Well the whole idea that nfb improves a stage before or after is incorrect. Any improvement in gain within a loop will improve the performance of the loop as a whole. More Vas gain will improve the whole amp, it will not improve the output stage.
The important idea is that the amp ALWAYS works in open loop, whatever the nfb around it. Nfb does nothing to the amp or its stages, it only manipulates the input signal to make the overall performance better. But the stages inside the amp do not change.
In your example, the nfb and the Vas gain improve the performance of the loop. Since the i/p stage is outside the loop, the i/p stage performance impact on the whole amp will not change, I agree. But that is not because it is before the Vas, it is because it is outside the loop.
If you would have an extra inverting stage inside the loop and connect the nfb to the +input (parallel feedback) and the -input to ground, the i/p stage is now inside the loop and the total loop is improved, even when the i/p stage is before the vas.
jan
The important idea is that the amp ALWAYS works in open loop, whatever the nfb around it. Nfb does nothing to the amp or its stages, it only manipulates the input signal to make the overall performance better. But the stages inside the amp do not change.
In your example, the nfb and the Vas gain improve the performance of the loop. Since the i/p stage is outside the loop, the i/p stage performance impact on the whole amp will not change, I agree. But that is not because it is before the Vas, it is because it is outside the loop.
If you would have an extra inverting stage inside the loop and connect the nfb to the +input (parallel feedback) and the -input to ground, the i/p stage is now inside the loop and the total loop is improved, even when the i/p stage is before the vas.
jan
That does not sound right to me, if different stages in the amp would interact/help/make/influence the amount of 'error-signal-corrected' then why are these stages missing from all formulae that (I have seen) calculate THD in (op)amp circuits? http://www.analog.com/library/analogDialogue/archives/43-09/Edch 1 op amps.pdf
Attachments
An exemple to substantiate my understanding.
Let say an IPS with 20dB gain and 1% THD.
Whatever the following stages , in the best case there s only
20dB NFB available for the IPS , that is , it wont do lower than 0.1% THD
once the amp s global loop is closed and the amp s total THD will be no less than 0.1%.
As for the formulae above , i dont see what it is related to ,
in respect to the question on the same post...
Let say an IPS with 20dB gain and 1% THD.
Whatever the following stages , in the best case there s only
20dB NFB available for the IPS , that is , it wont do lower than 0.1% THD
once the amp s global loop is closed and the amp s total THD will be no less than 0.1%.
As for the formulae above , i dont see what it is related to ,
in respect to the question on the same post...
...As for the formulae above , i dont see what it is related to ,
in respect to the question on the same post...
The point is, there is no allowance for the number (or type) of stages inside the amp that are mentioned in the THD formulae (or any other formulae related to distortion or feedback). (maybe the selected formulae was not the best one
)Yes. I have tried to make the best out of the dual LTP with JFET input (a-la Goldmund) topology. The problem with such topology is that the JFET is biased relatively low (JFET wants high current) so that the distortion of the IPS is very high. Even tho this is in the end corrected by closed loop FB, I believe that the sound quality will suffer (only perceivable by trained ears). Now I think this might be the reason why I don't like JFET input, because the local stage distortion is probably very high in many circuits.
I have temporarily dropped the dual LTP topology project because it is not easy to design the whole amp when the input bias current is above 10mA.
The point is, there is no allowance for the number (or type) of stages inside the amp that are mentioned in the THD formulae (or any other formulae related to distortion or feedback). (maybe the selected formulae was not the best one
The THD formula you showed is related only with input versus output, whether it is one stage or a thousand stages doesn't matter, it only cares with the final THD out of the last stage.
And? is there anything else that matters? Maybe my point was that I totally agree to Jan's point.
An exemple to substantiate my understanding.
Let say an IPS with 20dB gain and 1% THD.
Whatever the following stages , in the best case there s only
20dB NFB available for the IPS , that is , it wont do lower than 0.1% THD
once the amp s global loop is closed and the amp s total THD will be no less than 0.1%.
As for the formulae above , i dont see what it is related to ,
in respect to the question on the same post...
Wahab, if that i/p stage is followed with a Vas with 40db gain (asssume for the discussion that the Vas is distortion free).
You want an amp closed loop gain of 20 dB you have 40dB feedback gain to throw away as nfb and then your amp distortion goes down to 1% -40dB which is 0.01% if the ni/p stage is within the loop.
If the i/p stage is not within the loop the total amp distortion will remain 1%.
As FdW said, the formula for the nfb distortion reduction says nothing about each stage distortion reduction. The reason is that there is NO change to each stage distortion. How can the i/p stage suddenly distort less, there is no change to it!
As I said, the whole of the amp, all stages, continues to work open loop even when you close the nfb loop. Nfb only manipulates the input signal to make the amp overall distort less (and less gain) but each stage is NOT changed at all.
jan
The distorsion will be 0.1% even if the VAS gain is increased to 100db.
Fact is that when increasing VAS gain , the IPS will be less sollicited ,
wich will reduce its intrinsical THD , wich will lead to the wrong
assumption that this THD reduction is due to increased NFB.
More precisely , it modulate the IPS gain by instantly
adjusting its transconductance.
This is what happens with the 40db excess gain of your exemple ,
that will lead to an IPS CL gain of -20db , the VAS will have 40db
effective gain , hence the 20db CLG.
The feedback decreases the effective input voltage to the i/p stage, and the i/p stage distorts less with less input voltage.
The gain or transconductance looks like an S shape, so with lower levels you're more at the straight part in the middle.
If you want to be fancy and call that 'instantly adjusting the transconductance' that seems to me to make it needlessly more difficult to understand.
It suggests some mechanism to change the transconductance from the outside which is not the case.
There is nothing that changes in the i/p stage itself of course.
jan
The gain or transconductance looks like an S shape, so with lower levels you're more at the straight part in the middle.
If you want to be fancy and call that 'instantly adjusting the transconductance' that seems to me to make it needlessly more difficult to understand.
It suggests some mechanism to change the transconductance from the outside which is not the case.
There is nothing that changes in the i/p stage itself of course.
jan
Last edited:
Yes, agreed, but if you read from the start, the issue was different.
In fact there were several, one being that nfb cannot lower distortion of a stage before the gain stage, which is not true.
But I think I made my point, so I have no desire for yet another discussion.
BTW Isn't it time for your daily beer
jan
In fact there were several, one being that nfb cannot lower distortion of a stage before the gain stage, which is not true.
But I think I made my point, so I have no desire for yet another discussion.
BTW Isn't it time for your daily beer
jan
Input stage in SSA is in a deep class A, VAS input current is hundreds times lower than IPS quiescent current in that case both periods of a NFB signal completely affect/interact with the input signal in a linear manner. CFB has major advantage compared to other types of diff inputs and that is direct subtraction at the same junction as input. Input node is high impedance and CFB from the output low impedance, so naturally correct impedance matching as it should be. I don't see the need the output signal should be send to diff input via another high impedance device as with LTP-s.
All in all that natural impedance match and subtraction in a simplest possible way can be clearly heard as superior sound to more standard LTP IPS.
All in all that natural impedance match and subtraction in a simplest possible way can be clearly heard as superior sound to more standard LTP IPS.
Last edited:
Of course everyone will agree with Jan, even probably with everyone else. That's if you tried to read between the lines and never tried to find flaws in semantics. Once you tried to find flaws in somebody's statement, he might in turn try to find flaws in your statement.
I don't know the point of your question, but of course there is something matters here. THD is measured after the final stage, right? But it doesn't tell you the open loop condition. 0.01% open loop can sometimes be better than 0.001% closed loop. I mean, the condition before the NFB is applied is very important, and it isn't showed in the standard THD figure. I already posted my opinion about JFET-input double LTP topology, why I don't like it.
This is poppycock, why would every one agree with Jan? Why would every one agree with everyone else? And why would that be probable? Why would you search for faults (alone) and not for truth... I will not discuss this any further. (these are questions for myth-busters).
In the end the only thing of interest is what comes out (in relation to what was put in) the observer (listener/judge) is not interested in the number and type of stages in-between. There are people that claim that zero (stages in-between) is the best (and they may have a point) but I would say that any number of undetectable stages is fine also. It has nothing to do with type, it has to do with detectability and desirability (or total THD). P.s. the questions where rhetorical.
beer
After dropping my post I went right to Hoppe.
[...]
BTW Isn't it time for your daily beer
jan
After dropping my post I went right to Hoppe.
Hi , Edmond
I was talking effectively of the impossibility for a GNFB loop
comprising several stages to reduce distorsion of a given stage
using available loop gain provided by the following stages.
To simplify , the IPS gain used for GNFB can reduce distorsion of the VAS
as well as the one produced by the OS , while eventual loop gain quantity
that originate from the VAS can reduce distorsion for the VAS and the OS.
If the OS has also some gain that is used for GNFB , this loop gain
quantity could only reduce the distorsion of the OS.
That was the point , indeed , and we do not agree about it.
But I think I made my point, so I have no desire for yet another discussion.
BTW Isn't it time for your daily beer
jan
Well , i did settle for coffee , do not see any will of contradiction in it...
I know it was rhetorical. And now I know your point, exactly as I guessed.
This is nothing new. There has been 2 camps since long time ago. I belong to one camp, and you the other one
. Trying to find out which is wrong is useless. Trying to find out the truth is useful (such intention can be seen from open mindset, questioning own self).The problem is that THD has nothing to do (not 100%) with what is perceived or heard. Yes, we care only with what comes out in relation with what was put in. But remember, in music we never put in pure sine waves.
Yes, the observer is only interested in the quality of the sound. But observers have different experience in observing the sound. Some people find certain phenomena and have hard times to find the answer/reason for the existence of the phenomena, while some others do not find the phenomena exist.
So the blame is on those who see the phenomena because they cannot find the proper explanation. Sometimes, complex minds just make mistakes, and they can look like idiot minds, just because they stretch their mind.
Member
Joined 2009
Paid Member
I'm afraid that according to my understanding (I don't claim to have a perfect understanding of anything in this world) most, if not all of this, is completely wrong and will be confusing.
The thing about nfb is that it acts to reduce distortion from the composite of everything inside it's loop. The feedback loop is 'blind' to what is inside the loop, it 'cares' only about the net result of everything inside the loop that affects the signal at the output (i.e. wherever the nfb take off point is). There is no merit in discussion about the levels of distortion of some pieces of what is inside the loop when the loop is closed. It only makes sense to talk about the distortion of what each piece inside the loop contributes to the result when the feedback loop is open.
When the nfb loop is closed, the distortion produced by each piece inside the loop doesn't change since nothing physically has changed to these pieces (no doubt there are exemptions, but perhaps only under circumstances such as clipping). If the middle piece of the circuit adds 1% distortion to the signal passing through it with nfb loop open, it will still add 1% distortion to the signal passing through it with the nfb closed.
As has already been said, the nfb changes the input signal to the circuitry inside the nfb loop. Once that signal progresses through the circuitry inside the loop the nfb, closed or not, has no influence.
A rather unscientific way of thinking about it, although not entirely accurate, is to say that the gnu loop 'looks' at the distortion at the output, then applies a reverse copy of this distortion back to the input so as to cancel out the distortion generated by the circuitry inside the nfb loop. It doesn't 'see' or know what is inside the loop, it only 'sees' what appears at the output. Because it has to 'look' at the distortion at the output there must be some distortion at the output for it to work - so paradoxically it can never remove all the distortion (feedforward can pull that trick off, but it's harder to do).
Last edited:
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.