Better Audio with or without NFB (Negative Feedback) - for me the wrong Question

Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.
Since the last decades and even today this question regarded audio amplifier circuits is discussed controversial (unfortunately also discussed sometimes very emotionally but not technically correct).
After read a wide range of publications, documents and comments, also by this forum, I note, this is for me actually the wrong question.
The right question for me should read as follows:

Which character is to observe at the higher frequencies far outside the audible range?

In general there are only two different misbehaviours resp. two different characters

1) Low pass character, i. e. only level reduction but no visual deformation of the sine wave signal occurs even arround 1 MHz and more.

This character is to observe, both if I use no NFB as well as by use of NFB with only one voltage gain stage (VAS) in the feedback loop. If I have this kind of behaviour, I am sure, the amplifier sounds good in this case, even by relatively high levels of THD

2) No Low pass character, i. e. visual deformation of the sine wave signal to a triangle or saw tooth signal form arround frequencies between 50 KHz and 500 KHz (consequently high order distortion), but smaller level reduction character

This character is to observe, if I use NFB by more than one voltage gain stage (VAS) in the feedback loop (there are mostly two such stages with Cdom by the second stage). If I have this kind of behaviour, I am sure, the amplifier sounds harsh, even by extreme low levels of THD in the audible range. Only the low frequency range (below approximately 500 Hz) is more tight and clear than by the first character, because the open loop gain in this range is much more higher; this means, that this kind of topology is first choice for bass transducers with diaphragms, that have heavy weight (e. g. driver for subwoofer)

Both observed characters are completely independent of whether I use the output stage in "pure Class-A" mode, or I use the typical value of the quiescent current around 30 - 50 mA, i. e. "Class AB" mode.

So my conclusion is, if I want to use an amplifier for midrange/tweeter and full range, the first character is the best choice, and the second character (i. e. high open loop gain in the audible range) is only for low frequency applications the best choice.

In the above described, I have assumed, that power compression of the power supply is negligible, GND management would be performing in the correct manner, and the clipping behaviour is without overshoots and peaks through saturation

What do you think – are my observations correct, or I still overlook other important things?
 
Last edited:
Hi,

Sometimes I find on this forum is that you have to ask a question in a very simple way because we come from different places with different languages.

If I can understand your point...

a) the more gain stages within the feedback loop the worse the sound. Nelson would likely agree with you on this

b) the downside of multiple gain stages within a feedback loop gets worse at higher frequencies. Plus it's easier to make this problem visible in simulations if you use a very high frequency test signal.

Did I get this right ?
 
Hi,
b) the downside of multiple gain stages within a feedback loop gets worse at higher frequencies. Plus it's easier to make this problem visible in simulations if you use a very high frequency test signal.
please read this:
http://www.diyaudio.com/forums/pass...zero-o-os-0s-versus-later-aleph-versions.html
and download pdf file by #9

please not additional: for getting one voltage gain stage by mostly two voltage gain stage topology you can introduce source resistors (emitter resistors) by the first stage (long tailed pair), so that the open loop gain of this stage is only "unity gain".

this thread could be also of interest:
http://www.diyaudio.com/forums/pass...m-results-pass-x-series-us-pat-5376899-a.html
 
Last edited:
Negative feedback that is in phase with the input is good.
The problems start when the feedback phase shifts and causes oscilation.

An amplifier with negative feedback is essentially a servo system and all the rules of servos apply to the amplifier.

Critical damping is vital to get a good output.

Getting the gain right is important too.

Are you sure? :rofl:
 
Negative feedback, that is in phase with the input ???
I think, you mean negative feedback, that is in anti phase (-180 degrees) with the input. Otherwise I would have positive feedback.

He's right. Every amp with gNFB has two inputs. Your siganl goes into the noninverting input, and the feedback signal goes into the inverting input. You will have negative feedback then.

Most all SS amps these days have a differential amp as the first stage. Follow the signal: it always goes into the noninverting input.

With hollow state amps, it's a bit different: the NFB signal is almost always applied to a cathode, while the signal goes into the grid. It's still invert/noninvert operation.
 
Originally Posted by tiefbassuebertr - Negative feedback, that is in phase with the input ???
I think, you mean negative feedback, that is in anti phase (-180 degrees) with the input. Otherwise I would have positive feedback.
Depending on frequency.....
In phase at the LTP base , but exactly "antiphase" (-180) at the VAS "drive" (LTP collector) at low frequencies. At 25k it starts going out of phase (attachment 1).
If the amp goes too much further than 110 degrees before unity gain (600k - 1mhz) you MIGHT have an oscillator. If you can manage 180 before UG , you will have perfect positive feedback and have a dandy oscillator. :redhot::redhot:
OS
 

Attachments

  • plot.gif
    plot.gif
    14.2 KB · Views: 607
Last edited:
If the amp goes too much further than 110 degrees before unity gain (600k - 1mhz) you MIGHT have an oscillator. If you can manage 180 before UG , you will have perfect positive feedback and have a dandy oscillator. :redhot::redhot:
OS

Mostly incorrect. The stability criteria requires a loop gain phase shift of less than 180 exactly at the unity loop gain (ULG) frequency. If the phase goes over 180 before or after the ULG frequency, the amp is called "conditionally stable", the condition being the closed loop gain (CLG). For certain CLGs the amp is stable, for others it is not. On top of my head, a conditionally stable amp is my VSOP amp. While rock solid at the designed CLG, by measurements under the worst conditions (load, clipping, HF, etc...) and listening tests in the worse possible speakers (electrostatics), I have decided that conditional stability is not an option for audio. That's why that project was never finalized, other than as "lessons learned".
 
By syn08 - Mostly incorrect.
Well , that is good .. something to learn.

How does this relate to "phase margin" Andy c. told me 70-80 degrees would be sufficient and this is the plot I use as a baseline (attached - my next build). All those that I have actually built did not oscillate , so it must be close. Also , most of the class B amps at DIYA have a similar OLG plot when simulated and also do not oscillate when actually built.

Since you described "conditionally stable" , would there also be unconditionally stable and how would a design/plot (please show a plot) like this be characterized ?

OS
 

Attachments

  • SST3T.gif
    SST3T.gif
    22.5 KB · Views: 562
If the phase goes over 180 before or after the ULG frequency, the amp is called "conditionally stable", the condition being the closed loop gain (CLG).

I searched around and found that this is indeed the commonly-accepted definition of conditional stability. But it seems to me to not be a very useful definition. After all, what amplifier's loop gain never reaches a phase of -180 degrees at some sufficiently high frequency? Only contrived textbook cases that I know of. I suppose you could set an upper bound on how much the loop gain (and therefore the ULG frequency) would be allowed to increase by requiring stability at unity CLG (only attenuation or unity gain allowed in the feedback loop). That seems extreme though, as you'd need to set the ULG frequency of a power amp about 30x lower than if you specified some nominal phase margin (say 80 degrees) at the actual CLG (assumed to be 30x above).

OTOH, the idea that some conditionally-stable amps that are in a stable state can be made unstable by reducing the loop gain by a factor that's constant with frequency is disturbing to some. This could happen with two-pole compensation (TPC) (I know you know this, it's just for other readers). The reduction in loop gain could be due, say, to clipping or gain compression. I can't recall the compensation type that VSOP used. Was it TPC or some variant thereof? If so, was that the reason you soured on this approach?
 
Last edited:
Yes, for two-pole comp, the phase shift of the loop gain usually gets close to -180 deg at a frequency much less than the ULG frequency. Therefore scaling the loop gain by a constant less than 1 can make a stable TPC amp unstable. I guess what I was wondering is if that is what was bothering syn08 about the VSOP performance, or if it was something else, possibly related to potential increases of loop gain (which I can't picture well, except for small changes due to component tolerances).
 

GK

Disabled Account
Joined 2006
Hmmmmm...... if it is possible to make a TPC amp oscillate when clipping, I have not been able to do it with any of those that I've build. As I see it, when a stage in driven into clipping, it basically stops passing any AC signal at all, which doesn't seem to make for an effective oscillator.
 
I searched around and found that this is indeed the commonly-accepted definition of conditional stability. But it seems to me to not be a very useful definition. After all, what amplifier's loop gain never reaches a phase of -180 degrees at some sufficiently high frequency? Only contrived textbook cases that I know of. I suppose you could set an upper bound on how much the loop gain (and therefore the ULG frequency) would be allowed to increase by requiring stability at unity CLG (only attenuation or unity gain allowed in the feedback loop). That seems extreme though, as you'd need to set the ULG frequency of a power amp about 30x lower than if you specified some nominal phase margin (say 80 degrees) at the actual CLG (assumed to be 30x above).

OTOH, the idea that some conditionally-stable amps that are in a stable state can be made unstable by reducing the loop gain by a factor that's constant with frequency is disturbing to some. This could happen with two-pole compensation (TPC) (I know you know this, it's just for other readers). The reduction in loop gain could be due, say, to clipping or gain compression. I can't recall the compensation type that VSOP used. Was it TPC or some variant thereof? If so, was that the reason you soured on this approach?


An N stage amplifier has an asymptotic gain limit slope of N*20 dB/decade. This was important back in JC's glory times when, due to the lack of high Ft devices, designers were trying to squeeze the last drop of gain-bandwidth, hence the concept of "Maximum obtainable feedback". Obviously, having few stages with high gain was good (because less zeroes were required, to bring the phase back, and the gain to 20dB/decade, at ULG), therefore the whole old story about "counting the stages" and the corollary "simple is good". But this is not the case today, we don't need to go to those pathological limits.

VSOP was indeed TPC, but that was not the problem per se. The problem was that another zero was required to compensate for the input stage opamp pole (which was also dropping the phase well below 180 at the target ULG, when the output stage loop was unconditionally stable). One option I have found (and that was a bad decision) was to add a small cap across the global feedback resistor. While this indeed brought back the phase and made the amp rock solid with about 85 degs of phase margin (while still dipping under 180 before the ULG frequency), this had a negative impact on a) how the amp was able to handle difficult loads and b) how the amp was handling (exactly as you said) clipping and compression. I was not happy with both the measurements and listening tests, so I shelved the project and went to improve YAP to v2.1

One of the practical lesson learned is that TPC is (very) good for compensating single gain stage amps (typical VAS configuration). When the gain is spread across multiple stages, and in particular when multiple loops are involved, using TPC is not at all straightforward and there's a number of tradeoffs involved, well beyond the 2nd order system overshoot.

I am having a lot of fun identifying reasons for the myths and the legends in audio. Cable directionality myth was probably born when manufacturers decided to mark on the cables the direction where the ground was connected (in balanced cables, think on how a phono balanced cable is connected, you can't connect it backwards without running in ground loop troubles!). Now, counting the gain stages and promoting "simple is better" could be due to the past (50's-60') problems with achieving the maximum obtainable feedback and the difficulties to compensate for that. I am convinced there's a similar explanation for each of the audio myths and the legends that are still perpetrated today.

To address Glen's point, the clipping impact on the OLG is very difficult to estimate. I think both extremes (considering the amp at clipping as still a linear network, having the poles and zeroes altered by this border condition, and the non-linear model where the signal is effectively cut off) are failing short in providing a general behavioural model. All I can tell is that I have noticed experimentally (all other conditions being fairly identical) a large difference in clipping behaviour, between a simple Miller compensation and a TPC compensation. BTW, while TMC seems like it should be fairly similar to the Miller case, in fact it is not - and that's all that I can tell so far. Obviously, the best solution is to always use some sort of clipping protection circuit.

My conclusion is that indeed, clipping behaviour is a good argument for using as much as possible open loop linear circuits and low open loop gains. This is how you can get nice rounded clipping responses, without any special protection (look at Upupa's amps). But then, trading open loop linearity for open loop gain (perhaps by using local feedback, or something like multi-tanh) is just another trade-off... Probably a hypothetical gain x linearity product is as much a constant as gain x bandwidth in single pole systems.

As a side note, it would be very interesting to study the multi-tanh effect on the closed loop linearity. Obviously, for a constant tail current, multi-tanh linearizes the response, but this comes at the price of gain. What is the net impact on the open loop gain? Is it worth doing multi-tanh if you consider the OLG loss, available for NFB linearizing?

To may questions, not enough time...
 
Last edited:
Hmmmmm...... if it is possible to make a TPC amp oscillate when clipping, I have not been able to do it with any of those that I've build. As I see it, when a stage in driven into clipping, it basically stops passing any AC signal at all, which doesn't seem to make for an effective oscillator.

Yes, it is possible. I have somewhere on the hard drive a scope screen shot of a TPC amp bursting in oscillations at clipping, I'll search for and eventually post it.
 
Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.