Maker Pro
Maker Pro

anyone familiar with this topology?

J

John Larkin

Jan 1, 1970
0
But this reliability issue has nothing to do with the topology.


Another reason I like this opamp-Vcc current splitter thing (it really
deserves a decent name) is that you can know for sure what the maximum
drive is going to be into the power current mirrors. So even if the
loop goes wide open, you can fully define, by design, the output stage
peak power dissipation, and arrange for soa shutdown before the
silicon melts. That helps make the design reliable and fault-tolerant.
I've seen (er, designed) discrete amps that had great foldback current
limiting in the output stages, and consequently, when overloaded, blew
up the drivers.

Discretes don't have poles? I just use an opamp here that has 100x the
loop bandwidth, and don't worry about that part.

John
 
J

Jim Thompson

Jan 1, 1970
0
On Sun, 12 Sep 2004 08:26:58 GMT, "Kevin Aylward"


Discretes don't have poles? I just use an opamp here that has 100x the
loop bandwidth, and don't worry about that part.

John

Isn't it amusing how Kevin comes blaring out with his BS without ANY
hands-on experience with the topology under discussion?

Seems that more than a few of us old farts have actually built and
used this topology successfully in production quantities.

(Of course Kevin will retort with another of his "age" quips. I may
be old, but I'm not a mental midget. :)

Now I must 'fess up that I designed and used this topology at least
15-20 years before there was any simulator available to prove that it
"is bad news".

For my own amusement though I will try to accurately model the
topology. I suspect that it has far better useable gain-bandwidth
product than Kevin realizes.

...Jim Thompson
 
J

John Larkin

Jan 1, 1970
0
Isn't it amusing how Kevin comes blaring out with his BS without ANY
hands-on experience with the topology under discussion?

Seems that more than a few of us old farts have actually built and
used this topology successfully in production quantities.

(Of course Kevin will retort with another of his "age" quips. I may
be old, but I'm not a mental midget. :)

Now I must 'fess up that I designed and used this topology at least
15-20 years before there was any simulator available to prove that it
"is bad news".

For my own amusement though I will try to accurately model the
topology. I suspect that it has far better useable gain-bandwidth
product than Kevin realizes.

...Jim Thompson


One concern is that the V+ and V- currents might include a transient
shoot-through term, especially if this opamp's speed isn't grossly
faster than the rest of the loop. Does that happen with typical
opamps?

Even scarier is that some opamps draw a lot of current when railed,
which could fry the output transistors. When I do this, I usually clip
the inputs such that an open-loop event will not rail this stage.

My gradient amps have big mosfet output stages, but everything else is
opamps, five opamp gain stages typically around the loop. Some opamps
are inside the local fb loops of others. Lots of poles, happily
playing together.


John
 
J

Jim Thompson

Jan 1, 1970
0
On Sun, 12 Sep 2004 11:03:41 -0700, Jim Thompson



One concern is that the V+ and V- currents might include a transient
shoot-through term, especially if this opamp's speed isn't grossly
faster than the rest of the loop. Does that happen with typical
opamps?

Even scarier is that some opamps draw a lot of current when railed,
which could fry the output transistors. When I do this, I usually clip
the inputs such that an open-loop event will not rail this stage.

My gradient amps have big mosfet output stages, but everything else is
opamps, five opamp gain stages typically around the loop. Some opamps
are inside the local fb loops of others. Lots of poles, happily
playing together.


John

"Lots of poles"... careful there John, Kevin will tell you it can't be
done. But, of course, he has no clue about control systems.

...Jim Thompson
 
K

Ken Smith

Jan 1, 1970
0
Jim Thompson said:
For my own amusement though I will try to accurately model the
topology. I suspect that it has far better useable gain-bandwidth
product than Kevin realizes.

I think you may find that putting a resistor in series with the capacitor,
from the op-amp's output to the circuit output is helpful.

I haven't modeled this but I suspect that this will allow a higher value
capacitor and thus a bit more local feedback at the frequency of interest.
The resistor will prevent the unity gain frequency from being lowered.
 
J

Jim Thompson

Jan 1, 1970
0
I think you may find that putting a resistor in series with the capacitor,
from the op-amp's output to the circuit output is helpful.

I haven't modeled this but I suspect that this will allow a higher value
capacitor and thus a bit more local feedback at the frequency of interest.
The resistor will prevent the unity gain frequency from being lowered.

--

I'm trying to find a schematic from that phase of my life (~1970-73).

IIRC, I had an R to ground from the OpAmp output, plus an R from the
OpAmp output to the "boosted" output, with a cap across that second
resistor.

...Jim Thompson
 
J

John Woodgate

Jan 1, 1970
0
I read in sci.electronics.design that John Larkin <jjlarkin@highlandSNIP
techTHISnologyPLEASE.com> wrote (in <gg49k09175f79m3nhfnv50nv73u2ghf9ut@
4ax.com>) about 'anyone familiar with this topology?', on Sun, 12 Sep
2004:
Lots of poles, happily playing
together.

Yes, they've got a pretty good soccer team. And some good orchestras.
 
W

Winfield Hill

Jan 1, 1970
0
Genome wrote...
The proper way to do it is to give the op-amp a local gain of about
3 and ask it to drive a low value resistor. Then you take feedback
from the power stage output to the op-amps output. Add some emitter
degeneration for the darlingtons. Use a TL074. Implement a Vbe
multiplier with two of the op-amps driving each other and scale things
through the bias chain. Use one as a DC servo and use one as the main
amp. SEE ABSE

That's cute, for sure. But there are a multitude of "proper" ways.
For example, I prefer to add an emitter follower driving the output
transistors. This provides fast cutoff despite the Ccb capacitance,
and also tempco compensation for the Darlington bias (in some types
only one transistor works during the low-current quiescent mode).

Appropo to the scene, I have posted to abse an article scan sent me
by Martin Griffith, that he'd like to contribute to the discussion.
See "current-dumping-amplifier by GSchmidt, old Elektor article"
 
J

John Larkin

Jan 1, 1970
0
Genome wrote...

That's cute, for sure. But there are a multitude of "proper" ways.
For example, I prefer to add an emitter follower driving the output
transistors. This provides fast cutoff despite the Ccb capacitance,
and also tempco compensation for the Darlington bias (in some types
only one transistor works during the low-current quiescent mode).

Appropo to the scene, I have posted to abse an article scan sent me
by Martin Griffith, that he'd like to contribute to the discussion.
See "current-dumping-amplifier by GSchmidt, old Elektor article"


Win,

I think this is what you mean, a Darlington in which the first
transistor has a considerable idle current and the second has none...

|
+--------+
| |
c |
--------b |
e |
| c
+-------b
| e
r1 |
| r2
| |
+--------+
|


where r2 is optional, and there's less than 1 jd in r1, so the thing
is continuous at all times but the main transistor only kicks on at
higher currents. This can be thermally stable without deliberate
compensation. Is there a name for this?

And right, the first collector could be run somewhere else to reduce
capacitive feedback. This is a nice complement to the (still un-named)
opamp rail current thing.

John
 
T

Tim Shoppa

Jan 1, 1970
0
I still maintain that "the output section is running class C" by which I
mean
that the op-amp's output current needs to be greater than about

1.4V/680 - 1.4mA = 0.66mA

in either direction to turn on one of the output transistors.

1.4 mA is the typical per-section quiescent current of the TL081, I see
from the datasheet. But in real life (especially at elevated temps)
the current draw I'v measured always seems to be greater (2mA is actually
more typical) which would bias the output transistors a little into
conduction.

If that 2mA isn't enough, can't you just add an extra resistor between
the power pins of the op-amp to add extra quiscent current.

Tim.
 
J

John Larkin

Jan 1, 1970
0
1.4 mA is the typical per-section quiescent current of the TL081, I see
from the datasheet. But in real life (especially at elevated temps)
the current draw I'v measured always seems to be greater (2mA is actually
more typical) which would bias the output transistors a little into
conduction.

If that 2mA isn't enough, can't you just add an extra resistor between
the power pins of the op-amp to add extra quiscent current.

Yup. That adjusts the idle current.

John
 
K

Ken Smith

Jan 1, 1970
0
[...]
I think you may find that putting a resistor in series with the capacitor,
from the op-amp's output to the circuit output is helpful.

I haven't modeled this but I suspect that this will allow a higher value
capacitor and thus a bit more local feedback at the frequency of interest.
The resistor will prevent the unity gain frequency from being lowered.

--

I'm trying to find a schematic from that phase of my life (~1970-73).

IIRC, I had an R to ground from the OpAmp output, plus an R from the
OpAmp output to the "boosted" output, with a cap across that second
resistor.

That combination relies on the output impedance of the op-amp to set the
high frequency "output to supply current" ratio.

As far as a disturbance added at the system's output is concerned, the
output of most op-amps looks like a common base stage. The bases of the
transistors see a lowish impedance because of the way the compensation
circuit works. Some emitter resistance is usually in the op-amp's design
as part of the current limiting.
 
K

Ken Smith

Jan 1, 1970
0
[email protected] (Ken Smith) wrote in message


1.4 mA is the typical per-section quiescent current of the TL081, I see
from the datasheet. But in real life (especially at elevated temps)
the current draw I'v measured always seems to be greater (2mA is actually
more typical) which would bias the output transistors a little into
conduction.

Oh, thats bad, the idle current rising with temperature isn't a feature
I'd want in my circuit.

If that 2mA isn't enough, can't you just add an extra resistor between
the power pins of the op-amp to add extra quiscent current.

Or better yet a constant current diode so the circuit's idle current is
near constant with supply voltage.

You could also use some current mirror circuits to make the bias point of
the output not depend on the bias point of the op-amp. That would be more
complex though.
 
J

Jim Thompson

Jan 1, 1970
0
[snip]
As far as a disturbance added at the system's output is concerned, the
output of most op-amps looks like a common base stage. The bases of the
transistors see a lowish impedance because of the way the compensation
circuit works.

Eh? The bases of the output devices see the pole-splitting capacitor
(usually < 30pF, but multiplied by the current gain of the second
stage) in parallel with the impedance of a current source. Any
lowering of output impedance is strictly due to feedback.
Some emitter resistance is usually in the op-amp's design
as part of the current limiting.

True.

...Jim Thompson
 
K

Ken Smith

Jan 1, 1970
0
[snip]
As far as a disturbance added at the system's output is concerned, the
output of most op-amps looks like a common base stage. The bases of the
transistors see a lowish impedance because of the way the compensation
circuit works.

Eh? The bases of the output devices see the pole-splitting capacitor
(usually < 30pF, but multiplied by the current gain of the second
stage) in parallel with the impedance of a current source. Any
lowering of output impedance is strictly due to feedback.

Yes, local feedback like this:

ASCII ART of over simplified:

! !
O I1 !/
! -----! NPN
-----! ! !\e
! ! ! !
---Cc !------------ !
--- ! ! !
! !/ ! !/e
-----! Q1 -----! PNP
!\e !\
! !
GND GND

The fact that Cc provides local feedback of the collector voltage on Q1,
reduces the impedance its collector presents to the output pair. Near the
gain cross over frequency of the op-amp, this makes the impedance on the
base of the output devices lower than the series resistors in the emitter
resistors.
 
R

Rich Grise

Jan 1, 1970
0
If I were given unlimited funding to design the perfect current
splitter custom linear IC for my gradient amps, I'd probably wind up
with the opamp I'm using.
If I were given unlimited funding, I'd retire in Chaing Mai.

Cheers!
Rich
 
K

Kevin Aylward

Jan 1, 1970
0
Jim said:
Isn't it amusing how Kevin comes blaring out with his BS without ANY
hands-on experience with the topology under discussion?

Of course I have.
Seems that more than a few of us old farts have actually built and
used this topology successfully in production quantities.

That's not the point. Sure, its ok for some applications. Just like
shagging an ugly chick is, but it aint a grate choice.
(Of course Kevin will retort with another of his "age" quips. I may
be old, but I'm not a mental midget. :)

Now I must 'fess up that I designed and used this topology at least
15-20 years before there was any simulator available to prove that it
"is bad news".

For my own amusement though I will try to accurately model the
topology.

Indeed. Which illustrates a problem. Spice models don't model the
transfer function from inputs to the rails. Neither is is specked in the
data sheets. This means that worst case design for production quantities
is an issue.
I suspect that it has far better useable gain-bandwidth
product than Kevin realizes.

Its not the GBW, its how all of the poles come in.

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.
 
K

Kevin Aylward

Jan 1, 1970
0
John said:
Another reason I like this opamp-Vcc current splitter thing (it really
deserves a decent name) is that you can know for sure what the maximum
drive is going to be into the power current mirrors. So even if the
loop goes wide open, you can fully define, by design, the output stage
peak power dissipation, and arrange for soa shutdown before the
silicon melts. That helps make the design reliable and fault-tolerant.
I've seen (er, designed) discrete amps that had great foldback current
limiting in the output stages, and consequently, when overloaded, blew
up the drivers.

Yes, this is a standard design fault, seen it many times. However, its
easy to fix. One simply designs the thing correctly, i.e. pre-diver
currents need to be limited.
Discretes don't have poles?

They dont have as many, usually.
I just use an opamp here that has 100x the
loop bandwidth,

Ho hum. Not usually possible in a high performance designs. Secondly,
phase shift starts coming in 10 times earlier.

I just suspect many haven't actually tried doing a state of the art
audio amp. For example, if we consider a normal amp, but using an op-amp
in it, there is going to be two major roll offs, not one. ok, one could
use an uncompensated op amp but then...

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.
 
J

John Woodgate

Jan 1, 1970
0
I read in sci.electronics.design that Kevin Aylward
ueyonder.co.uk>) about 'anyone familiar with this topology?', on Mon, 13
Sep 2004:
Its not the GBW, its how all of the poles come in.

These days, it's the GWB and how the polls will come in. (;-)
 
W

Winfield Hill

Jan 1, 1970
0
John Woodgate wrote...
Kevin Aylward wrote ...

These days, it's the GWB and how the polls will come in. (;-)

Yes, and like Kevin said, all the polls will crowd together.
Not to mention all the pols.
 
Top