Musings about negative feedback

[Circlotron]

Negative feedback is always a bit of a problem because in a real-world amplifier it takes time for the signal to go through the amplifier and appear at the output, and it is this slightly delayed signal that is fed back to the input to be subtracted from the input signal, the difference of the two then being fed to the amplifier input. That means that among other things, the (delayed) feedback signal is never exactly suitable for the input signal - it would only be so if you had an amplifier with *zero* delay in the signal path.

What occurred to me however is that when a sinewave is fed into an amplifier for a distortion measurement, the feedback may have more validity than first appears because of what I would call the *self-similarity* of a sinewave. A sinewave can be represented by a rotating vector with a sine and a cosine component. The delay in the signal path of the amplifier simply applies a phase-shifted feedback signal, which due to the self-similarity of a sine, does an ok job subtracting from the input sinewave. Vectorially speaking, it is just a bit further around the circle but still the same shape,

However... That is all fine and good when we have a sinewave as a test signal, but what if we apply some other shape of waveform? There would be no moment-to-moment self-similarity of the input waveform anymore - the vector of the wave doesn't draw as a circle anymore. (bad description, I know) You can't move several degrees around the circumference and expect the new shape to have a constant radius.

What I am getting at is this - an amplifier may measure very good distortion figures with a sinewave, but if it were possible to measure distortion with a complex waveform (how far it deviates from ideal rather than spectral analysis method) then that may prove to be very interesting. It may shed some light on why many amplifiers measure good but sound bad.

OK, all you guys with Phd's and whatnot, shoot me down in flames.

 

[Frank Berry]

I believe you are correct.
That is why many amplifiers with large amounts of inverse feedback show poor squarewave response.
Most often, you'll see overshoot on the leading edge of the squarewave and undershoot on the lagging edge.
It's a symptom of the time delay through the amplifier.

 

[SY]

Depends on how you define "time delay." The minimum phase delay caused by rolloffs is exactly what feedback theory accounts for. The nonminimum phase delay, caused by signal transit time, is pretty irrelevant unless you're worried about making an amp flat into the tens of megahertz range.

Think about this: when you drive, how do you stay in your lane?

 

[Sch3mat1c]

Quite correct, Circ. The phase shift is indeed constant in the amplifier and manifests itself as a high-frequency rise, as ringing on the leading edge of a squarewave.

The key to avoiding this is to build a quick amplifier (basically, use a good OPT) and compensate so that rather than the process of "ohp, input wants to be high... better raise the output... oops not that fast, that's overshoot...go down.. oh not that far! etc.", instead it just rises slowly "oh what? huh you say you want to go up? uhhh okay whatever...".

Incidentially, the phase shift does NOT cause any distortion, only the above frequency artifact. It does cause IMD, but this is generally attributed to distoring a distorted signal (creating higher harmonics) and is avoided by reducing distortion to inaudible levels with levels >15dB of NFB.

Tim

 

[Mr Evil]

...The key to avoiding this is to build a quick amplifier...

The key is to make sure the amplifier has sufficient phase margin. If the gain is too high when the feedback becomes positive, even if the amp is stable, it will cause ringing. If the gain is low enough, then the feedback is no longer trying to correct the output and ringing is avoided. Building a fast amp may make this easier, but it doesn't solve anything by itself.

 

[serengetiplains]

Frank Berry has a PhD. And I like his answer. Time delay is time delay no matter how you slice it, and a horse out of the gate is a horse out of the gate. How's that for fancy thinking?

 

[Mr Evil]

No matter how you wrap your mind about it, wondering how it can work when it can only correct in retrospect, the fact remains that negative feedback works, and works well. You could try to avoid the problem by building circuits without feedback, but it's impossible, given how even a simple emitter-follower effectively has negative feedback.

The ringing that Frank mentioned can be completely avoided, even in high feedback amps. The fact that many amps do exhibit ringing is because the designer either doesn't understand that a well damped amplifier is desirable, or wanted the specs to look good (i.e. wide bandwidth and high slew rate rather than good control).

 

[serengetiplains]

No matter how you wrap your mind about it, wondering how it can work when it can only correct in retrospect, the fact remains that negative feedback works, and works well. You could try to avoid the problem by building circuits without feedback, but it's impossible, given how even a simple emitter-follower effectively has negative feedback.

"Works well" is not fact in my experience and the phrase, in any event, does not describe what any given form of feedback actually does. I'm happy to go on record to say emitter-follower feedback is not the same as local feedback is not the same as global feedback in sonically important ways of differing. Nor am I convinced that emitter- or cathode-follower feedback is "feedback," and the issue, to me, is not merely semantic. A circuit, so far as I know, might work all as a piece, meaning there might be no delay between current exiting and current returning to an active device, meaning the time and phase delay aspects of local and global feedback form no part of what emitter/cathode "feedback" does.

I'm personally interested in knowing why feedback changes the sound of an amplifier in the way it does. We all know at least the more significant of the sonic and electrical benefits of feedback. What accounts for its sonic drawbacks which I, overall, prefer to do without?

 

[ThorstenL]

Konnichiwa,

No matter how you wrap your mind about it, wondering how it can work when it can only correct in retrospect, the fact remains that negative feedback works, and works well.

From Evil Anime Charater to the evil Mr.....

The above is HIGHLY dependant on the definition and interpretation of the phrase "it works".

I would argue that the way it is commonly implememted NFB does not "work", by my own definition of the terms "works".

I can live with the fact that for you "NFB works". Can you live with the fact that for me "NFB potentially works BUT a long list of quite narrow conditions apply under which NFB works, otherwise it doesn't."

Sayonara

 

[ThorstenL]

Konnichiwa,

I'm happy to go on record to say emitter-follower feedback is not the same as local feedback is not the same as global feedback in sonically important ways of differing. Nor am I convinced that emitter- or cathode-follower feedback is "feedback," and the issue, to me, is not merely semantic.

I agree. I hate to see local degeneration and looped inverse feedback circuits to be lumped into one. It is easy to call the "NFB" as there are some similarities but technically speaking they differ more than they are the same.

My personal usage of NFB is strictly in the sense of "looped inverse feedback". Degeneration is degeneration, feedback (positive or negative) is feedback.

The common use indicates muddy thinking and poor language skills.

Sayonara

 

[Mr Evil]

"Works well" is not fact in my experience and the phrase, in any event, does not describe what any given form of feedback actually does. I'm happy to go on record to say emitter-follower feedback is not the same as local feedback is not the same as global feedback in sonically important ways of differing. Nor am I convinced that emitter- or cathode-follower feedback is "feedback," and the issue, to me, is not merely semantic. A circuit, so far as I know, might work all as a piece, meaning there might be no delay between current exiting and current returning to an active device, meaning the time and phase delay aspects of local and global feedback form no part of what emitter/cathode "feedback" does...

Of course there are different levels of feedback, but they are all the same fundamentally, they only differ in degree. In the emitter-follower example, you have to realise that there are actually two inputs, although it may not be immediately apparent. This arises because the output is a function of the voltage difference between the base and the emitter. The base is the normal input, and the emitter is the other and the same as the output, hence you can see that the amount of feedback is 100%.

Since it takes a finite amount of time for the signal to propogate through from the base to the emitter, it is clear that the feedback that appears at the output is also delayed relative to the input, thus the base-emitter voltage is formed from the input at the present, plus a function of the input in the past, just as with any other negative feedback.

This can manifest itself with ringing or even complete oscillation, just as with feedback around multiple stages in a complete amp. The difference is only in the amount of delay, and thus the frequency at which oscillations occur.

Konnichiwa,



From Evil Anime Charater to the evil Mr.....

The above is HIGHLY dependant on the definition and interpretation of the phrase "it works".

I would argue that the way it is commonly implememted NFB does not "work", by my own definition of the terms "works".

I can live with the fact that for you "NFB works". Can you live with the fact that for me "NFB potentially works BUT a long list of quite narrow conditions apply under which NFB works, otherwise it doesn't."

Sayonara

Of course there are places where negative feedback breaks down, or simply doesn't work at all. This is why as I said up there that an amplifier must have sufficient phase margin, so that there is effectively no feedback above where is ceases to work. Where the delay is small relative to the frequency then feedback does work well, forcing the output to be closer to proportional to the input.

Negative feedback is everywhere, such as SY's example of staying on course when driving. Likewise while typing this, my fingers don't know exactly where they are in relation to the keys; my eyes must monitor their progress and then corrections can be applied. Even though there is a significant delay, my fingers still reach the right keys at the right time. They don't fly off to one side and they don't oscillate - it works.

 

[hermanv]

Feedback can be more complex

Global feedback is an excellent way to prevent for example DC at the speaker binding posts of a transistor power amplifier.

To make this work correctly the feedback itself is filtered so only the lowest frequencies are applied all the way around the loop.

Middle frequencies can safely be fed back around a couple of stages and the higest frequencies should be handled with local feedback for best effect.

Like all other circuit details, wise application can help performance. These things are rarely as black and white as the various opinions on this and other forums would lead you to believe.

Look up TIMs and SID (Transient induced distortion and Slewing induced distortion) for lots more information on these topics.

There is no one and only correct answer.

 

[Sch3mat1c]

The common use indicates muddy thinking and poor language skills.

This from someone who clearly has no understanding of how local NFB is applied.

(Did you know that triodes have internal NFB? It's true! )

Tim

 

[serengetiplains]

(Did you know that triodes have internal NFB? It's true! )

Tim, does so-called internal NFB do anything electrically different than, say, global NFB across, say, 3 gain stages? I think that was Thorsten's point, which I think you missed.

 

[Sch3mat1c]

Global vs. internal or local NFB is just wiring (although triode NFB specifically is nonlinear).

Tim

 

[ThorstenL]

Konnichiwa,

Negative feedback is everywhere, such as SY's example of staying on course when driving. Likewise while typing this, my fingers don't know exactly where they are in relation to the keys; my eyes must monitor their progress and then corrections can be applied. Even though there is a significant delay, my fingers still reach the right keys at the right time. They don't fly off to one side and they don't oscillate - it works.

Actually, both your examples illustrate that NFB does NOT work. Instead of a smooth line the course of the car will be a coolection of little jinks and swerves, which AVERAGE OUT to a pretty straight line, but are not, UNLESS heavily averaged.

Equally, you would be surprised how jittery your fingers are if you look at it close enough.

I repeat, your consideration if "feedback works" or not depends upon the defintion of "it works".

This from someone who clearly has no understanding of how local NFB is applied.

Yes, I appraciate my rather large ignorance. Maybe you would care to enlighten me how local feedback (NOT DEGENERATION, which is something entierly different) is applied?

Did you know that triodes have internal NFB? It's true!

Actually, I am aware that some people have asserted that Triodes behave IN SOME WAYS like a Pentode with a local (anode to grid) feedback loop applied (which is true and can be experimentally observed) however they failed to mention the various areas where this was NOT the case and then incorrectly drew the conclusion that "Triodes are nothing but penthodes with internal feedback".

That conclusion is FALSE as the underpinning framework has holes big enough to drive a heavy battletank through them.

Repeating a false statement does not make it any truer.

Sayonara

 

[Sch3mat1c]

Actually, both your examples illustrate that NFB does NOT work. Instead of a smooth line the course of the car will be a coolection of little jinks and swerves, which AVERAGE OUT to a pretty straight line, but are not, UNLESS heavily averaged.

Yep. Nothing's perfect. Fortunately, nothing needs to be absolutely so. So we use NFB, especially because it's better than nothing.

Yes, I appraciate my rather large ignorance. Maybe you would care to enlighten me how local feedback (NOT DEGENERATION, which is something entierly different) is applied?

Can be done by shunt feedback (usually plate to grid, grounded cathode) or voltage feedback (cathode to grid, grounded plate).
In tetrodes and pentodes, the screen can also be involved; common uses include triode and UL mode.

That conclusion is FALSE as the underpinning framework has holes big enough to drive a heavy battletank through them.

Ok. So what have you to say about connecting screen to plate, especially the remarkable semblence of these curves to a triode of similar construction?

Hey, even Maxwell says it's so.

Tim

 

[SY]

I'm personally interested in knowing why feedback changes the sound of an amplifier in the way it does. We all know at least the more significant of the sonic and electrical benefits of feedback. What accounts for its sonic drawbacks which I, overall, prefer to do without?

Assuming you've made an apples-to-apples comparison (i.e., a reasonably well-designed feedback amp), it may be that you prefer distortion (many people do) or the frequency response changes concommitant to the inevitable high source Z of a typical no-loop-feedback amp. For many years, I said the same thing as you, filled numerous lab notebooks with experimental results, and found some interesting design tricks that let me get away without that awful feedback loop.

My head was cleared of such nonsense when a nasty old curmudgeon from University of Wisconsin told me I was full of it and I set up some experiments to prove him wrong. Once I started doing proper listening tests (comparing input to output with RIGOROUS level-matching and, where necessary, going blind), I found out the bad news that I preferred amps that acted as tone controls and aural exciters. Hmmmm...... that old guy had the last laugh.

 

[jeff mai]

I found out the bad news that I preferred amps that acted as tone controls and aural exciters. Hmmmm...... that old guy had the last laugh.

So what did you do about it? Did you start using amps you didn't prefer? Or did you start to prefer different amps? Either way it sounds sort of anti-fun for a pleasure activity.

 

[ThorstenL]

Konnichiwa,

Yep. Nothing's perfect. Fortunately, nothing needs to be absolutely so. So we use NFB, especially because it's better than nothing.

Well, Looped Inverse Feedback is NOT better than making an inherently more device and circuit. It is the lazy designers way out of doing his job properly.

Can be done by shunt feedback (usually plate to grid, grounded cathode) or voltage feedback (cathode to grid, grounded plate).

One of your solutions is to change the Amplification stage from common cathode/emitter/source to common anode/collector/drain. This is an application of looped inverse feedback, NOT "local" feedback. The other solution you present is again looped inverse feedback NOT "local" feedback. You should also be aware that I have many times applied either methode of lopped inverse feedback.

So, youy have NOT shown ANY methode of applying "local" feedback, you have merely illustrated that it is possible to apply looped inverse feedback around a single device. I am quite familiar with this, thank you very much.

In tetrodes and pentodes, the screen can also be involved; common uses include triode and UL mode.

More looped inverse feedback (UL Mode), though all but linear (never mind "Ultra-Linear").

Pseudo triode is an interesting concept. If it where actual lopped feedback, then the gain should be reduced to unity.

Therefore using sendary and tertiary grids as signal inputs (feedback or actual signals) is a rather specific case and one may argue that it is the application of looped feedback through additional mechinisms.

Ok. So what have you to say about connecting screen to plate, especially the remarkable semblence of these curves to a triode of similar construction?

I have nothing specifically to say. Merely that what happens within a valve happens directly through moderating the electron flow. To call this inverse looped feedback is the same as calling the Water Cycle feedback. You cannot call it degeneration either.

This behaviour is merely incident and inherent to the device.

Hey, even Maxwell says it's so.

He does? Where. I don't remember him writing:

"A triode is the exact analog of a pentode with anode to grid looped feedback."

I doubt he would ever write such nonsense anyway.

Yes, there are parallels in some areas and differences in other.

To conclude from some limited congruence under exclusion of any non-congruent areas that two different processes are the same is actually not science, but science fiction.

Sayonara