lumanauw said:
Well, if I can't dazzle you with brilliance, the only remaining option is to confuse you with BS 😀
lumanauw said:
I think you are right, if the fb network is purely resistive, then there is only phase shift in the main loop. And that is what counts, the total phase shift around the loop, fb network and forward path.
Or is it you mean intrinsic C that is inherent in every 1/4W resistor? Yes, also.
Or is it parasitic C of PCB tracks? Yes, also.
There is no conceoptual difference between Vas/Miller cap and output transistor B-E and B-C capacitance: it's all phase shift (for audio freqs). It all can be seen in the freq response and phase response characteristics of the amp. It is not that an amp has a certain freq/phase response on one hand, and then another secret delay mechanism buried in the transistors on the other hand. Again, David, I know you don't like it, but the black box analogy is also here very powerfull!
Jan Didden
Hi, Janneman,
What is the clue when you are giving the BS😀
I cannot stop thinking about a funnier stuff. IF this phase delay existed. A differential consist of 2 transistors with emitors tied together. The right (inverting input) sense at instance what happened at the output node. The left (non-inverting input) sense the input signal, also instantly. How come with the condition of emitors tied together, the base of the left has signal with different phase with from right transistor, and the whole thing works properly? These 2 transistors usually placed face to face with the other, with very small pcb track distance.
OK, in the case above, maybe it is happening. But how about this one. The JLH classA power amp, with singleton input stage. Input is fed to base, and feedback is fed to emitors.
In one transistor, the base and the emitor is having different phase? I cannot fit into this one. If the phase shift existed here, it have to be caused by transistor's internal capacitance. But FORR said this is a myth.
What is the clue when you are giving the BS😀
I cannot stop thinking about a funnier stuff. IF this phase delay existed. A differential consist of 2 transistors with emitors tied together. The right (inverting input) sense at instance what happened at the output node. The left (non-inverting input) sense the input signal, also instantly. How come with the condition of emitors tied together, the base of the left has signal with different phase with from right transistor, and the whole thing works properly? These 2 transistors usually placed face to face with the other, with very small pcb track distance.
OK, in the case above, maybe it is happening. But how about this one. The JLH classA power amp, with singleton input stage. Input is fed to base, and feedback is fed to emitors.
In one transistor, the base and the emitor is having different phase? I cannot fit into this one. If the phase shift existed here, it have to be caused by transistor's internal capacitance. But FORR said this is a myth.
lumanauw said:
David,
I am not sure I understand your concerns, let me try to answer you.
There is a difference between phase shift, caused by complex component networks, and absolute or inverse phase signals.
For neg feedback to work, there must be some stage where the input signal can be subtracted from the feedback signal. This subtraction happens if you have an adding stage, but have one signal inverted. Adding a + and - is like subtracting two +'s, right.
So, in the differential pair, the output is one of the collector currents, and that is the result of the differences in the two base voltages. So, if you put Vin at one base, and at the other base the nfb, the collector current comes from the difference. Suppose that the Vout is less than it should be. Then the nfb signal is also too low. That means one of the base signals is to low, so the difference between Vin and the nfb increases, the collector current used as output increases, and Vout increases to what it should be. In case the Vout is correct, the Vin and the nfb signal are the same, and there is no correcting collector current (assuming the diff stage gain is infinite, which it isn't so you need a small delta between the base voltages to maintain Vout).
Interestingly, with nfb the amp still works open loop! Say you have an opamp with a gain of 1.000.000, and you want stage gain of 10 times. You wrap nfb around in with R-ratio 1:10. Now, with Vin = 1V you get Vout = 10V, right. But what is the amp input voltage on the input pins? 10V/1.000.000 = 10uV. The amp itself works open loop! Same with that audio power amp diff input stage: the delta between the base voltages is Vout/OL gain. Isn't technology fascinating??
Jan Didden
Hi, Janneman,
How about the other one, the smaller phase shift, not the 0-180 one? For example lets consider the input stage of JLH amp, the singleton one. IF there is phase shift inherent in power amps forward path, If this amp is run openloop (without feedback), it will delay, say 10ns (this is too much, but for example)
If we use feedback (singleton feedback) the signal at base (input) and at emitor (feedback) will differ by 10ns. This could not be, the singleton differential wouldn't allow this delay (by the nature of how 1 transistor works). In the same transistor the signal at base and emitor becomes different by 10ns. Assume like FORR said, there is no effect of capacitance inside transistor. So in this singleton transistor, the B and E is instant relationship without any delay.
What happen in this condition? The open loop amp wants to delay by 10ns, but the input stage (the singleton differential) does not allow this at all. What will happen here?
There are 2 sides. The whole amp's forward transfer that wants to delay, but the differential doesn't allow delay. Who will win? Or the whole thing will be compromized by small delay but producing higher order harmonics?
How about the other one, the smaller phase shift, not the 0-180 one? For example lets consider the input stage of JLH amp, the singleton one. IF there is phase shift inherent in power amps forward path, If this amp is run openloop (without feedback), it will delay, say 10ns (this is too much, but for example)
If we use feedback (singleton feedback) the signal at base (input) and at emitor (feedback) will differ by 10ns. This could not be, the singleton differential wouldn't allow this delay (by the nature of how 1 transistor works). In the same transistor the signal at base and emitor becomes different by 10ns. Assume like FORR said, there is no effect of capacitance inside transistor. So in this singleton transistor, the B and E is instant relationship without any delay.
What happen in this condition? The open loop amp wants to delay by 10ns, but the input stage (the singleton differential) does not allow this at all. What will happen here?
There are 2 sides. The whole amp's forward transfer that wants to delay, but the differential doesn't allow delay. Who will win? Or the whole thing will be compromized by small delay but producing higher order harmonics?
Hi lumanauw !
You are forgetting that for voltageamplification you take the
voltage from collector of the amplifying transistor. The collector has
a quite big capacitance, for a smallsignal ~5pf (2n5401).
This slows down the voltageslewing. There is your phasehift !
That is also the reason why higher current increases the speed here.
In fact, this capacitance is the dominant factor in the speed of
a voltagegaintransistor. (ft gets nearly negligible)
It's correct that between B & E you have nearly no delay, but
you have a basecapacitance that slows down the signal on the
baseside. (Otherwise the vbe would vary)
In addition you have millercapacitance, this means the voltageslew
on the collector is coupled through this cap to the base.
Mike
You are forgetting that for voltageamplification you take the
voltage from collector of the amplifying transistor. The collector has
a quite big capacitance, for a smallsignal ~5pf (2n5401).
This slows down the voltageslewing. There is your phasehift !
That is also the reason why higher current increases the speed here.
In fact, this capacitance is the dominant factor in the speed of
a voltagegaintransistor. (ft gets nearly negligible)
It's correct that between B & E you have nearly no delay, but
you have a basecapacitance that slows down the signal on the
baseside. (Otherwise the vbe would vary)
In addition you have millercapacitance, this means the voltageslew
on the collector is coupled through this cap to the base.
Mike
lumanauw said:
Its not like that at all, David. Lets first assume the feedback and amp have no delay or phase shift. On one side of the adder (and it really is immaterial whether it is a singleton, JLH, diff stage or whatever, so lets not confuse the issue with those irrelevancies), you have Vin. On the other side, a fraction of Vout which will tend to be the same as Vin (assuming infinite gain) plus some distortion. The difference, coming out of the adder, is ... the distortion, but in inverse form to the Vout distortion. That is then amplifed so the distortion is cancelled in the output.
Now add phase shift/delay. The nfb signal is not exactly Vin + HD, so in the adder there will be an output which is not exactly the HD component, so cancellation of the HD is not perfect.
Two points:
- the nfb does work always, but not always (actually never) perfect;
- the nfb HD cancellation gets worse when freq rises, because the phase shift has more and more influence.
Jan Didden
Hi, Mike,
I also thinking the same with you, but FORR give me post #52. I think that the only capacitance that can be seriously enough is transistor's capacitance or miller capacitance. Resistor's internal capacitance or PCB track parasitic capacitance should be negligible to this.
Mike, which is sounds better, Symasym4 or Symasym5evolution? Or now you have another better sounding design?
Hi, Janneman,
This must be one of you BS, because I can understand clearly, and I agree with those 😀
You said HD cancelation gets worse with frequency rises. What is the connection about HD cancelation and number of stages inside a certain power amp? Does feedback works the same effectively in 2 stages power amp and 5 stages power amp for example?
Can I add 1 more question? Does Feedback make higher order harmonics/IM? If it does, what is the mechanism? Maybe there is a way to use feedback without producing higher order harmonics /IM.
lumanauw said:
Yes, feedback creates high order harmonics not present in the input signal. The order and magnitude of them depends on open-loop response of the circuit and on the non-linearities closed inside feedback loop. The case is too complicated to be explained in a few sentences. The cure is wide open loop bandwidth and minimizing of non-linearities before you use global NFB. Local nested FB, for example.
But we also have to speak about amplitude of high order harmonics. It makes difference, if they are -60 dB or - 130 dB 😉
I think that the effect of closing a feedback loop is being overlooked, since closed-loop phase shift works no longer in the same way as when the loop is open (where is the roll-off associated to the pole and the phase shift when the llop is closed?).
Note that one zero is just all what is needed in order to cancel any phase shift due to a pole, and this is just what feedback does when the loop is closed, it simulates a zero and cancels open-loop phase shift. Anyway, everything has instantaneous response in the audio band. Remember that phase shift does not cause any actual delay in the signal, altough a phase-shifted steady-state sine wave might look like delayed at first sight.
Also, a non linear mechanism (exponential, quadratic, logarithmic, etc...) is required in order to create new frequencies at the output that were not present at the input of the system. So there is no way for simple "summing" linear feedback to create any new harmonics, intermodulation, etc... that weren't already present in the open loop response. All that feedback is able to do is attenuate them.
Anyway, I think I would rather leave now that thread because it may start to grow uncontrolled and I don't like science fiction applied to electronics.
Note that one zero is just all what is needed in order to cancel any phase shift due to a pole, and this is just what feedback does when the loop is closed, it simulates a zero and cancels open-loop phase shift. Anyway, everything has instantaneous response in the audio band. Remember that phase shift does not cause any actual delay in the signal, altough a phase-shifted steady-state sine wave might look like delayed at first sight.
Also, a non linear mechanism (exponential, quadratic, logarithmic, etc...) is required in order to create new frequencies at the output that were not present at the input of the system. So there is no way for simple "summing" linear feedback to create any new harmonics, intermodulation, etc... that weren't already present in the open loop response. All that feedback is able to do is attenuate them.
Anyway, I think I would rather leave now that thread because it may start to grow uncontrolled and I don't like science fiction applied to electronics.
Of course the overall error is reduced but by modifying the gain characteristic feedback shifts lower order to higher harmonics. Because higher order harmonics are not masked well by the human ear this distortion imight be more audible although it's (much) lower in value
Feedback does not prevent the designer to look at the open loop characteristic of an amp. Open loop distortion of an amp should reduced as much as possible. This might be done by simulation. Applying global feedback then stabilises the complete system, defines the bandwidth and reduces distortion.
lumanauw said:
Hi lumanauw, i think you understood FORR wrong, these capacitances
cannot create delay, only slowdown/phasehift.
About symasym, i prefer no5, there is not no6 yet, but is sure to come.
Mike
lumanauw said:
Taske out the black box again! The fb doesn't "know" how many stages there are in the amp. The important properties are OL transfer function and fb factor. Whether you have 2 or 5 stages inside the black box doesn't matter!
Your other questions have been addressed by others and I agree fully to those remarks. Those are the second layers of the onion you meet when you peel back the basic, simple ones we have been discussing.😉
Jan Didden
bocka said:
Bocka,
Not sure I follow you here. What I can follow is that harmonic components in the output are re-inserted in the input and may give rise to harmonics-of-harmonics and sum- and difference frequencies with the input signal. This is caused by the same non-linearities that generated the original harmonic components in the first place. Is that what you mean?
Jan Didden
Hi Jan,
the output voltage of an open loop amp is
Uo = Vo * Uin
where Uin is the input voltage and Vo ist the open loop gain.
When you apply a feedback loop with a feedback factor K then the voltage of
Um = 1/K * Uo is fed back to the summing point, where
Uin = Up - Um
Up is now the input voltage.
This can be written as
Uo = Vo * Uin
= Vo * (Up - Um)
= Vo * (Up - 1/K * Uo)
Uo * (1 + 1/K * Vo ) = Up * Vo
So the closed loop output voltage is
Uo = Go(uin) / (1 + 1/K * Vo) * Up
= K / (1 + K / Vo)
Well known maths so far. Note that Vo is a function of the input voltage and not only of the ampitude but also of the frequency. To make this alittle bit clearer lets write
Vo = Vo(Up - Um)
Now the output voltage is
Uo = K / (1 + K / Vo(Up - Um)) * Up
If Vo >> K we have the typical output voltage of
Uo = K * Up
But we can see more. The closed loop characteristic is different than the scaled open loop characteristic
K1 * Vo(Up) != K / (1 + K / Vo(Up - Um)) * Up
We are not able to apply any constant factor for K1 that the formula above becomes true. Feedback introduces different harmonics which are not present in the open loop gain characteristic.
If you apply now a Taylor expansion of Vo of the term
(1 + K / Vo(Up - Um))
there are higher order terms present, if you are applying anything different than a constant for Vo. And this means higher order distortion.
the output voltage of an open loop amp is
Uo = Vo * Uin
where Uin is the input voltage and Vo ist the open loop gain.
When you apply a feedback loop with a feedback factor K then the voltage of
Um = 1/K * Uo is fed back to the summing point, where
Uin = Up - Um
Up is now the input voltage.
This can be written as
Uo = Vo * Uin
= Vo * (Up - Um)
= Vo * (Up - 1/K * Uo)
Uo * (1 + 1/K * Vo ) = Up * Vo
So the closed loop output voltage is
Uo = Go(uin) / (1 + 1/K * Vo) * Up
= K / (1 + K / Vo)
Well known maths so far. Note that Vo is a function of the input voltage and not only of the ampitude but also of the frequency. To make this alittle bit clearer lets write
Vo = Vo(Up - Um)
Now the output voltage is
Uo = K / (1 + K / Vo(Up - Um)) * Up
If Vo >> K we have the typical output voltage of
Uo = K * Up
But we can see more. The closed loop characteristic is different than the scaled open loop characteristic
K1 * Vo(Up) != K / (1 + K / Vo(Up - Um)) * Up
We are not able to apply any constant factor for K1 that the formula above becomes true. Feedback introduces different harmonics which are not present in the open loop gain characteristic.
If you apply now a Taylor expansion of Vo of the term
(1 + K / Vo(Up - Um))
there are higher order terms present, if you are applying anything different than a constant for Vo. And this means higher order distortion.
bocka said:
Fascinating! All of a sudden I understand much better that old graph showing that initially, applying fb, increases the level of each individual harmonic, until the fb is increased above a certain factor and then the harmonic components also decrease. Also, that graph shows clearly the emergence of harmonics not present without fb. Thanks, another puzzle solved.
Jan Didden
Eva said:
Non-linear open-loop circuit creates new harmonics (compared to input signal). When closed into feedback, these new harmonics are fed back to the input, resulting in further new harmonics not present at the output of the open-loop circuit. As a result, feedback usually well reduces lower order harmonics to negligible level, but can create very high order harmonics, depending on the open-loop non-linearity. You can simply measure it.
PMA (about high frequency distorsion)
"The cure is wide open loop bandwidth and minimizing of non-linearities before you use global NFB. Local nested FB, for example."
This is a thinking reminiscent of Ottala.
As wide open loop bandwith is only obtained by reducing the open loop at low frequencies, it is not an efficient cure. Making the circuit very linear before applying feedback is now universaly admitted as the good strategy.
MIKEB
"Lumanauw, I think you understood FORR wrong,
these capacitances cannot create delay, only slowdown/phasehift."
Thanks, MikeB, this is exactly what I mean.
~~~~~~~~ Forr
§§§
"The cure is wide open loop bandwidth and minimizing of non-linearities before you use global NFB. Local nested FB, for example."
This is a thinking reminiscent of Ottala.
As wide open loop bandwith is only obtained by reducing the open loop at low frequencies, it is not an efficient cure. Making the circuit very linear before applying feedback is now universaly admitted as the good strategy.
MIKEB
"Lumanauw, I think you understood FORR wrong,
these capacitances cannot create delay, only slowdown/phasehift."
Thanks, MikeB, this is exactly what I mean.
~~~~~~~~ Forr
§§§
Now this is turning funny. Thanks Bocka for providing a derivation.
My initial point got somehow lost in the semantics issue and I could not express it in a formula. So I made a philosophical / logical point if you will.
I used the word "delay" (and I still maintain that "phase" as a concept makes sense for a periodic signal only, and that what we're after is the information conatined in the time dependent arrival of the waveform, not the infinitesimal first electron arriving quasi-immediately at the summing point through feedback). The point I was seeking to express is that feedback comes "after the fact", and somewhere above (or in the previous thread) I called feedback a "nonlinear fix for a nonlinear problem".
In any equation of the form d'something'/dt you may make dt arbitrarily small, but not zero. Furthermore feddback at input requires prior output. So by definition feedback must contain an element of "delay" or if that word is taboo because of prior commitment to other uses, then let's call it "order of events" (in my book, another definition of time itself but anyway...).
As it happens Bocka gave a nice derivation on how in a typical feedback amp feedback must lead to a nonlinear relation, quite not unlike the logistic equation, notorious for producing either stability, periodicity (oscillation) or chaos ;-) - depending on parameters.
I find the whole subject interesting out of mostly theoretical aspects of what happens in feedback systems and I usually try to visualize it w/o the maths.
In any case I am not one of those who seek "shorter feedback paths" through making the resistors a little shorter. My point was completely different.
Note on "high order harmonics", off topic: practically speaking, what ppl often seem to overlook is that our most sensitive hearing occurs at ca. 3000 Hz. This is the 150th harmonic of 20 Hz. Funny that ppl usually don't pay much attention to HD at low frequencies, when due to Fletcher Munson the high order HD created at low frequencies should matter a great deal more than the ones created at 1000 Hz.
HD at low frequencies may explain the somehow puzzling observations that this or that op amp has "strong" or "woolly" bass when the FR is identical to the one next to it. A strong 5th order harmonic of a 40 Hz fundamental falls at 200 Hz, where hearing is likely up to 10 dB more sensitive than at 40 Hz. That should be enough to muddy up both the bass and midrange.
In other words if the available solution through feedback is to push the distortions into such high order as to drive them out of the audio range, one should have only distortions past 20 kHz, which is the 1000th harmonic of 20 Hz. Or, less stringently, taking 200 Hz as a base, supposedly the most energy rich spectrum of music, and freedom of HD until at least 5 kHz as a design target, we should target to only have HD beyond 5000/200=25th order. Multi amping makes this easier of course - then the higher order harmonics are not passed on to the higher frequency drivers.
My initial point got somehow lost in the semantics issue and I could not express it in a formula. So I made a philosophical / logical point if you will.
I used the word "delay" (and I still maintain that "phase" as a concept makes sense for a periodic signal only, and that what we're after is the information conatined in the time dependent arrival of the waveform, not the infinitesimal first electron arriving quasi-immediately at the summing point through feedback). The point I was seeking to express is that feedback comes "after the fact", and somewhere above (or in the previous thread) I called feedback a "nonlinear fix for a nonlinear problem".
In any equation of the form d'something'/dt you may make dt arbitrarily small, but not zero. Furthermore feddback at input requires prior output. So by definition feedback must contain an element of "delay" or if that word is taboo because of prior commitment to other uses, then let's call it "order of events" (in my book, another definition of time itself but anyway...).
As it happens Bocka gave a nice derivation on how in a typical feedback amp feedback must lead to a nonlinear relation, quite not unlike the logistic equation, notorious for producing either stability, periodicity (oscillation) or chaos ;-) - depending on parameters.
I find the whole subject interesting out of mostly theoretical aspects of what happens in feedback systems and I usually try to visualize it w/o the maths.
In any case I am not one of those who seek "shorter feedback paths" through making the resistors a little shorter. My point was completely different.
Note on "high order harmonics", off topic: practically speaking, what ppl often seem to overlook is that our most sensitive hearing occurs at ca. 3000 Hz. This is the 150th harmonic of 20 Hz. Funny that ppl usually don't pay much attention to HD at low frequencies, when due to Fletcher Munson the high order HD created at low frequencies should matter a great deal more than the ones created at 1000 Hz.
HD at low frequencies may explain the somehow puzzling observations that this or that op amp has "strong" or "woolly" bass when the FR is identical to the one next to it. A strong 5th order harmonic of a 40 Hz fundamental falls at 200 Hz, where hearing is likely up to 10 dB more sensitive than at 40 Hz. That should be enough to muddy up both the bass and midrange.
In other words if the available solution through feedback is to push the distortions into such high order as to drive them out of the audio range, one should have only distortions past 20 kHz, which is the 1000th harmonic of 20 Hz. Or, less stringently, taking 200 Hz as a base, supposedly the most energy rich spectrum of music, and freedom of HD until at least 5 kHz as a design target, we should target to only have HD beyond 5000/200=25th order. Multi amping makes this easier of course - then the higher order harmonics are not passed on to the higher frequency drivers.
Hi, EVA,
I'm not as sophisticated as you, but I happens to also think the same thing. If an amp in open loop mode have a phase shift. This phase shift is happening naturally, caused by any C inherent in the CCT.
If we make the same amp with feedback, the differential will not allow any phase shift, it will be wrong for the differential, because the differential works in instant phase, in both bases, input and output. What happened to the "natural phase shift" that naturally should be contained in the CCT, but becomes disabled by the differential pair? I'm not clear here.
Hi, Bocka,
Nice explenation 😀 So feedback will create higher order harmonics, regardless of the cct (simple or complicated) used.
Hi, FORR,
I understand now.😀 The C's are not making delay, but phase shift. Actually it is non-feedback amp that will suffer this Delay (or is it phase shift?😀), not the feedback amp. It is imposible to have delay if we use differential or singleton input stage. Do you agree that more stages in power amp (more B's tobe passed by signal) will give more phase shift? If it is true, then many stages power amp will always have more or higher order of "high order harmonics" than 2 stages power amp.
Hi, Janneman
How come you can get "picture" of the steps of feedback, when all happens at instance?
Hi, MBK,
I think if one amp has high order harmonics, it will be spreaded all over 20hz-20khz. You cannot make "clean or free from high order harmonics" for below 20khz, and it may occur above 20khz (for example), it is impossible to do this. The only way is to prevent high order harmonics to not happening at all.
I'm not as sophisticated as you, but I happens to also think the same thing. If an amp in open loop mode have a phase shift. This phase shift is happening naturally, caused by any C inherent in the CCT.
If we make the same amp with feedback, the differential will not allow any phase shift, it will be wrong for the differential, because the differential works in instant phase, in both bases, input and output. What happened to the "natural phase shift" that naturally should be contained in the CCT, but becomes disabled by the differential pair? I'm not clear here.
Hi, Bocka,
Nice explenation 😀 So feedback will create higher order harmonics, regardless of the cct (simple or complicated) used.
Hi, FORR,
I understand now.😀 The C's are not making delay, but phase shift. Actually it is non-feedback amp that will suffer this Delay (or is it phase shift?😀), not the feedback amp. It is imposible to have delay if we use differential or singleton input stage. Do you agree that more stages in power amp (more B's tobe passed by signal) will give more phase shift? If it is true, then many stages power amp will always have more or higher order of "high order harmonics" than 2 stages power amp.
How to do this? I assume the non-linear thing is the transistors gain curve. It is usually made linear by using RE degeneration. Is this how to do your advice?
Hi, Janneman
I think what is inside the feedback loop is at instance. I even cannot know what happens in the first question above (that I ask EVA in this post).
How come you can get "picture" of the steps of feedback, when all happens at instance?
Hi, MBK,
I think if one amp has high order harmonics, it will be spreaded all over 20hz-20khz. You cannot make "clean or free from high order harmonics" for below 20khz, and it may occur above 20khz (for example), it is impossible to do this. The only way is to prevent high order harmonics to not happening at all.
- Status
- Not open for further replies.