Using reverse biased zeners to correct for mosfett turnon voltagein a push pull stage

[Adrian Nievergelt]

Hi

I'm trying to use mosfets for a push pull current gain stage on a dc
coupled amplifier with opamp feedback. The feedback is supposed to
operate in the high MHz range, so I'm looking for ways to statically
reduce crossover distortion even before feedback to alleviate the load
on the opamp and get a better THD. After looking around for a way which
does not involve a lot of components (and with that a lot of parasitic
capacitances) I believe using zener diodes in reverse bias might be
"cleaner" than stacking multiple diodes in forward bias to get a level
shift.

Schematic: https://www.circuitlab.com/circuit/g64s7f/pushpullzener/

I haven't seen anyone else doing it this way and I suspect there to be a
reason. Where might I run into problems with this scheme? Is there an
easier (high frequency compatible) version I might have overlooked?

Thanks
Adrian

 

[mike]

Hi

I'm trying to use mosfets for a push pull current gain stage on a dc
coupled amplifier with opamp feedback. The feedback is supposed to
operate in the high MHz range, so I'm looking for ways to statically
reduce crossover distortion even before feedback to alleviate the load
on the opamp and get a better THD. After looking around for a way which
does not involve a lot of components (and with that a lot of parasitic
capacitances) I believe using zener diodes in reverse bias might be
"cleaner" than stacking multiple diodes in forward bias to get a level
shift.

Schematic: https://www.circuitlab.com/circuit/g64s7f/pushpullzener/

I haven't seen anyone else doing it this way and I suspect there to be a
reason. Where might I run into problems with this scheme? Is there an
easier (high frequency compatible) version I might have overlooked?

Thanks
Adrian

You just built a smoke generator.
You have no way to compensate for component variations or for temperature.
You'll have similar problems with series diodes.
Typical solution for bipolars would be emitter resistors.
It's more complicated with fets because of the variations in threshold
voltage.
You're describing a voltage amplifier circuit using the term "current
gain stage".
How many is "high" MHz.?
THD is certainly a valid term, if you put a number on "better", but it
may obscure your objective.
What are you trying to accomplish? Is your load always purely resistive
100 ohms?
The devil is in the details.

 

[legg]

Hi

I'm trying to use mosfets for a push pull current gain stage on a dc
coupled amplifier with opamp feedback. The feedback is supposed to
operate in the high MHz range, so I'm looking for ways to statically
reduce crossover distortion even before feedback to alleviate the load
on the opamp and get a better THD. After looking around for a way which
does not involve a lot of components (and with that a lot of parasitic
capacitances) I believe using zener diodes in reverse bias might be
"cleaner" than stacking multiple diodes in forward bias to get a level
shift.

Schematic: https://www.circuitlab.com/circuit/g64s7f/pushpullzener/

I haven't seen anyone else doing it this way and I suspect there to be a
reason. Where might I run into problems with this scheme? Is there an
easier (high frequency compatible) version I might have overlooked?

Thanks
Adrian

The dominant 'stray' capacitances will be Cgs and Cdg, regardless of
the number of parts you hang onto the gates, whether the power device
being controlled ia a mosfet or an IGBT (as your schematic actually
indicates).

With a very fast op amp and no local compensation, you'll run into the
situation where the Op amp will try to charge the gate capacitances
through the zener body diode, something that will not happen with a
diode string.

Cross-over distortion is not a 'static' characteristic; I'm guessing
you're just looking for a linear transfer and stable DC bias. Unless
you know why the first technique was used (with all those 'extra'
components), it's not really possible to speculate on the benefits or
advantages of a modification.

RL

 

[Joerg]

Hi

I'm trying to use mosfets for a push pull current gain stage on a dc
coupled amplifier with opamp feedback. The feedback is supposed to
operate in the high MHz range, so I'm looking for ways to statically
reduce crossover distortion even before feedback to alleviate the load
on the opamp and get a better THD. After looking around for a way which
does not involve a lot of components (and with that a lot of parasitic
capacitances) I believe using zener diodes in reverse bias might be
"cleaner" than stacking multiple diodes in forward bias to get a level
shift.

Schematic: https://www.circuitlab.com/circuit/g64s7f/pushpullzener/

I haven't seen anyone else doing it this way and I suspect there to be a
reason. Where might I run into problems with this scheme? Is there an
easier (high frequency compatible) version I might have overlooked?

As others have mention it's better done with BJT. FETs have way too
large a production tolerance in the thresholds. It'll either distort
badly or go phssst ... *PHUT* because of excessive quiescent current.

Another common trick of the trade is to have a resistor between opamp
output and load, to (somewhat) tide over the dead zone.

 

[Adrian Nievergelt]

You just built a smoke generator.
You have no way to compensate for component variations or for temperature.
You'll have similar problems with series diodes.
Typical solution for bipolars would be emitter resistors.
It's more complicated with fets because of the variations in threshold
voltage.
You're describing a voltage amplifier circuit using the term "current
gain stage".
How many is "high" MHz.?
THD is certainly a valid term, if you put a number on "better", but it
may obscure your objective.
What are you trying to accomplish? Is your load always purely resistive
100 ohms?
The devil is in the details.

I was afraid it would be something like this. I'll try to elaborate
some: With current gain stage I mean the push pull output, as it's
ideally a follower. The speed question is more difficult. The load for
the amp is a piezoceramic with a capaciance of around 90nF. The resistor
in the schematic was just for the purpose of my simulation. I use a
series resistor with the piezo to decouple the reactive load from the
piezo. Now the overall bandwidth should excell 1MHz piezo drive. In most
cases the output will not oscillate, but follow another controllers
signal. To achieve 1MHz output however I've found (simulation and board
testing) that i need quite fast opamps (400MHz GBP up) so that i can
afterwards silence the oscillation of the whole amplifier with
capacitive load with the series resistance. With deadbugging I've noted
that the THD with unbiased fets is around -35dB at 10kHz, which i would
like to reduce with biasing.

I hope that clears it up some.

 

[Adrian Nievergelt]

As Mike pointed out, this circuit's bias is unstable. Also, you'd need
very low voltage zeners for most recent MOSFETs. In addition, you'll
need a very high voltage op amp to make use of those +-30V supplies.

With 10k resistors on the gates, the MOSFET capacitance will prevent the
gate following your op amp very fast in the direction towards drain
potential, so if you really try to go at megahertz rates, you'll see a
lot of distortion due to slew rate limiting.

One cute method is to put resistors in series with the op amp's supply
pins, and drive the gates from there. (The sources go to the supplies,
i.e. the PMOS device goes on the high side.) In this case, you could
put the zeners in series with the supply pins, so you don't blow up the
op amp. A resistor to ground from the op amp output, plus one from the
op amp output to the booster output (the FET drains) gives you some
local feedback that makes it easier to frequency-compensate the whole
thing.

(Source resistors on the FETs will help stabilize the bias, either way.)

I wouldn't use the fancy method for a general-purpose amplifier, though,
because its quiescent current isn't too well controlled and because it
tends to be vulnerable to reactive loads, which can make it oscillate or
blow it up.

Cheers

Phil Hobbs

The opamp i use is in a suspended rail configuration, so up to +-100V
output swings are no problem.
I don't think i understand your suggestion yet, but it sounds
complicated to do as my supplied to the opamp are already floating.
Since my load is inherently purely reactive (dampened some by the
addition of a series resistance) the blowing up is always a concern.

 

[Adrian Nievergelt]

As others have mention it's better done with BJT. FETs have way too
large a production tolerance in the thresholds. It'll either distort
badly or go phssst ... *PHUT* because of excessive quiescent current.

Another common trick of the trade is to have a resistor between opamp
output and load, to (somewhat) tide over the dead zone.

I initially used BJTs as output instead of the mosfets, but as the
output load can spike to 6-10A easily at full speed and full swing i've
found that I achieve higher bandwidths with a fet output, at the cost of
some headroom and unfortunately also some crossover distortion.

 

[Joerg]

I was afraid it would be something like this. I'll try to elaborate
some: With current gain stage I mean the push pull output, as it's
ideally a follower. The speed question is more difficult. The load for
the amp is a piezoceramic with a capaciance of around 90nF. The resistor
in the schematic was just for the purpose of my simulation. I use a
series resistor with the piezo to decouple the reactive load from the
piezo. Now the overall bandwidth should excell 1MHz piezo drive. In most
cases the output will not oscillate, but follow another controllers
signal. To achieve 1MHz output however I've found (simulation and board
testing) that i need quite fast opamps (400MHz GBP up) so that i can
afterwards silence the oscillation of the whole amplifier with
capacitive load with the series resistance. With deadbugging I've noted
that the THD with unbiased fets is around -35dB at 10kHz, which i would
like to reduce with biasing.

I hope that clears it up some.

That is what I'd call a "serious load". You'll probably have to
pre-bias. Assuming that you transmit ultrasound pulse trains you should
have that running in a servo in order to mitigate temperature drift.

Then I'd add some sort of fast current sense so you can detect an
overload before molten solder and TO220 pieces splatter about. Or maybe
even a simple SWR-bridge so you can detect if the transducer loses
coupling to the medium, to avoid putting a big crack into the piezo.

Question: Why do you need to drive a tranducer with a linear amp? Even
PZT-5H with a good backing material isn't more than 50% in BW.

 

[Adrian Nievergelt]

That is what I'd call a "serious load". You'll probably have to
pre-bias. Assuming that you transmit ultrasound pulse trains you should
have that running in a servo in order to mitigate temperature drift.

Then I'd add some sort of fast current sense so you can detect an
overload before molten solder and TO220 pieces splatter about. Or maybe
even a simple SWR-bridge so you can detect if the transducer loses
coupling to the medium, to avoid putting a big crack into the piezo.

Question: Why do you need to drive a tranducer with a linear amp? Even
PZT-5H with a good backing material isn't more than 50% in BW.

Ideally E'd run it in a charge amplifying configuration to even
linearize the piezo motion which has some hysteresis. The reason for
using a linear amp is that I'm not interested in transmitting power, but
for the use as a nanopositioning device. It's a part of a homebuilt
atomic force microscope that is supposed to allow for at least 100kHz,
ideally even up to 500kHz topography feedback. I can already drive the
load with the version of the amplifier that I already have, however i
stagger around 300kHz and I'm currently redesigning it to allow for
ideally >1MHz operation.
I am fully well aware that driving a 100nF load at MHz is a more than
serious load and I know that doing full swings at that frequency is
utopic, but with 6-10A output transistors it's possible. The only thing
I'm as mentioned stuggling with is getting dc precision, bandwidth and
signal integrity to a compromise.
The actual integrated up power load on the amp is not that staggering,
but the peak loads are a bit extreme.

 

[Adrian Nievergelt]

Take a look at TCA0372. It's a 1 MHz, dual, 1 amp opamp, good for +-20
volt rails. Use both sections to bridge-drive both ends of the piezo
for more voltage swing, or run the sections in parallel with ballast
resistors if you prefer more current. Costs 50 cents.

I need +-60V upwards unfortunately, but cost is really not an issue, i
can spend as much as i want as long as i get it to work.

 

[Joerg]

I need +-60V upwards unfortunately, but cost is really not an issue, i
can spend as much as i want as long as i get it to work.

As much as you want? Now that is a nice situation to be in

Well, then ...

http://www.cirrus.com/en/pubs/proDatasheet/PA96U_G.pdf

Digikey had them in stock. But sit down before looking at the price. Of
course, the usual precautions for capacitive loading apply but that is
the case for any amplifier.

 

[Adrian Nievergelt]

That makes it a lot easier. How about a Cirrus (nee Apex) PA85?
http://www.cirrus.com/en/products/pa85.html

Cheers

Phil Hobbs

Yeees, that's just the thing. They're not fast enough. That's where I
started out, using www.cirrus.com/en/products/pa107.html, but with them
i never got the noise as far down as with my suspended rail highspeed
opamp. Also i've had trouble getting more than 200kHz closed loop
bandwith out of them. That's the reason i started building my own
amplifier in the first place.

 

[Joerg]

Yeees, that's just the thing. They're not fast enough. That's where I
started out, using www.cirrus.com/en/products/pa107.html, but with them
i never got the noise as far down as with my suspended rail highspeed
opamp. Also i've had trouble getting more than 200kHz closed loop
bandwith out of them. That's the reason i started building my own
amplifier in the first place.

Well, then the path looks pretty clear to me. You have to refine your
own design. Maybe start with stabilizing the bias.

Can you stop the scan of the microscope once in a while when it hapens
to be somewhere in the middle, or let it run towards the middle,
auto-correct the bias and then continue the scan? Essentially similar to
how blacklevel-clamping operated in the good old analog TV days. The
days when TV actually worked all the time.

 

[Adrian Nievergelt]

Well, then the path looks pretty clear to me. You have to refine your
own design. Maybe start with stabilizing the bias.

Can you stop the scan of the microscope once in a while when it hapens
to be somewhere in the middle, or let it run towards the middle,
auto-correct the bias and then continue the scan? Essentially similar to
how blacklevel-clamping operated in the good old analog TV days. The
days when TV actually worked all the time.

I think based on the suggestions i got here what i will try is going
with ofsetting at least a part of the bias statically. Since i have the
luxury of just having to get to run individual amps and not mass produce
them i could match the bias to the individual transistors. I might be
able to reduce the crossover distortion enough by forcing the deadband
from 8V down to 1-2V. As transistor offset voltage is linearly dependent
on temperature I might get away with it as long as i don't heat the zeners.
I'll have to think more about if the condition that the opamp, if fast
enough, will charge the transistors itself is not actually something i
want, as it will lift the slew limitation while simultaneously relieving
the capacitive load over the deadband on the opamp. I might be wrong,
but I don't yet see where there fallacy is there, so I'd be glad if
someone points it out before I do it.
The other way to go would maybe be to use a level shifter (bjt with two
resistors and a current source) to bias the push pull.

Thanks for all the suggestions though.

 

[Tim Williams]

But the cost, the cost

Meh, not so terrible:
http://webpages.charter.net/dawill/tmoranwms/Circuits_2010/Power_OTA.png
Just tedious.

All those mirrors would look better in an IC, of course. Add cascodes,
improved mirrors (Wilson, etc.) where desirable.

Simulations of this circuit show it goes great up to 10MHz or so, though the
phase shift is far too great at that frequency to close the loop. Still,
bandwidth in the 100s of kHz is easily achieved, and probably pushing it to
1 meg closed loop is possible. To get that bandwidth at low distortion will
require lots of loop gain somewhere, as the OP mentioned; compensating that
shall be left as an exercise for the end user.

Also, as posted elsewhere, cost is apparently no object

Tim

 

[Joerg]

I think based on the suggestions i got here what i will try is going
with ofsetting at least a part of the bias statically. Since i have the
luxury of just having to get to run individual amps and not mass produce
them i could match the bias to the individual transistors. I might be
able to reduce the crossover distortion enough by forcing the deadband
from 8V down to 1-2V. As transistor offset voltage is linearly dependent
on temperature I might get away with it as long as i don't heat the zeners.
I'll have to think more about if the condition that the opamp, if fast
enough, will charge the transistors itself is not actually something i
want, as it will lift the slew limitation while simultaneously relieving
the capacitive load over the deadband on the opamp. I might be wrong,
but I don't yet see where there fallacy is there, so I'd be glad if
someone points it out before I do it.
The other way to go would maybe be to use a level shifter (bjt with two
resistors and a current source) to bias the push pull.

Thanks for all the suggestions though.

Probably best to post a schematic here showing how you want to build it.
The risk with zeners like in the example you brought is that they charge
the gates fast but the discharge is slow. That does not make for a very
linear operation.

Also, the threshold drifts with temperture and since the FETs are the
parts that will become the hottest that needs to be reckoned with.

Personally, I'd servo it. Since the piezo is a capacitive device it
cannot have a DC part in its current so it should be possible to lock
the quiescent current.

 

[Joerg]

Meh, not so terrible:
http://webpages.charter.net/dawill/tmoranwms/Circuits_2010/Power_OTA.png
Just tedious.

All those mirrors would look better in an IC, of course. Add cascodes,
improved mirrors (Wilson, etc.) where desirable.

Simulations of this circuit show it goes great up to 10MHz or so, though
the phase shift is far too great at that frequency to close the loop.
Still, bandwidth in the 100s of kHz is easily achieved, and probably
pushing it to 1 meg closed loop is possible. To get that bandwidth at
low distortion will require lots of loop gain somewhere, as the OP
mentioned; compensating that shall be left as an exercise for the end
user.

Also, as posted elsewhere, cost is apparently no object

If one is willing to do it in discretes it is not such a cost concern.
After all, back in the days of fast analog scopes they had pretty high
voltage plate deflection amps and those had to be pretty darn stable in
the DC and linearity.

 

[Adrian Nievergelt]

Probably best to post a schematic here showing how you want to build it.
The risk with zeners like in the example you brought is that they charge
the gates fast but the discharge is slow. That does not make for a very
linear operation.

Also, the threshold drifts with temperture and since the FETs are the
parts that will become the hottest that needs to be reckoned with.

Personally, I'd servo it. Since the piezo is a capacitive device it
cannot have a DC part in its current so it should be possible to lock
the quiescent current.

The configuration i would try to build right now would be
https://www.circuitlab.com/circuit/trh76n/susprailamp/
with the zener voltages below the respective threshold voltages.

I don't exactly get what you mean by servo it, care to elaborate? Google
only turns up that servo seems to be the same as buffering.

 

[Tim Williams]

If one is willing to do it in discretes it is not such a cost concern.
After all, back in the days of fast analog scopes they had pretty high
voltage plate deflection amps and those had to be pretty darn stable in
the DC and linearity.

It's funny, back in the 10MHz Tek days, I bet they would've loved having
some of the later TV tubes -- video output amps and such, huge
transconductance (some of them comparable with JFETs in current capacity,
gain and cutoff voltage, at a few times higher terminal voltages). By my
estimate, 7KY6 for example (which is usually pretty cheap even these days)
can do 5MHz bandwidth pretty easily, which means they'd be able to do 10 or
20 with some effort.

But, the price for the cutting edge demands that one cannot wait, and thus
distributed amps and hybrid circuits were used.

But even those caught up, and the newest vertical deflection circuits were
quite involved indeed. Pull ups, pull downs, pulls up and down for the
pullups and pulldowns, etc. Double complementary double balanced and who
knows what else...

....Then they introduced distributed deflection plates and, with MCP, pretty
much finished the history of analog scopes.

Tim

 

[Adrian Nievergelt]

Big fets, and power supplies, and heat sinks are expensive. Opamps and
sip dc/dc converters are cheap.

https://dl.dropbox.com/u/53724080/Circuits/Pretty_Linear.JPG


That removes threshold voltage variations from the situation, has tons
of fast gate drive, is thermally stable, has very low distortion, and
swings all the way to the power rails.

It will need some details, like not blowing out the opamp inputs at
startup, and maybe current limiting, stuff like that.

Uhh... i _really_ like the idea of using low frequency integrators to
correct dynamically for any threshold voltage. The two additional opamps
will introduce a tiny bit of noise, but really nothing to write home
about. This should get nearly completely rid of any THD. I'll try if i
can get this to run in my suspended rail configuration to make it swing
120V+.
Thanks a lot for the suggestion. If you're interested I'll post some
results and actual schematics when i get them.