1] I usually take this to mean the increase in the effective capacitance
due to the far end of it being connected to an inverted signal.

So what? So a in a few cases the numbers work out such that it *may* be
effective in some special cases.

So, you were wrong, wrong, ever so wrong that's what. :> You overlooked
something, it happens.

As I said, go and do some typically spice simulations and it will be
very clear that this resister can only be used in special circumstances.

But, I don't need to do "some typically spice" simulations. I just need a
few weird examples to show that it works for those cases.

Oh dear...with all due respect, your phrasing here indicates that you
only have a basic understanding of all of this.

... and then there are lines like this. No way to win friends there
buddy. My phrasing is a bit odd in many situations for reasons I do not
wich to go into here.

*Of course* the Miller
effect is a capacitor multiplier, this is 101 electronics, but that is
not relevant to the idea of why Miller compensation is used. Even if one
had an enormous capacitor easily available one would still not simply
plonk it down at the input of the stage. The performance would be
dreadfull.

Yes that would be dreadful and I don't think anyone has suggested that.

There was more but I just deleted it. Your tone makes it not worth it,
but go back and check it and you will find you made an "oversimplifcation"
(error).

Really the "error" (and this is IMO the most common cause of flame wars
here) is making a definitive statement, rather than a suitably hedged
statement. We are (mostly) smart people, and all picky bastards too, so
the minute someone either forgets to qualify something, or makes an
over-bold assertion [eg "that never works in opamps" (not what KA said,
Im just too lazy to go back & look, so I'll paraphrase) vs "Usually that
doesnt help"] some picky bugger dashes off the minimum of 1 example that
technically disproves the "never" assertion.

mind you, if you try and hedge the things you say (to take into account
all these exceptions) then the verbiage quickly becomes unreadable.

Cheers
Terry


Cheers
Terry

 

[Terry Given]

Win, What do you sip on there in Massachusetts? We E-mailed back and
forth in regard to some typos ;-)

...Jim Thompson

Hash-a-chussetts?

ya better watch out if thats the case, 'cos Rich will be on his way

Cheers
Terry

 

[Kevin Aylward]

[.. about ponk..]

Why?
I might be an arrogant know it all, but out of 10,000 posts, you
would be hard pressed to find more than a couple of technically
er...misleading posts, of mine on technical maters.

Consider this very thread. You said that placing a resistor in series
with the compensation capacitor of an op-amp was "bad design".

No I didn't.

You
didn't qualify it with "usually" or "almost always".

Yes I did, to wit, my exact words were:

"..If they do, its often a poor design..."

"Often" was my qualifier, as was "poor".

I, usually, always cover my arse. As indeed I do here wth the word
"usually".

Do I get to call that 0.75 mistakes right here?


Nope

Yes, of course a cascode helps the input roll off, but it is not
completely effective. This cascode itself has Miller gain from its
emitter to collector. What saves one is that its the low impedance
re that drives this capacitor.

Please explain the term "Miller gain[1]" as you use it here.

Anywhere there is gain and an impedance connecting the gain output to
input.

The
transistor has a capacitance from collector to emitter, but this does
not go via a gain to the emitter. There is also the collector-base
capacitance which does unless the base sees a very low impedance to
ground.

Not at issue.

For the collector emitter one, the low "re" does matter, for
the other it drops out of the issue.

As I said, re is low so its effect is nothing like as significant as
when driving from the base. Of course it has some effect. The Miller
effect is, esentially, the same whether the input is from the base or
emitter, but one is Rs.Ccb/Av the other is re.Cce/Av. Both re and Cce
are usually much lower.

Side point. Miller gain has nothing to do with negative feedback. Its
gain that is important not the polarity. The gain from emitter to
collector is positive, but the magnitude of the impedance is still
reduced by (1+Av), in this case (1-Av).

[1] I usually take this to mean the increase in the effective
capacitance due to the far end of it being connected to an inverted
signal.

See above. Common misconception but the inversion is not necessary. It
the output swings many times more than the input, it will source/sink
additional current through the coupling impedance, whatever its sign.

So, you were wrong, wrong, ever so wrong that's what. :> You
overlooked something, it happens.

Nope. What part of "often" did you miss?

But, I don't need to do "some typically spice" simulations. I just
need a few weird examples to show that it works for those cases.

Experienced designers know that that it is usually not effective to use
such a technique, thats why we dismiss it it general. As I explained,
the Miller cap does a pole split. If one tries to insert a resister to
compensate for another HF pole, it reintroduces the parasitic poles at
the Miller stage that were previously clobbered. One negates the main
feature of the Miller stage, and gets one back to where one was. For the
technique to work it has to be compensating a pole that is much lower
then when the Mirrrr stage runs out of steam. In most/many cases the
extra poles are all at HF, so the technique has limited value.
Everything in an op-amp all tends to crap out at the same point, i.e.
related to the basic Ft of the process.

... and then there are lines like this. No way to win friends there
buddy. My phrasing is a bit odd in many situations for reasons I do
not wich to go into here.

You came across as someone who actually does not do this stuff 10,000's
of hours a year, and I mean 10,000's hours, i.e. hard simulations, as I
do. How can one phrase something that can be taken as personal, but
isn't? Again, with all due respect, your comments in this post, while
showing that you have a good basic understanding, show that you have
significant gaps in your knowledge. Its just the way it is. We all have
to learn sometimes.

Yes that would be dreadful and I don't think anyone has suggested
that.

There was more but I just deleted it. Your tone makes it not worth
it, but go back and check it and you will find you made an
"oversimplifcation" (error).

Not in my view. Were would that have been?

Kevin Aylward
[email protected]
http://www.anasoft.co.uk
SuperSpice, a very affordable Mixed-Mode
Windows Simulator with Schematic Capture,
Waveform Display, FFT's and Filter Design.