AMP with inverting output stage mosfet

Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.

fab

Member
Joined 2004
Paid Member
Hi

It seems to me that there is not so many inverting output stage in power amp.
This is an amp I have in mind and in the simulator but NOT built as prototype yet. It is an all symetrical amp with NO hi gain VAS (like SKA amp from amplifierguru) but a combined low gain VAS/output stage with local feedback (to reduce the gain variation due to load impedance variation). Low value of C39 and C40 has not the purpose to increase the input stage gain but to extend the open loop frequency bandwidth a little bit more. Q23 and Q24 are small signal transistors. Q34 and Q35 are medium power transistors. Output Mosfet are lateral ones. The input stage is a "low memory" one.

What is strange is that my simulator (Microcap 6) gives a different current in R44 (14 ma) and R45 (10 ma) due to different voltage across R80 and R81 and I do not know if in real life it will be like this or not?

Any comment on this amp is welcome.

Thanks
 

Attachments

  • amp_fab_novas_0e0.jpg
    amp_fab_novas_0e0.jpg
    97.3 KB · Views: 1,379

fab

Member
Joined 2004
Paid Member
simplified schematics for ease

This is a simplified version.

It is a low feedback design (< 35 db) since the input stage is not bootstrapped at audio frequencies like in the SKA amp. Based on the simulation, the phase shift at 20 KHz is less than 4 degree with input filter removed.

Changing the C39 and C40 to 47uF would increase the Open loop gain a lot.
 

Attachments

  • amp_fab_novas_0e0simplea.jpg
    amp_fab_novas_0e0simplea.jpg
    88.2 KB · Views: 1,132
Re: simplified schematics for ease

fab said:
This is a simplified version.

what I really like is:
separate supply for Input + VAS and Output Stage

Not often we see this
But this is the good way to do it.

John Curl, who is designer of real high end amplifiers,
told me once that he had used this in some of his top class amps.
He is a perfectionist, if anyone! And he knows ....

If we aim for best possible results, it wont cost much to add
at least one separate rectifier bridge + filter capacitors
for the sensitive input/VAS stages.


Even a very small, a few VA only extra transformer, for total separation,
is not a very high cost, compared to the other 'big stuff' in a power amp.
You can spend your money at things that is far less an improvement
than making vital parts in your amplifier circuit
get clean as possible supply!


Your simplified version, still looks a bit complicated to me.
But I am sure it is very good performance.
--------------------------

People, begin to do it like fab. And like lineup.
And like John Curl.
Please, use more often good separate supply for
The Heart of a typical feedback power amplifier - the input stage + VAS.
:cool:
 
fab said:

What is strange is that my simulator (Microcap 6) gives a different current in R44 (14 ma) and R45 (10 ma)
due to different voltage across R80 and R81
and I do not know if in real life it will be like this or not?


this is perfectly normal.
2SK134, N-MOSFET and 2SJ49, P-MOSFET
has different Volt Gate-Source at same current level.

So the voltage across these resistors HAVE TO BE different
to keep the output in balance = 0 volt output.

It is practically impossible to to make N-MOS and P-MOS true complementary.
Even NPN PNP bipolar pairs, are different.
So we end up with two different transistors, 2SK134 + 2SJ49.
Some qualities, data, are the same, but VGS / Current curves
are most often different.

For n-jfet and p-jfet, it seems it is even more difficult to get the P and N transistors alike.


lineup
 
Hi fab,
I've tried this design few years ago and I called it oscilator. If power oscilator is not what you are going to build, then at least you should reduce open loop gain of output stage by increasing value of R80 & R81 (but I didn't use C41 & C42 in feedback, these capacitors should reduce oscilations but there are other problems). It's very hard to tune tempetature compensation. In my opinion R44, R45, R80 and R81 should be the same value and use mosfet for thermal compensation (Q25). It's enough to use 200R in that places so the driver doesn't have to work so hard.
 
I have to partially agree with 'darkone'...
For one, your input stage has a VERY high +-60V power supply, yet it is only going to need a small part of it's maximum voltage swing to fully drive the output stage, because you have set a very high gain for the latter.

Setting a high gain for your BJT-MOS CFP output is a problem here. Darkone has pointed you in the right direction but may have missed the point - the temperature compensation you should be worrying about is not that of the MOSFETs, as these are laterals which will, to a degree, self-compensate, but of the driving BJTs. The problem here is not obvious - putting Q25, 34, 35 on the same heatsink should, in theory, take care of that. The non-obvious part where the practise differs from theory, is that any error in temperature compensation will be amplified by your output stage with gain. Please note that your output stage BJTs only see the difference in current between your top and bottom bank of MOSFETs as something to correct, as this is the only thing that will produce different voltages on R80 and R81. You could have 100A flowing straight from +60 to -60V and your BJTs in the output stage would know nothing about this - this is where you have potential for instability.

You could reduce this potential by including the MOSFET current into the local NFB of the CFP pair - add a resistor between the MOSFET drain + R82/R101 point, and output. This is really how a CFP should be made, because this point is a 'virtual emitter' of sorts for the CFP.
In addition, the efficiency of this feedback is reversely proportional to the gain you set for your output stage. If it was me, I would set it to about 2, that should be more than enough.

One more thing: your input stage is capable of saturating Q34 and Q35 on clipping, to the point where 60V less protection zener voltage will appear on the 4.7 ohm resistor in it's emitter, equating to a collector current of over 10A! What do you think will happen to the zeners, transistors, and the input stage when this happens?
 

fab

Member
Joined 2004
Paid Member
Thanks to all for your comments!

lineup said:


what I really like is:
separate supply for Input + VAS and Output Stage

Not often we see this
But this is the good way to do it.
...
People, begin to do it like fab. And like lineup.
And like John Curl.
Please, use more often good separate supply for
The Heart of a typical feedback power amplifier - the input stage + VAS.
:cool:

I use separate supply for several years now. See my thread at http://www.diyaudio.com/forums/showthread.php?postid=359323#post359323

However, if you look at this design, there is no traditional high gain VAS but a high input stage gain and some additional medium gain in ouptut stage like the SKA amp (amplifierguru) that so many peoples are in love with.


lineup said:


this is perfectly normal.
2SK134, N-MOSFET and 2SJ49, P-MOSFET
has different Volt Gate-Source at same current level.

So the voltage across these resistors HAVE TO BE different
to keep the output in balance = 0 volt output.

It is practically impossible to to make N-MOS and P-MOS true complementary.
Even NPN PNP bipolar pairs, are different.
So we end up with two different transistors, 2SK134 + 2SJ49.
Some qualities, data, are the same, but VGS / Current curves
are most often different.

For n-jfet and p-jfet, it seems it is even more difficult to get the P and N transistors alike.
lineup

OK, that makes sense since even if I use same VGSoff, I operate at higher VGS in linear operation thus maybe the generic model for N and P channels are different in Microcap simulator. So you suggest a potentiometer to adjust voltage?


DarkOne said:
Hi fab,
I've tried this design few years ago and I called it oscilator. If power oscilator is not what you are going to build, then at least you should reduce open loop gain of output stage by increasing value of R80 & R81 (but I didn't use C41 & C42 in feedback, these capacitors should reduce oscilations but there are other problems). It's very hard to tune tempetature compensation. In my opinion R44, R45, R80 and R81 should be the same value and use mosfet for thermal compensation (Q25). It's enough to use 200R in that places so the driver doesn't have to work so hard.

I am also worried of possible oscilation but how come SKA amp and latest Hafler amps (9500, 9505) do not oscillate ?
For the driver resistor since usually these mosfet are biased with about 1v to get 250 ma of current this gives only 13 ma mosfet drive current and I want to reduce the mosfet input capacitance ( I am used to have follower mosfet output stage, is it different in inverting mode?). The drivers are 2SA1380/2SC3502 5W.

ilimzn said:
I have to partially agree with 'darkone'...
For one, your input stage has a VERY high +-60V power supply, yet it is only going to need a small part of it's maximum voltage swing to fully drive the output stage, because you have set a very high gain for the latter.

Setting a high gain for your BJT-MOS CFP output is a problem here. Darkone has pointed you in the right direction but may have missed the point - the temperature compensation you should be worrying about is not that of the MOSFETs, as these are laterals which will, to a degree, self-compensate, but of the driving BJTs. The problem here is not obvious - putting Q25, 34, 35 on the same heatsink should, in theory, take care of that. The non-obvious part where the practise differs from theory, is that any error in temperature compensation will be amplified by your output stage with gain. Please note that your output stage BJTs only see the difference in current between your top and bottom bank of MOSFETs as something to correct, as this is the only thing that will produce different voltages on R80 and R81. You could have 100A flowing straight from +60 to -60V and your BJTs in the output stage would know nothing about this - this is where you have potential for instability.

You could reduce this potential by including the MOSFET current into the local NFB of the CFP pair - add a resistor between the MOSFET drain + R82/R101 point, and output. This is really how a CFP should be made, because this point is a 'virtual emitter' of sorts for the CFP.
In addition, the efficiency of this feedback is reversely proportional to the gain you set for your output stage. If it was me, I would set it to about 2, that should be more than enough.

One more thing: your input stage is capable of saturating Q34 and Q35 on clipping, to the point where 60V less protection zener voltage will appear on the 4.7 ohm resistor in it's emitter, equating to a collector current of over 10A! What do you think will happen to the zeners, transistors, and the input stage when this happens?


OK but If I use the R52 to R55 in the drains of the mosfet instead of the sources will it keep the benefit of local feedback on the mosfets?
Since I have no hi gain VAS, the input stage gain without bootstrapping is not high enough and the Q23 and Q24 voltage range is not so high because of R89 and R90 so this is why I used a gain of about 7 to get full voltage swing. Redusing the input/follower stages power supply voltage would not help this situation. Anyway. I plan to use 53V (regulating the 60V supply without additional transformer, modding a DH-200 amp). According to the simulator to drive 2 ohms load I need a local output gain of more than 6... What can I do?

For the satuarion issue, would 53V (instead of 60) for input/follower stage and R89/R90 would prevent it?
 
Originally posted by lineup

this is perfectly normal.
2SK134, N-MOSFET and 2SJ49, P-MOSFET
has different Volt Gate-Source at same current level.
So the voltage across these resistors HAVE TO BE different
to keep the output in balance = 0 volt output.
------------------------------------------------------

fab said:

OK, that makes sense since even if I use same VGSoff, I operate at higher VGS in linear operation thus maybe the generic model for N and P channels are different in Microcap simulator.

So you suggest a potentiometer to adjust voltage?

>potentiometer to adjust voltage

you should do nothing!!!
.. as i said, this is normal in symmetrical amplifiers

(which are often less symmetrical
than non-symmetrical amplifiers
due to non-complementary, so called 'complementary', devices )

if this voltage across resistors VGS is NOT different, R44 and R45
you will get DC-offset ( not 0V at output )
because at same positive current 2SJ49 and negative current 2SK134
the output current across load is ZERO current

now P-MOS need a different voltage across R44
than N-MOS needs voltage across R45
to output same current
-- because they have different Voltage VGS per Current through transistor

the same proportional voltage difference will be across R80 and R81
as across R44 and R45


hope you did get it now
Again -
Perfectly normal to have different voltage across R80/81 R44/45
Do nothing!



lineup
 
fab said:
Thanks to all for your comments!

I use separate supply for several years now. See my thread at http://www.diyaudio.com/forums/showthread.php?postid=359323#post359323

However, if you look at this design,
there is no traditional high gain VAS
but a high input stage gain and some additional medium gain in ouptut stage

Yes, I noticed the difference. Of gain in output stage, too.
This is a what these drain output do. Voltage gain.
This voltage gain is controlled, reduced, by a local feedback in output stage.

These amplifiers can get near to Rail output swing.
Even if driver stage has got same voltage supply level.
Now when using separate supply for stages
we are not restricted to use same voltage everywhere.
Which can be a good thing
.. for example in traditional amps, where output is taken from emitters.
Then VAS stage can have +10 voltage, for example
to allow more headroom for VAS to operate in.

--------------------------------


I did look at your earlier project - The Hafler Topic.
It is nice see someone has understood the benefits of separate supply for input/VAS and output.

There is in fact normally NO GOOD REASON, regarding amp quality, to use only one supply.

Face it .. many times, is only 1 resistor + 1 small signal transistor
in a normal CCS of LTP pair
that separates to the Output Heavy current rail of Output stage from input pair.
In the other rail there is only 1 resistor between!!! Many times.
The output stage supply.
With all its ripples and hum and variations in voltage level.

--------------------------------


What does it take someone to create a separate supply for input?


this it takes:
- ( eventual one extra little transformer, a couple of VA )
- 2 (or 4 )rectifier diodes, for example 1N4004, 1 Ampere
- 2 smaller electrolytic capacitors with suitable voltage rating
- 2 extra supply wires - or rails

Can you afford this?
I can.


Regards
lineup
 

fab

Member
Joined 2004
Paid Member
QUOTE]Originally posted by lineup
Originally posted by lineup
...

>potentiometer to adjust voltage

you should do nothing!!!
.....
hope you did get it now
Again -
Perfectly normal to have different voltage across R80/81 R44/45
Do nothing!

lineup
[/QUOTE]

Hi lineup

OK, I understand but my problem is the DC simulation that provides different currents for upper and lower drivers (V across R44 and R45). Some DC current seems to flow through R82 and R101 local feedback resistors. Maybe in real life it will not be the case....? I know I have seen an AMP (GrandMOS) (inverting gain with local feedback) with a pot but I will not use it for now do not worry...

_______________________________________

Getting back to the stability concerns, I will adjust the voltage swing of earlier stages to allow only a gain of 2 in the output section if I succeed. Have in mind that since I use lateral mosfet their gain ie limited to about [0.7-1.4] * load impedance,
as opposed to Vertical mosfet which has a much higher tranconductance (but a higher capacitance variation too). This is why I needed to have a high gain for Q34 and Q35. Again, this is still simulation only.

One intent is to have a good isolation between the main gain stage (input stage in this design) and the output stage, thus the folloer intermediate stage like in the SKA amp.

The intent of the local feedback is to reduce the open loop gain modulation against the load impedance dynamic variation (my speaker - rated 6 ohms nominal - has a load change from 3 to 18 ohms depending of frequency).

Another advantage is the use of separate supplies for driver and output stages respectively where in the SKA amp it is NOT possible.

I know that SKA and Hafler do not use CFP in output stage with mosfet but I know at least 3 designs that do it although they may not be commercially produced... except Grandmos amp.

I rather have the blessing of some peoples on this forum before building it.

Thanks again but other comments are welcome too.
 

fab

Member
Joined 2004
Paid Member
Modified to rev 0E2

I have followed the ilimzn and Darkone advise for the output stage as per attached simplified revised schematics. I did not include the mosfet for thermal compensation (what types N and P do you use for the VGS multiplier which will be a CFP also in the possible final design).

Please let me know the possible flaws or improvements or give me your "blessing" (like Darkfenriz....).
 

Attachments

  • amp_fab_novas_0e2simple.jpg
    amp_fab_novas_0e2simple.jpg
    92.3 KB · Views: 494
Hi fab!
Is this a design with CFP output ?
This schematic differs others design in using common collector VAS
I read thread "Bob Cordell interview..." and I were know that John Curl is a person which like low feedback design.
fab! How is distortion of your circuit ?
Bye!

fab! you try to use PNG or GIF format instead of JPEG. If content of image is about schematic, figure--> using PNG or GIf will give us a better result. If the content is about "space" such as portrait, lanscape, nature, .... JPEG will be the best choice
 

fab

Member
Joined 2004
Paid Member
Hi Thanh!

yes it is a CFP output with Mosfet.
It uses a buffer stage at the usual place of the VAS like in the SKA design from Amplifierguru. Mainly the voltage gain is in the input stage. For low feedback purpose the input stage is not bootstrapped as opposed the SKA amp. However, is low feedback design compatible with CFP output with mosfet?

I have no means to measure THD with my simulator (Microcap 6). Anyway, is THD estimate accurate (simulator)?

Since I have read this thread :http://www.diyaudio.com/forums/showthread.php?postid=949071#post949071 so I am not that confident anymore of the success. I need some cheep up, please ....and more important some wise advise!

Thanks for the tip on formats for saving image files.
 
at first i wondered why using 2SK134 / 2SJ49
I thought they were older mosfets

but if you have a look at my Attachment
you see 2SK134 / 2SJ49 have same data as 2SK1057 / 2SJ161
seems like they are same transistors, basically

I find this overview of 2SK/2SJ pairs, from HITACHI very good.
You can see the approximate capacitance, data for BOTH N- and P-device.

missing are the 2SK1530 / 2SJ201 pair ( 2SK1529 / 2SJ200 )
which I think is more powerful at the expense of a bit more capacitance

Toshiba datasheets for 2SK1530 and 2SJ201 at this page


lineup
 

Attachments

  • hitachi-sk-sj-overview-all.png
    hitachi-sk-sj-overview-all.png
    56 KB · Views: 376

fab

Member
Joined 2004
Paid Member
lineup said:
at first i wondered why using 2SK134 / 2SJ49
I thought they were older mosfets

but if you have a look at my Attachment
you see 2SK134 / 2SJ49 have same data as 2SK1057 / 2SJ161
seems like they are same transistors, basically

I find this overview of 2SK/2SJ pairs, from HITACHI very good.
You can see the approximate capacitance, data for BOTH N- and P-device.
lineup

yes 2SK134 / 2SJ49 are old TO-3 version that have been replaced by 2SK1058 / 2SJ162 in TO-3P package.

Thanks for this table for quick reference.
What do you suggest for VGS multiplier for best thermal compensation?
 
fab said:

Thanks for this table for quick reference.
What do you suggest for VGS multiplier for best thermal compensation?

I have no idea.
I only know that Nelson Pass told us to use a HEXFET
for compensation of IRFP240/IRFP9240 MOSFET ( HEXFET, Not Lateral MOS ).
For example IRF610, which is same type.


We should try to find transistor with same thermal characteristics as the output transistors.

And we should use some case model that can be attached to a heatsink.
So in this case maybe we should use Q25 to match the Q34 + Q35 bipolar pair.
As these two will decide bias current to MOSFET, M1 and M3.

I really don't know.
Which transistors should be put onto BIG heatsink?
And which should not ...
Either all on Big heatsink: Q25 + Q34 + Q35 + M1 + M3
or only these on Big heatsink: Q25 + M1 + M3


The most used bipolar for Q25 in amplifiers,
is the BD139, in TO126 case.
No doubt about this!
So I would try BD139 first.


Let's here what those with practical experience will say.

lineup
 
Hi Fab !
... go on and let us know how it is working in real life and if you like the sound.
Your inverting power stage is quite similiar to one of my Rookie designs. Have a look to Rookie G...
The only weak point of this output stage seems to be the fact that the openloop output impedance is high.
But with your additional gain from your complimentary differential input stage you might be able to overcome this, if you find a way to handle the oscialltion issues in real life.
Take especially care for capacitive loads in the range between 50pF and 5nF, because that's a common range of real speaker wires.
 
Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.