Feedback Question/Clarification

[Bonsai]

It is truly regrettable that this thread has morphed into a serious and considered discussion about feedback. I was looking forward to slinging mud and making facile arguments.

 

[lumanauw]

This slide #43 is very interesting.
-Can a process be improved by adding parts? Yes it can!
-A bunch of parts enclosed in one feedback loop = ONE process.

 

[jan.didden]

Refer to Table 1, the third and fourth data set. He breaks out the effects of NFB separately from the combined feedback, which is what caught my attention at your talk. As I had guessed at the time, the feedback is relatively low (11dB without the regeneration). Yet every harmonic is decreased.

My point was not an apples to apples comparison- it's an observation that what you have in your hand is never actually an apple, it's a tangerine. The conclusions drawn from a simple square-law model are probably not generally applicable to real, multistage amplifiers, whether transistor or tube. Low feedback factors may indeed be beneficial.

Stuart, as always, you are correct.

Jan Didden

 

[SY]

Jan, to be fair, pace Scroggie, although all harmonics in Miller's work are diminished, the lower order ones are diminished more.

Do you have the capability of repeating Baxandall's simulations but using a 3/2 law? That might be interesting...

 

[traderbam]

lumanauw wrote: This slide #43 is very interesting.

Please elaborate.

 

[jan.didden]

Jan, to be fair, pace Scroggie, although all harmonics in Miller's work are diminished, the lower order ones are diminished more.

Do you have the capability of repeating Baxandall's simulations but using a 3/2 law? That might be interesting...

It has been done , sort of.

BTW I noticed that your posts tend to become shorter than your sig line. Should I worry?

Jan Didden

 

[SY]

BTW I noticed that your posts tend to become shorter than your sig line. Should I worry?

Yes.

 

[jan.didden]

... just a snapshot, to wet your appetite.

(The FET case is appended to my previous post)

Jan Didden

 

[KSTR]

Jan, for me your PDFs won't open correctly....

... and Bruno's slides don't work also (to lazy to invoke those 100++ slides manually)

Klaus

 

[SY]

Work fine for me.

 

[jan.didden]

Jan, for me your PDFs won't open correctly....

... and Bruno's slides don't work also (to lazy to invoke those 100++ slides manually)

Klaus

Maybe you need to upgrade Acrobat Reader? SY, what version do you use? I have 8.8.1

Jan Didden

 

[lumanauw]

Please elaborate.

I'm not an expert in math or feedback theory, but what Bruno wrote agrees with Jan's "Black Box" theory. No matter how many parts with how many nested loops inside, if the whole thing is closed within 1 global feedback loop it means there's only 1 process of in and out that counts.
The implication of this, in the global feedback closed amps, the "number of parts that the signal goes through" doesn't make much sense if the loop is closed (as long as the group delay is not unreasonable, offcourse ). The next implication is that the more part cct can perform better (not worse) if the loop is closed.
The other trend is "all else being equal, the simpler, the better the sound"

 

[jan.didden]

Hi Jan,

Yes, the weighting is corresponding to the OLG of each stage.

But, then we have some (semantical) contradiction, somehow:


It seems one needs to qualify that "smaller". Smaller only in absolute amounts (say, injected PSU ripple etc). From the distortion perspective, the errors should be referred to the actual signal levels at each stage and then we have a "same in, same out" black-box situation which is "blind" to what actually happens inside...

- Klaus

Indeed. But there is still the matter of the distortion of a stage being distorted in following stages again. That has not been considered, has it?

Jan Didden

 

[jan.didden]

Jan, for me your PDFs won't open correctly....

... and Bruno's slides don't work also (to lazy to invoke those 100++ slides manually)

Klaus

Where are these slides from Bruno?

Jan Didden

 

[KSTR]

Here:
http://www.hypex.nl/docs/Bruno Masterclass/slides.htm

Problem solved, er, circumvented: wrote a C-quickie which generates a html which invokes all the slides

Klaus

 

[andy_c]

I'm sure that people like you and Brian can wield some heavy math tools around to show that his reasoning would fall apart under this and that condition.

Well, if it were simple enough for me to solve, I'll bet somebody smarter than me would have solved it long ago .

It's just that he and some others have a way to cut through the chase by knowing what to take in, what to ignore. I personally like that, it's a very efficient way to get close to your goal, and if you really need to be much closer, you can breake out the heavvy math.

Indeed! I hope I didn't come across as dissing Baxandall here. He was one of the greats, for many reasons in addition to what you've already said.

Also, Gilbert used the same technique in his article that we spent a lot of time discussing, related to the PIM issue. I am still convinced of the validity of his approach. 'Nuff said.

Thinking about Gilbert's paper, along with Hugh's concern about the integrity of the differencing made me remember a SPICE experiment I was messing with a while back. The idea was to do in SPICE something like what Gilbert did - assume a real-world input stage with the rest of the circuitry ideal. I wanted to see what contribution to the overall amp distortion was being made by just the input stage by itself.

There's a picture of part of the circuit below. I computed the small-signal input impedance of the VAS, along with its transconductance, and threw in some biasing and frequency compensation, along with an ideal VCVS for the output stage.

Simulated distortion was disappointing. I was wondering what was going on. Then I tried bootstrapping the input stage cascode. The distortion dropped by a lot. In the sim that I'll attach to the next post, there is a 16x improvement in distortion from the bootstrapped cascode. So what might appear to be a second-order effect - the common-mode input voltage - actually dominates.

I went back to Walt's AES preprint, where he was correlating, among other things, THD to SID. He had to use inverting mode in the op-amps to establish the correlation, because of the common-mode effects.

 

[andy_c]

Here's the sim of the idealized amp with real input stage. The THD is computed at 20 kHz, assuming it's a power amp putting out 80V peak. There are two .asc files, one with the base of the input stage cascode at a fixed voltage, the other bootstrapped. This shows the 16x improvement in distortion.

Edit:

With bootstrapping:
Total Harmonic Distortion: 0.000048%

Without:
Total Harmonic Distortion: 0.000787%

 

[GK]

And here is a non-idealised, complete design, in the form of a unity gain buffer.

This is going to be the input buffer for my differential phase spitter. I opted against a hybrid circuit for this section in the end, having treated myself to a early Christmas present by buying out Farnell Australia’s remaining U404 dual matched jfet stock.
The extra RC shunting the LTP output is required to get the open loop gain below 0dB.

THD-20 delivering 1Vrms into 500 ohms:
0.000014%
THD-20 delivering 7.07V rms (17dB above the power amps clipping level) into 500 ohms:
0.000143%

7.07Vrms is a reasonable common mode signal! I think that this demonstrates that, with proper design, CM and LTP distortion is pretty much a non-issue.

 

[GK]

Here is the simulated loop gain, phase margin and the 100kHz squarewave response.

 

[AndrewT]

Hi GK,
why did you choose not to insert an EF buffer between the q10 cascode and the driver pair?

All else looks much as you would be expected to produce.
Excellent front end to a power amp?