Linear Mode of IGBT FSOA/hotspots?

Klaus Kragelund

unread,

Apr 21, 2015, 7:51:02 AM4/21/15

to

Hi

For an application I am looking into using a large IGBT for inrush (charge of DC link capacitor from mains)

To reduce BoM costs, I would like to drive it in linear mode, monitoring the current and adjusting the gate voltage accordingly

There are however some talk about hotspots of a linear mode driven IGBT.

The IGBT is a 20A device. My intention is to drive it at 10mA, so 2000 times less than the max current

Does anyone know the workings of these hotspots, and if the issue is real or just academic?

Regards

Klaus

Tim Williams

unread,

Apr 21, 2015, 9:34:24 AM4/21/15

to

I'd feel much better with a MOSFET.

I don't think I've seen any IGBTs rated for any kind of DC aside from
"off".

I don't know the coefficients or how you'd decide that an IGBT is going to
be unstable; you're on your own here, in "test to destruction" land. It
should be at least broadly similar to BJT behavior, so if you have a BJT
of the same die area that's rated for such-and-such, then...

Thermal limits are by die area anyway, so you're wasting no more money on
a 20A IGBT than a 10A MOSFET, or a 5A linear MOSFET.

Is this supposed to be turned on solid, after the fact, as an SSR? A real
relay might start looking competitive, too...

Or maybe something a little stranger, like a bidirectional current
limiting diode on the AC side, to be bypassed with a TRIAC. Or use a dumb
resistor...

Tim

--
Seven Transistor Labs
Electrical Engineering Consultation
Website: http://seventransistorlabs.com

"Klaus Kragelund" <klau...@hotmail.com> wrote in message
news:6e0312b6-5b9c-4a36...@googlegroups.com...

Joerg

unread,

Apr 21, 2015, 10:53:38 AM4/21/15

to

It is generally not recommended. Some datasheets don't even state a
forward SOA, some like this one do but then it's for single-shot:

http://www.irf.com/product-info/datasheets/data/irgb4062dpbf.pdf

IOW the only way to find out if it is legit for a product is probably to
obtain the blessing from the manufacturer for brief linear operation.
Most likely they will not give it. Why not use a FET? I've never had
much use for IGBT because I could obtain much lower conduction losses
with FETs and they are cheap.

--
Regards, Joerg

http://www.analogconsultants.com/

Klaus Kragelund

unread,

Apr 21, 2015, 2:54:23 PM4/21/15

to

I've asked the manufacturer about this only for reference, since our internal expert on IGBT's has provided details that this is a very risky path to take.

The reason for doing it was to use a part that was in the circuit in advance, so now we have to add either a PTC in parallel with the IGBT or a MOSFET/resistor to do the limited inrush

Cheers

Klaus

Joerg

unread,

Apr 21, 2015, 3:11:49 PM4/21/15

to

It's a four-layer device and I think that's what makes it dicey. One
engineer (Jim T?) once said that using an IGBT in the linear range is
like walking at the edge of a cliff.

> The reason for doing it was to use a part that was in the circuit in
> advance, so now we have to add either a PTC in parallel with the IGBT
> or a MOSFET/resistor to do the limited inrush
>

Maybe you can replace the IGBT with a FET? If not I guess the added cost
has to be swallowed in order to achieve a robust design.

How does a PTC in parallel work in this scenario?

Jim Thompson

unread,

Apr 21, 2015, 3:20:03 PM4/21/15

to

On Tue, 21 Apr 2015 12:11:45 -0700, Joerg <ne...@analogconsultants.com>
wrote:

I don't remember those exact words... but they're accurate.

>
>
>> The reason for doing it was to use a part that was in the circuit in
>> advance, so now we have to add either a PTC in parallel with the IGBT
>> or a MOSFET/resistor to do the limited inrush
>>
>
>Maybe you can replace the IGBT with a FET? If not I guess the added cost
>has to be swallowed in order to achieve a robust design.
>
>How does a PTC in parallel work in this scenario?

Jim Thompson

unread,

Apr 21, 2015, 3:28:22 PM4/21/15

to

Reminds me... On one of the GenRad portable supplies I simply used a
relay and a resistor to bring up the caps until a POR released the
relay and activated the loop. That's how I thwarted the wise-ass who
killed my supply by rapidly toggling the ON/OFF switch >:-}

... ON from too short an OFF inserted a delay in the POR.

Joerg

unread,

Apr 21, 2015, 3:39:08 PM4/21/15

to

That's how I used to test ultrasound machines, rocking the circuit
breaker back and forth. Initially they'd evenutally blue-screen. "But
you can't do that!" ... "Can you tell the power grid in Romania that it
shouldn't do that?" ... a few months later the system was nearly
bullet-proof.

> ... ON from too short an OFF inserted a delay in the POR.
>

Jon Elson

unread,

Apr 21, 2015, 5:14:54 PM4/21/15

to

Klaus Kragelund wrote:

> Hi
>
> For an application I am looking into using a large IGBT for inrush (charge
> of DC link capacitor from mains)
>
> To reduce BoM costs, I would like to drive it in linear mode, monitoring
> the current and adjusting the gate voltage accordingly
>
> There are however some talk about hotspots of a linear mode driven IGBT.
>

It is not "some talk". I was advised by IR, years ago, you must NEVER allow
an IGBT to run in linear mode. You have to drive it hard on and hard off.
When in linear mode, they have the typical positive temp coefficient, so
that the hot spot hogs all the current, and goes into thermal runaway. As
the transistor is really a parallel array of thousands of tiny transistor
elements, they don't current share.

When hard saturated, they get into a region where the tempco is much less
positive, and may actually turn just a bit negative, allowing much better
current balancing across the device.

> The IGBT is a 20A device. My intention is to drive it at 10mA, so 2000
> times less than the max current
>
> Does anyone know the workings of these hotspots, and if the issue is real
> or just academic?

It is NOT academic, I've had the silicon die pass me like glowing bullets
when they failed. Why don't you just use a FET? They are always negative
tempco, and can work well in the linear range. The only advantage to an
IGBT is the saturation voltage can be quite low for high voltage devices.
Makes no sense to use an IGBT for linear operation.

Jon

Klaus Kragelund

unread,

Apr 21, 2015, 5:27:54 PM4/21/15

to

The IGBT is part of the power module, used for inrush. But, we need slow controlled inrush (by adding a PTC in parallel), to charge the DC link slowly in order to reduce peak ground currents via the Y capacitors to earth. They would otherwise trip the ground fault relay

Cheers

Klaus

Klaus Kragelund

unread,

Apr 21, 2015, 5:29:10 PM4/21/15

to

That is a common solution, but the relay does not like vibration

Cheers

Klaus

Jim Thompson

unread,

Apr 21, 2015, 5:31:31 PM4/21/15

to

Use a FET or a TRIAC.

Tim Williams

unread,

Apr 21, 2015, 5:38:23 PM4/21/15

to

"Jon Elson" <jme...@wustl.edu> wrote in message
news:DK-dnT20nO5UIKvI...@giganews.com...

> Why don't you just use a FET? They are always negative tempco

^ In the resistive region, of course!

All three (IGBT, MOS and BJT) exhibit NTC in linear/CC range, for the same
general reason as far as I know (charge carrier concentration varying with
temp).

> and can work well in the linear range.

Observing ratings, of course.

The historic claim of "freedom from 2nd breakdown" I think was only ever a
matter of degree: ancient lateral MOS was simply so ludicrously
space-inefficient that it would melt before reaching runaway.

Curiously, SuperJunction MOS usually meets full DC SOA. Not sure if
that's to do with a different tempco, or what.

> The only advantage to an
> IGBT is the saturation voltage can be quite low for high voltage
> devices.
> Makes no sense to use an IGBT for linear operation.
>
> Jon

A little curious what the application/goal and driving constraints are...
a very beefy MOSFET could be chosen here, which would have more than
enough die area to handle the precharge power, and low enough Rds(on) to
drop much less than the IGBT. A'course... such MOSFETs aren't cheap,
either. :) Based on the IGBT (initial?) choice, I'm guessing price is the
bigger motivator?

mrob...@att.net

unread,

Apr 21, 2015, 7:27:21 PM4/21/15

to

Klaus Kragelund <klau...@hotmail.com> wrote:
> The IGBT is part of the power module, used for inrush. But, we need
> slow controlled inrush (by adding a PTC in parallel), to charge the DC
> link slowly in order to reduce peak ground currents via the Y
> capacitors to earth.

I have a portable NiMH battery charger that runs on unleaded gasoline.
To control the inrush when the battery pack (274 V DC nominal) connects
to the inverter, it uses a small ceramic-housed resistor, switched in
and out by a relay at startup. It was built in late 2000, and I don't
know the exact cycle count on it, but it's probably on the order of a
few tens of thousands; no complaints from the relays, resistor, or
inverter. As far as I know, the latest models do it the same way, and
this design has sold about 2 million copies. On the other hand, it
does have the luxury of a pretty high BOM cost, probably US$10k-$15k or
more.

Matt Roberds

Phil Allison

unread,

Apr 21, 2015, 10:14:42 PM4/21/15

to

Klaus Kragelund wrote:
>
>
> For an application I am looking into using a large IGBT for inrush (charge of DC link capacitor from mains)

** That is best done with a suitable NTC thermistor.

> To reduce BoM costs, I would like to drive it
> in linear mode, monitoring the current and adjusting
> the gate voltage accordingly

** Is there a more expensive and idiotic way you might try instead ?

>
> There are however some talk about hotspots
> of a linear mode driven IGBT.
>
> The IGBT is a 20A device.
> My intention is to drive it at 10mA,
> so 2000 times less than the max current

** This is not April 1, is it ?

Cos 10mA is not an inrush surge of any kind.

> Does anyone know the workings of these hotspots, and if the issue is real or just academic?

** I know your posts are not for real.

... Phil

Klaus Bahner

unread,

Apr 22, 2015, 4:06:41 AM4/22/15

to

Contrary to what others have said, I have successfully used IGBTs in
linear mode. The device in question was an IXYS 32A high voltage
NPT-IGBT running at a few mA current with a max CE voltage of up to 2kV.

No problems at all, running for several thousand hours without problems.
So it is feasible, at least at those low currents. As a side note, I was
surprised back then how well the SPICE model reflected reality also in
linear mode. Obviiously, SPICE models don't tell you anything about SOA
etc. and of course the mentioned concerns are for real.

Another example of IGBTs running in linear mode were IGBT audio
amplifiers (don't ask me what that shall be good for, though). Although
some of them gained a bad reputation due to failing IGBTs - there were
also many which actually worked. Google for Alexander (spelling??) amp.

So in summary, you are in uncharted territory and you do it at your own
risk, this is, you have to do a lot of tests to figure out the safety
margins. But there are examples of working IGBT designs in linear mode.

Klaus

Winfield Hill

unread,

Apr 22, 2015, 9:16:41 AM4/22/15

to

Klaus Bahner wrote...

>
>
> Another example of IGBTs running in linear mode were IGBT audio
> amplifiers (don't ask me what that shall be good for, though). Although
> some of them gained a bad reputation due to failing IGBTs - there were
> also many which actually worked. Google for Alexander (spelling??) amp.

See Analog Devices AN-211. Alexander's amplifier used a pair
of special 250V IGBTs that aren't available now. In particular
Toshiba offered complementary IGBTs, gt20d101 and gt20d201.

The high Vgs voltages caused some extra trouble for Alexander.
He used a separate higher voltage for the VAS and driver and
a TL431 plus a thermally-coupled BJT, with gain for biasing.

--
Thanks,
- Win

M Philbrook

unread,

Apr 22, 2015, 5:27:37 PM4/22/15

to

In article <6e0312b6-5b9c-4a36...@googlegroups.com>,
klau...@hotmail.com says...

It's hard to know exactly what it'll do when you attempt a linear
region, doing so means you need close loop to maintain it and
yes, depending on the inrush time it can destroy it if you are pushing
it to the max.

A simpler solution, pulse it when inrush is detected.

Jamie

Robert Loos

unread,

Apr 23, 2015, 12:08:07 PM4/23/15

to

Am 21.04.2015 um 16:53 schrieb Joerg:

> IOW the only way to find out if it is legit for a product is probably to
> obtain the blessing from the manufacturer for brief linear operation.
> Most likely they will not give it. Why not use a FET? I've never had
> much use for IGBT because I could obtain much lower conduction losses
> with FETs and they are cheap.

I'm currently designing a constant current load and I found that BJTs
are the better choice. It is easier to prevent oscillations and in
linear mode, they can carry more current than even really big MOSFETs.
All MOSFETs I considered allow only about 12V at 4A while a cheap TIP102
allows 20V.
And yes, hot spots are real. I have seen transistors (both types) dying
without obvious reason and without getting feelable hot. The process
seems to be very fast.

If you are interested, a preview of this (hobby-) project is available at
www.loosweb.de/eload/doc/de/index.html
(unfortunately, only in german for now)
Under "Safe Operating Area" you can see a rare picture of the SOA in a
linear scale and how dramatically it affects the allowable current at
higher voltages.

Cheers

Robert

Joerg

unread,

Apr 23, 2015, 3:49:07 PM4/23/15

to

On 2015-04-23 8:33 AM, Robert Loos wrote:
> Am 21.04.2015 um 16:53 schrieb Joerg:
>
>> IOW the only way to find out if it is legit for a product is probably to
>> obtain the blessing from the manufacturer for brief linear operation.
>> Most likely they will not give it. Why not use a FET? I've never had
>> much use for IGBT because I could obtain much lower conduction losses
>> with FETs and they are cheap.
>
> I'm currently designing a constant current load and I found that BJTs
> are the better choice. It is easier to prevent oscillations and in

> linear mode, ...

If you are having trouble with that consider a ferrite bead right in
front of the gate. FETs are often faster which is why they can be dicey.
The major upside of FETs is that they need very little drive power
without having to resort to stacked arrangements such as Darlington.

they can carry more current than even really big MOSFETs.

That I doubt.

> All MOSFETs I considered allow only about 12V at 4A while a cheap TIP102
> allows 20V.

Old American saying: Then you need a bigger FET. For example this one:

http://www.fairchildsemi.com/datasheets/FQ/FQP19N20C.pdf

> And yes, hot spots are real. I have seen transistors (both types) dying
> without obvious reason and without getting feelable hot. The process
> seems to be very fast.
>
> If you are interested, a preview of this (hobby-) project is available at
> www.loosweb.de/eload/doc/de/index.html
> (unfortunately, only in german for now)

Nice. When I semi-retire some day I want to learn how to create a better
web site. Not so easy for an analog guy with next to nothing in
programming experience.

> Under "Safe Operating Area" you can see a rare picture of the SOA in a
> linear scale and how dramatically it affects the allowable current at
> higher voltages.
>

Yep, got to watch that SOA and also not push it to the limits. John
Larkin has some impressive photos of the results when one does push the
limits.

Robert Loos

unread,

Apr 24, 2015, 6:08:06 AM4/24/15

to

Am 23.04.2015 um 21:49 schrieb Joerg:

>> I'm currently designing a constant current load and I found that BJTs
>> are the better choice. It is easier to prevent oscillations and in
>> linear mode, ...
>
>
> If you are having trouble with that consider a ferrite bead right in
> front of the gate. FETs are often faster which is why they can be dicey.
> The major upside of FETs is that they need very little drive power
> without having to resort to stacked arrangements such as Darlington.

you mean something like a würth
http://katalog.we-online.de/pbs/datasheet/742792693.pdf
in series with the gate resistor?

My first approach was with a MOSFET (can't remember the exact type), I
turned it on and I had a fabulous power oscillator. I hacked the circuit
into LTspice and thought it would not oscillate but it showed exactly
the same curves than the real circuit. I selected a different FET and it
was quiet. Don't know what parameter of the FET makes this behavior.
This was the main reason I switched to BJTs. May be, with a different
BJT it will start oscillating again...

> they can carry more current than even really big MOSFETs.
>
>
> That I doubt.

ok.

>> All MOSFETs I considered allow only about 12V at 4A while a cheap TIP102
>> allows 20V.
>
>
> Old American saying: Then you need a bigger FET. For example this one:
>
> http://www.fairchildsemi.com/datasheets/FQ/FQP19N20C.pdf

what a cool device. 4A x 35V is just the maximum Pd. I should have asked
you earlier :-)

>> If you are interested, a preview of this (hobby-) project is available at
>> www.loosweb.de/eload/doc/de/index.html
>> (unfortunately, only in german for now)
>
>
> Nice. When I semi-retire some day I want to learn how to create a better
> web site. Not so easy for an analog guy with next to nothing in
> programming experience.

It is pain. I'm also not a html professional. The usual learning curve
is you create a layout that looks like you want, then you try a
different browser and start again. Some day you have a look at your site
with your android phone and the real work begins. At last, some
colleague tries it on his I-phone and it appears completely
non-functional...
If you want, you can use any modification of my css, there's nothing
secret about it. They are compressed on my site for performance reasons
but I can send you the human readable version. If you have special
questions, just drop me an email.

> Yep, got to watch that SOA and also not push it to the limits. John
> Larkin has some impressive photos of the results when one does push the
> limits.

Got a link?

Cheers

Robert

Chris Jones

unread,

Apr 24, 2015, 7:35:39 AM4/24/15

to

On 24/04/2015 19:59, Robert Loos wrote:
[...]

>
> My first approach was with a MOSFET (can't remember the exact type), I
> turned it on and I had a fabulous power oscillator. I hacked the circuit
> into LTspice and thought it would not oscillate but it showed exactly
> the same curves than the real circuit.

Are you driving the gate with an op-amp? If so, then you could try
swapping the op-amp for an OTA like the first stage of a LT1228 (take
the output from pin 1 not pin 6, and maybe add some extra capacitance on
the gate), or try some op-amp that is intended for driving capacitive loads.

The open-loop output impedance of most op-amps combined with the gate
capacitance of a big MOSFET (or even worse, the Cgd * Miller effect),
will give you a nearly 90 degree phase lag at many frequencies, that was
not allowed for by whoever designed the compensation of most op-amps.
Together with another nearly 90 degree lag in the compensation of the
op-amp, and a few degrees from other parasitic poles, and you have an
oscillator.

Chris

Winfield Hill

unread,

Apr 24, 2015, 9:40:32 AM4/24/15

to

Chris Jones wrote...

Yes there are twp poles, but generally it's easy
to add a zero in the loop to stabilize it.

--
Thanks,
- Win

Joerg

unread,

Apr 24, 2015, 11:08:51 AM4/24/15

to

On 2015-04-24 2:59 AM, Robert Loos wrote:
> Am 23.04.2015 um 21:49 schrieb Joerg:
>
>>> I'm currently designing a constant current load and I found that BJTs
>>> are the better choice. It is easier to prevent oscillations and in
>>> linear mode, ...
>>
>>
>> If you are having trouble with that consider a ferrite bead right in
>> front of the gate. FETs are often faster which is why they can be dicey.
>> The major upside of FETs is that they need very little drive power
>> without having to resort to stacked arrangements such as Darlington.
>
> you mean something like a würth
> http://katalog.we-online.de/pbs/datasheet/742792693.pdf
> in series with the gate resistor?
>

That one is a bit extreme but for slow stuff it would work. Although I
never use gate resistors you may have to place 100ohms or so if driving
the FET with an opapmp. As Chris wrote many of those cannot take a
capacitive load directly and a FET is always a capacitive load. You have
to close the loop for higher frequencies before the bead/resistor, for
example using 1000pF from opamp output to negative input.

BJT do not present much of a capacitive load. But you can't blame a FET
for that to oscillate. That would be as if I blamed my mountain bike for
the tree that was "in the way" last year :-)

> My first approach was with a MOSFET (can't remember the exact type), I
> turned it on and I had a fabulous power oscillator.

How did you turn it on?

> ... I hacked the circuit

> into LTspice and thought it would not oscillate but it showed exactly
> the same curves than the real circuit. I selected a different FET and it
> was quiet. Don't know what parameter of the FET makes this behavior.
> This was the main reason I switched to BJTs. May be, with a different
> BJT it will start oscillating again...
>

Can you post the LTSpice circuit file? It's ASCII so you can just copy
that into a post here.

>> they can carry more current than even really big MOSFETs.
>>
>>
>> That I doubt.
>
> ok.
>
>>> All MOSFETs I considered allow only about 12V at 4A while a cheap TIP102
>>> allows 20V.
>>
>>
>> Old American saying: Then you need a bigger FET. For example this one:
>>
>> http://www.fairchildsemi.com/datasheets/FQ/FQP19N20C.pdf
>
> what a cool device. 4A x 35V is just the maximum Pd. I should have asked
> you earlier :-)
>

With BJT as well as FETs it's not great to burn that much power in a
single device. Gets hot, unless you do that only for a short time. Best
to spread into 2-3 devices on a large heat sink.

>>> If you are interested, a preview of this (hobby-) project is
>>> available at
>>> www.loosweb.de/eload/doc/de/index.html
>>> (unfortunately, only in german for now)
>>
>>
>> Nice. When I semi-retire some day I want to learn how to create a better
>> web site. Not so easy for an analog guy with next to nothing in
>> programming experience.
>
> It is pain. I'm also not a html professional. The usual learning curve
> is you create a layout that looks like you want, then you try a
> different browser and start again. Some day you have a look at your site
> with your android phone and the real work begins. At last, some
> colleague tries it on his I-phone and it appears completely
> non-functional...

I'd have to rely on other people's smart phones because I don't have one
and right now don't want one. Other than the occasional email I would
have little use for it.

> If you want, you can use any modification of my css, there's nothing
> secret about it. They are compressed on my site for performance reasons
> but I can send you the human readable version. If you have special
> questions, just drop me an email.
>

Thanks, maybe I do that. But first I have to reach semi-retirement :-)

I don't really need a web site, currently it's mainly used for people to
find me back. Old clients after many years and so on. But later I want
to start writing circuit design guidelines for the next generations
because most universities do not teach analog properly anymore. That
would mean proper schematic example linking, maybe PDFs, et cetera.

>> Yep, got to watch that SOA and also not push it to the limits. John
>> Larkin has some impressive photos of the results when one does push the
>> limits.
>
> Got a link?
>

Can't find it. John, if you are reading this, can you post the link for
your blown FET collection?

Robert Loos

unread,

Apr 27, 2015, 4:08:07 AM4/27/15

to

Am 24.04.2015 um 17:08 schrieb Joerg:
...

>> you mean something like a würth
>> http://katalog.we-online.de/pbs/datasheet/742792693.pdf
>> in series with the gate resistor?
>>
>
> That one is a bit extreme but for slow stuff it would work. Although I
> never use gate resistors you may have to place 100ohms or so if driving
> the FET with an opapmp. As Chris wrote many of those cannot take a
> capacitive load directly and a FET is always a capacitive load. You have
> to close the loop for higher frequencies before the bead/resistor, for
> example using 1000pF from opamp output to negative input.

I thought so. I just picked the one with the highest impedance we have
in our stock :-)

> BJT do not present much of a capacitive load. But you can't blame a FET
> for that to oscillate. That would be as if I blamed my mountain bike for
> the tree that was "in the way" last year :-)

Glad that you are well up again.

> How did you turn it on?

Well, I just plugged the wall wart... see the spice file for details.

> Can you post the LTSpice circuit file? It's ASCII so you can just copy
> that into a post here.

Here it comes:

Version 4
SHEET 1 2184 788
WIRE 1712 -320 912 -320
WIRE 1712 -208 1712 -320
WIRE 480 -160 480 -224
WIRE 912 -96 912 -320
WIRE 288 -32 240 -32
WIRE 336 -32 288 -32
WIRE 480 -16 480 -80
WIRE 480 -16 400 -16
WIRE 512 -16 480 -16
WIRE 544 -16 512 -16
WIRE 640 -16 624 -16
WIRE 704 -16 640 -16
WIRE 800 -16 704 -16
WIRE 864 -16 800 -16
WIRE -128 0 -432 0
WIRE 64 0 -48 0
WIRE 144 0 64 0
WIRE 288 0 144 0
WIRE 336 0 288 0
WIRE 144 32 144 0
WIRE 640 48 640 -16
WIRE 704 64 704 -16
WIRE 1712 64 1712 -128
WIRE 240 128 240 -32
WIRE 336 128 240 128
WIRE 480 128 480 -16
WIRE 480 128 400 128
WIRE 640 160 640 128
WIRE 704 160 704 128
WIRE 144 192 144 96
WIRE 240 288 240 128
WIRE 64 320 64 0
WIRE -720 336 -720 288
WIRE -432 336 -432 0
WIRE 240 416 240 368
WIRE 816 416 240 416
WIRE 912 416 912 0
WIRE 912 416 816 416
WIRE -720 448 -720 416
WIRE 912 448 912 416
WIRE -432 736 -432 416
WIRE -176 736 -176 576
WIRE -176 736 -432 736
WIRE 64 736 64 400
WIRE 64 736 -176 736
WIRE 912 736 912 528
WIRE 912 736 64 736
WIRE 1712 736 1712 144
WIRE 1712 736 912 736
WIRE 912 768 912 736
FLAG 912 768 0
FLAG -720 448 0
FLAG -720 288 P
FLAG 704 160 0
FLAG 512 -16 QU1
FLAG 144 192 0
FLAG 368 -48 P
FLAG 640 160 0
FLAG 288 -32 InM
FLAG 288 0 InP
FLAG 800 -16 Gate
FLAG 368 16 0
FLAG 480 -224 0
FLAG 816 416 Sense
SYMBOL nmos 864 -96 R0
SYMATTR InstName M1
SYMATTR Value Si4466DY
SYMBOL res 896 432 R0
SYMATTR InstName R1
SYMATTR Value 3R4
SYMATTR SpiceLine tol=5 pwr=3
SYMBOL voltage 1712 48 R0
WINDOW 123 0 0 Left 2
WINDOW 39 0 0 Left 2
SYMATTR InstName V1
SYMATTR Value PULSE(12 13 0 10u 10u 10m 20m)
SYMBOL voltage -720 320 R0
SYMATTR InstName V2
SYMATTR Value 12
SYMBOL voltage -176 480 R0
WINDOW 123 0 0 Left 2
WINDOW 39 0 0 Left 2
SYMATTR InstName V4
SYMATTR Value SINE(2.5 2.5 10)
SYMBOL voltage -432 320 R0
WINDOW 123 0 0 Left 2
WINDOW 39 0 0 Left 2
SYMATTR InstName V5
SYMATTR Value PULSE(0.01 2.5 0 1n 1n {Tperiod/2} {Tperiod})
SYMBOL res 640 -32 R90
WINDOW 0 0 56 VBottom 2
WINDOW 3 32 56 VTop 2
SYMATTR InstName R9
SYMATTR Value 56
SYMBOL cap 688 64 R0
SYMATTR InstName C1
SYMATTR Value 1f
SYMBOL res -32 -16 R90
WINDOW 0 0 56 VBottom 2
WINDOW 3 32 56 VTop 2
SYMATTR InstName R10
SYMATTR Value 8k2
SYMBOL res 48 304 R0
SYMATTR InstName R2
SYMATTR Value 2k2
SYMBOL res 224 272 R0
SYMATTR InstName R3
SYMATTR Value 1k
SYMBOL cap 400 112 R90
WINDOW 0 0 32 VBottom 2
WINDOW 3 32 32 VTop 2
SYMATTR InstName C2
SYMATTR Value 1µ
SYMBOL cap 128 32 R0
SYMATTR InstName C6
SYMATTR Value 4n7
SYMBOL Opamps\\LT2078 368 -80 R0
SYMATTR InstName U1
SYMBOL res 624 32 R0
SYMATTR InstName R5
SYMATTR Value 1t
SYMBOL res 464 -176 R0
SYMATTR InstName R11
SYMATTR Value 1t
SYMBOL ind 1696 -224 R0
SYMATTR InstName L1
SYMATTR Value 10µ
TEXT -752 648 Left 2 !.tran .3
TEXT 1016 488 Left 2 ;R-Sense
TEXT 936 536 Left 2 ;3W
TEXT -744 -104 Left 2 !.param Tperiod=.1

I hope there are no special parts in it.
There are some teraohm resistors and femtofarad capacitors in it, I used
them to play around. For example, if you reduce the pullup at the opamps
output to 2k or less, oscillation stops.

> With BJT as well as FETs it's not great to burn that much power in a
> single device. Gets hot, unless you do that only for a short time. Best
> to spread into 2-3 devices on a large heat sink.

Of course. But my design will only operate in the range of 10..20W so
this is no problem.

> I'd have to rely on other people's smart phones because I don't have one
> and right now don't want one. Other than the occasional email I would
> have little use for it.

Seems we have the same problem. I bought an android phone to keep me a
little bit up to date. But of the three calls I get per year, two are
from my wife when she get stuck somewhere with her car...
But it is great to find a pizza nearby :-)

>> If you want, you can use any modification of my css, there's nothing
>> secret about it. They are compressed on my site for performance reasons
>> but I can send you the human readable version. If you have special
>> questions, just drop me an email.
>>
>
> Thanks, maybe I do that. But first I have to reach semi-retirement :-)

May come faster than you think. I know what I'm talking about.

> I don't really need a web site, currently it's mainly used for people to
> find me back. Old clients after many years and so on. But later I want
> to start writing circuit design guidelines for the next generations
> because most universities do not teach analog properly anymore. That
> would mean proper schematic example linking, maybe PDFs, et cetera.

I would highly appreciate if you started writing a little bit earlier :-)
A web site is the cheapest form of advertisement.

Cheers

Robert

Joerg

unread,

Apr 27, 2015, 4:56:12 PM4/27/15

to

On 2015-04-27 1:03 AM, Robert Loos wrote:
> Am 24.04.2015 um 17:08 schrieb Joerg:
> ...
>>> you mean something like a würth
>>> http://katalog.we-online.de/pbs/datasheet/742792693.pdf
>>> in series with the gate resistor?
>>>
>>
>> That one is a bit extreme but for slow stuff it would work. Although I
>> never use gate resistors you may have to place 100ohms or so if driving
>> the FET with an opapmp. As Chris wrote many of those cannot take a
>> capacitive load directly and a FET is always a capacitive load. You have
>> to close the loop for higher frequencies before the bead/resistor, for
>> example using 1000pF from opamp output to negative input.
>
> I thought so. I just picked the one with the highest impedance we have
> in our stock :-)
>

But the opamp won't like a direct capacitive load.

>> BJT do not present much of a capacitive load. But you can't blame a FET
>> for that to oscillate. That would be as if I blamed my mountain bike for
>> the tree that was "in the way" last year :-)
>
> Glad that you are well up again.
>

That happened the evening before a flight to Europe. I was on a narrow
trail, it got late, so I shifted up and stepped on it. 15 miles an hour,
18, 19, 20 ... *KAPOW* ... big dust plume wafted off ... I was majorly
scraped up and at the airport some people looked as if I had some sort
of disease.

>> How did you turn it on?
>
> Well, I just plugged the wall wart... see the spice file for details.
>
>> Can you post the LTSpice circuit file? It's ASCII so you can just copy
>> that into a post here.
>
> Here it comes:
>

[LTSPice netlist]

> I hope there are no special parts in it.

No, but the opamp is the problem.

> There are some teraohm resistors and femtofarad capacitors in it, I used
> them to play around. For example, if you reduce the pullup at the opamps
> output to 2k or less, oscillation stops.
>

You've used an expensive micropower opamp in there. Those are no good to
drive any serious loads directly and this is a serious load. For
example, if you replace it with the OP07 (I tried it because it happened
to be in the LTSPice library) nothing oscillates anymore. But even then
you have to simulate and measure the phase margin so your circuit won't
be too marginal.

http://www.linear.com/docs/3884

The graph on page 6 bottom right shows the problem. The output already
starts to collapse at a mere 2mA. It doesn't have much oomph, wrong part
for the job.

[...]

>> I'd have to rely on other people's smart phones because I don't have one
>> and right now don't want one. Other than the occasional email I would
>> have little use for it.
>
> Seems we have the same problem. I bought an android phone to keep me a
> little bit up to date. But of the three calls I get per year, two are
> from my wife when she get stuck somewhere with her car...
> But it is great to find a pizza nearby :-)
>

It's even more extreme here. A few years ago was the first time in our
life that we ordered take-out pizza. Normally we always bake our own,
outdoors over a wood fire.

>>> If you want, you can use any modification of my css, there's nothing
>>> secret about it. They are compressed on my site for performance reasons
>>> but I can send you the human readable version. If you have special
>>> questions, just drop me an email.
>>>
>>
>> Thanks, maybe I do that. But first I have to reach semi-retirement :-)
>
> May come faster than you think. I know what I'm talking about.
>

I wish. Then I could bike more.

>> I don't really need a web site, currently it's mainly used for people to
>> find me back. Old clients after many years and so on. But later I want
>> to start writing circuit design guidelines for the next generations
>> because most universities do not teach analog properly anymore. That
>> would mean proper schematic example linking, maybe PDFs, et cetera.
>
> I would highly appreciate if you started writing a little bit earlier :-)

Time is the problem right now. I did reduce my business hours a little
already but the free hours are often filled with volunteer work such as
nursing home visits. We feel that it's the right thing to do.

> A web site is the cheapest form of advertisement.
>

I really don't need any advertizing but yes, in the past the simple web
site I have has helped obtain interesting new assignments. It helped me
get out of medical device design to a large extent which began to become
a bit boring and that field was also largely destroyed by Obamacare.

Robert Loos

unread,

Apr 28, 2015, 12:08:06 PM4/28/15

to

Am 24.04.2015 um 15:40 schrieb Winfield Hill:

> Yes there are twp poles, but generally it's easy
> to add a zero in the loop to stabilize it.

Winfield, it's easy for you!
But I'm afraid I was too often away from important lessons...
Can you give one or two sentences for dummies about how to do that?

Cheers

Robert

Tim Williams

unread,

Apr 28, 2015, 3:12:56 PM4/28/15

to

"Robert Loos" <12...@baer-gmbh.com> wrote in message
news:kik41c-...@baer-gmbh.com...

Instead of a feedback C from out to -in, use an R+C. That's the basic
approach; it works quite well with 1/2 order systems (giving 2/3 order
loops that can occasionally be tweaked to reasonably optimal response).

Tim

--
Seven Transistor Labs, LLC
Electrical Engineering Consultation and Contract Design
Website: http://seventransistorlabs.com

Winfield Hill

unread,

Apr 29, 2015, 3:36:20 AM4/29/15

to

Tim Williams wrote...
>
> Robert Loos wrote ...

>> schrieb Winfield Hill:
>>
>>> Yes there are twp poles, but generally it's easy
>>> to add a zero in the loop to stabilize it.
>>
>> Winfield, it's easy for you!
>> But I'm afraid I was too often away from important lessons...
>> Can you give one or two sentences for dummies about how to do that?
>
> Instead of a feedback C from out to -in, use an R+C. That's the
> basic approach; it works quite well with 1/2 order systems
> (giving 2/3 order loops that can occasionally be tweaked to
> reasonably optimal response).
>
> Tim

Right, see Robert, I said it was easy!

--
Thanks,
- Win

Robert Loos

unread,

Apr 29, 2015, 9:08:08 AM4/29/15

to

Am 27.04.2015 um 22:56 schrieb Joerg:

> No, but the opamp is the problem.

...

> You've used an expensive micropower opamp in there. Those are no good to
> drive any serious loads directly and this is a serious load. For
> example, if you replace it with the OP07 (I tried it because it happened
> to be in the LTSPice library) nothing oscillates anymore. But even then
> you have to simulate and measure the phase margin so your circuit won't
> be too marginal.

hm. With the OP07 it does not oscillate but it also does not work since
the OP07 has no rail to rail input. If I add a negative supply, it
oscillates again, only at a higher frequency.
Ferrite bead at the gate won't help either.

There is a video on the LT-site:
http://www.linear.com/solutions/4449
that shows how to measure the phase margin. I tried this with the
original circuit, with the feedback C reduced to 200pF and it stays well
above 50 degrees. But the transient simulation oscillates.

Here the Netlist:

Version 4
SHEET 1 2184 788
WIRE 1712 -320 912 -320
WIRE 1712 -208 1712 -320
WIRE 480 -160 480 -224
WIRE 912 -96 912 -320
WIRE 288 -32 240 -32
WIRE 336 -32 288 -32
WIRE 480 -16 480 -80
WIRE 480 -16 400 -16
WIRE 512 -16 480 -16
WIRE 544 -16 512 -16
WIRE 640 -16 624 -16
WIRE 704 -16 640 -16
WIRE 800 -16 704 -16
WIRE 864 -16 800 -16
WIRE -128 0 -432 0
WIRE 64 0 -48 0
WIRE 144 0 64 0
WIRE 288 0 144 0
WIRE 336 0 288 0
WIRE 144 32 144 0

WIRE 240 48 240 -32

WIRE 640 48 640 -16
WIRE 704 64 704 -16
WIRE 1712 64 1712 -128

WIRE 640 160 640 128
WIRE 704 160 704 128

WIRE 240 176 240 128
WIRE 288 176 240 176
WIRE 336 176 288 176
WIRE 480 176 480 -16
WIRE 480 176 400 176

WIRE 144 192 144 96

WIRE 240 288 240 176

WIRE 64 320 64 0

WIRE -752 336 -752 288
WIRE -608 336 -608 288

WIRE 240 416 240 368
WIRE 816 416 240 416
WIRE 912 416 912 0
WIRE 912 416 816 416

WIRE -752 448 -752 416
WIRE -608 448 -608 416

WIRE 912 448 912 416

WIRE -432 736 -432 0

WIRE 64 736 64 400

WIRE 64 736 -432 736

WIRE 912 736 912 528
WIRE 912 736 64 736
WIRE 1712 736 1712 144
WIRE 1712 736 912 736
WIRE 912 768 912 736
FLAG 912 768 0

FLAG -752 448 0
FLAG -752 288 P

FLAG 704 160 0
FLAG 512 -16 QU1
FLAG 144 192 0
FLAG 368 -48 P
FLAG 640 160 0
FLAG 288 -32 InM
FLAG 288 0 InP
FLAG 800 -16 Gate

FLAG 480 -224 0
FLAG 816 416 Sense

FLAG 288 176 FB
FLAG -608 448 0
FLAG -608 288 M
FLAG 368 16 M

SYMBOL nmos 864 -96 R0
SYMATTR InstName M1
SYMATTR Value Si4466DY
SYMBOL res 896 432 R0
SYMATTR InstName R1
SYMATTR Value 3R4
SYMATTR SpiceLine tol=5 pwr=3
SYMBOL voltage 1712 48 R0
WINDOW 123 0 0 Left 2
WINDOW 39 0 0 Left 2

SYMATTR InstName V3
SYMATTR Value 12
SYMBOL voltage -752 320 R0

SYMATTR InstName V2
SYMATTR Value 12

SYMBOL res 640 -32 R90
WINDOW 0 0 56 VBottom 2
WINDOW 3 32 56 VTop 2
SYMATTR InstName R9
SYMATTR Value 56
SYMBOL cap 688 64 R0
SYMATTR InstName C1
SYMATTR Value 1f
SYMBOL res -32 -16 R90
WINDOW 0 0 56 VBottom 2
WINDOW 3 32 56 VTop 2
SYMATTR InstName R10
SYMATTR Value 8k2
SYMBOL res 48 304 R0
SYMATTR InstName R2
SYMATTR Value 2k2
SYMBOL res 224 272 R0
SYMATTR InstName R3
SYMATTR Value 1k

SYMBOL cap 400 160 R90

WINDOW 0 0 32 VBottom 2
WINDOW 3 32 32 VTop 2
SYMATTR InstName C2

SYMATTR Value 200p

SYMBOL cap 128 32 R0
SYMATTR InstName C6
SYMATTR Value 4n7

SYMBOL res 624 32 R0
SYMATTR InstName R5
SYMATTR Value 1t
SYMBOL res 464 -176 R0
SYMATTR InstName R11
SYMATTR Value 1t
SYMBOL ind 1696 -224 R0
SYMATTR InstName L1
SYMATTR Value 10µ

SYMBOL voltage 240 32 R0
WINDOW 3 41 61 Left 2
WINDOW 123 41 89 Left 2

WINDOW 39 0 0 Left 2

SYMATTR Value 0
SYMATTR InstName V1
SYMATTR Value2 AC 1
SYMBOL voltage -608 432 R180
WINDOW 0 24 96 Left 2
WINDOW 3 -65 10 Left 2
SYMATTR InstName V4
SYMATTR Value 0

SYMBOL Opamps\\LT2078 368 -80 R0
SYMATTR InstName U1

TEXT -752 648 Left 2 !.ac oct 12 100 10meg

TEXT 1016 488 Left 2 ;R-Sense
TEXT 936 536 Left 2 ;3W
TEXT -744 -104 Left 2 !.param Tperiod=.1

TEXT -512 -240 Left 2 ;According to the LTspice video, plot the ratio of
V(fb)/V(inm). The phase margin is\nthe phase at 0dB.

I've added a negative supply that I set to 0 for the LT2078.
Something very interesting happens if you raise it. Up to 9.99V the
curve stays nearly the same. But at 10V it looks completely different!
Can this be real? A problem in LTspice?
What I find also curious is that the oscillation is nearly independent
of the feedback C. You cannot stop is even with arbitrary big C. Not
even the frequency changes.
Also the inductance I added in the load path to simulate line inductance
does not significantly change it. What is the timing factor in this
circuit? A different opamp changes it. But how? Such a simple circuit
but this babe drives me crazy.

> It's even more extreme here. A few years ago was the first time in our
> life that we ordered take-out pizza. Normally we always bake our own,
> outdoors over a wood fire.

*smack*

Cheers

Robert

Robert Loos

unread,

Apr 29, 2015, 9:08:08 AM4/29/15

to

I'm ashamed. Thank you both!

Cheers

Robert

Joerg

unread,

Apr 29, 2015, 4:40:24 PM4/29/15

to

On 2015-04-29 4:30 AM, Robert Loos wrote:
> Am 27.04.2015 um 22:56 schrieb Joerg:
>
>> No, but the opamp is the problem.
> ...
>> You've used an expensive micropower opamp in there. Those are no good to
>> drive any serious loads directly and this is a serious load. For
>> example, if you replace it with the OP07 (I tried it because it happened
>> to be in the LTSPice library) nothing oscillates anymore. But even then
>> you have to simulate and measure the phase margin so your circuit won't
>> be too marginal.
> hm. With the OP07 it does not oscillate but it also does not work since
> the OP07 has no rail to rail input. If I add a negative supply, it
> oscillates again, only at a higher frequency.
> Ferrite bead at the gate won't help either.
>
> There is a video on the LT-site:
> http://www.linear.com/solutions/4449
> that shows how to measure the phase margin. I tried this with the
> original circuit, with the feedback C reduced to 200pF and it stays well
> above 50 degrees. But the transient simulation oscillates.
>

.AC is a small signal analysis method and does not work well for power
stuff, it has to be measured.

The OP07 was just one example and if I go above 120ohms it no longer
oscillates even with negative supply. For this kind of application you
need an opamp with a more powerful output or have a complementary pair
follower. I only used the OP07 because it was in the LTSpice directory.
Bottomline this is never going to work well with the LT2078 unless you
make it very slow. Why do you want a micropower opamp in there?

[LtSpice netlist]

> I've added a negative supply that I set to 0 for the LT2078.
> Something very interesting happens if you raise it. Up to 9.99V the
> curve stays nearly the same. But at 10V it looks completely different!
> Can this be real? A problem in LTspice?

Not likely. Opamps and other models are at least to some extent
behavioral models and there can be parameters under which they fall off
the cliff. It shouldn't with this opamp though because I believe it can
go to +/-18V.

> What I find also curious is that the oscillation is nearly independent
> of the feedback C. You cannot stop is even with arbitrary big C. Not
> even the frequency changes.

Strange. If I use, for example, 120ohms it is stable until about 0.02uF.
At 0.01uF it oscillates. Kind of what I'd expect. Here it is with 0.1uF:

Version 4
SHEET 1 2184 788
WIRE 1712 -320 912 -320
WIRE 1712 -208 1712 -320
WIRE 480 -160 480 -224
WIRE 912 -96 912 -320
WIRE 288 -32 240 -32
WIRE 336 -32 288 -32
WIRE 480 -16 480 -80
WIRE 480 -16 400 -16
WIRE 512 -16 480 -16
WIRE 544 -16 512 -16
WIRE 640 -16 624 -16
WIRE 704 -16 640 -16
WIRE 800 -16 704 -16
WIRE 864 -16 800 -16
WIRE -128 0 -432 0
WIRE 64 0 -48 0
WIRE 144 0 64 0
WIRE 288 0 144 0
WIRE 336 0 288 0
WIRE 144 32 144 0

WIRE 368 48 368 16

WIRE 640 48 640 -16
WIRE 704 64 704 -16
WIRE 1712 64 1712 -128

WIRE 368 144 368 128

WIRE 640 160 640 128
WIRE 704 160 704 128

WIRE 144 192 144 96

WIRE 240 224 240 -32
WIRE 336 224 240 224
WIRE 480 224 480 -16
WIRE 480 224 400 224
WIRE 240 288 240 224

WIRE 64 320 64 0

WIRE -720 336 -720 288
WIRE -432 336 -432 0

WIRE 240 416 240 368
WIRE 816 416 240 416
WIRE 912 416 912 0
WIRE 912 416 816 416

WIRE -720 448 -720 416

WIRE 912 448 912 416

WIRE -432 736 -432 416
WIRE -176 736 -176 576
WIRE -176 736 -432 736

WIRE 64 736 64 400

WIRE 64 736 -176 736

WIRE 912 736 912 528
WIRE 912 736 64 736
WIRE 1712 736 1712 144
WIRE 1712 736 912 736
WIRE 912 768 912 736
FLAG 912 768 0

FLAG -720 448 0
FLAG -720 288 P

FLAG 704 160 0
FLAG 512 -16 QU1
FLAG 144 192 0
FLAG 368 -48 P
FLAG 640 160 0
FLAG 288 -32 InM
FLAG 288 0 InP
FLAG 800 -16 Gate
FLAG 480 -224 0
FLAG 816 416 Sense

FLAG 368 144 0

SYMBOL nmos 864 -96 R0
SYMATTR InstName M1
SYMATTR Value Si4466DY
SYMBOL res 896 432 R0
SYMATTR InstName R1
SYMATTR Value 3R4
SYMATTR SpiceLine tol=5 pwr=3
SYMBOL voltage 1712 48 R0
WINDOW 123 0 0 Left 2
WINDOW 39 0 0 Left 2

SYMATTR InstName V1
SYMATTR Value PULSE(12 13 0 10u 10u 10m 20m)

SYMBOL voltage -720 320 R0

SYMATTR InstName V2
SYMATTR Value 12

SYMBOL voltage -176 480 R0

WINDOW 123 0 0 Left 2
WINDOW 39 0 0 Left 2

SYMATTR InstName V4
SYMATTR Value SINE(2.5 2.5 10)

SYMBOL voltage -432 320 R0

WINDOW 123 0 0 Left 2
WINDOW 39 0 0 Left 2

SYMATTR InstName V5
SYMATTR Value PULSE(0.01 2.5 0 1n 1n {Tperiod/2} {Tperiod})

SYMBOL res 640 -32 R90
WINDOW 0 0 56 VBottom 2
WINDOW 3 32 56 VTop 2
SYMATTR InstName R9

SYMATTR Value 120

SYMBOL cap 688 64 R0
SYMATTR InstName C1
SYMATTR Value 1f
SYMBOL res -32 -16 R90
WINDOW 0 0 56 VBottom 2
WINDOW 3 32 56 VTop 2
SYMATTR InstName R10
SYMATTR Value 8k2
SYMBOL res 48 304 R0
SYMATTR InstName R2
SYMATTR Value 2k2
SYMBOL res 224 272 R0
SYMATTR InstName R3
SYMATTR Value 1k

SYMBOL cap 400 208 R90

WINDOW 0 0 32 VBottom 2
WINDOW 3 32 32 VTop 2
SYMATTR InstName C2

SYMATTR Value 0.1µ

SYMBOL cap 128 32 R0
SYMATTR InstName C6
SYMATTR Value 4n7
SYMBOL res 624 32 R0
SYMATTR InstName R5
SYMATTR Value 1t
SYMBOL res 464 -176 R0
SYMATTR InstName R11
SYMATTR Value 1t
SYMBOL ind 1696 -224 R0
SYMATTR InstName L1
SYMATTR Value 10µ

SYMBOL Opamps\\OP07 368 -80 R0
SYMATTR InstName U1
SYMBOL voltage 368 144 R180

WINDOW 0 24 96 Left 2

WINDOW 3 24 16 Left 2

WINDOW 123 0 0 Left 2
WINDOW 39 0 0 Left 2
SYMATTR InstName V3

SYMATTR Value 6
TEXT -752 648 Left 2 !.tran .3

TEXT 1016 488 Left 2 ;R-Sense
TEXT 936 536 Left 2 ;3W
TEXT -744 -104 Left 2 !.param Tperiod=.1

> Also the inductance I added in the load path to simulate line inductance
> does not significantly change it. What is the timing factor in this
> circuit? A different opamp changes it. But how? Such a simple circuit
> but this babe drives me crazy.
>

The inductance is 10uH, very small. The opamp can change it because they
differ in drive capability. That makes a huge difference when driving
capacitive or other hard loads. Small signal analysis doesn't help that
much with it. For example, the slew rate matters, big time. This is why
at university I never really bought the small signal theory they taught
us. Full confession: It resulted in me skipping class and just writing
in the exam what they wanted to hear.

>> It's even more extreme here. A few years ago was the first time in our
>> life that we ordered take-out pizza. Normally we always bake our own,
>> outdoors over a wood fire.
> *smack*
>

Wood-fire baked pizza tastes beyond compare. We've never looked back,
haven't baked one in the oven since years because it simply isn't as good.

Robert Loos

unread,

Apr 30, 2015, 9:08:08 AM4/30/15

to

Am 29.04.2015 um 22:40 schrieb Joerg:

>> There is a video on the LT-site:
>> http://www.linear.com/solutions/4449
>> that shows how to measure the phase margin. I tried this with the
>> original circuit, with the feedback C reduced to 200pF and it stays well
>> above 50 degrees. But the transient simulation oscillates.
>>
>
> .AC is a small signal analysis method and does not work well for power
> stuff, it has to be measured.

I found a way to simulate it: the good phase margin is just because the
input is grounded. If you insert a 2V voltage source the plot shows a
terrible phase and it is obvious that it must oscillate.

> The OP07 was just one example and if I go above 120ohms it no longer
> oscillates even with negative supply. For this kind of application you
> need an opamp with a more powerful output or have a complementary pair
> follower. I only used the OP07 because it was in the LTSpice directory.
> Bottomline this is never going to work well with the LT2078 unless you
> make it very slow. Why do you want a micropower opamp in there?

Historic reasons. In my first board I used a TLC272 (without thinking
much about it, just because it was on stock) but the high offset made it
impossible to set the current to 0.
Then I was searching for a pin compatible type with rail to rail input
and low offset that is easy to get and ended up with the LT2078.
Since I'll have to make a new board anyway, I will now search for a type
with rail to rail input, low offset and high drive capability that is
easy to get :-)

>> What I find also curious is that the oscillation is nearly independent
>> of the feedback C. You cannot stop is even with arbitrary big C. Not
>> even the frequency changes.
>
> Strange. If I use, for example, 120ohms it is stable until about 0.02uF.
> At 0.01uF it oscillates. Kind of what I'd expect. Here it is with 0.1uF:

...
The gate resistor is the clue. If you plot the .ac analysis (with the 2V
source) you can see how the phase improves. 220R would be even better.

> The inductance is 10uH, very small. The opamp can change it because they
> differ in drive capability. That makes a huge difference when driving
> capacitive or other hard loads. Small signal analysis doesn't help that
> much with it. For example, the slew rate matters, big time. This is why
> at university I never really bought the small signal theory they taught
> us. Full confession: It resulted in me skipping class and just writing
> in the exam what they wanted to hear.

10uH is just the last value I used. I tried bigger values (up to whole
Henries) just to see what happens. LTspice is fun! You can try any
nonsense and it does not smell :-)

> Wood-fire baked pizza tastes beyond compare. We've never looked back,
> haven't baked one in the oven since years because it simply isn't as good.

That I believe unseen.

Unfortunately I'm now away from this thread. I'm going on a cruise
tomorrow :-)
Don't know if it will still be on our news server when I come back.

Joerg, I'm deeply grateful for your help. I feel I have learned a lot in
the past days, not only from you but also from the simulations.

Thanks to everybody who contributed!

Cheers

Robert

Joerg

unread,

Apr 30, 2015, 7:01:26 PM4/30/15

to

On 2015-04-30 5:37 AM, Robert Loos wrote:
> Am 29.04.2015 um 22:40 schrieb Joerg:
>
>>> There is a video on the LT-site:
>>> http://www.linear.com/solutions/4449
>>> that shows how to measure the phase margin. I tried this with the
>>> original circuit, with the feedback C reduced to 200pF and it stays well
>>> above 50 degrees. But the transient simulation oscillates.
>>>
>>
>> .AC is a small signal analysis method and does not work well for power
>> stuff, it has to be measured.
>
> I found a way to simulate it: the good phase margin is just because the
> input is grounded. If you insert a 2V voltage source the plot shows a
> terrible phase and it is obvious that it must oscillate.
>

As I said, it makes no sense to drive this with a micropower amplifier.
It can't do it. Put an OP07 in there and all is well even at 2V.

Trying to drive a big FET with an LT2078 is like trying to pull a 5-ton
trailer up a steep hill with a Fiat 500.

>> The OP07 was just one example and if I go above 120ohms it no longer
>> oscillates even with negative supply. For this kind of application you
>> need an opamp with a more powerful output or have a complementary pair
>> follower. I only used the OP07 because it was in the LTSpice directory.
>> Bottomline this is never going to work well with the LT2078 unless you
>> make it very slow. Why do you want a micropower opamp in there?
>
> Historic reasons. In my first board I used a TLC272 (without thinking
> much about it, just because it was on stock) but the high offset made it
> impossible to set the current to 0.
> Then I was searching for a pin compatible type with rail to rail input
> and low offset that is easy to get and ended up with the LT2078.
> Since I'll have to make a new board anyway, I will now search for a type
> with rail to rail input, low offset and high drive capability that is
> easy to get :-)
>

Coincidentally there is a very similar discussion going on in
de.sci.electronics, subject "Schneller Strompuls / LED-Blitz". Since you
understand German, take a look. Timm is now using the THS4631 for
controlling a big old FET in a fast manner.

>>> What I find also curious is that the oscillation is nearly independent
>>> of the feedback C. You cannot stop is even with arbitrary big C. Not
>>> even the frequency changes.
>>
>> Strange. If I use, for example, 120ohms it is stable until about 0.02uF.
>> At 0.01uF it oscillates. Kind of what I'd expect. Here it is with 0.1uF:
>
> ...
> The gate resistor is the clue. If you plot the .ac analysis (with the 2V
> source) you can see how the phase improves. 220R would be even better.
>

Yeah but at some point you start losing drive. Depends on how fast you
have to swing that gate.

>> The inductance is 10uH, very small. The opamp can change it because they
>> differ in drive capability. That makes a huge difference when driving
>> capacitive or other hard loads. Small signal analysis doesn't help that
>> much with it. For example, the slew rate matters, big time. This is why
>> at university I never really bought the small signal theory they taught
>> us. Full confession: It resulted in me skipping class and just writing
>> in the exam what they wanted to hear.
>
> 10uH is just the last value I used. I tried bigger values (up to whole
> Henries) just to see what happens. LTspice is fun! You can try any
> nonsense and it does not smell :-)
>

But the occasional *PHUT* can be fun as well. When our Skyworks rep was
at my office he said "If I ride my mountain bike and there wasn't a
spectacular crash then it probably was a boring ride".

>> Wood-fire baked pizza tastes beyond compare. We've never looked back,
>> haven't baked one in the oven since years because it simply isn't as
>> good.
>
> That I believe unseen.
>
> Unfortunately I'm now away from this thread. I'm going on a cruise
> tomorrow :-)
> Don't know if it will still be on our news server when I come back.
>

Well, if the cruise lasts more than three months, maybe not :-)

Have fun. And easy on that big buffet.

> Joerg, I'm deeply grateful for your help. I feel I have learned a lot in
> the past days, not only from you but also from the simulations.
>
> Thanks to everybody who contributed!
>

As a SW engineer once put it (and the bible does), we are here to serve :-)

Linear Mode of IGBT FSOA/hotspots? - feasable?

Klaus Kragelund

Tim Williams

Joerg

Klaus Kragelund

Joerg

Jim Thompson

Jim Thompson

Joerg

Jon Elson

Klaus Kragelund

Klaus Kragelund

Jim Thompson

Tim Williams

mrob...@att.net

Phil Allison

Klaus Bahner

Winfield Hill

M Philbrook

Robert Loos

Joerg

Robert Loos

Chris Jones

Winfield Hill

Joerg

Robert Loos

Joerg

Robert Loos

Tim Williams

Winfield Hill

Robert Loos

Robert Loos

Joerg

Robert Loos

Joerg