Class AB common source mosfet power amplifier

[John Devereux]

Hi,

OK, I have been simulating away but am now all spiced out!

I am trying to come up with a linear (i.e. not switched) power mosfet
amplifier that works with the output device drains tied together. This
is mainly so they can be bolted to the same heatsink without using
insulators, although an extended voltage swing would also be good.

I.e. generally like this:

.. V+ --------------------------------------------
.. |
.. [R] (optional)
.. |
.. |<
.. ------|| P-Channel
.. |-
.. Drive |--------------------[LOAD]--->GND
.. |-
.. ------|| N-Channel
.. |>
.. |
.. |
.. [R] (optional)
.. |
.. V- --------------------------------------------

Any ideas?

This is not for audio, so don't need super low distortion - but I
don't want any cross-over distortion to be *too* obvious either.

Frequency range is up to a few kHz, voltage rails will be +/- 80V max.

Thanks,

 

[John Larkin]

Hi,

OK, I have been simulating away but am now all spiced out!

I am trying to come up with a linear (i.e. not switched) power mosfet
amplifier that works with the output device drains tied together. This
is mainly so they can be bolted to the same heatsink without using
insulators, although an extended voltage swing would also be good.

I.e. generally like this:

. V+ --------------------------------------------
. |
. [R] (optional)
. |
. |<
. ------|| P-Channel
. |-
. Drive |--------------------[LOAD]--->GND
. |-
. ------|| N-Channel
. |>
. |
. |
. [R] (optional)
. |
. V- --------------------------------------------

Any ideas?

This is not for audio, so don't need super low distortion - but I
don't want any cross-over distortion to be *too* obvious either.

Frequency range is up to a few kHz, voltage rails will be +/- 80V max.

Thanks,

That's the topology I use in my NMR gradient amps, which range from
+-12 volts at 3 amps up to +-180 at 100 amps. I've posted pics of the
big fet/heatsink assembly a couple of times.

I use an opamp per fet (up to 32 of them!) to close a loop on the
voltage drop across the source resistors. That gives very tight
control of each fet's current and lets me use small sense resistors.

The driver has to intelligently split positive and negative drives and
maintain idle currents, and shift the drives up to the power rails.
That's not trivial to do right.

Note that this config is like an ldo regulator: it has a high
open-loop output impedance, so loop dynamics is a lot more sensitive
to loads than a more conventional source-output amp.

John

 

[Vladimir Vassilevsky]

Hi,

OK, I have been simulating away but am now all spiced out!

I am trying to come up with a linear (i.e. not switched) power mosfet
amplifier that works with the output device drains tied together. This
is mainly so they can be bolted to the same heatsink without using
insulators, although an extended voltage swing would also be good.

I.e. generally like this:

. V+ --------------------------------------------
. |
. [R] (optional)
. |
. |<
. ------|| P-Channel
. |-
. Drive |--------------------[LOAD]--->GND
. |-
. ------|| N-Channel
. |>
. |
. |
. [R] (optional)
. |
. V- --------------------------------------------

Any ideas?

This is not for audio, so don't need super low distortion - but I
don't want any cross-over distortion to be *too* obvious either.

Frequency range is up to a few kHz, voltage rails will be +/- 80V max.

This topology can be problematic because of the wild variation of the
gain depending on the load impedance and current. I assume you are
running in the class AB, right? Nevertheless it can be worked out
however you may have to add the heavy correction capacitors between
drains and gates (and maybe some nonlinear elements too it the goal is
the high performance).

But why this topology? It is unlikely that you need R-R operation at
+/-80V. Why not a trivial pnp-npn emitter follower?

BTW, if you can drop the rail voltage to somewhat +/-50V, you can
probably use an audio IC like TDA7294.




Vladimir Vassilevsky
DSP and Mixed Signal Design Consultant
http://www.abvolt.com

 

[John Devereux]

Hi,

OK, I have been simulating away but am now all spiced out!

I am trying to come up with a linear (i.e. not switched) power mosfet
amplifier that works with the output device drains tied together. This
is mainly so they can be bolted to the same heatsink without using
insulators, although an extended voltage swing would also be good.

I.e. generally like this:

. V+ --------------------------------------------
. |
. [R] (optional)
. |
. |<
. ------|| P-Channel
. |-
. Drive |--------------------[LOAD]--->GND
. |-
. ------|| N-Channel
. |>
. |
. |
. [R] (optional)
. |
. V- --------------------------------------------

Any ideas?

This is not for audio, so don't need super low distortion - but I
don't want any cross-over distortion to be *too* obvious either.

Frequency range is up to a few kHz, voltage rails will be +/- 80V max.

This topology can be problematic because of the wild variation of the
gain depending on the load impedance and current. I assume you are
running in the class AB, right?
Right

Nevertheless it can be worked out however you may have to add the
heavy correction capacitors between drains and gates (and maybe some
nonlinear elements too it the goal is the high performance).
OK


But why this topology? It is unlikely that you need R-R operation at
+/-80V. Why not a trivial pnp-npn emitter follower?

I don't *need* rail to rail. As I said, it is mainly so I can bolt the
devices to the same heatsink without insulation which would reduce the
heat transfer. Power mosfets have the drain connected to the tab, so
this implies the topology above.

BTW, if you can drop the rail voltage to somewhat +/-50V, you can
probably use an audio IC like TDA7294.

Not for this - the final thing will have 8 mosfets paralleled for each
shown above, and will also use a copy driven in antiphase to get full
bridge drive. 32 mosfets in all,~10kW pulsed, > 1kW rms!

 

[John Devereux]

Hi,

OK, I have been simulating away but am now all spiced out!

I am trying to come up with a linear (i.e. not switched) power mosfet
amplifier that works with the output device drains tied together. This
is mainly so they can be bolted to the same heatsink without using
insulators, although an extended voltage swing would also be good.

I.e. generally like this:

. V+ --------------------------------------------
. |
. [R] (optional)
. |
. |<
. ------|| P-Channel
. |-
. Drive |--------------------[LOAD]--->GND
. |-
. ------|| N-Channel
. |>
. |
. |
. [R] (optional)
. |
. V- --------------------------------------------

Any ideas?

This is not for audio, so don't need super low distortion - but I
don't want any cross-over distortion to be *too* obvious either.

Frequency range is up to a few kHz, voltage rails will be +/- 80V max.

Thanks,

That's the topology I use in my NMR gradient amps, which range from
+-12 volts at 3 amps up to +-180 at 100 amps. I've posted pics of the
big fet/heatsink assembly a couple of times.

I use an opamp per fet (up to 32 of them!) to close a loop on the
voltage drop across the source resistors. That gives very tight
control of each fet's current and lets me use small sense resistors.

That was my first idea, then I got diverted by all the audio power
amplifier designs out there! You even have the same number of mosfets
(my final version will be a full bridge configuration).

The driver has to intelligently split positive and negative drives and
maintain idle currents, and shift the drives up to the power rails.
That's not trivial to do right.

That's where I started looking for other ideas...

Note that this config is like an ldo regulator: it has a high
open-loop output impedance, so loop dynamics is a lot more sensitive
to loads than a more conventional source-output amp.

Makes sense, thanks.

I do have a common drain version - that simulates OK - but common
source would simplify construction.

 

[Joerg]

Hi,

OK, I have been simulating away but am now all spiced out!

I am trying to come up with a linear (i.e. not switched) power mosfet
amplifier that works with the output device drains tied together. This
is mainly so they can be bolted to the same heatsink without using
insulators, although an extended voltage swing would also be good.

I.e. generally like this:

. V+ --------------------------------------------
. |
. [R] (optional)
. |
. |<
. ------|| P-Channel
. |-
. Drive |--------------------[LOAD]--->GND
. |-
. ------|| N-Channel
. |>
. |
. |
. [R] (optional)
. |
. V- --------------------------------------------

Any ideas?

This is not for audio, so don't need super low distortion - but I
don't want any cross-over distortion to be *too* obvious either.

Frequency range is up to a few kHz, voltage rails will be +/- 80V max.

This topology can be problematic because of the wild variation of the
gain depending on the load impedance and current. I assume you are
running in the class AB, right?
Right

Nevertheless it can be worked out however you may have to add the
heavy correction capacitors between drains and gates (and maybe some
nonlinear elements too it the goal is the high performance).
OK

But why this topology? It is unlikely that you need R-R operation at
+/-80V. Why not a trivial pnp-npn emitter follower?

I don't *need* rail to rail. As I said, it is mainly so I can bolt the
devices to the same heatsink without insulation which would reduce the
heat transfer. Power mosfets have the drain connected to the tab, so
this implies the topology above.

BTW, if you can drop the rail voltage to somewhat +/-50V, you can
probably use an audio IC like TDA7294.

Not for this - the final thing will have 8 mosfets paralleled for each
shown above, and will also use a copy driven in antiphase to get full
bridge drive. 32 mosfets in all,~10kW pulsed, > 1kW rms!

Wow! Who's playing lead guitar during that concert?

 

[John Devereux]

John Devereux wrote:
[...]

Not for this - the final thing will have 8 mosfets paralleled for each
shown above, and will also use a copy driven in antiphase to get full
bridge drive. 32 mosfets in all,~10kW pulsed, > 1kW rms!

Wow! Who's playing lead guitar during that concert?

Yes - should drive a decent size sub-woofer!

Its actually for a vibrator. (Not that sort! A voice-coil actuator for
vibration testing).

 

[Vladimir Vassilevsky]

John Larkin wrote:

I use an opamp per fet (up to 32 of them!) to close a loop on the
voltage drop across the source resistors. That gives very tight
control of each fet's current and lets me use small sense resistors.

So the output stage operates as the current source and the dynamic
response is completely determined by the parameters of the load.
Assuming the unconditional stability, closing the global feedback loop
by the output voltage would be problematic because there is only ~45
degrees of phase available for everything.

The driver has to intelligently split positive and negative drives and
maintain idle currents, and shift the drives up to the power rails.
That's not trivial to do right.

Yes, this is not very trivial as this schematic is prone to the
different kinds of ill behavior.

Note that this config is like an ldo regulator: it has a high
open-loop output impedance, so loop dynamics is a lot more sensitive
to loads than a more conventional source-output amp.

Vladimir Vassilevsky
DSP and Mixed Signal Design Consultant
http://www.abvolt.com

 

[John Larkin]

Hi,

OK, I have been simulating away but am now all spiced out!

I am trying to come up with a linear (i.e. not switched) power mosfet
amplifier that works with the output device drains tied together. This
is mainly so they can be bolted to the same heatsink without using
insulators, although an extended voltage swing would also be good.

I.e. generally like this:

. V+ --------------------------------------------
. |
. [R] (optional)
. |
. |<
. ------|| P-Channel
. |-
. Drive |--------------------[LOAD]--->GND
. |-
. ------|| N-Channel
. |>
. |
. |
. [R] (optional)
. |
. V- --------------------------------------------

Any ideas?

This is not for audio, so don't need super low distortion - but I
don't want any cross-over distortion to be *too* obvious either.

Frequency range is up to a few kHz, voltage rails will be +/- 80V max.

This topology can be problematic because of the wild variation of the
gain depending on the load impedance and current. I assume you are
running in the class AB, right?
Right

Nevertheless it can be worked out however you may have to add the
heavy correction capacitors between drains and gates (and maybe some
nonlinear elements too it the goal is the high performance).
OK


But why this topology? It is unlikely that you need R-R operation at
+/-80V. Why not a trivial pnp-npn emitter follower?

I don't *need* rail to rail. As I said, it is mainly so I can bolt the
devices to the same heatsink without insulation which would reduce the
heat transfer. Power mosfets have the drain connected to the tab, so
this implies the topology above.

BTW, if you can drop the rail voltage to somewhat +/-50V, you can
probably use an audio IC like TDA7294.

Not for this - the final thing will have 8 mosfets paralleled for each
shown above, and will also use a copy driven in antiphase to get full
bridge drive. 32 mosfets in all,~10kW pulsed, > 1kW rms!

One other note: mosfets spec'd for switching use have a bad habit of
exploding when used in linear mode. We tested a lot of "300 watt"
power fets, bolted to a solid copper block, and most of them couldn't
dissipate 300 watts for 100 milliseconds when running at higher drain
voltages. Imagine loud noises. It has to do with nonuniform heating or
something. IXYS has some fets specially designed to not explode in
this mode.

So test your candidate fets to destruction to make sure you have
margin! Or go switchmode.

John

 

[John Larkin]

John Larkin wrote:



So the output stage operates as the current source and the dynamic
response is completely determined by the parameters of the load.
Assuming the unconditional stability, closing the global feedback loop
by the output voltage would be problematic because there is only ~45
degrees of phase available for everything.

We're driving an NMR gradient coil with a current source, and feeding
back on sensed current. So having a high-z output stage makes us
nearly insensitive to load impedance. So use a low-z stage if you're
feeding back on voltage, and a high-z stahe if you feed back on
current.

Yes, this is not very trivial as this schematic is prone to the
different kinds of ill behavior.

Well, we did do it right.

John

 

[John Devereux]

[...]

I don't *need* rail to rail. As I said, it is mainly so I can bolt the
devices to the same heatsink without insulation which would reduce the
heat transfer. Power mosfets have the drain connected to the tab, so
this implies the topology above.



Not for this - the final thing will have 8 mosfets paralleled for each
shown above, and will also use a copy driven in antiphase to get full
bridge drive. 32 mosfets in all,~10kW pulsed, > 1kW rms!

One other note: mosfets spec'd for switching use have a bad habit of
exploding when used in linear mode. We tested a lot of "300 watt"
power fets, bolted to a solid copper block, and most of them couldn't
dissipate 300 watts for 100 milliseconds when running at higher drain
voltages. Imagine loud noises. It has to do with nonuniform heating or
something. IXYS has some fets specially designed to not explode in
this mode.

The datasheets do show SOA for various duration pulses, which I will
respect.

So test your candidate fets to destruction to make sure you have
margin!

Will do.

Or go switchmode.

That seemed even more complicated (and I do not need to save power).

 

[Winfield Hill]

OK, I have been simulating away but am now all spiced out!

I am trying to come up with a linear (i.e. not switched) power mosfet
amplifier that works with the output device drains tied together. This
is mainly so they can be bolted to the same heatsink without using
insulators, although an extended voltage swing would also be good.

I.e. generally like this:

. V+ --------------------------------------------
. |
. [R] (optional)
. |
. |<
. ------|| P-Channel
. |-
. Drive |--------------------[LOAD]--->GND
. |-
. ------|| N-Channel
. |>
. |
. |
. [R] (optional)
. |
. V- --------------------------------------------

This is not for audio, so don't need super low distortion - but I
don't want any cross-over distortion to be *too* obvious either.

Frequency range is up to a few kHz, voltage rails will be +/- 80V max.

This topology can be problematic because of the wild variation of the
gain depending on the load impedance and current. I assume you are
running in the class AB, right? Nevertheless it can be worked out
however you may have to add the heavy correction capacitors between
drains and gates (and maybe some nonlinear elements too it the goal is
the high performance).

See ftp://ftp.rowland.org/pub/hill/ris-496-1.pdf

Actually, properly done, this can be a very stable
amplifier configuration. Here's an example of
such an amplifier I designed, with a little help
from Tony Williams, R.I.P. It has been discussed
here on s.e.d. several times, including when I
was doing the design. It uses BJTs instead of
MOSFETs (which I came to regret, SOA, ouch!),
but you may find it useful to study anyway.

 

[Phil Allison]

"John Devereux"

Hi,

OK, I have been simulating away but am now all spiced out!

I am trying to come up with a linear (i.e. not switched) power mosfet
amplifier that works with the output device drains tied together. This
is mainly so they can be bolted to the same heatsink without using
insulators, although an extended voltage swing would also be good.

I.e. generally like this:

. V+ --------------------------------------------
. |
. [R] (optional)
. |
. |<
. ------|| P-Channel
. |-
. Drive |--------------------[LOAD]--->GND
. |-
. ------|| N-Channel
. |>
. |
. |
. [R] (optional)
. |
. V- --------------------------------------------

Any ideas?

This is not for audio, so don't need super low distortion - but I
don't want any cross-over distortion to be *too* obvious either.

Frequency range is up to a few kHz, voltage rails will be +/- 80V max.

** For heaven's sake, forget stupid switching fets and use complementary
LATERAL mosfets for linear apps.

The case is the source, they operate in parallel like a dream, are not prone
to thermal runaway and inherently protect themselves if the chip goes
overtemp.

Egs:

http://www.magnatec-uk.co.uk/pdf/magnatec/BUZ905D.pdf

http://www.magnatec-uk.co.uk/pdf/magnatec/BUZ900D.pdf


These are 16A, 250W dual chip devices.

Same numbers without the "D" are single chip versions.



....... Phil

 

[John Devereux]

"John Devereux"

Hi,

OK, I have been simulating away but am now all spiced out!

I am trying to come up with a linear (i.e. not switched) power mosfet
amplifier that works with the output device drains tied together. This
is mainly so they can be bolted to the same heatsink without using
insulators, although an extended voltage swing would also be good.

I.e. generally like this:

. V+ --------------------------------------------
. |
. [R] (optional)
. |
. |<
. ------|| P-Channel
. |-
. Drive |--------------------[LOAD]--->GND
. |-
. ------|| N-Channel
. |>
. |
. |
. [R] (optional)
. |
. V- --------------------------------------------

Any ideas?

This is not for audio, so don't need super low distortion - but I
don't want any cross-over distortion to be *too* obvious either.

Frequency range is up to a few kHz, voltage rails will be +/- 80V max.

** For heaven's sake, forget stupid switching fets and use complementary
LATERAL mosfets for linear apps.

They seem quite rare, likely expensive, and I was not too worried
about distortion.

The case is the source, they operate in parallel like a dream, are not prone
to thermal runaway and inherently protect themselves if the chip goes
overtemp.

....but you make a good case!

Egs:

http://www.magnatec-uk.co.uk/pdf/magnatec/BUZ905D.pdf

http://www.magnatec-uk.co.uk/pdf/magnatec/BUZ900D.pdf


These are 16A, 250W dual chip devices.

Same numbers without the "D" are single chip versions.

Thanks Phil - I had not realised you could get anything with case
connected to source. I will investigate these tomorrow.

 

[John Devereux]

OK, I have been simulating away but am now all spiced out!

I am trying to come up with a linear (i.e. not switched) power mosfet
amplifier that works with the output device drains tied together. This
is mainly so they can be bolted to the same heatsink without using
insulators, although an extended voltage swing would also be good.

I.e. generally like this:

. V+ --------------------------------------------
. |
. [R] (optional)
. |
. |<
. ------|| P-Channel
. |-
. Drive |--------------------[LOAD]--->GND
. |-
. ------|| N-Channel
. |>
. |
. |
. [R] (optional)
. |
. V- --------------------------------------------

This is not for audio, so don't need super low distortion - but I
don't want any cross-over distortion to be *too* obvious either.

Frequency range is up to a few kHz, voltage rails will be +/- 80V max.

This topology can be problematic because of the wild variation of the
gain depending on the load impedance and current. I assume you are
running in the class AB, right? Nevertheless it can be worked out
however you may have to add the heavy correction capacitors between
drains and gates (and maybe some nonlinear elements too it the goal is
the high performance).

See ftp://ftp.rowland.org/pub/hill/ris-496-1.pdf

Actually, properly done, this can be a very stable
amplifier configuration. Here's an example of
such an amplifier I designed, with a little help
from Tony Williams, R.I.P. It has been discussed
here on s.e.d. several times, including when I
was doing the design. It uses BJTs instead of
MOSFETs (which I came to regret, SOA, ouch!),
but you may find it useful to study anyway.

Wow, thanks, I will study it. Am I right in saying this is a variant
of the circuit warned against here (fig 16)?

<http://sound.westhost.com/amp_design.htm#nogood>.

(Hey, it's like one of your "bad circuits" sections!)

 

[Phil Allison]

"John Devereux"
"Phil Allison"

They seem quite rare, likely expensive, and I was not too worried
about distortion.


...but you make a good case!


Thanks Phil - I had not realised you could get anything with case
connected to source. I will investigate these tomorrow.

** Farnell have them.




....... Phil

 

[John Larkin]

OK, I have been simulating away but am now all spiced out!

I am trying to come up with a linear (i.e. not switched) power mosfet
amplifier that works with the output device drains tied together. This
is mainly so they can be bolted to the same heatsink without using
insulators, although an extended voltage swing would also be good.

I.e. generally like this:

. V+ --------------------------------------------
. |
. [R] (optional)
. |
. |<
. ------|| P-Channel
. |-
. Drive |--------------------[LOAD]--->GND
. |-
. ------|| N-Channel
. |>
. |
. |
. [R] (optional)
. |
. V- --------------------------------------------

This is not for audio, so don't need super low distortion - but I
don't want any cross-over distortion to be *too* obvious either.

Frequency range is up to a few kHz, voltage rails will be +/- 80V max.

This topology can be problematic because of the wild variation of the
gain depending on the load impedance and current. I assume you are
running in the class AB, right? Nevertheless it can be worked out
however you may have to add the heavy correction capacitors between
drains and gates (and maybe some nonlinear elements too it the goal is
the high performance).

See ftp://ftp.rowland.org/pub/hill/ris-496-1.pdf

Doggone, U6, with its power pins cascoded into the complementary HV
transistors, is hauntingly familiar. Not to mention D3 and D4, except
that I like to add LED's there, too.

John

 

[Tim Williams]

That seemed even more complicated (and I do not need to save power).

LOL. So, I'm reading this thread, and the guy is asking if he can solve
this problem where the transistors all have to be mounted on the same
heatsink (or two). And then another guy chimes in with a solution that
completely removes the problem of heat, and the guy shoots it down...

;-)

C'mon...switching ain't all that hard...hell, I did it with tubes!
http://webpages.charter.net/dawill/tmoranwms/Elec_Compound.html
Rak(on) ~= 100 ohms on that 38HE7.

Tim

 

[D from BC]

Hi,

OK, I have been simulating away but am now all spiced out!

I am trying to come up with a linear (i.e. not switched) power mosfet
amplifier that works with the output device drains tied together. This
is mainly so they can be bolted to the same heatsink without using
insulators, although an extended voltage swing would also be good.

I.e. generally like this:

. V+ --------------------------------------------
. |
. [R] (optional)
. |
. |<
. ------|| P-Channel
. |-
. Drive |--------------------[LOAD]--->GND
. |-
. ------|| N-Channel
. |>
. |
. |
. [R] (optional)
. |
. V- --------------------------------------------

Any ideas?

This is not for audio, so don't need super low distortion - but I
don't want any cross-over distortion to be *too* obvious either.

Frequency range is up to a few kHz, voltage rails will be +/- 80V max.

Thanks,

I found an old schematic I created in 1997.
I quickly drew it in Ltspice..
http://www.members.shaw.ca/chainsaw/SED/somoldamp.jpg
351Kbyte

Don't ask about the output network.. I've completely forgotten why I
did that and it's looks weird to me now..


D from BC

 

[Winfield]

Wow, thanks, I will study it. Am I right in saying this is
a variant of the circuit warned against here (fig 16)?
<http://sound.westhost.com/amp_design.htm#nogood>.
(Hey, it's like one of your "bad circuits" sections!)

Don't be silly. First, in reality the configurations are
totally different. Second, do the calculations and see the
damaging effect in OUT1 of a too-small emitter resistor.
My output stages are mirrors - do you see any mirrors in
either OUT1 or OUT2 --- no I didn't think so. That's OK,
I knew you were pulling my leg.

Go read the circuit discussions in s.e.d.