MOSFET output stage

[Eeyore]

Fig 8: 4KW for 100 usec, which isn't as frightening as 330 watts CW.

You just reminded me. There was one bipolar design I developed of a range of powers which was used
extensively across a range of products including rack mount amplifiers and powered mixers. We must
have built tens of thousands of the amp modules.

Late in the proving process I ran one up with no fan. The heatsink reached 150C and there was a
strong smell of hot aluminium plus creaking noises from thermal expansion before I took pity on it
and powered it down. Worked fine the next day. TO-3 devices you see. Can't beat them.

Graham

 

[Eeyore]

The opamp power supply can be a cheap isolated dc/dc converter, +-12
volts maybe, floating on the output rail, so the opamps never see a
lot of swing. Of course, a real circuit needs some protections for
overload and startup conditions, but that's just a few diodes. The
opamp inputs need never go more than a few tenths of a volt above or
below the output rail.

The floating opamp supply allows one to truly saturate the fets and
swing the output all the way to both supply rails. (I assume some
number of N-channel and P-channel opamp-composite fets in a real amp.)
That pays for a dinky DC-DC sip thingie all by itself.

Yes you need some high side drive for that. I've used that in both my mosfet
and bipolar designs (to oversome multiple Vbe's (some of the drivers ran off
of the high side rail) AND a Baker clamp to stop the last voltage gain stage
transistor saturating).

You probably do only need input protection. Might need to be clever about
start up / turn-off conditions. Maybe a gate control ? I used that one on my
amps too. 100% effective against power-out damage even if the previous signal
chain generates a 'thump'.

Graham

 

[Eeyore]

point. I'm wondering if one might get a brief >dead band. What would be ideal would be if the
power device never fully >turned off and left say 10mA Iq.

Assuming some number of composite N-channel and P-channel fet
thingies, as a complementary follower, some bias voltage has to be
applied between the pseudo-gates (N-bank and P-bank opamp inputs) to
set the idle current. When the mess starts to drive a load in one
direction, some means ought to make sure the other bank doesn't go
off, but maintains its bias current. That takes a little more "analog
logic", still floating with all the rest of the stuff. That's not a
big deal.

I wish it were that simple. It's defeated me for years. Maybe a new set of eyes can discover the
trick ?

for fast overloads and back that up with a >> >> digital fet power dissipation simulation that
provides the real >> >> protections.

You can do that, simulate the Tj's, once for each bank...

Tnfet = Theatsink + K * (lowpass filter of) { (V+ - Vout) * Iout }

Tpfet = Theatsink + K * (lowpass filter of) { (V- - Vout) * Iout }

(except you have to get all the signs right)

where K relates to Theta-junction-heatsink and the lowpass filter
simulates the thermal mass of the silicon.

Absolutely. You can get all the thermal time constants in there and so on.

My NMR amps needed a uP and ADC and display anyhow, so the fancy
protections were pretty much just more code. The code runs at a few
KHz. A real PITA, but free in production.

My big amps display everything... temperatures, power supply voltages,
TRMS load current, output power, error messages, tons of stuff.

Nice. The pro-audio market wouldn't buy it though. Price (and weight) is currently everything,
although there are a very few high end niche products with ethernet (or proprietary) control and
monitoring around.

Graham

 

[Eeyore]

extensively across a range of products including rack mount amplifiers and powered >mixers. We
musthave built tens of thousands of the amp modules.
strong smell of hot aluminium plus creaking noises from thermal expansion before I took >pity on it
and powered it down. Worked fine the next day. TO-3 devices you see. Can't beat >them.

We tested a bunch of TO-247 power mosfets to destruction, in various
sadistic ways. One was pure temperature. Vgs-th drops as temp goes up.
The fets seemed to turn on hard, with 0 gate voltage, at 300C, but
recovered when cooled. After 330C, they died, on hard, and didn't
recover.

Were you estimating Tj ?

It's hard to buy TO-3 fets any more.

More's the shame. Tj max is typically rated 50C higher. And there are TWO bolts to hold then down with
too. Need I mention the advantages of that ?

Graham

 

[Eeyore]

used extensively across a range of products including rack mount amplifiers and >> >powered mixers. We
musthave built tens of thousands of the amp modules.
strong smell of hot aluminium plus creaking noises from thermal expansion before I >> >took pity on it

And my colleagues who were looking slightly anxious.

No, we were heating the fets externally.

Sorry, I didn't read it properly.

How about by self-heating ?

too. Need I mention the advantages of that ?

But you can put a lot more silicon into a TO-247, especially the
version without the mounting hole.

Well, at least clips or mounting bars won't bend the tab !

Semelab/Magnatec make lateral fets with 2 matched dies in TO-3. Beat that ! 250W true continuous Pd.
http://www.magnatec-uk.co.uk/latmos.shtml

And quads too by the look of it !
http://www.magnatec-uk.co.uk/mosdata.shtml

BUZ901X4S 200V 32A 500W SOT227
BUZ906X4S -200V -32A 500W SOT227

Graham

 

[MooseFET]

Indeed. You'll need lossy ballast resistors. Laterals are different that
way.

Yes, but once the STW55NM60 or equiv. is biased up, it makes a nice
200V at about 100KHz. Its not exactly audio but is sure isn't really
RF either.

 

[Kevin Aylward]

John Larkin wrote:

Eeyore wrote:
John Larkin wrote:
Eeyore wrote:
Learn something about LATERAL mosfets that were designed for
audio. I've already given part number and links to data sheets.
That doesn't really matter. The transfer function only needs to
be continuous so that you can close a loop around it, and the fet
needs to be able to stand the peak power dissipation. That can
easily be done with vertical "switching" type fets. A modern
FLOOD architecture [1] works great with most any kind of fet.
Lots of things have changed in the last few decades.

John

[1] Of course you've never heard the term before. I just invented
it.
Fine. Can you elaborate some more on it ? Laterals have some truly
lovely features for audio. The only downside being a slightly
highish Ron. Not really a problem when (as I have) used as many as
6 in parallel (12 mosfets per channel / 24 per amp). They also
match beautifully with no need for source balance resistors (so
some of the Ron loss 'goes away').

An opamp per fet, closing a local loop, feedback from the fet
source, makes each fet look like a perfect unity-gain, fast,
zero-offset device.
Interesting idea. I'll have to chew that one over. I can see
possible problems fron op-amp output overshoot.

I have a simple embodiment of that concept here, done a while ago, in
virtual land;-)

http://www.kevinaylward.co.uk/ee/circuits/VeryLowDistortionAmp2.jpg

Its a push/pull gain loop around the output devices, forcing them to
be unity gain followers.

You can get lower distortion, at the expense of speed, because you
have to compensate earlier.

Common mode feedback at the second stage, allows for enormous dc/lf
gain. cascodes to allow the use of fast small transistors to do all
the main work. Emitter follower buffer to reduce the current swing
in the input pair, as per doug self. Spice says it should be in the
< 0.0001% , 20Khz range, maybe...

Kevin Aylward

Kevin - an interesting circuit, and I appreciate what you have done
with the output stage, but I'm still wondering why you didn't include
it within the global feedback loop - that could only have made it
better, lower output impedance, more load insensitive etc etc etc.

It is.!!!

I think the schematic is not as clear as it should be.. I have a zero volt
source near the output devices in the feedback circuit to calculate LG. The
overall loop feedback passes through this source!!!

Regards

Kevin Aylward
www.blonddee.co.uk
www.kevinaylward.co.uk

 

[Kevin Aylward]

It is always a good idea to lookuop what we are talking about.
Indeed.


Some people are really good at that stuff, sort of reasonin gI mean.

But if you look at the combination MOSFET - opamp, as John Larkin is
describing in an other post in this thread, then you can treat that
as one 'perfect MOSFET'. Of course it is not really perfect, but you
can make that as stable or unstable as you want. Then basically what
you do is chaining stable blocks together. If you then apply
feedback, you have to use the phase characteristic of all those, and,
as long as you prevent too high frequencies from circulating, it
should be stable, and largely independent of the load.

If we consider calculating the actual LG=B(s).A(s) of the whole system, then
it is identical whther or not there is local feedback around the 2nd stage
or not.

Yes, you can analyse as you describe here, but the result must be identical,
as breaking all loops (if it can actually be done that is). So, it can't be
because there is a local loop that there is no zobel network.

I note the
TDA9274 has one capacitor to roll of in the driver... This is normal,
at least in the amps I designed.


You mean 'open loop gain?'

No. I mean the loop gain. The loop gain is the open loop gain X the beta
factor (e.g. resistive divider). It is the gain around the loop that
matters.

Yes, but he second amp would have gain 1.

But thats my point. Its irrelevent that the closed lop gain of the second
stage is unity when the loop is closed. When you do the stability analyisis
you need to break all the loops. As I said, if you break the loop directly
at the output of the second amp, which breaks both feedback loops at once,
it is obvious that there overal loop gain is not effected by having the 2nd
stage. It is the same loop gain

I was not suggesting to widen bandwidth, although strong local
feedback would of course widen the bandwidth of a stage,

But not the BW of the final, overall amp, so the BW of the internal stages
is irrelevant if it is the result of feedback.

What the local loop can buy you is reduced distortion at lower frequencies.

I ran these two circuits up quite a while ago. One has the UGB at the
output, one doesn't.

http://www.kevinaylward.co.uk/ee/circuits/VeryLowDistortionAmp1.jpg
http://www.kevinaylward.co.uk/ee/circuits/VeryLowDistortionAmp2.jpg

The UGB version had to be compensated earlier in frequency, i.e the non UGB
version was significantly faster. Unfortunately, I cant remember much of the
data, and I seem to have lost the SS files;-)

I think one had 0.0001% at 20khz.

Honestly, I have to think about this a bit, maybe run it in spice.
Fact remains that the TDA9274 is the only amp I know that needs no
Boucherot circuit

I would wager it's "non-optimum" designed. You give up a bit if the output
load is not defined, usually. Without the zobel, the load reflected to the
gain stages is all over the place. My guess is that they had a design goal
of minimising the number of external components, which is standard practice
in designing ics, but consequently, gave up some potential performance
improvement.

It seems to exist for US bankers ATM.

Kevin Aylward
www.kevinaylward.co.uk
www.blonddee.co.uk

 

[Kevin Aylward]

John Larkin wrote:
Eeyore wrote:
John Larkin wrote:
Eeyore wrote:

Learn something about LATERAL mosfets that were designed for
audio. I've already given part number and links to data sheets.

That doesn't really matter. The transfer function only needs to
be continuous so that you can close a loop around it, and the fet
needs to be able to stand the peak power dissipation. That can
easily be done with vertical "switching" type fets. A modern
FLOOD architecture [1] works great with most any kind of fet.
Lots of things have changed in the last few decades.

John

[1] Of course you've never heard the term before. I just invented
it.

Fine. Can you elaborate some more on it ? Laterals have some truly
lovely features for audio. The only downside being a slightly
highish Ron. Not really a problem when (as I have) used as many as
6 in parallel (12 mosfets per channel / 24 per amp). They also
match beautifully with no need for source balance resistors (so
some of the Ron loss 'goes away').


An opamp per fet, closing a local loop, feedback from the fet
source, makes each fet look like a perfect unity-gain, fast,
zero-offset device.

Interesting idea. I'll have to chew that one over. I can see
possible problems fron op-amp output overshoot.

I have a simple embodiment of that concept here, done a while ago, in
virtual land;-)

http://www.kevinaylward.co.uk/ee/circuits/VeryLowDistortionAmp2.jpg

Its a push/pull gain loop around the output devices, forcing them to
be unity gain followers.

You can get lower distortion, at the expense of speed, because you
have to compensate earlier.

Common mode feedback at the second stage, allows for enormous dc/lf
gain. cascodes to allow the use of fast small transistors to do all
the main work. Emitter follower buffer to reduce the current swing
in the input pair, as per doug self. Spice says it should be in the
< 0.0001% , 20Khz range, maybe...

Not exactly short of current mirrors ! ;~)

What gave you the idea ?

Graham

The loop around the output came about from studying a patent that could not
work as claimed, around 1983. This patent claimed distortion reduction by
eliminating miller effect. It had a standard class A gain stage, but used a
cascode. However, the cascode transistor base was connected to the output in
a local feedback loop. This was the claim for distortion reduction. However,
to1st order, the voltage on the base of a cascode transistor, don't effect
anything. The input is current feed, irespective of the base potential,
therfore the could not be any signal feedback. This led me to consider a way
of doing it properly. I also noted that as the base was connected to the
output, the emmiter of the cascode, and the collecter of the driving
transister. hence the collecter still swings the full output, hence, millor
effect is still there. So, that patent was complete nonsense, and my final
circuit was completly different, but it followed from the thought flow of
that patent.

The key point in this approach was to get a push pull drive to both
outputs.,i.e. to avoid low turn off resisters, which kills gain and drive.

Kevin Aylward
www.blonddee.co.uk
www.kevinaylward.co.uk

 

[Eeyore]

I would wager it's "non-optimum" designed. You give up a bit if the output
load is not defined, usually. Without the zobel, the load reflected to the
gain stages is all over the place. My guess is that they had a design goal
of minimising the number of external components, which is standard practice
in designing ics, but consequently, gave up some potential performance
improvement.

Absolutely. I always use an isolating inductor twixt amp and load and the series
RC to GND to define an accurate HF load.

Never heard it called a "Boucherot circuit" though. Zobel network is the popular
one here, although technically isn't that across the load itself ?

There's a lot to be said for impedance compensation of speaker drivers too.

Graham

 

[Eeyore]

Actually the "right" place to connect the feedback sensing resistor is
right out at the speaker itself, via a third sensing wire.

You don't realise just how true that is. I've tuned pcb layouts for THD just by
moving that node. PCB layout guys look perplexed but thankfully usually do it. In
fact there's loads of layout tricks the PCB guys are hopeless at, loops in
particular.

In my 1200B design, all the power fets were mounted on a separate pcb to which
power, drive and feedback wires were attached, manually soldered. If you didn't
get the feedback wire bang in the centre up went the 2nd harmonic THD.

Graham

 

[Eeyore]

Kevin Aylward wrote:
Eeyore wrote:
John Larkin wrote:
Eeyore wrote:
John Larkin wrote:
Eeyore wrote:
Learn something about LATERAL mosfets that were designed for
audio. I've already given part number and links to data sheets.
That doesn't really matter. The transfer function only needs to
be continuous so that you can close a loop around it, and the fet
needs to be able to stand the peak power dissipation. That can
easily be done with vertical "switching" type fets. A modern
FLOOD architecture [1] works great with most any kind of fet.
Lots of things have changed in the last few decades.

John

[1] Of course you've never heard the term before. I just invented
it.
Fine. Can you elaborate some more on it ? Laterals have some truly
lovely features for audio. The only downside being a slightly
highish Ron. Not really a problem when (as I have) used as many as
6 in parallel (12 mosfets per channel / 24 per amp). They also
match beautifully with no need for source balance resistors (so
some of the Ron loss 'goes away').
An opamp per fet, closing a local loop, feedback from the fet
source, makes each fet look like a perfect unity-gain, fast,
zero-offset device.
Interesting idea. I'll have to chew that one over. I can see
possible problems fron op-amp output overshoot.
I have a simple embodiment of that concept here, done a while ago, in
virtual land;-)

http://www.kevinaylward.co.uk/ee/circuits/VeryLowDistortionAmp2.jpg

Its a push/pull gain loop around the output devices, forcing them to
be unity gain followers.

You can get lower distortion, at the expense of speed, because you
have to compensate earlier.

Common mode feedback at the second stage, allows for enormous dc/lf
gain. cascodes to allow the use of fast small transistors to do all
the main work. Emitter follower buffer to reduce the current swing
in the input pair, as per doug self. Spice says it should be in the
< 0.0001% , 20Khz range, maybe...

Kevin Aylward


Kevin - an interesting circuit, and I appreciate what you have done
with the output stage, but I'm still wondering why you didn't include
it within the global feedback loop - that could only have made it
better, lower output impedance, more load insensitive etc etc etc.

It is.!!!

I think the schematic is not as clear as it should be.. I have a zero volt
source near the output devices in the feedback circuit to calculate LG. The
overall loop feedback passes through this source!!!

Ah, is that what it was? I thought you were putting actual voltage
sources in there that would be replaced by some small circuit in an
actual design. I hope you are connecting the sensing point of the
feedback resistor to the final summed speaker connection, not some
random point in amongst the bunch of fets (teaching granny to suck
eggs?) ;-)

Actually the "right" place to connect the feedback sensing resistor is
right out at the speaker itself, via a third sensing wire.

Better still, have differential sensing, to compensate for the volt drop in the
groundy side wire too !

Graham

 

[Jan Panteltje]

But not the BW of the final, overall amp, so the BW of the internal stages
is irrelevant if it is the result of feedback.

What the local loop can buy you is reduced distortion at lower frequencies.

I ran these two circuits up quite a while ago. One has the UGB at the
output, one doesn't.

http://www.kevinaylward.co.uk/ee/circuits/VeryLowDistortionAmp1.jpg

Now that gets complicated, Q4 and the 2 diodes in the emitter...

http://www.kevinaylward.co.uk/ee/circuits/VeryLowDistortionAmp2.jpg

The UGB version had to be compensated earlier in frequency, i.e the non UGB
version was significantly faster. Unfortunately, I cant remember much of the
data, and I seem to have lost the SS files;-)

I think one had 0.0001% at 20khz.

Yes, very good. Why use dotted paper ;-)

 

[Jan Panteltje]

Actually the "right" place to connect the feedback sensing resistor is
right out at the speaker itself, via a third sensing wire.

Ha, why did I never think of that...
This will eliminate my massive gold feed rods to the woofer.

You will need 2 sensing wires, and a diff amp.

 

[Jan Panteltje]

Except you don't want an audio amp to foldback, just shut down when it sees an 'impossible'
load.

Graham

That depends, if you just current limit, and have thermal protection, then
that takes time, perhaps enough time to melt some silicon.

If an amp is designed for 4 Ohm minimum load, and somebody connects 2 speakers
in parallel (for example), then current limit will step in.
The voltage drop will be small over the 2 Ohm (or lower, in case of a near short),
so say the transistors will have near full voltage at max current.
It is clear that when current limiting, then you can lower the reference for the current limit
circuit for low output voltages.
This reduces dissipation.

In fact (but this may hurt audio-people's egos perhaps), an audio amp
is nothing but 2 symmetrical regulated power supplies ;-)
In such a power supply one often also uses fold back for protection against
bad loads, example 7805 regulator...

 

[Kevin Aylward]

You don't realise just how true that is. I've tuned pcb layouts for
THD just by moving that node. PCB layout guys look perplexed but
thankfully usually do it. In fact there's loads of layout tricks the
PCB guys are hopeless at, loops in particular.

Yes indeed. This is exactly what happened on the 1st PCB prototype of the
MOSFET 1000. Ian did the layout. When it came back, I was only getting 0.02%
thd. Now..this is *exactly* how it happened. I looked at the routing,
immediately noticed that the feedback point was picked up along a high
current trace, took a piece of wire and jumpered across the offending trace,
and thd dropped down to 0.002%. Like, simple putting a wire in || with
another one (that had a tee in it) worked magic!


Kevin Aylward
www.blonddee.co.uk
www.kevinaylward.co.uk

 

[Eeyore]

On a sunny day it happened Don Pearce wrote


Ha, why did I never think of that...
This will eliminate my massive gold feed rods to the woofer.

You will need 2 sensing wires, and a diff amp.

LOL ! See one of my other posts.

Graham

 

[MooseFET]

   Tell that to WWVB, who transmits at 60 kHz.

Good point. I wasn't radiating, at least not on purpose.

A lot of echo sounders work up at 500KHz so 100KHz wasn't really too
high to be called sound too.

 

[Eeyore]

If you really think that's important (and I don't), why not just put the
amplifiers near the speakers? That way, there won't be any nasty
stability problems to deal with.

They're called active speakers and are widely used by many professionals.

Graham

 

[Kevin Aylward]

They're called active speakers and are widely used by many
professionals.

Indeed.

I don't actually like powered speakers though. They need *two* leads. Just
more hassle in setting up for the gig.


Kevin Aylward
www.blonddee.co.uk
www.anasoft.co.uk