Feedback in audio esp wrt op-amps.

[Eeyore]

There was part of a thread a while back about how adding negative feedback can
create higher order harmonic distortion products than exist open-loop in an
amplifier stage.

This made me think about the application of op-amps in audio generally. Negative
feedback is used primarily to linearise the transfer function and is used in
huge quantites as much as 80dB @ 1 kHz for example.

Since this amount of NFB is not required to provide an accurate gain setting, it
struck me that it's somewhat counter productive. If instead the open-loop
transfer characteritic was made more linear by degeneration of the open-loop
gain for example, when NFB is applied, the overall result should be largely
similar (i.e. no worse) but would presumably also suffer less from the creation
of these new distortion products .

Comments ?

Graham

 

[William Sommerwerck]

There was part of a thread a while back about how adding negative feedback
can create higher order harmonic distortion products than exist open-loop
in
an amplifier stage.

This made me think about the application of op-amps in audio generally.
Negative feedback is used primarily to linearise the transfer function and
is used in huge quantites as much as 80dB @ 1 kHz for example.

Since this amount of NFB is not required to provide an accurate gain
setting,
it struck me that it's somewhat counter productive. If instead the
open-loop
transfer characteritic was made more linear by degeneration of the
open-loop
gain for example, when NFB is applied, the overall result should be
largely
similar (i.e. no worse) but would presumably also suffer less from the
creation
of these new distortion products .

Comments ?

That negative feedback linearizes the transfer function at the expensive of
adding higher-order harmonics has been long-known. What you say is perfectly
logical.

However, the presence of higher-order harmonics is not the only factor, but
their amplitude. Below a certain percentage (I'm sure Arny will be able to
tell us what that is), they're inaudible.

A good op amp can be used as a buffer and be sonically transparent, its
output indistinguishable from its input.

 

[MooseFET]

On Aug 19, 8:39 am, Eeyore <rabbitsfriendsandrelati...@hotmail.com>
wrote:

There was part of a thread a while back about how adding negative feedback can
create higher order harmonic distortion products than exist open-loop in an
amplifier stage.

This made me think about the application of op-amps in audio generally. Negative
feedback is used primarily to linearise the transfer function and is used in
huge quantites as much as 80dB @ 1 kHz for example.

Since this amount of NFB is not required to provide an accurate gain setting, it
struck me that it's somewhat counter productive. If instead the open-loop
transfer characteritic was made more linear by degeneration

Degeneration is NFB. It is just applied locally. What you really
want is to go with a topology that is naturally more linear,

of the open-loop
gain for example, when NFB is applied, the overall result should be largely
similar (i.e. no worse) but would presumably also suffer less from the creation
of these new distortion products .

Comments ?

One huge problem with including a lot of local NFB is that it makes
the overall system harder to close. Local feedback often creates 2
pole systems with modest Q values within the system. When you go to
close the loop, you have to keep a good phase margin so you are forced
to use a lower overall loop gain.

Try spice modeling a thing like this:


Vcc
---------------+------------
!
\
/
Vbias \
! !
/ +--------------
\ ! !
! !/ e !
---!!--+------! PNP ---
!\ ---
+--Out !
\ !
/ !
\ !
! !
+--------------
!
V D1
---
!
GND

Change D1 to be a resistor and back and you will see quite a
difference in the amount of degeneration needed to get the same
distortion values for a modest signal of lets say 10mV in.

 

[Eeyore]

MooseFET wrote:

Eeyore wrote:

There was part of a thread a while back about how adding negative feedback can
create higher order harmonic distortion products than exist open-loop in an
amplifier stage.

This made me think about the application of op-amps in audio generally. Negative
feedback is used primarily to linearise the transfer function and is used in
huge quantites as much as 80dB @ 1 kHz for example.

Since this amount of NFB is not required to provide an accurate gain setting, it
struck me that it's somewhat counter productive. If instead the open-loop
transfer characteritic was made more linear by degeneration

Degeneration is NFB. It is just applied locally. What you really
want is to go with a topology that is naturally more linear,

Sorry I didn't make that clearer.

Yes, I'm referring to the reduction of overall loop feedback.

Graham

 

[Eeyore]

MooseFET wrote:

One huge problem with including a lot of local NFB is that it makes
the overall system harder to close.

That's not my experience. Quite the reverse actually. But then I do tend to
incorporate internal lead-lag compensation. This results in a far BETTER phase margin.

Graham

 

[D from BC]

On Sun, 19 Aug 2007 16:39:55 +0100, Eeyore
<rabbitsfriendsandrelations@hotmail.com> wrote:

There was part of a thread a while back about how adding negative feedback can
create higher order harmonic distortion products than exist open-loop in an
amplifier stage.

This made me think about the application of op-amps in audio generally. Negative
feedback is used primarily to linearise the transfer function and is used in
huge quantites as much as 80dB @ 1 kHz for example.

Since this amount of NFB is not required to provide an accurate gain setting, it
struck me that it's somewhat counter productive. If instead the open-loop
transfer characteritic was made more linear by degeneration of the open-loop
gain for example, when NFB is applied, the overall result should be largely
similar (i.e. no worse) but would presumably also suffer less from the creation
of these new distortion products .

Comments ?

Graham

Just speaky from some audio hobby work....

*Like with most things in electronics, there are frequency limits. I
think feedback decreases with frequency. The harmonic distortion
becomes an ultrasonic problem.
*Feedback is a correction signal.. If nothing messes up this process
then all's well.
*For large signals, doesn't every semiconductor naturally distort?
Developing the best linear open loop design may not be enough.

D from BC

 

[Eeyore]

D from BC wrote:

Just speaky from some audio hobby work....

*Like with most things in electronics, there are frequency limits. I
think feedback decreases with frequency. The harmonic distortion
becomes an ultrasonic problem.
*Feedback is a correction signal.. If nothing messes up this process
then all's well.
*For large signals, doesn't every semiconductor naturally distort?
Developing the best linear open loop design may not be enough.

You need to learn more.

I appreciate your interest but your grasp of the issues is beginner level.

Graham

 

[Eeyore]

William Sommerwerck wrote:

There was part of a thread a while back about how adding negative feedback
can create higher order harmonic distortion products than exist open-loop
in an amplifier stage.

This made me think about the application of op-amps in audio generally.
Negative feedback is used primarily to linearise the transfer function and
is used in huge quantites as much as 80dB @ 1 kHz for example.

Since this amount of NFB is not required to provide an accurate gain
setting, it struck me that it's somewhat counter productive. If instead the
open-loop transfer characteritic was made more linear by degeneration of the

open-loop gain for example, when NFB is applied, the overall result should
be
largely similar (i.e. no worse) but would presumably also suffer less from
the
creation of these new distortion products .

Comments ?

That negative feedback linearizes the transfer function at the expensive of
adding higher-order harmonics has been long-known. What you say is perfectly
logical.

However, the presence of higher-order harmonics is not the only factor, but
their amplitude. Below a certain percentage (I'm sure Arny will be able to
tell us what that is), they're inaudible.

There's more than a little discussion about what level that is, and indeed it's
known that audibility varies according to harmonic number.

A good op amp can be used as a buffer and be sonically transparent, its
output indistinguishable from its input.

As a buffer it has 100% NFB and I hope that's the case.. As a gain stage with
say 40dB of voltage gain that isn't the case however.

Really, part of what I'm saying is that the classic op-amp isn't really the
ideal gain stage for audio circuits if you want to produce totally 'technically
blameless' performance.

Graham

 

[MooseFET]

On Aug 19, 9:57 am, Eeyore <rabbitsfriendsandrelati...@hotmail.com>
wrote:

MooseFET wrote:
One huge problem with including a lot of local NFB is that it makes
the overall system harder to close.

That's not my experience. Quite the reverse actually. But then I do tend to
incorporate internal lead-lag compensation. This results in a far BETTER phase margin.

This means that you have lowered the outter loop gain in the process.
If the internal part looks kind of like this:

--!!-/\/\--
! !
---/\/\-+--!-\ !
! >------+---
!+/

The amplification stage you are placing the NFB around must have a
great enough bandwidth to make the feedback determine the responce.

The local feedback has all of the problems a global feedback has with
creating upper harmonics. Your global feedback is now at a lower gain
and thus can't remove them. This is just a case of the lack of a free
lunch.

The pole and zero inside the loop is a good thing to do to improve the
phase margin when you have other poles in the system. It allows you
to determine where the gain crossover happens and the phase at the
crossover. It is a method of lowering the overall loop gain. It
doesn't however get rid of the harmonics issue. It also is something
that you can only do a few times inside the loop. When the system
starts to look like 3 of those in series, you are back in trouble.

In audio stuff, you generally want to put the pole-zero thing near the
output, ideally enclosing the output. This makes the system apply a
low pass filter to any distortion products that the feedback can't get
rid of.

 

[MooseFET]

On Aug 19, 9:43 am, "William Sommerwerck" <grizzledgee...@comcast.net>
wrote:

There was part of a thread a while back about how adding negative feedback
can create higher order harmonic distortion products than exist open-loop
in
an amplifier stage.
This made me think about the application of op-amps in audio generally.
Negative feedback is used primarily to linearise the transfer function and
is used in huge quantites as much as 80dB @ 1 kHz for example.
Since this amount of NFB is not required to provide an accurate gain
setting,
it struck me that it's somewhat counter productive. If instead the
open-loop
transfer characteritic was made more linear by degeneration of the
open-loop
gain for example, when NFB is applied, the overall result should be
largely
similar (i.e. no worse) but would presumably also suffer less from the
creation
of these new distortion products .
Comments ?

That negative feedback linearizes the transfer function at the expensive of
adding higher-order harmonics has been long-known. What you say is perfectly
logical.

However, the presence of higher-order harmonics is not the only factor, but
their amplitude. Below a certain percentage (I'm sure Arny will be able to
tell us what that is), they're inaudible.

A good op amp can be used as a buffer and be sonically transparent, its
output indistinguishable from its input.

Even at reasonable gains, there are many that will perform well enough
that nobody will hear the difference. Power amplifiers are the place
where it gets very hard to keep distortion low at reasonable
efficiencies.

 

[MooseFET]

On Aug 19, 10:03 am, D from BC <myrealaddr...@comic.com> wrote:

On Sun, 19 Aug 2007 16:39:55 +0100, Eeyore



rabbitsfriendsandrelati...@hotmail.com> wrote:
There was part of a thread a while back about how adding negative feedback can
create higher order harmonic distortion products than exist open-loop in an
amplifier stage.

This made me think about the application of op-amps in audio generally. Negative
feedback is used primarily to linearise the transfer function and is used in
huge quantites as much as 80dB @ 1 kHz for example.

Since this amount of NFB is not required to provide an accurate gain setting, it
struck me that it's somewhat counter productive. If instead the open-loop
transfer characteritic was made more linear by degeneration of the open-loop
gain for example, when NFB is applied, the overall result should be largely
similar (i.e. no worse) but would presumably also suffer less from the creation
of these new distortion products .

Comments ?

Graham

Just speaky from some audio hobby work....

*Like with most things in electronics, there are frequency limits. I
think feedback decreases with frequency.
Yes it typically does generally decrease. It also has a phase shift.

If you add feedforward, you can have a band in which the feedback
increases with frequency.

The harmonic distortion
becomes an ultrasonic problem.

One problem is that ultasonic things can interact on any nonlinear
part of the system. This can lead to frequencies that are things like
7*F1 - 9*F2 in the circuit. It is like someone injected a signal at
that frequency into that point in the circuit. How the system
responds to it determines whether it will be heard or not.

*Feedback is a correction signal.. If nothing messes up this process
then all's well.
*For large signals, doesn't every semiconductor naturally distort?
Developing the best linear open loop design may not be enough.

D from BC

 

[MooseFET]

On Aug 19, 10:03 am, D from BC <myrealaddr...@comic.com> wrote:
[...]

*For large signals, doesn't every semiconductor naturally distort?
Developing the best linear open loop design may not be enough.

For some reason my cursor went away. This makes it harder to edit
what I'm typing.

Making the "best linear open loop" is for all practical purposes never
enough.
You need a very linear open loop design with a low enough phase shift
to make the NFB work and ideally to have a lowpass effect applied to
any distortion that is created.

You also have to trade off performance against water cooling. A
simple class A power MOSFET common source stage can be used as an
example. If you use about 10 power MOSFETs in parallel, have each one
passing about 0.5 Amps, and run with a 50V supply, you will have a
circuit that is darn linear for a 1mV input signal.

 

[whit3rd]

On Aug 19, 8:39 am, Eeyore <rabbitsfriendsandrelati...@hotmail.com>
wrote:

...about the application of op-amps in audio generally. Negative
feedback is used primarily to linearise the transfer function and is used in
huge quantites as much as 80dB @ 1 kHz for example.

All amplifiers have characteristic curves; the gain isn't completely
linear.
Feedback components can be (very linear) resistors. So you get
some combination of amplification and negative feedback in most useful
low-distortion amplifiers.

A single transistor can have power gain of 10,000; a single vacuum
tube
or MOSFET can have more. Giving up gain for linearity is a good
trade.
It's never perfect (even resistors are distortion sources, if you have
signals
at 2 Hz and the self-heating of the resistors isn't insignificant),
but it's good
enough. Listen. Enjoy.

 

[William Sommerwerck]

A good op amp can be used as a buffer and be sonically transparent,
its output indistinguishable from its input.

As a buffer it has 100% NFB and I hope that's the case. As a gain stage
with say 40dB of voltage gain that isn't the case however.

Really, part of what I'm saying is that the classic op-amp isn't really
the ideal gain stage for audio circuits if you want to produce totally
"technically blameless" performance.

That's certainly true. But does it matter what type of circuit or components
you use if the performance is audibly blameless?

 

[MooseFET]

On Aug 19, 2:42 pm, "William Sommerwerck" <grizzledgee...@comcast.net>
wrote:

A good op amp can be used as a buffer and be sonically transparent,
its output indistinguishable from its input.
As a buffer it has 100% NFB and I hope that's the case. As a gain stage
with say 40dB of voltage gain that isn't the case however.
Really, part of what I'm saying is that the classic op-amp isn't really
the ideal gain stage for audio circuits if you want to produce totally
"technically blameless" performance.

That's certainly true. But does it matter what type of circuit or components
you use if the performance is audibly blameless?

It also has to work for a reasonable time, be easy to manufacture and
look good.

 

[Eeyore]

MooseFET wrote:

Eeyore wrote:
MooseFET wrote:
One huge problem with including a lot of local NFB is that it makes
the overall system harder to close.

That's not my experience. Quite the reverse actually. But then I do tend to
incorporate internal lead-lag compensation. This results in a far BETTER phase margin.

This means that you have lowered the outter loop gain in the process.

Yes. As stated in my introduction, the pursuit of super-fabulously high DC (or AC) open loop
gain makes no sense for audio. Besides, gain is cheap these days. I have no objection to the
introduction of another gain stage for example. I'd rather have a sensible amount of very
linear and well defined gain than oodles of 'poor quality' gain.

If the internal part looks kind of like this:

--!!-/\/\--
! !
---/\/\-+--!-\ !
! >------+---
!+/

The amplification stage you are placing the NFB around must have a
great enough bandwidth to make the feedback determine the responce.

Yes and yes. Many IC op-amps used for audio have GBPs in the 10MHz region so this isn't too
difficult even if using one of those inside the loop (which I did in a recent design ).
Discrete stages suitably degenerated can have higher GBPs than that.

The local feedback has all of the problems a global feedback has with
creating upper harmonics. Your global feedback is now at a lower gain
and thus can't remove them. This is just a case of the lack of a free
lunch.

I'm not sure I 'get that' entirely. I see where you're coming from and that would lead one
to imagine that pursuit of linearity in individual stages was a pointless pursuit and you
might as well have tons of non-linear gain and I know that's not the case, not least because
the very hugh gain system has to be stable and that tends to lead to rolling off the gain
(and the advantage of NFB) from very low frequencies.

The pole and zero inside the loop is a good thing to do to improve the
phase margin when you have other poles in the system. It allows you
to determine where the gain crossover happens and the phase at the
crossover. It is a method of lowering the overall loop gain. It
doesn't however get rid of the harmonics issue. It also is something
that you can only do a few times inside the loop. When the system
starts to look like 3 of those in series, you are back in trouble.

3 of them would be a bit much. I've not used more than 2 inside the loop in fact.

In audio stuff, you generally want to put the pole-zero thing near the
output, ideally enclosing the output. This makes the system apply a

low pass filter to any distortion products that the feedback can't get
rid of.

Interesting idea.

Graham

 

[Eeyore]

whit3rd wrote:

Eeyore wrote:

...about the application of op-amps in audio generally. Negative
feedback is used primarily to linearise the transfer function and is used in
huge quantites as much as 80dB @ 1 kHz for example.

All amplifiers have characteristic curves; the gain isn't completely
linear.

I know. That's why I said the individual stages should be degenerated to linearise
them. This results in a lower gain but this may not be a problem in practice as
long as GBP is maintained.

Feedback components can be (very linear) resistors. So you get
some combination of amplification and negative feedback in most useful
low-distortion amplifiers.

A single transistor can have power gain of 10,000;

In every audio amplifier stage I know, POWER gain is of little importance. Voltage
gain is what's required. Cuurent gain can be readily added where needed by using
emiiter followers.

a single vacuum tube or MOSFET can have more. Giving up gain for linearity is a
good
trade.

That was indeed my point wrt giving up some of that *open-loop* gain in a gain
block.

Graham

 

[Eeyore]

William Sommerwerck wrote:

A good op amp can be used as a buffer and be sonically transparent,
its output indistinguishable from its input.

As a buffer it has 100% NFB and I hope that's the case. As a gain stage
with say 40dB of voltage gain that isn't the case however.

Really, part of what I'm saying is that the classic op-amp isn't really
the ideal gain stage for audio circuits if you want to produce totally
"technically blameless" performance.

That's certainly true. But does it matter what type of circuit or components
you use if the performance is audibly blameless?

You can (and people do) argue forever about what is or isn't subjectively
audible. The '990C' discrete op-amp was mentioned in another thread for example.
With THD of 0.06% (-64dB) under some conditions it strikes me that those
distortion products could easily be audible yet ppl leapt to its defence.

If it can be shown that the defects must be inaudible from first principles
(such as distortion below 100dB for example) you're on firmer ground IMHO.

Graham

 

[MooseFET]

On Aug 19, 4:34 pm, Eeyore <rabbitsfriendsandrelati...@hotmail.com>
wrote:

MooseFET wrote:
Eeyore wrote:
MooseFET wrote:
One huge problem with including a lot of local NFB is that it makes
the overall system harder to close.

That's not my experience. Quite the reverse actually. But then I do tend to
incorporate internal lead-lag compensation. This results in a far BETTER phase margin.

This means that you have lowered the outter loop gain in the process.

Yes. As stated in my introduction, the pursuit of super-fabulously high DC (or AC) open loop
gain makes no sense for audio.

You do need merely "high gain" however. This high gain needs to be
true at the frequencies of interest so the GBP does have to be at
least some reasonable amount.

Very high values of loop gain makes for very large amounts of
reductions in the harmonics within the band. This can argue for much
more gain than it would normally appear you need if you only needed
enough gain to be sure that the feedback resistors were what was
setting the gain.

Besides, gain is cheap these days. I have no objection to the
introduction of another gain stage for example. I'd rather have a sensible amount of very
linear and well defined gain than oodles of 'poor quality' gain.

Adding stages adds to the phase shifts. This is another "no free
lunch situation". When you increase the number of stages you also
want to increase the bandwidths of most of the stages to keep the
phase shift near the gain cross over within reason.


Keeping this for reference later:

--!!-/\/\--
! !
---/\/\-+--!-\ !
! >------+---
!+/

[.. stuff we have covered and agree on ...

The local feedback has all of the problems a global feedback has with
creating upper harmonics. Your global feedback is now at a lower gain
and thus can't remove them. This is just a case of the lack of a free
lunch.

I'm not sure I 'get that' entirely. I see where you're coming from and that would lead one
to imagine that pursuit of linearity in individual stages was a pointless pursuit and you
might as well have tons of non-linear gain and I know that's not the case, not least because
the very hugh gain system has to be stable and that tends to lead to rolling off the gain
(and the advantage of NFB) from very low frequencies.

No what I am pointing out is that local feedback is not a good
substitute for a naturally more linear stage. Consider this sort of a
situation:

---------- ------ ---------- ------
Signal--->! Subtract !---! Gain !--->! Subtract !--->! Bad !--+----

---------- ------ ---------- ! gain ! !

^ ^ ------ !
! ! !
------------------------+---------------------

You can trade back and forth how much subtracting you do in the two
subtraction circuits but you can't really fix the "bad gain" section.
I am for not leaving out the idea that the good gain has a limited
bandwidth and assuming all of the bandwidth limiting happens in the
"bad gain". I think this makes the idea obvious in it simple form.
To take a bit of a real example, consider a dreadful output stage that
works like this:

----+--------+---
! !
\ \
R1 / / R2
\ \
! !
! !/e
+-------! PNP
! !\
!/ !
! NPN !
!\e ! R3
----------+----/\/\---
--- mirror for other half

R3 is providing a measure of local feedback. R2 also is doing so.
This stage will still be a horror story. Adding a diode in series
with R1 to match to the E-B drop of the PNP makes it much less so.
The diode makes the PNP act much more like a linear current mirror and
thus reduces the natural distortion.

Since the transistors used in power stages are usually slower than the
others in the design. The output is almost always where the pole you
didn't design in lives.

The pole and zero inside the loop is a good thing to do to improve the
phase margin when you have other poles in the system. It allows you
to determine where the gain crossover happens and the phase at the
crossover. It is a method of lowering the overall loop gain. It
doesn't however get rid of the harmonics issue. It also is something
that you can only do a few times inside the loop. When the system
starts to look like 3 of those in series, you are back in trouble.

3 of them would be a bit much. I've not used more than 2 inside the loop in fact.

Trust me on this: Don't put three inside the loop. Reconsider the
design if you find yourself going there. Two is ok. One plus a
feedforwards is ok but three always seems to mean trouble.

In audio stuff, you generally want to put the pole-zero thing near the
output, ideally enclosing the output. This makes the system apply a
low pass filter to any distortion products that the feedback can't get
rid of.

Interesting idea.

It only works up to a point. It also requires largish (mechanically)
parts be involved. You have a capacitor and a resistor with fairly
large swings on them and are working at lowish impedances.

 

[Eeyore]

MooseFET wrote:

Eeyore wrote:
MooseFET wrote:
Eeyore wrote:
MooseFET wrote:
One huge problem with including a lot of local NFB is that it makes
the overall system harder to close.

That's not my experience. Quite the reverse actually. But then I do tend to
incorporate internal lead-lag compensation. This results in a far BETTER phase margin.

This means that you have lowered the outter loop gain in the process.

Yes. As stated in my introduction, the pursuit of super-fabulously high DC (or AC) open loop
gain makes no sense for audio.

You do need merely "high gain" however. This high gain needs to be
true at the frequencies of interest so the GBP does have to be at
least some reasonable amount.

10MHz seems to work reasonably well but 120dB gain at LF is not a requirement.

Very high values of loop gain makes for very large amounts of
reductions in the harmonics within the band. This can argue for much
more gain than it would normally appear you need if you only needed
enough gain to be sure that the feedback resistors were what was
setting the gain.

That's sort of what I'm after.

Besides, gain is cheap these days. I have no objection to the
introduction of another gain stage for example. I'd rather have a sensible amount of very
linear and well defined gain than oodles of 'poor quality' gain.

Adding stages adds to the phase shifts.

Needn't be a very significant phase shift. Plus, if the 'natural' phase shift of the existing
stages is reduced through degeneration, that's all fine.

This is another "no free lunch situation". When you increase the number of stages you also
want to increase the bandwidths of most of the stages to keep the
phase shift near the gain cross over within reason.

Oh yes and degeneration will do that of course.

Graham