Feedback in audio esp wrt op-amps.

[William Sommerwerck]

Sometimes when I hear the golden earers talk I'm surprised

that I can make out any music at all when listening with my
Cantons fed from an old Sony amp through particularly
oxygen-rich cables.

I think if you owned better equipment, your views of what constitute good
and bad reproduction might change.

 

[William Sommerwerck]

["Followup-To:" header set to sci.electronics.design.]
William Sommerwerck wrote:

I'm not sure that's right. My memory (which could be faulty) is that this
can be shown mathematically.
I'll ask around (I know a few people in high places) and see if I can get
a reference.

If you want real information, don't ask people in high places. Ask techs.

And their qualifications for making a valid mathematical analysis would
be...?

 

[neon]

negative F/B does not linearize nothing just reduce gain to a level where it is usefull in inplementation. on a transistor amplifier the transfer charateristics is strickly related to loading and the linearity of beta itself.

 

[Arny Krueger]

"Scott Dorsey wrote

The EQ section alone on a Neve V series (and derivatives) has 18 op-amp
stages.

Can't find a schematic for that one, but I'm looking at the schematic of a
Neve 83022EQ which seems to be representative.

http://www.danalexanderaudio.com/neveinfo/83049/83022EQ.jpg

There are a ton of op amps, but they aren't all cascaded on the signal path.
For example, 16 op amps are in 4 state-variable filters each composed of 4
stages, plus a helper amplifier.

In actual use, the full bandwidth and amplitude of the output signal of the
equalizer rarely if ever flows through all 16 op amps.

The state variable filters are typically used as hi pass, lo pass, shelving,
peaking or nulling filters, so only a fraction of the audio band is affected
by each. When each parametric section's boost/cut control is centered as it
often is, very little of the output signal passes through them.

There are 5 op amps with gain either -1 or +1, cascaded across the top of
the schematic. They are always in the signal path of the eq. They each pass
the entire audio band. However, it looks like it may be possible for the
whole eq to be bypassed.

My analog parametric eqs include individual bypass switches for each
section, and a bypass the whole eq. I can see maybe 20 ops amps actually
interposed full-band and full-signal in a record/play signal path, but 100
seems like a reach.

I've done experiements where we built up a string of 20 unity and 10 dB
stages, using fairly primitive op amps like TL074s. No reliable detection
in level-matched, bias-controlled tests, using very clean sources, very
clean monitors, and a variety of listeners who were either audio engineers
and/or audiophiles, and thought they would hear a difference.

 

[Arny Krueger]

Wouldn't you get 4 kHz (2nd harmonic of 2kHz) rather than 3 kHz?

You get both third and fourth. The 4th is another 46dB or so down, or about
138 dB down from the fundamental. I felt safe ignoring it. ;-)

I think that the third harmonic is actually due to the modulation of the DC
term from the first time through. The fourth harmonic is the second harmonic
of the second harmonic, of course.

 

[Arny Krueger]

I tried my harmonic of the harmonic argument again. Sometimes it
works sometimes not.

It all comes out if you do the math, which involves a few simple trig
identities.

It also comes out if you simulate it in Matlab or Audition/CEP. I did my
simulation in CEP using Edit, Mix, Paste and appropriate choice of the mix
and modulate options.

The same basic technique can be used to create music with controlled amounts
of various orders of added nonlinear distortion. Here is worked-out
example:

http://www.pcabx.com/technical/nonlinear/

 

[Arny Krueger]

Dammit, I've got that paper around here *someplace*.

The paper I'm referring to is by an English author, I think not Reg
Williamson and I think not Self, showing the generation of higher
harmonics
on the application of moderate amounts of feedback in a simple FET circuit
which produces only low-order harmonics without feedback. As the feedback
is
increased the high harmonics get smaller; they're at their worst in
low-feedback circuits. The measurements were real, not simulations.

Meanwhile, as I looked for that $%^$# article, I found this:

www.ucop.edu/research/micro/98_99/98_074.pdf

It's a theoretical discussion of the generation of higher-order IM
products
in feedback amps. The theory is supplemented by simulations, but
unfortunately not by real-world measurements, and the authors note that
their models are oversimplified.

Really? I see an article about sample-and-holds, and the like.

 

[Eeyore]

Can't find a schematic for that one, but I'm looking at the schematic of a
Neve 83022EQ which seems to be representative.

http://www.danalexanderaudio.com/neveinfo/83049/83022EQ.jpg

Yes at a casual glance it looks much the same.

There are a ton of op amps, but they aren't all cascaded on the signal path.
For example, 16 op amps are in 4 state-variable filters each composed of 4
stages, plus a helper amplifier.

In actual use, the full bandwidth and amplitude of the output signal of the
equalizer rarely if ever flows through all 16 op amps.

Depending on the cut and boost, the signal may be affected by all of them.

Graham

 

[Eeyore]

You get both third and fourth. The 4th is another 46dB or so down, or about
138 dB down from the fundamental. I felt safe ignoring it. ;-)

I think that the third harmonic is actually due to the modulation of the DC
term from the first time through. The fourth harmonic is the second harmonic
of the second harmonic, of course.

Where does this DC term come from ?

Graham

 

[john jardine]

Dammit, I've got that paper around here *someplace*.

Meanwhile, can anyone help my blocked memory? Who the hell wrote that paper?

Peace,
Paul

Baxendall? in Wireless world magazine about 35 years ago.
Seem to remember the example was a diff amp pair. Article hinged on power
series expansions.

 

[Arny Krueger]

Yes at a casual glance it looks much the same.

Pretty typical for a 4-section parametric eq, plus/minus some details.

Depending on the cut and boost, the signal may be affected by all of them.

No doubt, but it is not the same as every ounce of signal going through all
of them cascaded, no matter what.

And, the channel strips are not usually cascaded, either. This one nets out
to being like 5-6 stages cascaded full time, more if you use EFX.

 

[Scott Dorsey]

100 op amps on parallel channels is a far different situation than 100 *ALL
in series* with the signal.
Of course in the real world the situation is somewhere in between those
extremes.

Pop the cover on an SSL 4000 some time...
--scott

 

[Arny Krueger]

Where does this DC term come from ?

A DC term is a natural consequence of a second order nonlinearity. Comes out
of the trig identity for X squared:

Sine squared(x) = 1/2 - 1/2 Cos (2x) = (1 - Cos (2x) ) /2

http://en.wikipedia.org/wiki/Trigonometric_identity#Power-reduction_formulae

Please see "Power-reduction formulae" for second and third orders. As I
recall the CRC tables have them for several orders beyond 3. Or, you can
derive them from the formulae for orders 2 and 3.

 

[Robert Latest]

I think if you owned better equipment, your views of what constitute good
and bad reproduction might change.

Oh, it absolutely would. No question about it. That wouldn't have to do with
the way the equipment reproduces the sound though. There's more to hearing
than what reaches the ear. What reaches the eye and leaves the wallet has to
do with it as well; I wish people would start acknowledging that.

robert

 

[MooseFET]

So for audio, put the gain crossover way out of band. Right?

That tends to happen if you have a high amount of feedback at the
normal audio frequencies. You want to put the gain crossover high and
use a large amount of feedback so it works out nicely.

 

[MooseFET]

Cool...
Maybe call it a distortion loop.

+-<<<--------------------------------------<+
| |
sine>--summation-------nonlinear transfer (inverting)->+
|
Not completely containing a signal to cancel out the
nonlinear transfer. So some 2nd harmonic gets to pass through the
nonlinear transfer again to make...the 4th....and so and so on..
(IIRC that would be the harmonic generation sequence for a 2nd order
nonlinear transfer.)

Take 2 tone and then there's the intermodulation products.
What a painful thing to think about...

Now add some noise and follow it around. I'm sure your head will
explode. You will discover that the signal modulates the noise and
intermixes with it. The peak in the noise near the gain cross over
gets mixed down with the harmonics of the signal that also land
there. If you make many very accurate frequency measurements on the
signal after the signal has been through such a process, you will find
that there is an increased low frequency modulation of the signal.

Significant magnitudes???

If it can be measured it can be called significant. Someone will
care.

 

[Mark]

[.....]

Cool...
Maybe call it a distortion loop.

+-<<<--------------------------------------<+
| |
sine>--summation-------nonlinear transfer (inverting)->+
|
Not completely containing a signal to cancel out the
nonlinear transfer. So some 2nd harmonic gets to pass through the
nonlinear transfer again to make...the 4th....and so and so on..
(IIRC that would be the harmonic generation sequence for a 2nd order
nonlinear transfer.)

Take 2 tone and then there's the intermodulation products.
What a painful thing to think about...

Now add some noise and follow it around. I'm sure your head will
explode. You will discover that the signal modulates the noise and
intermixes with it. The peak in the noise near the gain cross over
gets mixed down with the harmonics of the signal that also land
there. If you make many very accurate frequency measurements on the
signal after the signal has been through such a process, you will find
that there is an increased low frequency modulation of the signal.

Significant magnitudes???

If it can be measured it can be called significant. Someone will
care.

Cheerleader in electronics...
"2,4,6,8 what distortion do I hate."

D from BC- Hide quoted text -

- Show quoted text -- Hide quoted text -

- Show quoted text -

Is this the article?

http://stereophile.com/news/10065/

Someone mentioned a perfect second order ONLY device that open loop
produces ONLY 2x. When you put neg feedback around it you could get
the "harmonic of the harmonic" i.e. 4th harmonic which wasn't there
before. OK maybe in this special case. But this is a theoretical
math excersize then, any practical device that has a second order non-
linearity will also have high order terms and the neg feedback will
reduce those.

If you start with a hypothetical perfect 2nd order device, I MIGHT be
ready to concede that neg feedback might produce some small level of
4th order that wasn't there before. Someone needs to simulate this
case.

This may be an interesting mental exersize, but it has very little
connection to actual practice. In practice using any REAL amplifer,
neg feedback REDUCES all the harmonics. (another exception someone
mentioned would be those harmonics near the gain crossover frequency
if the neg feedback causes the gain to peak a few dB then the harmonic
could also be increased a few dB. Again in paractice, this is well
above 20 kHz. If there is any large amount of peaking, then the
system is only marginally stable.

Neg feedback is your friend.

Mark



Mark

 

[Mark]

If you start with a hypothetical perfect 2nd order device, I MIGHT be
ready to concede that neg feedback might produce some small level of
4th order that wasn't there before. Someone needs to simulate this
case.

OK I ran the sim...yes you are correct adding neg feedback to a
perfect 2nd order device creates higher order harmonics 3rd 4th etc
that were not there before.

Some PSPICE code for those that want to play....

Neg Feedback Amp does neg feedback create high order distortion

..TRAN 1uS 10ms

*transient analysis sine wave
Vin 1 0 Sin(0 1 1KHz)

Eamp 2 0 poly(1) (1,2) 0 100 -10 ;with 100% neg feedback
*Eamp 2 0 poly(1) (1,0) 0 100 -10 ;with NO neg feedback

Rloadin 1 0 600
Rloadout 2 0 600

..probe

..end

Small amounts of feedback created the most distortion. As I increased
the closed loop gain, as expceted all the distortion levels were
reduced.

In most any real amplifier, there will be high order non-linearities
in the device and adding neg feedback will reduce them. (with the
exceptions near the crossover frequency noted in the previous post)

Thank you for the interesting observation.

Mark

 

[D from BC]

[.....]

Cool...
Maybe call it a distortion loop.

+-<<<--------------------------------------<+
| |
sine>--summation-------nonlinear transfer (inverting)->+
|
Not completely containing a signal to cancel out the
nonlinear transfer. So some 2nd harmonic gets to pass through the
nonlinear transfer again to make...the 4th....and so and so on..
(IIRC that would be the harmonic generation sequence for a 2nd order
nonlinear transfer.)

Take 2 tone and then there's the intermodulation products.
What a painful thing to think about...

Now add some noise and follow it around. I'm sure your head will
explode. You will discover that the signal modulates the noise and
intermixes with it. The peak in the noise near the gain cross over
gets mixed down with the harmonics of the signal that also land
there. If you make many very accurate frequency measurements on the
signal after the signal has been through such a process, you will find
that there is an increased low frequency modulation of the signal.

Significant magnitudes???

If it can be measured it can be called significant. Someone will
care.

Cheerleader in electronics...
"2,4,6,8 what distortion do I hate."

D from BC- Hide quoted text -

- Show quoted text -- Hide quoted text -

- Show quoted text -

Is this the article?

http://stereophile.com/news/10065/

Someone mentioned a perfect second order ONLY device that open loop
produces ONLY 2x. When you put neg feedback around it you could get
the "harmonic of the harmonic" i.e. 4th harmonic which wasn't there
before. OK maybe in this special case. But this is a theoretical
math excersize then, any practical device that has a second order non-
linearity will also have high order terms and the neg feedback will
reduce those.

If you start with a hypothetical perfect 2nd order device, I MIGHT be
ready to concede that neg feedback might produce some small level of
4th order that wasn't there before. Someone needs to simulate this
case.

This may be an interesting mental exersize, but it has very little
connection to actual practice. In practice using any REAL amplifer,
neg feedback REDUCES all the harmonics. (another exception someone
mentioned would be those harmonics near the gain crossover frequency
if the neg feedback causes the gain to peak a few dB then the harmonic
could also be increased a few dB. Again in paractice, this is well
above 20 kHz. If there is any large amount of peaking, then the
system is only marginally stable.

Neg feedback is your friend.

Mark



Mark

Feedback does the job but like with most things in electronics...you
don't get something for nothing.
Usually something else gets fk'd when there's a large benefit.
So that's why there's some feedback bashing.
Trust nothing..
D from BC

 

[Scott Dorsey]

Neg feedback is your friend.

It absolutely is. However, in the 1970s it was regarded as a cure-all that
could fix all ills, and it's not. The resulting sonic issues were severe,
and the current backlash you see in the community against the use of feedback
is mostly a reaction to that. This is a shame, since feedback is a useful
tool.
--scott