[William Sommerwerck]

I ask the following, hoping you will carefully consider it before
responding... How do you know that reducing harmonic and IM
distortion to extremely low levels with test signals necessarily
produces a comparable reduction in such distortions with
program material?

Its called the laws of physics.

Mozart
Oh? Which laws?
/Mozart


In a broader sense, how do you know that an amplifier's behavior
with a complex signal (ie, music) is accurately predicted by its
behavior with simple signals?

Its called the laws of physics. There is nothing special about music
signals. Unfortunately, there are just too many unqualified individuals
around making all sorts of dubious claims that this is not so.

Again, which laws? You can't cite the law of superposition (do you even know
what that is?), because it doesn't apply. Keep reading.


If one takes a competentally designed DC coupled amplifier, and say,
gets 0.01% IM when driven with say, equal levels of steady state 19Khz
and 20Khz to just below clipping and at very small signals, its pretty
much
inconceivable that the amp is not going to be a piece of wire with gain.

Not in my book. Keep reading.


In principle, that one test is not measuring much, but in reality, it
does.
Other aspects, all things being equal, go hand in hand.

Do you know what a tautology is? You talk about competently designed
amplifiers, but /that's/ the very thing we're trying to define! You haven't
told us what a competently defined amplifier IS! Simply having low measured
distortion with static signals DOES NOT demonstrate that an amplifier is
linear.


The route [SIC!] of your question is, do the bias conditions of an amp
change significantly, when a signal is applied such that it makes the
steady state distortion tests invalid? Well, not if the amp is designed
not to do that.

What does this have to do with the issue?


If an amplifier is linear, then a simple signal is all that is required to
predict the results for any signal.

Precisely. So how do you determine whether it's linear?

This is a perfect example of circular reasoning. If an amp has low measured
distortion, it has to be linear, and therefore behavies identically with
simple or complex signals. But its linearity with simple signals doesn't
prove its linearity with complex signals!

I urge you to find an intelligent teacher and have him explain the logic of
this to you.


This is a provable mathematical fact. If the amplifier is non-linear, then
sure, more is required. However, if the amplifier is "linear enough", than
the simple tests are enough.

Prove it.

I invite anyone who believes that low meaured distortion equates
subjectively with low perceived distortion to find a Crown K-1 or K-2
amplifier and simply /listen/ to it. After you've recovered, measure it with
standard test signals, and tell us what there is in the measurements that
predicts why it sounds as awful as it does.

[Arny Krueger]

"William Sommerwerck" wrote in
message

Again, which laws? You can't cite the law of
superposition (do you even know what that is?),

Gratiutious insult.

because it doesn't apply. Keep reading.

Superposition is a highly exact model of real-world operation once a system
is sufficiently linear and stable.

In the case of audio, "sufficiently" is several times higher than the 0.01%
that Kevin used as his example.

Your problem Bill is that you have lived in the imaginary world of high end
audiophilia way too long. A few good DBTs would have straigtened you out a
few decades back, but you never did your homework.

[Kevin Aylward[_4_]]

"William Sommerwerck" wrote in message
...

I ask the following, hoping you will carefully consider it before
responding... How do you know that reducing harmonic and IM
distortion to extremely low levels with test signals necessarily
produces a comparable reduction in such distortions with
program material?

Its called the laws of physics.

Mozart
Oh? Which laws?
/Mozart


In a broader sense, how do you know that an amplifier's behavior
with a complex signal (ie, music) is accurately predicted by its
behavior with simple signals?

Its called the laws of physics. There is nothing special about music
signals. Unfortunately, there are just too many unqualified individuals
around making all sorts of dubious claims that this is not so.

Again, which laws? You can't cite the law of superposition (do you even
know
what that is?), because it doesn't apply. Keep reading.

Oh dear...

If one takes a competently designed DC coupled amplifier, and say,
gets 0.01% IM when driven with say, equal levels of steady state 19Khz
and 20Khz to just below clipping and at very small signals, its pretty
much
inconceivable that the amp is not going to be a piece of wire with gain.

Not in my book.

I would suggest that you are reading the wrong books.

Keep reading.

Why?

In principle, that one test is not measuring much, but in reality, it
does.
Other aspects, all things being equal, go hand in hand.

Do you know what a tautology is?

My favourite tautology is one of my own.

http://www.kevinaylward.co.uk/replicators/index.html

"That which is mostly observed, is that which replicates the most."

Its amazingly useful. For instance, it tells us that, statistically, men
will have sex at every reasonable opportunity and women won't.

You talk about competently designed
amplifiers, but /that's/ the very thing we're trying to define! You haven't
told us what a competently defined amplifier IS! Simply having low measured
distortion with static signals DOES NOT demonstrate that an amplifier is
linear.

By and large, it does.

Subject to a few reasonable mathematical conditions, we can express any
amplifier by a Taylor power series for its output voltage with respect to
input voltage. Actually, to be more exact, we actually need a Volterre
series, but the net effect is still a power series expansion.

The point being that if the system is nonlinear at all, it is provable
that there must be harmonic and intermodulation distortion. Conversely, if
the distortion for steady state signals is sufficiently low, then the system
must be linear, subject to conditions I already mentioned.

For the Taylor approach, we have:

Vout = sum An.Vin^n

If Vin = Vpk.Sin(wt), then the power terms generate a series harmonic
frequencies, via relations such as sin^2(x) = (1-cos(x)/2. If the system is
linear, then no harmonic frequencies are
generated.

The route [SIC!] of your question is, do the bias conditions of an amp
change significantly, when a signal is applied such that it makes the
steady state distortion tests invalid? Well, not if the amp is designed
not to do that.

What does this have to do with the issue?

A lot.

For example, in principle, one could construct an amplifier that used its
own output to charge up a supply capacitor, that was used as a supply for
itself. In a steady state test the amplifier might be biased up correctly
with the right voltages and currents from that supply. However, it may be
that it takes 100 cycles for that that supply capacitor to charge up, hence
transient busts of sine pulses, might well show distorted sine bursts.

So, to solve that sort of problem, don't do it.

If an amplifier is linear, then a simple signal is all that is required
to
predict the results for any signal.

Precisely. So how do you determine whether it's linear?

er... feed in a sine wave and see if other frequencies get generated.

This is a perfect example of circular reasoning. If an amp has low measured
distortion, it has to be linear, and therefore behavies identically with
simple or complex signals. But its linearity with simple signals doesn't
prove its linearity with complex signals!

See above, by and large it does. Linearity is independent of the waveform,
by definition. Except for the type of situation I described above, which can
be eliminated by correct design, if there are no harmonics, then the system
is linear.

I urge you to find an intelligent teacher and have him explain the logic of
this to you.

Ahmmm...I think you might do well to take a course on Signal Processing.

This is a provable mathematical fact. If the amplifier is non-linear,
then
sure, more is required. However, if the amplifier is "linear enough",
than
the simple tests are enough.

Prove it.

I invite anyone who believes that low measured distortion equates
subjectively with low perceived distortion

It is a known fact that adding small amounts of distortion, can actuality
make things sound "cleaner". So, there is some truth in what you say here.


to find a Crown K-1 or K-2
amplifier and simply /listen/ to it. After you've recovered, measure it
with
standard test signals, and tell us what there is in the measurements that
predicts why it sounds as awful as it does.

One needs to do distortion tests at various amplitudes. It is certainly
possible to quote misleading figures. One needs to account for 1mW and 100W
levels. If at all levels, its at the 0.01% IMD at 20khz/19khz, its a
straight piece of wire with gain.



Regards

Kevin Aylward B.Sc.
www.kevinaylward.co.uk

[Scott Dorsey]

Kevin Aylward wrote:
The point being that if the system is nonlinear at all, it is provable
that there must be harmonic and intermodulation distortion. Conversely, if
the distortion for steady state signals is sufficiently low, then the system
must be linear, subject to conditions I already mentioned.

Right, but how low is low?

Hint: the THD number is no longer useful. Back in the days when amplifiers
were all similar topologies and distortion spectra were similar, it was very
useful for comparing amplifiers. This is no longer the case.

For the Taylor approach, we have:

Vout = sum An.Vin^n

If Vin = Vpk.Sin(wt), then the power terms generate a series harmonic
frequencies, via relations such as sin^2(x) = (1-cos(x)/2. If the system is
linear, then no harmonic frequencies are
generated.

Right. But generation of harmonics depends on the nonlinearities involved.

I can build you a box with 2% distortion that you will have a hard time
identifying as being in-circuit or out of circuit. I can built you another
box with 0.01% distortion which is painfully obvious. The first box is
mostly third harmonic, the next box is mostly sixth.

But note ALSO, and this is the most important part, that the subtraction
test does not detect only harmonics. If there is any group delay in the
channel, or any small frequency response changes, products will appear in
the differential product. These are the result of distortions which are
not terribly audible, maybe not audible at all in the case of the group
delay, but which become very prominant in the subtraction test.

The subtraction test is a useful tool but only when the differential
product is analyzed. Just listening to it and measuring the amplitude
is not sufficient and is apt to be misleading.

Likewise the THD measure is not useful for comparing systems of different
distortion spectra. Folks are working on weighted measures like the
Geddes-Lee test for that but we're not there yet.

I invite anyone who believes that low measured distortion equates
subjectively with low perceived distortion

It is a known fact that adding small amounts of distortion, can actuality
make things sound "cleaner". So, there is some truth in what you say here.

Mostly that's high order even stuff, and it's the principle by which
the Aural Exciter and the BBE boxes operate. It's also a lot of why
there were problems with much of the early-seventies solid state amps
that resulted in an odd split of listeners' opinions.

One needs to do distortion tests at various amplitudes. It is certainly
possible to quote misleading figures. One needs to account for 1mW and 100W
levels. If at all levels, its at the 0.01% IMD at 20khz/19khz, its a
straight piece of wire with gain.

Maybe, but at which point does that happen?
--scott

--
"C'est un Nagra. C'est suisse, et tres, tres precis."

[Anahata]

On Sun, 30 Jan 2011 10:26:54 -0500, Scott Dorsey wrote:

The subtraction test is a useful tool but only when the differential
product is analyzed. Just listening to it and measuring the amplitude
is not sufficient and is apt to be misleading.

I seem to remember first encountering the subtraction test in an article
in Wireless Word (if it still called that then) in the 1960s by Peter
Baxandall. In the application where he used it, on an audio power
amplifier, the difference signal, with only some simple passive HF phase/
delay compensation to get a null, amplified with the same gain as the
amplifier under test, was inaudible.

I agree that audible results from a subtraction test need to be carefully
analysed, but if the result is silence, it surely tells us the amplifier
is good enough for all practical purposes.

There was some audiophile reaction that suggested that a distortion
component too low to be inaudible by itself could still somehow audibly
affect the sound when mixed to the signal, but all the work on auditory
masking doesn't tend to support this theory.

--
Anahata
-+- http://www.treewind.co.uk
Home: 01638 720444 Mob: 07976 263827

[Scott Dorsey]

Anahata wrote:
I seem to remember first encountering the subtraction test in an article
in Wireless Word (if it still called that then) in the 1960s by Peter
Baxandall. In the application where he used it, on an audio power
amplifier, the difference signal, with only some simple passive HF phase/
delay compensation to get a null, amplified with the same gain as the
amplifier under test, was inaudible.

Right... but then comes the question of how much gain can you add to make
it audible.

I agree that audible results from a subtraction test need to be carefully
analysed, but if the result is silence, it surely tells us the amplifier
is good enough for all practical purposes.

I'll buy that. You have to have an actual speaker load on the thing, though.
not just a pure resistance.

There was some audiophile reaction that suggested that a distortion
component too low to be inaudible by itself could still somehow audibly
affect the sound when mixed to the signal, but all the work on auditory
masking doesn't tend to support this theory.

I don't buy that. But I do buy the theory that someone else may have a
lower threshold of hearing than I have.
--scott

--
"C'est un Nagra. C'est suisse, et tres, tres precis."

[Kevin Aylward[_4_]]

"Scott Dorsey" wrote in message ...

Kevin Aylward wrote:
The point being that if the system is nonlinear at all, it is provable
that there must be harmonic and intermodulation distortion. Conversely, if
the distortion for steady state signals is sufficiently low, then the
system
must be linear, subject to conditions I already mentioned.

Right, but how low is low?

Hint: the THD number is no longer useful.

I don't agree. THD is very, very useful. It is one of the most useful
measures of the linearity of an amplifier. However, one needs to apply it
correctly. I actually prefer IMD, because IMD identifies components not
harmonically related to either input signal, hence, more detectable.

Back in the days when amplifiers
were all similar topologies and distortion spectra were similar, it was
very
useful for comparing amplifiers. This is no longer the case.

I don't agree, and don't see any support for this claim.

I am not really discussing comparing amplifiers on a 1% verses 0.5%,
especially, at single operating levels. Its trivial that one can have a
single measurement at 0.1% and have it sounding much worse that an amp with
1%. I am referring to the design of a straight piece of wire with gain,
where attention has been made for distortion at all operating levels.

For example, one can do a single measurement at say, full power and obtain
low distortion, yet have gross x-over distortion, such that the amplifier
sounds dreadful on real music peaking at full power. However, this is a
cheat. When, I stated all IMD products I mean, all IMD at any operating
level. Lower levels distortion for such an amp can be huge, which is where a
music signal spends most of its time.

For the Taylor approach, we have:

Vout = sum An.Vin^n

If Vin = Vpk.Sin(wt), then the power terms generate a series harmonic
frequencies, via relations such as sin^2(x) = (1-cos(x)/2. If the system
is
linear, then no harmonic frequencies are
generated.

Right. But generation of harmonics depends on the nonlinearities involved.

Sure, its inherent in the Taylor expansion that the specific value of the
co-efficient result in different spectrums of harmonics.

I can build you a box with 2% distortion that you will have a hard time
identifying as being in-circuit or out of circuit. I can built you another
box with 0.01% distortion which is painfully obvious. The first box is
mostly third harmonic, the next box is mostly sixth.

Sure, it is possible to deliberately construct pathological/poor amplifiers
that have strange characteristics to prove a point for the purposes of
oneupmanship debate, but I am discussing real amplifiers, that are
competently designed. I could certainly construct an amplifier with a very
low on off spot level distortion, yet sound bad on audio tests.

As I already claimed, if an amplifier has 0.01% IMD, at all operating
levels, I don't accept that one can distinguish that amplifier from a
straight piece of wire. You would have to cite some evidence of actual tests
to support your view here. All evidence I am aware of, contradicts your
claim for general amplifiers so designed.


Likewise the THD measure is not useful for comparing systems of different
distortion spectra. Folks are working on weighted measures like the
Geddes-Lee test for that but we're not there yet.

I think you miss my point. Sure 0.5% second sounds way different that 0.5%
third. This is all well understood, and has been understood for many years.
My claim is not about the differences of larger level distortions, but a
claim that if the thd/imd distortion is *sufficiently* low, at *all*
operating levels, than the amplifier is a straight piece of wire with gain.

One needs to do distortion tests at various amplitudes. It is certainly
possible to quote misleading figures. One needs to account for 1mW and
100W
levels. If at all levels, its at the 0.01% IMD at 20khz/19khz, its a
straight piece of wire with gain.

Maybe, but at which point does that happen?

I cant say exactly, but the above figure is used because, it pretty much
impossible to design an amplifier that has pathological characteristics that
would generate say, larger distortion at lower frequencies, and a figure
that would minimise disagreements. I personally believe the figure is much
higher. Full power IMD at 19kh/20khz is a very severe test.

An amplifier specifically designed to be a straight piece of wire should
have massive amounts of feedback, such that lower frequencies, distortion is
effectively non existent. For example, my Studiomaster MOSFET 1000 design
had 0.005% THD at 20khz at 300W/8 ohm. Sure, at 500W/4ohm it went up to
0.03%, but at that frequency, a 4 ohm speaker load is way more than 4 ohms.
I don't actually know the distortion at 1khz because the output measured
0.0015% THD with the input measuring 0.0018% THD!

So, a pint of Guinness for the explanation for this apparent distortion
reduction characteristic.

The point, is that all things being equal, when one designs a decent DC
coupled power amp, things all go hand in hand. This is not true for say, the
design of tube guitar amps.


Regards

Kevin Aylward B.Sc.

www.kevinaylward.co.uk

[Scott Dorsey]

Kevin Aylward wrote:
"Scott Dorsey" wrote in message ...

Right. But generation of harmonics depends on the nonlinearities involved.

Sure, its inherent in the Taylor expansion that the specific value of the
co-efficient result in different spectrums of harmonics.

I can build you a box with 2% distortion that you will have a hard time
identifying as being in-circuit or out of circuit. I can built you another
box with 0.01% distortion which is painfully obvious. The first box is
mostly third harmonic, the next box is mostly sixth.

Sure, it is possible to deliberately construct pathological/poor amplifiers
that have strange characteristics to prove a point for the purposes of
oneupmanship debate, but I am discussing real amplifiers, that are
competently designed. I could certainly construct an amplifier with a very
low on off spot level distortion, yet sound bad on audio tests.

Real amplifiers like... say... a Dynaco ST120. Okay, maybe that's not
competently designed but it was typical of the design in the era and it
has a whole lot of high order even harmonic trash coming out of it.

As I already claimed, if an amplifier has 0.01% IMD, at all operating
levels, I don't accept that one can distinguish that amplifier from a
straight piece of wire. You would have to cite some evidence of actual tests
to support your view here. All evidence I am aware of, contradicts your
claim for general amplifiers so designed.

It's possible. I can provide a cite, though, showing amps with 0.01% THD
in bench tests that show audible distortion. Of course, there are plenty
with that level (and a lot with much higher levels) that are not.

Likewise the THD measure is not useful for comparing systems of different
distortion spectra. Folks are working on weighted measures like the
Geddes-Lee test for that but we're not there yet.

I think you miss my point. Sure 0.5% second sounds way different that 0.5%
third. This is all well understood, and has been understood for many years.
My claim is not about the differences of larger level distortions, but a
claim that if the thd/imd distortion is *sufficiently* low, at *all*
operating levels, than the amplifier is a straight piece of wire with gain.

Right. But at what point is it "sufficiently low?" 0.1%? 0.01%? 0.001?

Remember, 0.01% THD means the total sum of harmonics is 40 dB down. That's
pretty far down. But if it's all concentrated in one harmonic product,
it's going to be audible at 40 dB down.

I cant say exactly, but the above figure is used because, it pretty much
impossible to design an amplifier that has pathological characteristics that
would generate say, larger distortion at lower frequencies, and a figure
that would minimise disagreements. I personally believe the figure is much
higher. Full power IMD at 19kh/20khz is a very severe test.

The larger distortion at lower frequencies thing is actually very common
and a result of using undersized electrolytic coupling caps. Doug Self
actually has a nice discussion of this in his amp design handbook. Turns
out that the capacitor values required for low distortion at low frequencies
are a lot higher than the values required for flat response.

My experience is that it can also be a sign of pollution on power supply
rails due to insufficient decoupling.

An amplifier specifically designed to be a straight piece of wire should
have massive amounts of feedback, such that lower frequencies, distortion is
effectively non existent. For example, my Studiomaster MOSFET 1000 design
had 0.005% THD at 20khz at 300W/8 ohm. Sure, at 500W/4ohm it went up to
0.03%, but at that frequency, a 4 ohm speaker load is way more than 4 ohms.
I don't actually know the distortion at 1khz because the output measured
0.0015% THD with the input measuring 0.0018% THD!

Massive amounts of feedback can be a good thing when the poles and zeroes
are in the right place, but it can also be trouble when driving bizarre
speaker loads, and a lot of speaker loads are bizarre. But yes, I agree
that feedback is a wonderful thing. I also agree that the system needs to
be linear in the first place before you add feedback.

So, a pint of Guinness for the explanation for this apparent distortion
reduction characteristic.

Bandlimiting, I would guess first off.

The point, is that all things being equal, when one designs a decent DC
coupled power amp, things all go hand in hand. This is not true for say, the
design of tube guitar amps.

Oh, this is absolutely true.
--scott

--
"C'est un Nagra. C'est suisse, et tres, tres precis."

[PStamler]

On Feb 1, 9:17*am, "Kevin Aylward"
wrote:
I don't actually know the distortion at 1khz because the output measured
0.0015% THD with the input measuring 0.0018% THD!

So, a pint of Guinness for the explanation for this apparent distortion
reduction characteristic.

Perhaps the amplifier generates harmonics in opposite poarity to those
in the signal generator, and they partially cancel?

Peace,
Paul

[PStamler]

Another limitation of continuous distortion tests -- of any sort -- is
that they don't measure certain forms of misbehavior that amplifiers
exhibit when driving speakers. A good example is what happens when
loudspeakers are fed a succession of low-frequency impulses in rapid
succession; the speaker can react in a way that draws excessive
current from an amplifier. A nominal 6-ohm speaker can act more like 2
ohms. Couple that with an amplifier that has a saggy power supply
after several impulses, and you can get real problems that *don't*
show up on continuous-signal tests. And to those who call this an
artificial torture test, not likely to happen in the real world, I
point to kickdrums.

Bob Cordell has a discussion of this in his excellent new book on
solid-state power amplifier design. He identifies other problems not
caught by the standard THD and IMD tests.

I'm not defending mysticism here, just saying that one particular set
of tests won't completely characterize the performance of an audio
device.

Peace,
Paul

[Kevin Aylward[_4_]]

"Scott Dorsey" wrote in message ...

Kevin Aylward wrote:
"Scott Dorsey" wrote in message ...

Right. But generation of harmonics depends on the nonlinearities
involved.

Sure, its inherent in the Taylor expansion that the specific value of the
co-efficient result in different spectrums of harmonics.

I can build you a box with 2% distortion that you will have a hard time
identifying as being in-circuit or out of circuit. I can built you
another
box with 0.01% distortion which is painfully obvious. The first box is
mostly third harmonic, the next box is mostly sixth.

Sure, it is possible to deliberately construct pathological/poor
amplifiers
that have strange characteristics to prove a point for the purposes of
oneupmanship debate, but I am discussing real amplifiers, that are
competently designed. I could certainly construct an amplifier with a very
low on off spot level distortion, yet sound bad on audio tests.

Real amplifiers like... say... a Dynaco ST120. Okay, maybe that's not
competently designed but it was typical of the design in the era and it
has a whole lot of high order even harmonic trash coming out of it.

I don't know about this amp. It might have slewing problems. So many earlier
amps just did not have full power high frequency response, which could be an
issue on some types of audio signals.


Likewise the THD measure is not useful for comparing systems of
different
distortion spectra. Folks are working on weighted measures like the
Geddes-Lee test for that but we're not there yet.

I think you miss my point. Sure 0.5% second sounds way different that 0.5%
third. This is all well understood, and has been understood for many
years.
My claim is not about the differences of larger level distortions, but a
claim that if the thd/imd distortion is *sufficiently* low, at *all*
operating levels, than the amplifier is a straight piece of wire with
gain.

Right. But at what point is it "sufficiently low?" 0.1%? 0.01%? 0.001?

Somewhere above 0.01% in my view. One has to note that speakers also have a
lot a distortion.


The larger distortion at lower frequencies thing is actually very common
and a result of using undersized electrolytic coupling caps. Doug Self
actually has a nice discussion of this in his amp design handbook. Turns
out that the capacitor values required for low distortion at low
frequencies
are a lot higher than the values required for flat response.

I am quite aware of Douglas Self. He is very knowledgeable. Well, "low
distortion" here is a bit undefined. Typically, its going to be 0.001%
over typical ranges of voltages on "small signal" coupling capacitors. So in
general, coupling capacitors is not a real problem. Douglas does present
some data (http://www.douglas-self.com/ampins/dipa/dipa.htm#2) on a speaker
output coupling capacitors, showing a midband rise to 0.003%, which is a
significant increases on his "blameless amplifier", but this is still so
low, that is not audibly significant in my view. I used a 5Hz LF point for
my Mosfet 1000 amp to avoid this potential issue though.

Actually, Douglas is a bit of a conundrum for me. He clearly know a lot, but
his "blameless amp" is not the "best/optimum" architecture in my view, if
"best/optimum" can actually be defined. His second stage is not
differential, and uses a fixed constant current load. A differential,
current mirror loaded second stage gives inherent lower distortion for that
second stage, and gives a somewhat "better" push pull drive to the next
stage. A fixed current load mean that the gain transistor is always tugging
at that current source.


My experience is that it can also be a sign of pollution on power supply
rails due to insufficient decoupling.

An amplifier specifically designed to be a straight piece of wire should
have massive amounts of feedback, such that lower frequencies, distortion
is
effectively non existent. For example, my Studiomaster MOSFET 1000 design
had 0.005% THD at 20khz at 300W/8 ohm. Sure, at 500W/4ohm it went up to
0.03%, but at that frequency, a 4 ohm speaker load is way more than 4
ohms.
I don't actually know the distortion at 1khz because the output measured
0.0015% THD with the input measuring 0.0018% THD!

Massive amounts of feedback can be a good thing when the poles and zeroes
are in the right place, but it can also be trouble when driving bizarre
speaker loads, and a lot of speaker loads are bizarre.

One has to be careful in the design of the loop, but using the right output
network to disconnect the load at HF, usually works ok for pretty much any
load. I could certainly run my Mosfet 1000 into a 2uf capacitor. In fact, I
was quite chuffed that I managed with only a 22nf/10ohm output damper. Many
amps use a 0.1uf. Testing at 200Khz full power is more problematic when
driving a 0.1uf

But yes, I agree
that feedback is a wonderful thing. I also agree that the system needs to
be linear in the first place before you add feedback.

Yes and no. I agree that feedback won't help for slew rate limiting,
essentially, because the gain tends to zero. If there aint enough current,
then there is not enough current. It can help sometimes to have lower open
loop distortion, but I have done quite a few actual measurements on this,
and as far as the numbers go, having more local feedback, usually results in
more total distortion.

For example, suppose one puts in emitter resisters in the input pair. This
linearises the input pair a fair bit, but because, all things being equal,
it also reduces the overall loop gain, resulting in larger distortion of the
whole amplifier. The input pair does not usually dominate the distortion of
a power amp. In the bigger picture, having some emitter degradation can
effectively reduce distortion at HF, in the sense that it can allow the
amplifier to be designed to have a higher slew rate by reducing the loop
gain.


So, a pint of Guinness for the explanation for this apparent distortion
reduction characteristic.

Bandlimiting, I would guess first off.

THD equipment measurers "distortion and noise". The oscillator signal was
around 1V, the amp output was around 50V...

So, I estimate that the amp was dong 0.001%

The point, is that all things being equal, when one designs a decent DC
coupled power amp, things all go hand in hand. This is not true for say,
the
design of tube guitar amps.

Oh, this is absolutely true.

One common fallacy is that tube guitar amps (essentially class B) when
overloaded generate even distortion. One only has to apply a signal and look
on a scope to see that this is nonsense. One actually sees huge x-over
distortion. Grids start taking current and nasty things start happening.

Regards

Kevin Aylward B.Sc.

www.kevinaylward.co.uk

[Scott Dorsey]

Kevin Aylward wrote:
"Scott Dorsey" wrote in message ...

Real amplifiers like... say... a Dynaco ST120. Okay, maybe that's not
competently designed but it was typical of the design in the era and it
has a whole lot of high order even harmonic trash coming out of it.

I don't know about this amp. It might have slewing problems. So many earlier
amps just did not have full power high frequency response, which could be an
issue on some types of audio signals.

It has severe slewing problems. Also a lot of other interesting issues...
the power supply is undersized so noise from the output stage on the rails
winds up in the input stage... and that noise is mostly even harmonics.
The 1 KHz THD numbers are great and the supply stays more or less steady,
but put something with high and low frequencies into it at the same time
and the supply rails look like the Swiss Alps on a scope.

It's also a unipolar supply with all the issues that brings up.

My claim is not about the differences of larger level distortions, but a
claim that if the thd/imd distortion is *sufficiently* low, at *all*
operating levels, than the amplifier is a straight piece of wire with
gain.

Right. But at what point is it "sufficiently low?" 0.1%? 0.01%? 0.001?

Somewhere above 0.01% in my view. One has to note that speakers also have a
lot a distortion.

Yes, and the speaker distortion in almost every case is going to swamp the
amplifier distortion. Still, you admit that the difference between .1% and
..01% distortion is audible... and the problem is that it's audible on
speakers that might be rated for as much as 5% distortion. That's because
the distortion spectra of the speakers and the amplifiers are different.

I am quite aware of Douglas Self. He is very knowledgeable. Well, "low
distortion" here is a bit undefined. Typically, its going to be 0.001%
over typical ranges of voltages on "small signal" coupling capacitors. So in
general, coupling capacitors is not a real problem. Douglas does present
some data (http://www.douglas-self.com/ampins/dipa/dipa.htm#2) on a speaker
output coupling capacitors, showing a midband rise to 0.003%, which is a
significant increases on his "blameless amplifier", but this is still so
low, that is not audibly significant in my view. I used a 5Hz LF point for
my Mosfet 1000 amp to avoid this potential issue though.

The problem is that it builds up. You get .001% distortion from a coupling
cap, then you put a couple hundred of them in the signal path and the end
result is a lot of distortion. Not a huge issue for a small power amp but
a big one for a mixing console.

But... let's take a degenerate case to show why a single scalar is
not useful. Pick a Phase Linear 700 for instance.... non-complementary
output stage built with TV horizontal output transistors not known for
their linearity. Lots of crossover distortion because the bottom part
of the curve is so ragged. Lots and lots of feedback which helps linearize
the amp to a great extent.

Full power THD on the amp is a really nice low number, something in the .001%
range as I recall. However, when you run the amp down to the one or two watt
level, the dead band resulting from the crossover issues becomes a much
larger portion of your signal and the total distortion percentage goes through
the roof.

Yeah, you could model the whole system as a Volterra series and publish it
on the datasheet if you wanted to, but that probably wouldn't help potential
purchasers any more than the single scalar does.

actually, Douglas is a bit of a conundrum for me. He clearly know a lot, but
his "blameless amp" is not the "best/optimum" architecture in my view, if
"best/optimum" can actually be defined. His second stage is not
differential, and uses a fixed constant current load. A differential,
current mirror loaded second stage gives inherent lower distortion for that
second stage, and gives a somewhat "better" push pull drive to the next
stage. A fixed current load mean that the gain transistor is always tugging
at that current source.

He actually talks about the advantages and disadvantages of making the
second stage differential in tha latest edition of his book. He claims it
doesn't actually buy a reduction in distortion in the long run. He is an
odd character and he is set in his ways in a lot of cases, but although he
sometimes has bizarre opinions he always has good numbers to back them up.

Massive amounts of feedback can be a good thing when the poles and zeroes
are in the right place, but it can also be trouble when driving bizarre
speaker loads, and a lot of speaker loads are bizarre.

One has to be careful in the design of the loop, but using the right output
network to disconnect the load at HF, usually works ok for pretty much any
load. I could certainly run my Mosfet 1000 into a 2uf capacitor. In fact, I
was quite chuffed that I managed with only a 22nf/10ohm output damper. Many
amps use a 0.1uf. Testing at 200Khz full power is more problematic when
driving a 0.1uf

A 2 uF capacitor is.... about like a stacked pair of original Quad ESLs....
granted that's a pretty crazy load but to my mind you can never have enough
margins....

Yes and no. I agree that feedback won't help for slew rate limiting,
essentially, because the gain tends to zero. If there aint enough current,
then there is not enough current. It can help sometimes to have lower open
loop distortion, but I have done quite a few actual measurements on this,
and as far as the numbers go, having more local feedback, usually results in
more total distortion.

For example, suppose one puts in emitter resisters in the input pair. This
linearises the input pair a fair bit, but because, all things being equal,
it also reduces the overall loop gain, resulting in larger distortion of the
whole amplifier. The input pair does not usually dominate the distortion of
a power amp. In the bigger picture, having some emitter degradation can
effectively reduce distortion at HF, in the sense that it can allow the
amplifier to be designed to have a higher slew rate by reducing the loop
gain.

Yes. It is a constant juggling act. That is what makes it interesting.
--scott

--
"C'est un Nagra. C'est suisse, et tres, tres precis."

[Kevin Aylward[_4_]]

"PStamler" wrote in message
...

Another limitation of continuous distortion tests -- of any sort -- is
that they don't measure certain forms of misbehavior that amplifiers
exhibit when driving speakers. A good example is what happens when
loudspeakers are fed a succession of low-frequency impulses in rapid
succession; the speaker can react in a way that draws excessive
current from an amplifier. A nominal 6-ohm speaker can act more like 2
ohms.

I think this is probably a bit excessive at a 3:1 ratio, but sure, one can
generate pulse waveforms into a simple model of a speaker and show larger
peak spike currents than that indicated Vpk/RDC with certain waveforms, e.g.
see http://www.epanorama.net/documents/a...impedance.html for the
model. A simple model, for square wave inputs would give short pulse spikes
of twice Vpk/Rdc, i.e. a simple HP filter on RDC and Cmes type effect.

This can still be evaluated from simple steady state, e.g. bursts of sine
and square waves.

Couple that with an amplifier that has a saggy power supply
after several impulses, and you can get real problems that *don't*
show up on continuous-signal tests. And to those who call this an
artificial torture test, not likely to happen in the real world, I
point to kickdrums.

I think this may miss a basic point. By construction, one needs to operate
the amplifier within its non overload region. By construction, an amplifier
needs to be designed such that it can reach its rated power into a rated
impedance e.g. 4 ohms or 8 ohms. By construction, speakers rated at say, 4
ohm should have a minimum resistance of 4 ohms. Typically they may be a bit
larger, e.g. 4.5. However, a decent amp always has a certain amount of
transient overload current available. My mosfet 1000 could do about twice
rated current continuously, until the thermal came in. So, one can indeed
use steady state tests, they just need to be done into about 1/2 the rated
load resistance!

Actually, the kick drum is not a great example for your point. To achieve
the above twice nominal continuous peak currents requires continuous fast
edges for the main waveform. The basic frequencies of drums are relatively
low so this effect is not so severe.

A saggy power supply is actually better. If measurements have been made in
steady state, the power supply will be at its lowest point, such that an
amplifier rated at for steady state conditions, will be able to more volts
for short busts.

Bob Cordell has a discussion of this in his excellent new book on
solid-state power amplifier design. He identifies other problems not
caught by the standard THD and IMD tests.

The point I am making is that, if one competently designs an amplifier with
very low THD/IMD, despite the fact that, in principle there may be other
issues, they are all part an parcel of achieving such an amp.

I'm not defending mysticism here, just saying that one particular set
of tests won't completely characterize the performance of an audio
device.

In principle possibly, but in practice, THD/IMD over all operating
frequencies and output levels (including overload), and some burst tests
will, if one is designing a straight piece of wire with gain, should be
enough. Achieving such 0.01% levels has actually been achievable with
trivial designs for at least 30 years.

Regards

Kevin Aylward B.Sc.

www.kevinaylward.co.uk

[Anahata]

On Wed, 02 Feb 2011 15:17:27 +0000, Kevin Aylward wrote:

an amplifier needs to be designed such that it can reach its rated power
into a rated impedance e.g. 4 ohms or 8 ohms.

With some amplifiers that only works for a pure resistive load.

By construction, speakers
rated at say, 4 ohm should have a minimum resistance of 4 ohms.
Typically they may be a bit larger, e.g. 4.5.

In practice, the crossover/speaker combination often goes below the
nominal impedance of the speaker. For a start, the DC resistance is
usually less, but the RC network may present a low impedances at one
frequency.

However, a decent amp
always has a certain amount of transient overload current available. My
mosfet 1000 could do about twice rated current continuously, until the
thermal came in.

That's good conservative design, but not always found in commercially
available amplifiers. I've seen (and heard!) a supposedly professional
grade power amplifier go into foldback limiting with a speaker whose
impedance was very reactive at some frequencies.

This is all beside the point though, as I think Paul was suggesting
audible LF artifacts appearing at the output as a result, perhaps, of
power supply or bias chain modulation, not failure of the amplifier to
deliver the required current. If a repeated mid/HF toneburst causes
thumping noises in the speakers, that's a kind of distortion, but it's
not necessarily caused by current limiting, and nor can it be revealed by
steady state tone tests.

--
Anahata
--/-- http://www.treewind.co.uk
+44 (0)1638 720444

[Kevin Aylward[_4_]]

"Scott Dorsey" wrote in message ...

Kevin Aylward wrote:
"Scott Dorsey" wrote in message ...

Real amplifiers like... say... a Dynaco ST120. Okay, maybe that's not
competently designed but it was typical of the design in the era and it
has a whole lot of high order even harmonic trash coming out of it.

I don't know about this amp. It might have slewing problems. So many
earlier
amps just did not have full power high frequency response, which could be
an
issue on some types of audio signals.

It has severe slewing problems. Also a lot of other interesting issues...
the power supply is undersized so noise from the output stage on the rails
winds up in the input stage... and that noise is mostly even harmonics.
The 1 KHz THD numbers are great and the supply stays more or less steady,
but put something with high and low frequencies into it at the same time
and the supply rails look like the Swiss Alps on a scope.

It's also a unipolar supply with all the issues that brings up.

My claim is not about the differences of larger level distortions, but
a
claim that if the thd/imd distortion is *sufficiently* low, at *all*
operating levels, than the amplifier is a straight piece of wire with
gain.

Right. But at what point is it "sufficiently low?" 0.1%? 0.01%?
0.001?

Somewhere above 0.01% in my view. One has to note that speakers also have
a
lot a distortion.

Yes, and the speaker distortion in almost every case is going to swamp the
amplifier distortion. Still, you admit that the difference between .1% and
.01% distortion is audible...

Actually, I claim that below 0.01%, at all power levels, it is not
detectable. I don't really make a clear claim for figures above that, but I
am of the opinion that if IMD at *all* reasonable power levels and
frequencies is below 0.05%, it is also not detectable. I other words,
someone would have to give me proof that they could detect such distortions.

and the problem is that it's audible on
speakers that might be rated for as much as 5% distortion. That's because
the distortion spectra of the speakers and the amplifiers are different.

I am quite aware of Douglas Self. He is very knowledgeable. Well, "low
distortion" here is a bit undefined. Typically, its going to be 0.001%
over typical ranges of voltages on "small signal" coupling capacitors. So
in
general, coupling capacitors is not a real problem. Douglas does present
some data (http://www.douglas-self.com/ampins/dipa/dipa.htm#2) on a
speaker
output coupling capacitors, showing a midband rise to 0.003%, which is a
significant increases on his "blameless amplifier", but this is still so
low, that is not audibly significant in my view. I used a 5Hz LF point for
my Mosfet 1000 amp to avoid this potential issue though.

The problem is that it builds up. You get .001% distortion from a coupling
cap, then you put a couple hundred of them in the signal path and the end
result is a lot of distortion. Not a huge issue for a small power amp but
a big one for a mixing console.

Well, 100 in in the path is a probably bit of an over estimate, no one
really wants to spend that sort of build cost, but sure, no one sets the
turnover at 20Hz if the whole system is to be specked at 20Hz - 20Khz +/-1db
anyway.

But... let's take a degenerate case to show why a single scalar is
not useful. Pick a Phase Linear 700 for instance.... non-complementary
output stage built with TV horizontal output transistors not known for
their linearity. Lots of crossover distortion because the bottom part
of the curve is so ragged. Lots and lots of feedback which helps linearize
the amp to a great extent.
Full power THD on the amp is a really nice low number, something in the
.001%
range as I recall. However, when you run the amp down to the one or two
watt
level, the dead band resulting from the crossover issues becomes a much
larger portion of your signal and the total distortion percentage goes
through
the roof.
Yeah, you could model the whole system as a Volterra series and publish it
on the datasheet if you wanted to, but that probably wouldn't help
potential
purchasers any more than the single scalar does.

And that's my point. I don't suggest a single point at full output. One
needs at least two points to include a value that reflects x-over
distortion. One also needs to know that the amp does not have some
pathological behaviour that makes an implication of performance not actually
specified, false.

I don't think anyone really cares today about the numbers for the area I am
interested in. i.e. P.A amplifiers for live band work. Many manufactures
don't quote any/many numbers nowadays. 30 years ago, it was more of a
numbers game. Today, most commercial PA amplifiers are worse than that was
achieved 30 years ago.

actually, Douglas is a bit of a conundrum for me. He clearly know a lot,
but
his "blameless amp" is not the "best/optimum" architecture in my view, if
"best/optimum" can actually be defined. His second stage is not
differential, and uses a fixed constant current load. A differential,
current mirror loaded second stage gives inherent lower distortion for
that
second stage, and gives a somewhat "better" push pull drive to the next
stage. A fixed current load mean that the gain transistor is always
tugging
at that current source.

He actually talks about the advantages and disadvantages of making the
second stage differential in tha latest edition of his book. He claims it
doesn't actually buy a reduction in distortion in the long run.

I am a great believer in differential whenever possible. In principle, it
can help with the issue you note above about distortion on the PS rails. I
also prefer the push/pull drive to the buffer stages that such a topology
gives.

He is an
odd character and he is set in his ways in a lot of cases, but although he
sometimes has bizarre opinions he always has good numbers to back them up.


Actually, I just discovered he lists me twice on his "notable letters to
Electronics World" section at:

http://www.douglas-self.com/ampins/l.../lettersWW.htm


Kevin Aylward B.Sc.
www.kevinaylward.co.uk

[Kevin Aylward[_4_]]

"anahata" wrote in message
...

On Wed, 02 Feb 2011 15:17:27 +0000, Kevin Aylward wrote:

an amplifier needs to be designed such that it can reach its rated power
into a rated impedance e.g. 4 ohms or 8 ohms.

With some amplifiers that only works for a pure resistive load.

Well, a poor amplifier might current limit at say, 20% above rated output,
but realistically I don't think that is a "competent" design. In practice,
one needs a certain margin just to account for component tolerances.

By construction, speakers
rated at say, 4 ohm should have a minimum resistance of 4 ohms.
Typically they may be a bit larger, e.g. 4.5.

Sure, a speaker might be as low as 3 ohms for a 4 ohm rating, possibly.
Although wiring it up might well add a 0.1 - 0.5 ohms or so.

In practice, the crossover/speaker combination often goes below the
nominal impedance of the speaker. For a start, the DC resistance is
usually less, but the RC network may present a low impedances at one
frequency.

Actually, very very, unlikely for small signal impedance of the speaker
itself. I have never seen a plot of loudspeaker impedance go below its DC
resistance. However, with specific pulse wave shapes applied, pulse currents
larger than that implied by V/Rdc can still occur.

Its debatable just what the max pulse can be. One has to time an
asymmetrical waveform just right to get the real peak, which theoretically
could be quite a large ratio. In practice, music signals just don't have the
right characteristics to do this.

However, a decent amp
always has a certain amount of transient overload current available. My
mosfet 1000 could do about twice rated current continuously, until the
thermal came in.

That's good conservative design, but not always found in commercially
available amplifiers. I've seen (and heard!) a supposedly professional
grade power amplifier go into foldback limiting with a speaker whose
impedance was very reactive at some frequencies.

I am not an advocate of VI limiters. I prefer a hard current limit, with
thermal cut-out.

This is all beside the point though, as I think Paul was suggesting
audible LF artifacts appearing at the output as a result, perhaps, of
power supply or bias chain modulation, not failure of the amplifier to
deliver the required current. If a repeated mid/HF toneburst causes
thumping noises in the speakers, that's a kind of distortion, but it's
not necessarily caused by current limiting, and nor can it be revealed by
steady state tone tests.

It can be revealed by simple, easily available pulse wave forms though, e.g.
burst sine and square. My point is that there is nothing magic about music
signals. Everything of relevance can be tested with standard test equipment.


Kevin Aylward B.Sc.
www.kevinaylward.co.uk

[Scott Dorsey]

Kevin Aylward wrote:
"Scott Dorsey" wrote in message ...

The problem is that it builds up. You get .001% distortion from a coupling
cap, then you put a couple hundred of them in the signal path and the end
result is a lot of distortion. Not a huge issue for a small power amp but
a big one for a mixing console.

Well, 100 in in the path is a probably bit of an over estimate, no one
really wants to spend that sort of build cost, but sure, no one sets the
turnover at 20Hz if the whole system is to be specked at 20Hz - 20Khz +/-1db
anyway.

Try around 220 in an SSL 4000 with a typical configuration. That includes
mike in, channel strip, master module, send, return, channel strip,
master module, and main out. This is how the problems build up. Every
op-amp module has a little distortion, every op-amp has a blocking cap,
and the problems blow up.

I was with an AES tour of the mastering room at Brooklyn Phono, and
one of the fellows in the tour of an audiophile nature asked the
mastering engineer about the Neumann cutting amps. "Are there any
capacitors in the signal path?" Paul, the engineer went blank for
a second just at the nature of the question. "Oh yeah, there must be
a million of them" he said. A million is exaggerating, but there are
a whole lot of stages in there and they are all AC-coupled.

And that's my point. I don't suggest a single point at full output. One
needs at least two points to include a value that reflects x-over
distortion. One also needs to know that the amp does not have some
pathological behaviour that makes an implication of performance not actually
specified, false.

If you're going to do that, why not just look at a full spectrum? It
tells you so much more information.

I don't think anyone really cares today about the numbers for the area I am
interested in. i.e. P.A amplifiers for live band work. Many manufactures
don't quote any/many numbers nowadays. 30 years ago, it was more of a
numbers game. Today, most commercial PA amplifiers are worse than that was
achieved 30 years ago.

Well, in that application, speaker distortion levels are still insanely
high, too. And, issues like efficiency and shipping weight become paramount
in ways that they are not for studio amplifiers.
--scott

--
"C'est un Nagra. C'est suisse, et tres, tres precis."

[Kevin Aylward[_4_]]

"Scott Dorsey" wrote in message ...

Kevin Aylward wrote:
"Scott Dorsey" wrote in message ...

The problem is that it builds up. You get .001% distortion from a
coupling
cap, then you put a couple hundred of them in the signal path and the
end
result is a lot of distortion. Not a huge issue for a small power amp
but
a big one for a mixing console.

Well, 100 in in the path is a probably bit of an over estimate, no one
really wants to spend that sort of build cost, but sure, no one sets the
turnover at 20Hz if the whole system is to be specked at 20Hz - 20Khz
+/-1db
anyway.

Try around 220 in an SSL 4000 with a typical configuration. That includes
mike in, channel strip, master module, send, return, channel strip,
master module, and main out.

???

That's 8. What else that can add up to 220 ?

Anyway...

This is how the problems build up. Every
op-amp module has a little distortion, every op-amp has a blocking cap,
and the problems blow up.

Sure there is a build up, in principle. In fact, it was a basic problem in
analogue telephony in the 20s-30s. It was, pretty much why feedback was
invented by Black at Bell labs. To get the final 5% - 10% crap distortion we
all love, when the phone system goes through 100s of repeaters, needed
0.01% distortion for each amplifier, and this at 60Khz-120Khz modulated
carrier frequencies.

And that's my point. I don't suggest a single point at full output. One
needs at least two points to include a value that reflects x-over
distortion. One also needs to know that the amp does not have some
pathological behaviour that makes an implication of performance not
actually
specified, false.

If you're going to do that, why not just look at a full spectrum? It
tells you so much more information.

The point is that, by and large, HF always has the most distortion i.e. the
max distortion. If this max is low enough, then its superfluous what it
actually is at lower frequencies as it will be lower.

Of course, in practice one does do tests at all frequencies as a check.

Actually, today with the extremely, extremely expensive Cadence software I
use at work in my day job, physical measurements are also quite superfluous.
One can design completely in the virtual world, and I usually do. Running a
Spectre R.F. steady state transient analysis using either harmonic balance
or shooting method, can take less than a minute on a full amplifier to
obtain full frequency spectrums of distortion, and THD and IMD and time
domain plots. I wish I had access to this kit 30 years ago. Its truly
wonderful. e.g. running parameter sweeps of AB bias current and emitter
resisters all in one go. Simulation tools have really came a long way.
Unfortunately, my own SuperSpice (www.anasoft.co.uk) does not do these
methods. Getting distortion by the usual methods needs really long transient
runs.

I don't think anyone really cares today about the numbers for the area I
am
interested in. i.e. P.A amplifiers for live band work. Many manufactures
don't quote any/many numbers nowadays. 30 years ago, it was more of a
numbers game. Today, most commercial PA amplifiers are worse than that was
achieved 30 years ago.

Well, in that application, speaker distortion levels are still insanely
high, too. And, issues like efficiency and shipping weight become
paramount
in ways that they are not for studio amplifiers.


Indeed, that is exactly my current priority. I want a loud, light and small
speaker system to fit my MPV car. Its the thing I dislike the most about
doing gigs. Carrying gear, and trying to pack it all in.

Regards

Kevin Aylward B.Sc.

www.kevinaylward.co.uk

[Scott Dorsey]

Kevin Aylward wrote:
"Scott Dorsey" wrote in message ...

Try around 220 in an SSL 4000 with a typical configuration. That includes
mike in, channel strip, master module, send, return, channel strip,
master module, and main out.

???

That's 8. What else that can add up to 220 ?

Each one of those sections has a dozen or two op-amps in the signal path,
in part because there's a whole lot of fancy routing stuff in there. Nobody
every got an SSL because it sounded good, they got it because it had very
powerful routing.

This is how the problems build up. Every
op-amp module has a little distortion, every op-amp has a blocking cap,
and the problems blow up.

Sure there is a build up, in principle. In fact, it was a basic problem in
analogue telephony in the 20s-30s. It was, pretty much why feedback was
invented by Black at Bell labs. To get the final 5% - 10% crap distortion we
all love, when the phone system goes through 100s of repeaters, needed
0.01% distortion for each amplifier, and this at 60Khz-120Khz modulated
carrier frequencies.

Precisely. It's not a new problem.

And that's my point. I don't suggest a single point at full output. One
needs at least two points to include a value that reflects x-over
distortion. One also needs to know that the amp does not have some
pathological behaviour that makes an implication of performance not
actually
specified, false.

If you're going to do that, why not just look at a full spectrum? It
tells you so much more information.

The point is that, by and large, HF always has the most distortion i.e. the
max distortion. If this max is low enough, then its superfluous what it
actually is at lower frequencies as it will be lower.

I worry about all measurable products, because as mentioned above they have
a tendency to build up throughout the signal chain. Also I worry about things
like amplifiers because they need to be measured with an actual real-world
load and they usually are not.

Actually, today with the extremely, extremely expensive Cadence software I
use at work in my day job, physical measurements are also quite superfluous.
One can design completely in the virtual world, and I usually do. Running a
Spectre R.F. steady state transient analysis using either harmonic balance
or shooting method, can take less than a minute on a full amplifier to
obtain full frequency spectrums of distortion, and THD and IMD and time
domain plots. I wish I had access to this kit 30 years ago. Its truly
wonderful. e.g. running parameter sweeps of AB bias current and emitter
resisters all in one go. Simulation tools have really came a long way.
Unfortunately, my own SuperSpice (www.anasoft.co.uk) does not do these
methods. Getting distortion by the usual methods needs really long transient
runs.

Simulation gives me the willies. It's good for a first cut design, but
every time I have dealt with a simulated design it has always needed some
tweaking when actually implemented, because the simulation is never quite
faithful.

On the other hand, it's possible to do simulations and see what the
consequences of part variations are very easily, and it's hard to do that
in the real world without actually going into production.
--scott
--
"C'est un Nagra. C'est suisse, et tres, tres precis."

[Kevin Aylward[_4_]]

"Scott Dorsey" wrote in message ...


Actually, today with the extremely, extremely expensive Cadence software I
use at work in my day job, physical measurements are also quite
superfluous.
One can design completely in the virtual world, and I usually do. Running
a
Spectre R.F. steady state transient analysis using either harmonic balance
or shooting method, can take less than a minute on a full amplifier to
obtain full frequency spectrums of distortion, and THD and IMD and time
domain plots. I wish I had access to this kit 30 years ago. Its truly
wonderful. e.g. running parameter sweeps of AB bias current and emitter
resisters all in one go. Simulation tools have really came a long way.
Unfortunately, my own SuperSpice (www.anasoft.co.uk) does not do these
methods. Getting distortion by the usual methods needs really long
transient
runs.

I should clarify, class B x-over distortion is problematic without periodic
steady state analysis. General class A, small signal distortion is perfectly
ok in spice.

Simulation gives me the willies. It's good for a first cut design, but
every time I have dealt with a simulated design it has always needed some
tweaking when actually implemented, because the simulation is never quite
faithful.

I have to disagree here. I design relatively complex analogue/mixed mode
system on chip i.c.s. It might have say 10,000 analogue transistors, doing
all sorts of things. Mask and fab costs ( $100k) and turnaround times ( 2
months ) are such that the chips need to be fully functional 1st time pass,
and production ready second time pass. This is indeed usually achieved as
the norm. In fact, its not unknown to get 2% of sales (say $10M) bonus for a
1st time production ready pass!

My experience is that errors occur, not because of simulation errors, but
simply because a certain simulation was not run at all, or a specification
was overlooked. Models and tools are so good nowadays, that the often quoted
tweaking on the bench is simply not necessary, and impossible for an ic
design. I haven't bench tweaked a design for over 20 years. I have found
faults on the bench, but I have always been able to replicate them in
simulation, and fix them entirely in the virtual world.

On the other hand, it's possible to do simulations and see what the
consequences of part variations are very easily, and it's hard to do that
in the real world without actually going into production.

Yes.

Kevin Aylward B.Sc.
www.kevinaylward.co.uk