Hi wg_ski,
EXACTLY !!!!
Hi Globulator,
Rise time has a fixed relationship to frequency. Look it up. This will help you understand where people go wrong in their thinking.
EXACTLY !!!!
Hi Globulator,
Rise time has a fixed relationship to frequency. Look it up. This will help you understand where people go wrong in their thinking.
Rise time (in the time domain) is related to phase shift (in the frequency domain).
An exponential rise time is caused by the nonlinear (arctangent) phase shift among the harmonics for a first order low pass filter
Linear phase shift (which is a true time delay) preserves both the waveform shape and the time relationships among the harmonics.
Linear (or zero) phase shift will have faster rise time than arctangent phase shift, but will cause symmetrical Gibbs effect apparent "ringing".
This is not ringing per se, but rather is the result of the limited number of harmonics being added together without relative phase shifts.
Amplifier “ringing” isn’t a Gibbs effect, it is the natural response (the solution to the differential equation with the RHS set to zero) asserting itself. Producing a frequency that is physically present in the step response excitation (which mathematically extends to infinity) at a relatively high value compared to frequency surrounding it. This obviously requires frequency content going into the amplifier way above 20 kHz. A digital source will not (but may have Gibbs effect of its own) and the rise time limitations of analog (audio, not lab instrument) sources will keep anything produced well below the noise floor. Unless something is broken. An amplifier being driven into slew rate limitation with an audio band signal is broken. One could say that an amplifier that is clipping is broken too, and for the purposes of undistorted reproduction at that level, it is.
The 'loop flight time' (LFT) in a feedback amplifier is about 5-10 nanoseconds - that's how long it takes the signal to go from the input, through the amplifier and all the way around to the feedback input. It is as good as instant. But LFT and phase shift are not the same thing. All circuits with reactive components (capacitance and/or inductance) have phase shift and tube amps (and some class D amps) exhibit much greater phase shifts than linear amplifiers.
If you feed a sinewave of say 50 kHz into a feedback amplifier and then look at the output you will see it is shifted wrt the input, and so therefore is the feedback signal. You may be tempted to say 'this is a delay' but it is not. The instant the sinewave at the input starts to go positive, the feedback signal will also be going positive but phase shifted so it will be at a very small value initially- the only delay being the LFT of 5-10 nanoseconds which will have zero effect on the amplifier behaviour or performance. If this was not the case, the amplifier would be running open loop and would just not work (see next paragraph for why). Phase shift wrt feedback and is a very well understood mechanism within control theory and it is dealt with through stability analysis and compensating the amplifier or system to ensure the stability criteria for a closed loop system are met - another part of control theory that is mature and well understood. If this stuff did not work, rockets would not fly straight and aeroplanes would be falling out of the sky every day. Luckily, for the most part things work really well and only go wrong when engineers don't follow the rules.
You can get a real 'big' delay in amplifier feedback of 1-3 us if the amplifier clips, or if it goes into slew rate limiting. In these conditions, the amplifier runs open loop and there is no feedback during the event. When the feedback signal is restored, the amplifier should recover gracefully without any ringing or overshoot. Slewing distortion and TIM are very well understood issues and the industry has known how to deal with this competently for at least 30 years.
Unfortunately, a lot of nonsense has been sprouted about this and the subject has been thrashed to death here on diyAudio over the years. If people tell you feedback goes 'round and round' or that there's a 'delay between input and output in a feedback amplifier' that causes problems, don't believe a word of it.
(To your oscillator point: you create an oscillator by ensuring the loop phase shift is >180 degrees and the gain > 0dB. This has nothing to do with the LFT of the oscillator feedback path)
As an aside, Mr. Bean aka Rowan Atkinson has an MSc in control theory from Oxford. Thats right. Get an MSc in control theory and then go out and become a funny man.
Last edited:
Getting a degree in control theory will turn you into a funny man, and be ready for the funny farm. It was one class in my undergrad years that I got a “B” in, but was one of the most USEFUL. Nothing worthwhile is easy.
Re: "marginal" stability,
At the risk of stating the obvious, the beta of bipolar transistors is prone to modulation, and usually multiple types at the same time. And that's just one transistor. Vce, collector current, and almost certainly drift with temperature as well. The effects from multiple transistors could multiply. So those theoretical open-loop gain vs frequency plots are just an optimistic snapshot, while real gain could bounce up and down.
Meaning that, as the gain rolls off at high frequencies, a 2nd pole increasing the slope could push the phase shift to 180. Conceivably, that critical area could bounce around a gain of 1. So it could briefly ring and then recover, in a signal-dependent manner. Compensation techniques generally try to either add a 3rd pole at a much lower frequency, so the region with phase reversal is pushed below 0dB with a suitable margin, or a zero is added (RC shelf filter), achieving similar ends with different means.
IMO, when referring to the frequency domain, there's nothing wrong with talking about a feedback process evolving "round and round". For one thing, the frequency domain by definition already talks about 'frequencies', which are self-similar throughout time. So, processing frequencies in a loop does not imply that there's a repetitive loop or sequence of events in the time domain.
At the risk of stating the obvious, the beta of bipolar transistors is prone to modulation, and usually multiple types at the same time. And that's just one transistor. Vce, collector current, and almost certainly drift with temperature as well. The effects from multiple transistors could multiply. So those theoretical open-loop gain vs frequency plots are just an optimistic snapshot, while real gain could bounce up and down.
Meaning that, as the gain rolls off at high frequencies, a 2nd pole increasing the slope could push the phase shift to 180. Conceivably, that critical area could bounce around a gain of 1. So it could briefly ring and then recover, in a signal-dependent manner. Compensation techniques generally try to either add a 3rd pole at a much lower frequency, so the region with phase reversal is pushed below 0dB with a suitable margin, or a zero is added (RC shelf filter), achieving similar ends with different means.
This would be a fairer comment, if people would at least occasionally explain how NFB is able to generate higher-order harmonics from a simpler non-linearity without some iterative process, while sticking to their preferred time domain mode of thinking.
IMO, when referring to the frequency domain, there's nothing wrong with talking about a feedback process evolving "round and round". For one thing, the frequency domain by definition already talks about 'frequencies', which are self-similar throughout time. So, processing frequencies in a loop does not imply that there's a repetitive loop or sequence of events in the time domain.
Hi abstract,
No.
You are exaggerating minor effects far past the point they could possibly affect an audio signal. Some folks without a reasonable understanding of how circuits really work will take a reported factor and blow it way out of proportion. Take effects that occur at RF frequencies, then they apply them way down to very low frequencies. You can't do that.
Here are a few ways to become permanently confused.
No.
You are exaggerating minor effects far past the point they could possibly affect an audio signal. Some folks without a reasonable understanding of how circuits really work will take a reported factor and blow it way out of proportion. Take effects that occur at RF frequencies, then they apply them way down to very low frequencies. You can't do that.
Here are a few ways to become permanently confused.
- Read advertising copy from manufacturers, "white papers" being the most dangerous.
- Read subjective listening tests not well controlled where someone attributes what they heard to technical ideas that are not applied correctly
- Read posts on the internet from untrained armchair critics.
Which effects are you referring to? Where did I take some RF effect and apply it at very low frequencies?
The whole idea of compensation is to ensure that changes in loop gain/phase are well managed and meet the stability criteria over a whole range of variables covering load, signal levels and everything else. That’s precisely why you simulate and then test.
I can sure you if a jet liner can be designed, built and tested to ensure it is stable in flight, an audio amp is a trivial application of control theory. We should not kid ourselves otherwise.
I can sure you if a jet liner can be designed, built and tested to ensure it is stable in flight, an audio amp is a trivial application of control theory. We should not kid ourselves otherwise.
I consider a jet liner good if, in the event of power loss, it can glide for 100s of km. No control theory required to keep it stable.
Hi abstract,
I don't think you have any real experience. You're pushing maybe, could be. In the real world, no decent design made today suffers from anything close to the situations you are suggesting. So in a practical sense you're barking up the wrong tree.
Stop taking small possibilities and trying to apply them to practical amplifiers. If you had some small variations in beta and open loop frequency, feedback established limits far away from those points. Go crunch some numbers and come back with your findings.
In the real, practical world, we measure amplifiers under load and anything you are suggesting sticks out like a sore thumb in some of our common tests. So maybe the easiest thing for you to do is get access to some really good testing equipment, and with a skilled operator run some tests on various amplifiers. At some point you'll get a feel for what you see in the measurements and how an amplifier sounds. Shocking .. but true.
only if those plots come from a simulator. We test circuits in real life under load and compensate them to be stable, which means those effects are not affecting the circuit.
That would be marginally stable. You would see other issues and any designer avoids running without enough phase margin.
I don't think you have any real experience. You're pushing maybe, could be. In the real world, no decent design made today suffers from anything close to the situations you are suggesting. So in a practical sense you're barking up the wrong tree.
Stop taking small possibilities and trying to apply them to practical amplifiers. If you had some small variations in beta and open loop frequency, feedback established limits far away from those points. Go crunch some numbers and come back with your findings.
In the real, practical world, we measure amplifiers under load and anything you are suggesting sticks out like a sore thumb in some of our common tests. So maybe the easiest thing for you to do is get access to some really good testing equipment, and with a skilled operator run some tests on various amplifiers. At some point you'll get a feel for what you see in the measurements and how an amplifier sounds. Shocking .. but true.
The pilot would be doing the control bit in that situation since it still requires some kind of control action to keep it stable. What would happen if the pilot did nothing? That said, a computer controller in general provides better response to unforeseen perturbations and there are plenty of examples where human control did not work very well (watch 'Aircraft Accident Investigation' for some of this stuff)
Control theory as applied to amplifiers is not an immature area of engineering and is not fraught with all sorts of problems engineers or designers don't know about or don't know how to deal with. Here's some info on the subject - one dating back to the early 1970s. Analog computers - which is where opamps originated - were used in fire control and guidance control systems starting in the 1940's already, so feedback and stability analysis is a well understood branch of engineering.
https://hifisonix.com/wp-content/uploads/2010/10/James-E-Solomon-Opamp-Tutorial.pdf
https://linearaudio.net/sites/linearaudio.net/files/volume1bp.pdf
Here's why people think feedback is a problem. In particular, a guy by the name of Martin Colloms pushed all sorts of nonsense about it - and shame on Stereophile for allowing this stuff to be published:-
https://www.stereophile.com/reference/70/index.html
Here's a short history of why in audio there were struggles in the early days to get to grips with feedback in solid state amplifiers.
https://hifisonix.com/technical/the-case-for-feedback/
Given the initial velocity and mass, wind shear (read:applied energy), and mgh (read: more applied energy), using control theory, one can reliably predict the cork screw trajectory straight into (but somewhat angled) the ground.
The (broken) Pass link in the hifisonix appears to be the same one I linked to, earlier.
I find that to be disingenuous. The output stages discussed had good linearity in the open loop. Then, stacking 2-3 similar stages in series was obviously illustrative. An inquiring mind could have picked up on some interesting points like the output of higher-order distortions in the experimental setup, and maybe even asked: how do they sound?
Identifying the sound quality would be an early step to determining how low the level needs to be for it to become 'blameless'. I suppose Pass' crime was noticing the diminishing returns, leading to people like me wondering: just how awful is that residual distortion, for it to require a mind-boggling 1-thousandth of 1% equivalent THD level (referring to the amount of feedback) to render it inaudible?
Hi abstract,
The problem is that if anyone writes an article that doesn't know what they are talking about, other's read it and accept it at face value. In order to satisfy your questions, I am afraid you will need to learn the material and earn practical experience. Many of us have done that.
We can't afford to answer each and every question time and time again for various people. At some point you have to become responsible for your own knowledge and follow your interest to actually learn. Reading what others wrote is the cheap way around it. Using their unproved ideas to argue a point with people in the industry doesn't seem to make much sense. To repeat opinions and thought experiments as fact to disprove others doesn't benefit anyone.
The problem is that if anyone writes an article that doesn't know what they are talking about, other's read it and accept it at face value. In order to satisfy your questions, I am afraid you will need to learn the material and earn practical experience. Many of us have done that.
We can't afford to answer each and every question time and time again for various people. At some point you have to become responsible for your own knowledge and follow your interest to actually learn. Reading what others wrote is the cheap way around it. Using their unproved ideas to argue a point with people in the industry doesn't seem to make much sense. To repeat opinions and thought experiments as fact to disprove others doesn't benefit anyone.
You cannot hear 1/1000th of 1%. That’s the disingenuous bit of this whole discussion as is the notion that feedback is bad, or goes round and round. Enquiring minds have looked into this and found absolutely nothing, other than humans find high levels of higher order harmonics less euphonic than lower order harmonics at the same level.
It’s clear you won’t be convinced by anything I or anyone else on this thread can say.
Blocked.
It’s clear you won’t be convinced by anything I or anyone else on this thread can say.
Blocked.