Crossover distortion and NFB

[Ken Smith]

On Sun, 24 Oct 2004 01:00:56 +0000, Joerg wrote: [...]
betas are on, the driving stage (op-amp or other) experiences lower load
impedance than when one of the transistors is off. This non-linearity is
bound to have repercussions. But if you can get the bias just exactly
right, you would never have both transistors on, and you wouldn't have a
dead zone, either.

Anyway, he concludes that class AB is not a good idea. Either go fully
class A, or just go class B (again, with his definition of class B).

I disagree with this suggestion. What you want is for the output stage's
voltage to remain a linear function of the driver transistor's current. If
the current in one transistor stops exactly at the point where the current
in the other starts, there is a voltage at which the current in the
transistor is 0.3mA. Assuming an 8 ohm load, the voltage gain of the
output section will be 0.5 at that point. Obviously this can't be true
and have the system linear. This implies that some idle current must be
flowing so the stage must be AB.

 

[Walter Harley]

Far better

I'm wondering what they look like, not how they look. That is, if the
primary distortion mode is indeed crossover, I would expect to see the
residual be close to zero for most of the wave, with a pair of spikes at
each zero crossing. When corrected with NFB, the spikes should be steeper
and shorter in time. Is that what you see? What does the residual sound
like, in the two cases?

Indeed. We are talking the 6% THD caused by crossover as compared with
real
equipment with a THD figure of 0.01%

Sorry, I didn't quite follow your answer there. Can you clarify?

Indeed. Which expains why I was puzzled when I was accused of being a
troll
when I first posed this question.

Blame the goat fanciers.

For example why is 0.1% THD as a result of clipping inaudible to most
people
yet 0.1% as a result of crossover distortion is blatanly obvious to anyone
with
even the clothest of ears? They are still odd harmonics.

Quite so. At this point, you can look to several sorts of explanation, none
very satisfying: for instance, explanations from the standpoint of
evolutionary biology, from the standpoint of psychoacoustics, or from the
standpoint of sensory biology.

I was genuinely asking a real question.

I'm interested in the answer too. Personally I've always just gone for
bias, because the idea of asking an amp to slew as rapidly as it can just
seems like asking for trouble; if nothing else, there's going to be a spike
on the power supply that will be hard to keep out of the sensitive
circuitry. But I agree that your findings are counterintuitive and, if
experiments and math bear them out (which you assert they have done), that
is always fascinating.

I still feel like the answer is in the residuals.

 

[Rich Grise]

Rich Grise wrote:

[snip]

It's because they eat goats

And was it the beer that made me laugh or was it Rich?

What does it matter which? Isn't the laughter the goal, after all? ;-)

I'll never forget Kermit the Frog, one of my heroes, in a uniquely
poignant(sp) moment, in "The Muppet Movie." The line was, "Millions
Of People Happy..." And I swear on my Mother's grave, that sock puppet
had stars in his eyes, theatrically speaking.

But that has always resonated with me as a pretty cool life goal:
"Millions of people happy."

So, I try to aspire to "class clown", while also attempting to promote
my own particular brand of trans-meta-para-dyno-infra-philosopho-ergono-
radio-femto-gefilte-preposto-scientific quasiknowledge, that can only be
found within, underneath the noise of the voices that are already
directing you, _especially vehemently,_ to resist your own awakeness,
which means the uncreation of the evil, which to [ERRRNT - LOST CARRIER]

And I betcha "not get killed by some wacko empire builder's armies" is
only the tip of the iceberg as to possibilities for happiness worldwide.

I know it's right up there in the top thing of my agenda. %-}

Cheers!
Rich

 

[Kevin Aylward]

On Sun, 24 Oct 2004 01:00:56 +0000, Joerg wrote: [...]
betas are on, the driving stage (op-amp or other) experiences lower
load impedance than when one of the transistors is off. This
non-linearity is bound to have repercussions. But if you can get the
bias just exactly right, you would never have both transistors on,
and you wouldn't have a dead zone, either.

Anyway, he concludes that class AB is not a good idea. Either go
fully class A, or just go class B (again, with his definition of
class B).

I disagree with this suggestion. What you want is for the output
stage's voltage to remain a linear function of the driver
transistor's current.

Not exactly. The goal is for the output voltage to follow the input
voltage linearly. The driver current is not a linear function of the
base emitter voltage, its exponential.

If the current in one transistor stops exactly
at the point where the current in the other starts, there is a
voltage at which the current in the transistor is 0.3mA. Assuming an
8 ohm load, the voltage gain of the output section will be 0.5 at
that point. Obviously this can't be true and have the system linear.

The point here is that the transistor isn't a switch. It has to have a
transition section.

This implies that some idle current must be flowing so the stage must
be AB. --

One certainly wants enough current to keep things up and running. A 1ma,
25 ohm re is just not cricket for a 100W amp.

Kevin Aylward
[email protected]
http://www.anasoft.co.uk
SuperSpice, a very affordable Mixed-Mode
Windows Simulator with Schematic Capture,
Waveform Display, FFT's and Filter Design.

 

[Kevin Aylward]

On Sun, 24 Oct 2004 01:00:56 +0000, Joerg wrote:

[snip]

This is why all linear audio amps
need a certain amount of quiescent current. Class D doesn't, of
course, but that's a whole other topic.

Have you ever read Douglas Self's book, _Audio Power Amplifier Design
Handbook_?

Some of this is apparantly here,
http://www.dself.dsl.pipex.com/ampins/dipa/dipa.htm

In it he argues that if the output transistors are not conducting
half the time for a sine wave, then it shouldn't be called "class B."
He reserves the term "class B" for when there is an essentially
seamless transition from the upper transistor to the lower one, but
no period where they both conduct. So if the bases of the output
transistors are connected together, it is NOT class B, according to
Self. This makes some sense.

Yes. There is some inconsistent terminology used here. The classic
non-biased push-pull pair might better be class C. No one does though.

But the interesting thing is that he measures the THD and residuals
for class B (his definition) and Class AB open loop output stages, and
the class B stage has lower THD than class AB.

Maybe. Why Self claims that this possibility may be a "vital fact is
little known" is beyond me:

5.3 DISTORTION 3.
"THD increases as the bias advances into AB operation. This is due to
so-called "gm-doubling" (ie the voltage-gain increase caused by both
devices conducting simultaneously in the centre of the output-voltage
range, in the Class-A region) putting edges into the distortion residual
that generate high-order harmonics much as under-biasing does. This
vital fact is little known, presumably because gm-doubling distortion is
at a relatively low level and is obscured in most amplifiers by other
distortions"

This is very well known in the industry. For example, it's a fundamental
issue in designing op-amps with rail to rail inputs. However, it a bit
more complicated than this. The increase in gm due to bias increase,
generally gives better linearity and can produce, in principle, a
*lower* THD. Its not clear from Self's paper how he actually determined
that the optimal point occurred when the transistors were actually on
the edge of switch over, which infact they cant do in such a manner
anyway. These measurements should really have been performed open loop
to make the effect more noticeable. For example, even if the thd was
initially higher, the higher gm might possibly make the system more
stable (less effect of re.Cload), allowing for more feedback which could
reduce the distortion. You simply don't want to operate output devices
at too low a current. Everything collapses.

Secondly, I have done quite a few measurements myself on this, and I can
state that for mosfets, the x-over distortion, by and large, can keep
going down as you move more and more into AB->A. In general, there is no
general claim that can be made here on the "best" bias point.

Anyway, he seems to make a habit of these "little known phenomena"
claims, like on the same paper,

"It seems to be little-known that electrolytic capacitors generate
distortion when they have a significant AC voltage across them. It is
even less well known that non-electrolytic show a similar effect in
applications like Sallen & Key high-pass filters."

Again, this is a fundamental issue in designing speaker x-overs. That's
why one often uses polypropylene caps. I can't imagine anyone engaged in
such matters not being aware of this.

This makes a certain
amount of sense, too, because when both output transistors with their
finite betas are on, the driving stage (op-amp or other) experiences
lower load impedance than when one of the transistors is off.

The effect is fundamentally due to gm (transconductance) doubling, not
hfe.

The voltage gain is:

Av = gm.Rl/(1 + gm.RL)

Change is gm effect the voltage gain.

This
non-linearity is bound to have repercussions. But if you can get the
bias just exactly right, you would never have both transistors on,
and you wouldn't have a dead zone, either.

Anyway, he concludes that class AB is not a good idea.

I dont see that he actually makes that conclusion in that paper.

Either go fully
class A, or just go class B (again, with his definition of class B).

I dont see this as being practical really. In general, you can't have
production amps tweaked up like this. For starters, bias current
temperature compensation never tracks perfectly. The effect he documents
is like "14b 0.00103%, and for 14c 0.00153%". A difference of 0.0005%,
with the speaker generating, say 1% is hardly a good argument for not
overbiasing at all. The main reason for not significantly overbising is
simply to avoid wasting power.


Kevin Aylward
[email protected]
http://www.anasoft.co.uk
SuperSpice, a very affordable Mixed-Mode
Windows Simulator with Schematic Capture,
Waveform Display, FFT's and Filter Design.

 

[Kevin Aylward]

Far better


Exactly my feelings. But measurements don't bear out my gut feelings.
And other posters proved mathematically (with maths that is way
beyond my abilities) that NFB cannot reduce the overall crossover
distortion, it can only move it elsewhere.


It does indeed


I did, many times


Indeed. We are talking the 6% THD caused by crossover as compared
with real equipment with a THD figure of 0.01%


But it's during the silent period, and ears are logarithmic. But how
quickly?


Indeed. Which expains why I was puzzled when I was accused of being a
troll when I first posed this question.

For example why is 0.1% THD as a result of clipping inaudible to most
people yet 0.1% as a result of crossover distortion is blatanly
obvious to anyone with even the clothest of ears? They are still odd
harmonics.

Having said that. The one who accused me of being a troll has since
been shown to be a cock.

A sinewave with 5% THD as a result of clipping actually doesn't sound
so bad. Some people can't even hear it. But add THD as a result of
crossover and it's blantantly obvious and sounds awful.

I was genuinely asking a real question.

THD is an average measurement, i.e. its the (root) mean square error. It
takes no account of the instantaneous error. Near the 0-xing point the
relative error is huge, e.g.the output should be 0.5, yet is producing
0. This is a 100% instantaneous error. At clipping say, 10V output, with
desired output of 10.5 (similar THD) the instantaneous error is only
0.5/10 = 5%.

One might thus conclude that the ear/brain can detect instantaneous
errors in the time domain. Even if this is only by means of analysing
the structure in the frequency domain, why should one spectrum of
distortion sound the same as another one that is completely different?
Not, why should all rms measurements sound the same just becase the
value is the same.

Suppose we have a group of people with heights, gassusain distributed.
Now have another group where everyone is the same, except for a few
giants and dwarfs. Such groups can easily have the same standard
deviation, yet obviously be very different and easily distinguished.

I don't see any rational reason why an average measurement *should* be
enough to determine the effect of the details of the what is being
measured. If the ear/brain was a thermal sensor, sure, but it isn't.


Kevin Aylward
[email protected]
http://www.anasoft.co.uk
SuperSpice, a very affordable Mixed-Mode
Windows Simulator with Schematic Capture,
Waveform Display, FFT's and Filter Design.

 

[Ken Smith]

Ken Smith wrote: [...]

I disagree with this suggestion. What you want is for the output
stage's voltage to remain a linear function of the driver
transistor's current.

Not exactly. The goal is for the output voltage to follow the input
voltage linearly. The driver current is not a linear function of the
base emitter voltage, its exponential.

In the designs I was thinking of, the compensation is done at the driver
stage. This makes the driver's transfer quite linear at high frequencies.

 

[Jim Thompson]

Seems something odd with the THD measurements. A spicing with a X10 opamp
and 1000Hz, 200mVpp input gave ...

Output V Output V
Buffer out of loop. Buffer in loop.

1000Hz 240mVpp 700mVpp (fundamental)
3000Hz 83mVpp 0.5mVpp
5000Hz 13mVpp 0.4mVpp
7000Hz 5mVpp 0.8mVpp
9000Hz 3mVpp 0.5mVpp

THD ... 35% 0.16%

All H.F harmonics up to 1Mhz were similar and of a low level, sufficient not
to contribute to the major player THD harmonics.
regards
john

Thanks, John, I was about to do that experiment also. Gibbo's
comments just seemed ridiculous to me.

...Jim Thompson

 

[Tam/WB2TT]

:

[snip]

There is basically something wrong here. >
Obviously

There are probably 100 million
amplifiers out there that work the way you describe

No I doubt it. They all use biasing to get rid of 99.9% of the THD as a
result
of crossover.

I mentioned biasing diodes. Why would you not use them? Run SWCAD on your
design, and do an FFT on the output.

Tam

 

[Tam/WB2TT]

Rich Grise wrote:

Summat....

You're close!

Gibbo

Why not measure the THD of your signal source, using the same
procedure.Might be interesting.

Tam

 

[Joerg]

Hi Mac,

Have you ever read Douglas Self's book, _Audio Power Amplifier Design
Handbook_?

No, not yet.

In it he argues that if the output transistors are not conducting half the
time for a sine wave, then it shouldn't be called "class B." He reserves
the term "class B" for when there is an essentially seamless transition
from the upper transistor to the lower one, but no period where they
both conduct. So if the bases of the output transistors are connected
together, it is NOT class B, according to Self. This makes some sense.

There has to be a seamless transition but it really doesn't matter much
how you get there. Full bias and lots of quiescent current, less bias
and feedback to haul the amp through the region fast enough, or
something in between.

But the interesting thing is that he measures the THD and residuals for
class B (his definition) and Class AB open loop output stages, and
the class B stage has lower THD than class AB. This makes a certain amount
of sense, too, because when both output transistors with their finite
betas are on, the driving stage (op-amp or other) experiences lower load
impedance than when one of the transistors is off. This non-linearity is
bound to have repercussions. But if you can get the bias just exactly
right, you would never have both transistors on, and you wouldn't have a
dead zone, either.

One of the problems of "just exactly right" is to hold it there even
when the amp gets hot. In RF amplifiers it used to be customary to
fasten diodes on the heat sink close to the final transistors so their
curves would move along the same pattern as the transistor Vbe.

Anyway, he concludes that class AB is not a good idea. Either go fully
class A, or just go class B (again, with his definition of class B).

I am always a bit careful with strong conclusions in books. It reminds
me of many, many publications advocating to split ground planes,
something that rarely works in the practical world ;-)

Regards, Joerg

 

[Joerg]

Hi Kevin,

... The effect he documents is like "14b 0.00103%, and for 14c 0.00153%". A difference of 0.0005%,
with the speaker generating, say 1% is hardly a good argument for not overbiasing at all. The main reason for not significantly overbising is simply to avoid wasting power.

That's exactly the point. Besides the speaker, which source of audio can
truly claim 0.001%?

Regards, Joerg

 

[Ken Smith]

Besides the speaker, which source of audio can
truly claim 0.001%?

How about an SRS DS360? Its about in there. Listening to it would get
boring quickly though.

 

[Joerg]

Hi Chris,

I have an old radio (circa 1930) that I keep because it fits in with the
surroundings (I live in a narrowboat with a vintage style stern and like to
keep some of the tradition) but the sound is beautiful. But hardly HiFi.

Well, how could that be HiFi? In 1930 there was no FM so your radio will
only feature the AM band, since you live in Europe probably also long
wave and then maybe a portion of short wave. I have one of those as
well, an old 'Sachsenwerk'. It states the dial position in meters
instead of kilohertz. Most of the stations listed on that dial are long
gone. The main problem is that the black paint of which the lettering
behind the dial glass consists is starting to flake off.

Living on a boat must be pretty romantic. If that is within an old
marina I am sure there are also plenty of nice pubs around.

Regards, Joerg

 

[Joerg]

Hi Ken,

How about an SRS DS360? Its about in there. Listening to it would get
boring quickly though.

That sure is a nice generator. I don't know what it costs, probably in
the vicinity of a nicely equipped car.

Yes, listening would be boring but maybe it has a fast enough interface
to use it as a synthesizer for a rendition of a Led Zeppelin piece.
There you'd have to add some distortion back in to make it all sound
good ;-)

Regards, Joerg

 

[john jardine]

A few years ago I posted a similar question and got...

A) Told to stop trolling.

B) Told to stop being a cock

Or

C) Told to go back to school

I'm still no wiser. I did all three and I'm still no wiser.

On audio, take the output of an opamp and feed it into dual emitter followers
with no overall NFB.

The distortion looks horrendous on a scope and sounds it. Take the NFB from
outside the emitter followers and it looks great on a scope, sounds better, but
the distortion as measured by a THD meter is the same. The harmonics have moved
(much higher in frequency) but the total %age is the same.

Even if you only count the distortion products within the audio band (which
gets lower with each passing year for me) the overall THD is only slightly
lower. Yet the difference in sound quality is dramatic.

Now I know about logarithmic ears etc. But why was I told to stop trolling?

I'm only posting this because everyone else is talking about Dubbya or Kerry
which I suppose means that, technically, it *is* trolling but surely it's more
interesting. Especially for some of us outside USA who don't give a flying fu*k
who you get lumbered with. Whoever it is, our kissass PM will be stuck right up
him anyway.

Gibbo

Seems something odd with the THD measurements. A spicing with a X10 opamp
and 1000Hz, 200mVpp input gave ...

Output V Output V
Buffer out of loop. Buffer in loop.

1000Hz 240mVpp 700mVpp (fundamental)
3000Hz 83mVpp 0.5mVpp
5000Hz 13mVpp 0.4mVpp
7000Hz 5mVpp 0.8mVpp
9000Hz 3mVpp 0.5mVpp

THD ... 35% 0.16%

All H.F harmonics up to 1Mhz were similar and of a low level, sufficient not
to contribute to the major player THD harmonics.
regards
john

 

[john jardine]

Harmonics should read "rms" not "pp"

 

[Mac]

On Sun, 24 Oct 2004 01:00:56 +0000, Joerg wrote:

[snip]

This is why all linear audio amps
need a certain amount of quiescent current. Class D doesn't, of
course, but that's a whole other topic.

Have you ever read Douglas Self's book, _Audio Power Amplifier Design
Handbook_?

Some of this is apparantly here,
http://www.dself.dsl.pipex.com/ampins/dipa/dipa.htm

I scanned the article, and it looks very much like the book. I think many
of the figures are exactly the same.

[snip]

Maybe. Why Self claims that this possibility may be a "vital fact is
little known" is beyond me:

5.3 DISTORTION 3.
"THD increases as the bias advances into AB operation. This is due to
so-called "gm-doubling" (ie the voltage-gain increase caused by both
devices conducting simultaneously in the centre of the output-voltage
range, in the Class-A region) putting edges into the distortion residual
that generate high-order harmonics much as under-biasing does. This
vital fact is little known, presumably because gm-doubling distortion is
at a relatively low level and is obscured in most amplifiers by other
distortions"

This is very well known in the industry. For example, it's a fundamental
issue in designing op-amps with rail to rail inputs. However, it a bit
more complicated than this. The increase in gm due to bias increase,
generally gives better linearity and can produce, in principle, a
*lower* THD. Its not clear from Self's paper how he actually determined
that the optimal point occurred when the transistors were actually on
the edge of switch over, which infact they cant do in such a manner
anyway. These measurements should really have been performed open loop
to make the effect more noticeable. For example, even if the thd was
initially higher, the higher gm might possibly make the system more
stable (less effect of re.Cload), allowing for more feedback which could
reduce the distortion. You simply don't want to operate output devices
at too low a current. Everything collapses.

Secondly, I have done quite a few measurements myself on this, and I can
state that for mosfets, the x-over distortion, by and large, can keep
going down as you move more and more into AB->A. In general, there is no
general claim that can be made here on the "best" bias point.

Self takes a dim view of MOSFET output stages for power amps.

Anyway, he seems to make a habit of these "little known phenomena"
claims, like on the same paper,

"It seems to be little-known that electrolytic capacitors generate
distortion when they have a significant AC voltage across them. It is
even less well known that non-electrolytic show a similar effect in
applications like Sallen & Key high-pass filters."

Again, this is a fundamental issue in designing speaker x-overs. That's
why one often uses polypropylene caps. I can't imagine anyone engaged in
such matters not being aware of this.

Maybe he just means he hasn't heard people in audio circles talk about it?
I don't know. Maybe he originally wrote about it a long time ago?

The effect is fundamentally due to gm (transconductance) doubling, not
hfe.

OK, right. Thanks. (My mistake, not Self's).

The voltage gain is:

Av = gm.Rl/(1 + gm.RL)

Change is gm effect the voltage gain.


I dont see that he actually makes that conclusion in that paper.

No, but he does in the book.

I dont see this as being practical really. In general, you can't have
production amps tweaked up like this. For starters, bias current
temperature compensation never tracks perfectly. The effect he documents
is like "14b 0.00103%, and for 14c 0.00153%". A difference of 0.0005%,
with the speaker generating, say 1% is hardly a good argument for not
overbiasing at all. The main reason for not significantly overbising is
simply to avoid wasting power.

I can easily believe what you are saying, and I certainly have no
experience to the contrary.

Kevin Aylward
[email protected]
http://www.anasoft.co.uk
SuperSpice, a very affordable Mixed-Mode
Windows Simulator with Schematic Capture,
Waveform Display, FFT's and Filter Design.

Thanks, Kevin!

--Mac

 

[Mac]

On Sun, 24 Oct 2004 01:00:56 +0000, Joerg wrote: [...]
betas are on, the driving stage (op-amp or other) experiences lower load
impedance than when one of the transistors is off. This non-linearity is
bound to have repercussions. But if you can get the bias just exactly
right, you would never have both transistors on, and you wouldn't have a
dead zone, either.

Anyway, he concludes that class AB is not a good idea. Either go fully
class A, or just go class B (again, with his definition of class B).

I disagree with this suggestion. What you want is for the output stage's
voltage to remain a linear function of the driver transistor's current. If
the current in one transistor stops exactly at the point where the current
in the other starts, there is a voltage at which the current in the
transistor is 0.3mA. Assuming an 8 ohm load, the voltage gain of the
output section will be 0.5 at that point. Obviously this can't be true
and have the system linear. This implies that some idle current must be
flowing so the stage must be AB.
--

Well, really what you want, I think, is to adjust the quiescent current of
the open loop circuit for lowest THD. At any rate, this was Self's
approach.

The system will not be perfectly linear no matter what you do, so your
argument is weak. Or at least, it isn't any more compelling than Self's
argument.

--Mac

 

[Kevin Aylward]

On Sun, 24 Oct 2004 01:00:56 +0000, Joerg wrote:

[snip]

This is why all linear audio amps
need a certain amount of quiescent current. Class D doesn't, of
course, but that's a whole other topic.

Have you ever read Douglas Self's book, _Audio Power Amplifier
Design Handbook_?

Some of this is apparantly here,
http://www.dself.dsl.pipex.com/ampins/dipa/dipa.htm

I scanned the article, and it looks very much like the book. I think
many of the figures are exactly the same.

[snip]

Maybe. Why Self claims that this possibility may be a "vital fact is
little known" is beyond me:

5.3 DISTORTION 3.
"THD increases as the bias advances into AB operation. This is due to
so-called "gm-doubling" (ie the voltage-gain increase caused by both
devices conducting simultaneously in the centre of the output-voltage
range, in the Class-A region) putting edges into the distortion
residual that generate high-order harmonics much as under-biasing
does. This vital fact is little known, presumably because
gm-doubling distortion is at a relatively low level and is obscured
in most amplifiers by other distortions"

This is very well known in the industry. For example, it's a
fundamental issue in designing op-amps with rail to rail inputs.
However, it a bit more complicated than this. The increase in gm due
to bias increase, generally gives better linearity and can produce,
in principle, a *lower* THD. Its not clear from Self's paper how he
actually determined that the optimal point occurred when the
transistors were actually on the edge of switch over, which infact
they cant do in such a manner anyway. These measurements should
really have been performed open loop to make the effect more
noticeable. For example, even if the thd was initially higher, the
higher gm might possibly make the system more stable (less effect of
re.Cload), allowing for more feedback which could reduce the
distortion. You simply don't want to operate output devices at too
low a current. Everything collapses.

Secondly, I have done quite a few measurements myself on this, and I
can state that for mosfets, the x-over distortion, by and large, can
keep going down as you move more and more into AB->A. In general,
there is no general claim that can be made here on the "best" bias
point.

Self takes a dim view of MOSFET output stages for power amps.

Yes I know, but he is misguided on this.

The basic argument against mosfets is that they are not as linear in an
emitter/source follower configuration because of their much lower gm.
This is correct. However, this is compensated by other advantages.
Mosfets are much faster with much lower capacitances. This results in a
simpler driver with less phase shift, and a allows for much larger
feedback. This larger feedback gets the distortion very low. At the
larger powers, say 500 watts, these aspects are very significant.
Typically you need 3 stages of emitter follower drivers, for the
bipolar, against only one driver stage for the mosfet. For example, a
modern high performance 50Mhz ft power transistor (MJL4281A) would have
an effective input capacitance of 1.5nf with a 4ohm load, a mosfet might
be only 250pf. In addition mosfets are much more resilient to overload.

The proof is in the pudding. All of the plots on Self's amp page show
THD at 20khz for his reference amps at over 0.01%. Without trying to
blow my own trumpet here, a 500W per chan mosfet amp I designed (i.e.
small modifications of standard circuits,
http://www.studiomaster.com/hp5.html) was < 0.005% at similar 8 ohm
loads.

The main argument for bipoler is that they can be cheaper.


Kevin Aylward
[email protected]
http://www.anasoft.co.uk
SuperSpice, a very affordable Mixed-Mode
Windows Simulator with Schematic Capture,
Waveform Display, FFT's and Filter Design.