lumanauw said:
David, (if you don't mind my commenting on this),
The Vbe of Q4, Q5 keep the Vcb of T6, T7 almost constant, so you could say that Q4, Q5 are driven from a constant voltage. That implies low drive impedance to Q4, Q5 bases.
BTW, there is an error in this scheme: with the feedback network as drawn, the gain is 1.
Jan Didden
Hi Lumanauw,
I tried non-complementary circuits similar to this but gave up.
There is a dynamic imbalance in differential/mirror current draw at high AF due to Cbc loading as T5 swings when T2 does not.
This can increase the risk of output stage cross-conduction.
Fixed capacitor balancing at T2 can help, but not cure.
I tried output rail cascoding the T5 limb too, but became satisfied that lack of improvement failed to justify the additional complexity.
Cheers ......... Graham.
I tried non-complementary circuits similar to this but gave up.
There is a dynamic imbalance in differential/mirror current draw at high AF due to Cbc loading as T5 swings when T2 does not.
This can increase the risk of output stage cross-conduction.
Fixed capacitor balancing at T2 can help, but not cure.
I tried output rail cascoding the T5 limb too, but became satisfied that lack of improvement failed to justify the additional complexity.
Cheers ......... Graham.
HI Johan Potgieter,
---Constant voltage gain, at least into practical loads, was there before NFB.---
Historically, I'm not sure at all of that.
What we can say is that negative feedback tames non-linearities and sets gain in a large bandwidth precisely with only two resistors which are dependable components.
---Constant voltage gain, at least into practical loads, was there before NFB.---
Historically, I'm not sure at all of that.
What we can say is that negative feedback tames non-linearities and sets gain in a large bandwidth precisely with only two resistors which are dependable components.
Hi, Janneman,
Thanks for the explenation.
If it is not constant Vcb, like ordinary current mirror loaded differential (right current mirror is made diode connection), it will be high impedance drive?
Hi, Graham,
I'm not so clear. Isn't that T4+T5 forming differential VAS? I see that T4 does not experience full swing (because V2 is common source), while V1 is follower (T5 experience full swing). Do you mean Cbc of T4 is different from Cbc of T5? How important is this Cbc imbalance to T2-T3 working POV?
I happens to auditioned 1 commercial amplifier that uses exactly this scheme. There is something not right about the sound, I wanted to know what causes it.
Thanks for the explenation.
I tought that to get low drive impedance for VAS base, putting R (instead of current mirror) to upper rail is the only way to get low drive impedance. This is new to me. If the source is collector-collector, it is still possible to make it low impedance drive by making constant voltage Vcb like your explenation?
If it is not constant Vcb, like ordinary current mirror loaded differential (right current mirror is made diode connection), it will be high impedance drive?
Hi, Graham,
I'm not so clear. Isn't that T4+T5 forming differential VAS? I see that T4 does not experience full swing (because V2 is common source), while V1 is follower (T5 experience full swing). Do you mean Cbc of T4 is different from Cbc of T5? How important is this Cbc imbalance to T2-T3 working POV?
I happens to auditioned 1 commercial amplifier that uses exactly this scheme. There is something not right about the sound, I wanted to know what causes it.
janneman said:
Hi Jan,
I looked at this rather carefully (as I always do with your posts!), and I have a problem. The collectors of T6, T7 are indeed a quite high source impedance by virtue of current feedback through R6, R7, and so are the collectors connected to them from the bottom input pair (forgot to note the numbers).
Now, the fact that the Vbe of a transistor is somewhat constant, has to do with the "internal" r(b-e) etc. of that transistor. That will be so whether it is fed from say 1K or 1M.ohm. Surely that cannot mean that the 1M is no longer a high (source) impedance or a current source? Or what am I missing?
Regards.
Graham Maynard said:
G’day Graham.
The cross-conduction limitation of that design is due to the fact that the output MOSFET’s are driven from a relatively high impedance, with only a resistive ‘pull-down’ to fight the large input capacitance when one of the MOSFET's is being driven to cut-off.
It would have virtually noting to do with the miller capacitance of T5, whose effects are small in comparison. That would explain why cascoding T5 made no difference.
Such a design topology might be passably OK for class A operation, but far from optimal for class AB, but you would need to use low value gate-source resistors for the output MOSFET's and run the second differential amplifier quite hot.
Cheers,
Glen
Hi Forr,
Well, one is talking relatively in a sense (as was brought home by the quip from estuart!). With proper design one could get a frequency response within 2 - 3dB over the audio band without NFB; that was in fact achieved in the better amplifiers of the days before NFB (triodes, 2A3 p.p. output stage with a quite fat output transformer and such - that was before my time but the information is available).
But the influence of NFB on the 3 separate points that I tried to define is of course quite interrelated. I was simply trying to say that long after a say 1dB frequency response has been achieved by NFB in any half-decent circuit, it might still not be sufficient to achieve the desired feedback goal. Ergo, one chooses an NFB factor with the latter rather than the former in view. That is then also why the question is often asked, why showing a frequency response of say 10 Hz - 50 KHz (or wider)? - with the simple answer that that is the automatic by-product of the degree of NFB used to achieve a certain low distortion figure.
But NFB is of course also used to obtain mainly frequency characteristics, as in a pre-amplifier.
Regards.
Well, one is talking relatively in a sense (as was brought home by the quip from estuart!). With proper design one could get a frequency response within 2 - 3dB over the audio band without NFB; that was in fact achieved in the better amplifiers of the days before NFB (triodes, 2A3 p.p. output stage with a quite fat output transformer and such - that was before my time but the information is available).
But the influence of NFB on the 3 separate points that I tried to define is of course quite interrelated. I was simply trying to say that long after a say 1dB frequency response has been achieved by NFB in any half-decent circuit, it might still not be sufficient to achieve the desired feedback goal. Ergo, one chooses an NFB factor with the latter rather than the former in view. That is then also why the question is often asked, why showing a frequency response of say 10 Hz - 50 KHz (or wider)? - with the simple answer that that is the automatic by-product of the degree of NFB used to achieve a certain low distortion figure.
But NFB is of course also used to obtain mainly frequency characteristics, as in a pre-amplifier.
Regards.
Johan Potgieter said:
Johan,
Well, the way I look at it, Q4,5 are driven from a circuit node that does vary very little in voltage with the current that it sources into Q4,5. That's the definition of a voltage source.
Sometimes circuits have an unexpected twist..😉
Jan Didden
Hi Glen.
When I tried that kind of amplifier circuit my output devices were a conventional bipolar class-AB arrangement, not Mosfet.
Yes for Mosfets the turn off could be improved, such that an inverted pull down PNP with bootstraped and temperature sensed maximum bias arrangement could be implemented, but that would insert another path device within the NFB loop.
Hi lumanauw.
There was something else I did not like about the circuit which I could not see any way around. Where the input differential might over-react via NFB due to unavoidable delay of the devices being driven, the mirror could momentarily reverse bias the base-emitter junctions of either T4 or T5.
Cheers ....... Graham.
When I tried that kind of amplifier circuit my output devices were a conventional bipolar class-AB arrangement, not Mosfet.
Yes for Mosfets the turn off could be improved, such that an inverted pull down PNP with bootstraped and temperature sensed maximum bias arrangement could be implemented, but that would insert another path device within the NFB loop.
Hi lumanauw.
There was something else I did not like about the circuit which I could not see any way around. Where the input differential might over-react via NFB due to unavoidable delay of the devices being driven, the mirror could momentarily reverse bias the base-emitter junctions of either T4 or T5.
Cheers ....... Graham.
Hi, Janneman,
You just tought me one important lesson (maybe yourself don't realize it 😀). To see a problem from a bigger scope.
I see that CCT only from collector-collector, not realizing what is the bigger picture. Driven from voltage source=low source impedance.
But something still annoying. I use the "inverting amplifier" POV. While in the inverting amplifier the inverting node is said to be sitting at 0V (virtually), it happens because the non-inverting node is sitting at 0V (and the whole amp tries to make the same of inverting node and non-inverting node). There is no real connection to 0V at the inverting node here. Theoritically it is true, that the inverting node is at virtual 0V, but in real practice inverting amplifiers exhibit many strange behavior especially in HF.
Can I use the same POV for this case? The CCT that I asked is "virtually" driven by low impedance?
Hi, Graham,
There is something not right about the sound. I will try to check whether it is caused by some possible reasons you mentioned.
You just tought me one important lesson (maybe yourself don't realize it 😀). To see a problem from a bigger scope.
I see that CCT only from collector-collector, not realizing what is the bigger picture. Driven from voltage source=low source impedance.
But something still annoying. I use the "inverting amplifier" POV. While in the inverting amplifier the inverting node is said to be sitting at 0V (virtually), it happens because the non-inverting node is sitting at 0V (and the whole amp tries to make the same of inverting node and non-inverting node). There is no real connection to 0V at the inverting node here. Theoritically it is true, that the inverting node is at virtual 0V, but in real practice inverting amplifiers exhibit many strange behavior especially in HF.
Can I use the same POV for this case? The CCT that I asked is "virtually" driven by low impedance?
Hi, Graham,
There is something not right about the sound. I will try to check whether it is caused by some possible reasons you mentioned.
lumanauw said:
Actually, it was Einsteins' lesson: 'Things should be made as simple as possible, but not simpler'.
You remember the 'black box' approach? Same thing.. 😀
Jan Didden
lumanauw said:
David,
There is nothing mystical about the non-inverting 'virtual earth' input. Remember, that the opamp doesn't know what is going on outside. It has been built to amplify the difference between it's two inputs by it's open loop gain. There is nothing we can do to change that.
So, in a feedback amp, if the non-inverting node is grounded, the inverting input is equal to Vo/Aol. Because the Aol is often, say 1 million (at low freq), the inverting input is only a few microvolts. We say then it is virtual earth, but it is not zero of course, otherwise Vo would be zero.
And with rising freq, the Aol drops, and the inverting input level rises also. That means that the actual input (diff between V+ and V-) rises, it becomes a greater and greater part of the system Vin. That greater part is amplified with that Aol non-linearity so that non-linearity has more and more influence on the Vo with rising freq. That's why distortion rises with freq. Simple, really.
That is also why we have the residual distortion even with high feedback, and that is why we have those high-order intermodulation products: what ever is at those two inputs, will be amplified by the open loop gain with all it's open-loop non-linearity.
Jan Didden
janneman said:
Sorry, Jan, I still have a step or two to take here!
The definition you gave is only strictly correct when looking "back" into the source, ignoring the nature of the load. The load here (a diode junction), is holding the voltage across it relatively constant by its internal nature. If that were to be different, the voltage over it would have changed in accordance with load characteristics. That is the definition of a current source.
Let us just check definitions. As I have it: A voltage source has a very low internal impedance, as a current source has a high internal impedance, irrespective of what happens over the (outside) load.
A different approach: You would agree that a current source will be happy feeding a short-circuit - in that case there is no voltage variation over the load with current. That indicates a low (=0) load resistance, not a low source resistance. In a manner of thinking, an adequate voltage source would have blown that short circuit load open with mismatch, not just supplied what it "wanted".
I would again offer consideration of a supply with a 1 meg internal resistance, feeding T4,5. The behaviour would be no different . That does not "convert" the 1 meg to a very low value; the source still has an internal resistance of 1 meg, i.e. it remains a current source.
It might be that stating that the feeding circuit node "vary very little in voltage with the current that it sources" needs a little further consideration.
Thanks for your patience!
My take on the drive to T4/5 in the post#820 illustration is that it is floating at high impedance, purely of differential current in nature, and such that T4/5 variation of Vbe with current becomes irrelevent.
This being a reason for T4/5 Cbc differences having a greater consequence for attempted NFB activities within the closed loop bandwidth. Also a reason why the Vbe potentials can become momentarily reversed.
This being a reason for T4/5 Cbc differences having a greater consequence for attempted NFB activities within the closed loop bandwidth. Also a reason why the Vbe potentials can become momentarily reversed.
Johan Potgieter said:
Hello Johan,
That's a pretty well reasoned post, I really can't disagree with it. I think the weak point in my reasoning is that the low impedance is caused by the load (B-E) rather than intrinsically by the source, as you correctly pointed out.
Jan Didden
Jan,
...at this stage, after having struggled through a different barbed thread, where comparatively little science survived in the end :🙁:
Thanks for the respectful reply, also evident from your other posts in general. Why do folks appear so keen to judge others by their worst performance, instead of noticing successes?
Just thought I would mention it ....
Regards
...at this stage, after having struggled through a different barbed thread, where comparatively little science survived in the end :🙁:
Thanks for the respectful reply, also evident from your other posts in general. Why do folks appear so keen to judge others by their worst performance, instead of noticing successes?
Just thought I would mention it ....
Regards
On the need for output coils
With the audibility of output coils thread effectively dead, I thought it would be a good idea to introduce a discussion of the NEED for output coils into this thread, since in many respects the need for output coils is an issue largely related to negative feedback.
The other thread of course touched on this issue in several places, but the discussion was very dispersed, and clouded by the central contrversial issue of the that thread, namely the audibility of the coil. I would suggest that discussion of the audibility be considered off-topic here so that we can focus more on the design aspects and less on subjective disagreements.
Indeed, for the purposes of this discussion, let's assume that the coil may be audible under some conditions, and that the coil is something that we would rather avoid or minimize if we can.
Let me start the ball rolling with a few comments and observations.
Some of us designers may be using them almost from a historical basis, and they may not be needed at all or as much as we think, so the need, and those conditions under which they are needed, should be re-examined.
In a related question, some have suggested that modern amplifiers that use faster output transistors may be less prone to destabilization by a non-isolated capacitive load, or, again, the size of the needed coil may be less.
By the same token, those who claim that they do not need an output coil may not be designing as conservatively for the worst case load. In other words, it may not be that their design is better or different, it just may be a different stability criteria in regard to what bad load they will tolerate.
This segways into the need for a common way of assessing the stability of an amplifier into a commonly accepted load template. For example, perhaps we can agree that the amplifier must be adequately stable into the following set of test loads:
1000 pF at the output terminals with and without 8 ohm load.
0.01 uF at the output terminals with and without 8 ohm load.
0.1 uF at the output terminals with and without 8 ohm load.
2 uF in series with 0.5 ohm with and without 8 ohm load.
In the same test, perhaps the stability template for the above conditions should be something like this:
Apply a 1 V p-p square wave at the input, bypassing the input LPF network, and view the square wave at the output, prior to the output coil (if one is used). First cycle overshoot should not exceed 20 percent. Each cycle of ringing should be half the value of the prior, starting with the peak-to-peak value of the cycle after the initial overshoot being no more than 10% peak-to-peak.
This would level the playing field for results among different amplifiers.
The above template suggestions are just meant to get the ball rolling.
There should also be some discussion that recognizes that not all of the stability issues are in regard to the global feedback loop. For example, it is well-known that emitter followers don't usually like to look directly into a capacitive load, so local VHF instability must be considered as well. This could be especially important with high-ft BJTs and MOSFETs in light of ineveitable stray inductances in the output stage layout.
Finally, there needs to be discussion of the possibility of destabilization under large-signal conditions and even clipping.
Cheers,
Bob
With the audibility of output coils thread effectively dead, I thought it would be a good idea to introduce a discussion of the NEED for output coils into this thread, since in many respects the need for output coils is an issue largely related to negative feedback.
The other thread of course touched on this issue in several places, but the discussion was very dispersed, and clouded by the central contrversial issue of the that thread, namely the audibility of the coil. I would suggest that discussion of the audibility be considered off-topic here so that we can focus more on the design aspects and less on subjective disagreements.
Indeed, for the purposes of this discussion, let's assume that the coil may be audible under some conditions, and that the coil is something that we would rather avoid or minimize if we can.
Let me start the ball rolling with a few comments and observations.
Some of us designers may be using them almost from a historical basis, and they may not be needed at all or as much as we think, so the need, and those conditions under which they are needed, should be re-examined.
In a related question, some have suggested that modern amplifiers that use faster output transistors may be less prone to destabilization by a non-isolated capacitive load, or, again, the size of the needed coil may be less.
By the same token, those who claim that they do not need an output coil may not be designing as conservatively for the worst case load. In other words, it may not be that their design is better or different, it just may be a different stability criteria in regard to what bad load they will tolerate.
This segways into the need for a common way of assessing the stability of an amplifier into a commonly accepted load template. For example, perhaps we can agree that the amplifier must be adequately stable into the following set of test loads:
1000 pF at the output terminals with and without 8 ohm load.
0.01 uF at the output terminals with and without 8 ohm load.
0.1 uF at the output terminals with and without 8 ohm load.
2 uF in series with 0.5 ohm with and without 8 ohm load.
In the same test, perhaps the stability template for the above conditions should be something like this:
Apply a 1 V p-p square wave at the input, bypassing the input LPF network, and view the square wave at the output, prior to the output coil (if one is used). First cycle overshoot should not exceed 20 percent. Each cycle of ringing should be half the value of the prior, starting with the peak-to-peak value of the cycle after the initial overshoot being no more than 10% peak-to-peak.
This would level the playing field for results among different amplifiers.
The above template suggestions are just meant to get the ball rolling.
There should also be some discussion that recognizes that not all of the stability issues are in regard to the global feedback loop. For example, it is well-known that emitter followers don't usually like to look directly into a capacitive load, so local VHF instability must be considered as well. This could be especially important with high-ft BJTs and MOSFETs in light of ineveitable stray inductances in the output stage layout.
Finally, there needs to be discussion of the possibility of destabilization under large-signal conditions and even clipping.
Cheers,
Bob
IMO, that's a pretty rigorous test, if you bypass the LPF in the input. I'll be curious how my amp does, and it wouldn't surprise me if the answer is "not all that great". Should the test also include a similar sequence of inductances in series with a test load?
Hi Bob,
You are addressing two issues:
1. The need for an output coil.
We can only answer this question if we know the worst case impedance of a loud speaker in terms of a equivalent L-R-C network.
2. How to qualify and quantify the stability of an amplifier.
Your proposal seems to be a very good starting point and indeed, one should take notice about HF local instability and stray inductance, especially those who are simulating amp's. So, don't forget these evil inductances, because just these are the culprits, who make an amp HF unstable if loaded capacitively.
Cheers, Edmond
You are addressing two issues:
1. The need for an output coil.
We can only answer this question if we know the worst case impedance of a loud speaker in terms of a equivalent L-R-C network.
2. How to qualify and quantify the stability of an amplifier.
Your proposal seems to be a very good starting point and indeed, one should take notice about HF local instability and stray inductance, especially those who are simulating amp's. So, don't forget these evil inductances, because just these are the culprits, who make an amp HF unstable if loaded capacitively.
Cheers, Edmond
Conrad Hoffman said:
Yes, you are right, it probably should. That would verify that a design that did not have an R-C Zobel in it would be stable into an inductive load. Maybe 1 mH in series with a 4 or 8 ohm load would do.
Bob