Bob Cordell Interview: Negative Feedback

Jump to Latest
#661
Hi JCX,

Your circuit in post#658 is a good illustration of why U2 cannot linearly assist U1 when the load is a reactive LS, this because input controlled voltage correction cannot predict phase variable load current output requirements.
____________________________________________________

All of the other simulation examinations being carried out here also *need* to be done with a virtual reactive load, because when the output stage has its own loop - the global response becomes modified *by the load*, especially where that local loop is not linear.


Cheers ......... Graham.
 
#663
andy_c said:


Well, I picked 2 MHz because I thought that's what your amp does 🙂.

Regarding gain variations, I think the big hit is with input stage gm. I don't think I'd use undegenerated FETs here. Just using a 2N5457 as an example, its gm varies by a factor of 5 in the data sheet (1000-5000 umhos). I don't think I'd want that even with single-pole comp: five-to-one GBW variation - yuck! With regard to VAS output impedance, this variation causes weird things to happen at low frequencies (peaking in open-loop gain, etc), but this effect goes away at high frequencies. So I think if gm is well controlled, things should be okay.
Click to expand...


Hi Andy. Yes, my amp does have the 2 MHz gain crossover, but I didn't want to use as an example one that was quite so agressive.

I agree completely that we don't want big production variations in GBW. I use modest degeneration with my input FET pairs and don't have much of a problem in that regard. Five-to-one gm variation at a given drain current is pretty bad for a FET.

My reluctance to have the phase margin dip below 35-40 degrees anywhere in frequency is not solely based on stability against production GBW variations. I just philosophically don't like to come close to a conditional stability situation. I agree, its just a matter of personal preference as to how agressive we go with DPC.

Bob
 
#666
Bob Cordell said:


I agree completely that we don't want big production variations in GBW. I use modest degeneration with my input FET pairs and don't have much of a problem in that regard. Five-to-one gm variation at a given drain current is pretty bad for a FET.

Bob
Click to expand...

General point, the gm of a JFET operated at constant drain current varies very little. This is due to the fact that the quantity Vp/sqrt(Idss) is fairly process invariant. Check the 2N5457 graphs again (OnSemi) operated at 1mA all three Idss grades take ~-1V delta Vgss to go down to ~100uA or roughly the same gm. Of course you can't operate a 1mA Idss one at 5.8mA. Toshiba datasheets have a better presentation of this where they plot gm vs Id for all the grades on top of each other. They lie roughly on each other the lower Idss ones just ending sooner.
 
#668
Amusing ppm's

ingrast said:

... In fact I find rather amusing to quote simulation results in the ppm range. I am afraid too much faith is put on device models, and while it can be argued simulations nonetheless allow at least for comparative tests with the same devices, I am afraid still there are higher order non included effects whose infuence is being ignored yet may be significant at this resolution level.
Rodolfo
Click to expand...

Hi ingrast,

Amusing ppm's? Don't forget the people who are well aware of the shortcomings of spice. The Gummel-Poon model, for example, doesn't offer enough accuracy (Early effect) for sub ppm freaks. However, most of these thd killers add their own models or make their amp's insensitive to these uncovered effects by design.

Cheers,
 
#670
Andrew,

it is not that difficult to decode.
Quasi-complementary output stage, low bias current, cross-over distortion, all resulting in non-linear Zout, loss of control near to zero-crossing, the higher frequency the worse.

Sound miracle for a lot of people 😉 . A lot of low budget commercial amps would not be any better.

FYI, THD is under 0.05% (mostly crossover distortion).
 
#672
andy_c The "Mike" loop gain is calculated by having the loop gain probe right at the output of the output stage. [/B][/QUOTE] [QUOTE]Originally posted by mikeks said:

Not mine.
Correct measurement of Loop gain in multi-loop systems, of at least one common node, discussed here.
Click to expand...

In other words, there exist two definition of a global loop function, one output related and one input related, right?
 
#673
decoding that X-Y plot.

different height of the left and right loops = quasi?
high central spike = crossover distortion?
central wobbles = loss of control near zero crossing?
? = low bias current?
? = non linear Zout?
slope from left to right = ??

not that difficult:xeye:
 
#674
jcx said:


Bode's "maximum feedback" design critieria is I believe more relevant to tube amps:


Cherry makes this point in his discussion of his nested differentiating feedback loops which are "conditionally stable" by Bode's criteria

As a speculation I would guess some of the practical determinants in successful amp topologies are driven by the nonlinear /clipping recovery behavior - after all linear theory points to maximizing global loop gain as the most "efficient" approach for reducing distortion with negative feedback, so why look elsewhere?

the choices of local vs global feedback (and compensation) have to satisfy linear stability but there is also this far more complex nonlinear performance that is seldom discussed - probably the preference for miller-dominant pole compensation + unity gain output stages is driven by this

The relative dominance of audio power amps by only a few fundamental topologies probably reflects in part the dismissal of those that have great linear performance but self destruct from these nonlinear oscillations/latch up effects at the least provocation

Bob’s input clipping (and input bandwidth limiting = slew rate limiting) may be useful for expanding the range of low distortion audio power amplifier topologies that could be used - although conditionally stable designs likely will require complicated internal current/saturation/gain limiting "nonlinear compensation" circuitry for safe operation

While Cherry’s ndf amplifiers don’t seem to have carried the market, Halcro’s designs which also play in the “conditional stability” end of high feedback design do seem to have some success
Click to expand...


Thanks. These are all very good points.

Bob
 
#675
mikeks said:




Thus it becomes apparent the Edmand's modification merely provides us with a single-pole characteristic for the major loop, while delivering the equivalent double-pole compensated characteristic in respect of the second and output stages.

In other words the input stage is deprived of the enhanced loop gain provided by the equivalent DPC network at HF, while the total feedback applied about the TIS and the output stage remain completely unchanged at the frequencies of interest compared with ordinary DPC: a complete waste of time effort money etc.
Click to expand...


Mike,
Your first paragraph here seems like a reasonable description of the scheme. I agree about the input stage not enjoying the same loop gain boost as with DPC, but I would not then say that the scheme is a complete waste of time and effort. In an amplifier where the biggest distortion problem is in the output stage, and where the designer does not want a DPC global NFB charactersistic, is this scheme not useful?

Cheers,
Bob
 
#676
PMA said:
Hi Graham and Rodolfo,

FYI, I have made the same measurement on a chipamp. Just to be aware of what one can expect:
.........
Click to expand...

It turns out this is a surprisingly powerfull test !!!

Only drawbacks I see are it does not exercise the amplifier at other output voltages but near 0 V, and high frequency forward path performance is checked only for feedback designs, where it is embedded in the loop response.

Yet it is surprising how crossover defects are readily unmasked!

Rodolfo
 
#677
Re: Amusing ppm's

estuart said:


Hi ingrast,

Amusing ppm's? Don't forget the people who are well aware of the shortcomings of spice. The Gummel-Poon model, for example, doesn't offer enough accuracy (Early effect) for sub ppm freaks. However, most of these thd killers add their own models or make their amp's insensitive to these uncovered effects by design.

Cheers,
Click to expand...


Edmond,

So we agree simultation results must be interpreted in proper context.

They obviously excel for testing concepts (and save charred silicon), and for comparative evaluation of alternatives as long as one does not go all the way to expect -180 dB performance just because the simulator says so.

As Bob has hammered countles times, "the proof is in the pudding" and "the devil is in the details", i.e. a real, working measured circuit.

Rodolfo
 
#678
scott wurcer said:


General point, the gm of a JFET operated at constant drain current varies very little. This is due to the fact that the quantity Vp/sqrt(Idss) is fairly process invariant. Check the 2N5457 graphs again (OnSemi) operated at 1mA all three Idss grades take ~-1V delta Vgss to go down to ~100uA or roughly the same gm. Of course you can't operate a 1mA Idss one at 5.8mA. Toshiba datasheets have a better presentation of this where they plot gm vs Id for all the grades on top of each other. They lie roughly on each other the lower Idss ones just ending sooner.
Click to expand...


Exactly.

Bob
 
#679
Hi PMA,

Quite a spike; +/-1A should not be difficult.

I used to know what would cause all the different deviations and looping error traces, but it looks as though there is an internal failure to adequately control one half of a complementary output stage at crossover.

Yes THD figures are meaningless when load angles become dynamically shifted.

I found your simulated and real world reverse testing of the Symasym interesting. For the real world impedances did not fully invoke the crossover distortions shown in simulation.
In this regard maybe lower accuracies for step size and interpolation would give more realistic results than simulating with maximum resolution.

Hi Rodolfo,

Yes indeed a powerful tool, for if anything shows up here it can only become worse with voltage amplitude swing, as when a fixed potential is applied to the amplifier input.
Or maybe standard 10kHz on amplifier under test input with output being observed, whilst a separate amplifier drives the earthy end of its 8 ohm load at say 100Hz, then try swopping the generators over.

Cheers ........ Graham.
 
#680
ingrast said:


It turns out this is a surprisingly powerfull test !!!

Only drawbacks I see are it does not exercise the amplifier at other output voltages but near 0 V, and high frequency forward path performance is checked only for feedback designs, where it is embedded in the loop response.

Yet it is surprising how crossover defects are readily unmasked!

Rodolfo
Click to expand...


Yes, these types of tests are neat. The IIM test that Otala came up with was one of the first of this class of test I had experience with wherein the output of the amplifier is reverse-excited by another amplifier through an 8 -ohm resistor. Another test in this class of tests that I like is to apply the 19 kHz and 20 kHz twin-tone test back-fed to the amplifier and look at the spectrum at the amplifier output. The fact that the test signal is attenuated by the low output impedance of the amplifier under test helps dynamic range of the test a lot.

Cheers,
Bob
 
You must log in or register to reply here.
Top Bottom