I meant time delay, as in so many nanoseconds. Phase lag that increases linearly with frequency is time delay. Well below a pole, the phase shift vs. frequency of that pole is somewhat linear, so it looks like time delay.
Cheers,
Bob
I don't disagree with what you are saying (ok, at least high level), so let me reformulate:
Do you think there's any parameter in a closed loop system that cannot be evaluated by the loop gain analysis?
To me it's a "nay". That's all I'm saying, a loop gain analysis will always provide a correct and complete prediction of the closed loop behaviour.
In theory, if you know everything about the OL circuit and you know exactly how you will change it when you make it CL then you should be able to fully predict the CL behaviour.
In practice, it is not possible and so I am recommending relying only on the CL performance.
You seem to be muddling up common English and engineering terminology. And this is my point. The sentence "Phase lag that increases linearly with frequency is time delay." is an engineering no-no. So I recommend not using the word delay in the NFB context when describing phase shift. Because folk really do get confused.
using the open loop transfer function caracteristics,a nyquist plot
allow to define the closed loop stability behaviours of a negative feeddback
circuit..
thus, applying negative feedback doesn t change the intrinsical transfer
function of a system....
OK, I am having trouble understanding your objection. I think you understand what I am trying to say. Let me know how you would state it so I can better understand.
I do like to use terms that are commonly understood, and maybe I am taking too much liberty here. On the other hand, your objection may be based on a philosophical one of yours, in that you philosophically disagree with using the concept of delay in looking at how feedback works.
Let me state one more time, that I am referring to simple delay, as if it were physical delay. If I inserted 50 feet of perfect wire into the loop, it might introduce on the order of 50 ns of delay (what some call flat delay). This is the kind of delay I am talking about. That 50 ns of delay will add 18 degrees of phase lag to the loop at a frequency of 1 MHz. In other words, it will detract from phase margin by 18 degrees if the gain crossover frequency is at 1 MHz. Do you agree with this assertion?
Moreover, I am saying that three or four added poles at some frequency(s) well above 1 MHz will largely approximate the delay described above.
Cheers,
Bob
Bob,
The handy thing about using maths and science is that one don't have to faff about with philosophical interpretations. One can, instead, say this is X and is not Y and prove it. Provided the rules and terminologies are clear.
A pole is not a time delay (Laplace: e^(-ts)). The roll-off and associated phase shift due to a pole is not caused by a time delay.
You may quantify the phase shift of a sine wave that a pole causes and convert this to time displacement. But this time displacement is nothing to do with a time delay in the system.
Is this clear?
Brian
"A time delay will cause a linearly increasing phase lag with frequency" is how I would restate what I think you are expressing.
The handy thing about using maths and science is that one don't have to faff about with philosophical interpretations. One can, instead, say this is X and is not Y and prove it. Provided the rules and terminologies are clear.
A pole is not a time delay (Laplace: e^(-ts)). The roll-off and associated phase shift due to a pole is not caused by a time delay.
You may quantify the phase shift of a sine wave that a pole causes and convert this to time displacement. But this time displacement is nothing to do with a time delay in the system.
Is this clear?
Brian
"A time delay will cause a linearly increasing phase lag with frequency" is how I would restate what I think you are expressing.
Last edited:
I have no problem with your re-statement that "a time delay will cause a linearly increasing phase lag with frequency".
However, I never suggested that a single pole was the same as a pure time delay. This is where you seem to be missing the boat.
What I said, or meant to say, is that a group of multiple poles can cause a nearly linearly increasing phase lag with frequency (over a limited range of frequencies depending on the tolerance you allow for the approximation) that emulates a time delay with reasonable accuracy. To the feedback loop, this gives the appearance of a constant time delay, well in the vicinity of the gain crossover frequency. If you simulate three poles at 5 MHz, 10 MHz, and 20 MHz, and look at the phase vs. frequency +/- one octave about 1 MHz you will see what I mean.
Once again, I NEVER suggested that a single pole was the same as a time delay.
Cheers,
Bob
80% accurate.
This is a great question and probably best answered by Andyc. There is a great thread about simulators somewhere around here.
A real circuit contains many, many more "components" than you tend to see in a typical LT Spice model and the "typical" device model parameters will not be identical to the parts you actually use, so it's a garbage-in, garbage-out thing.
So my opinion is that simulation is hugely useful but it is no substitute for measuring the real circuit.
This is where we disagree. Care to explain why would theory be here different to practice? Again, I'm talking about measurements, not simulations.
Last edited:
If by 'emulates' you mean it looks to a human observer as if a sinewave is delayed in time, then ok.Bob Cordell said:
No, no. Disagree. The feedback loop does not "see" a time delay. Trying to personify it, it is more like a passenger in a car feeling the suspension. This is an issue of responsiveness. If the loop contained an actual delay then the loop would respond differently.
This is all very nitty-gritty but I come back to my original point that novices get confused between the concepts of inertial phase shift of sine waves and time delays. In science and engineering it is important to get the concept right. So I am very careful not to use the word "delay" when talking about sinusoid phase shifts cause by reactive effects.
Is it rutting season? If you want to assert that in your omniscience you can measure an OL circuit and then predict exactly what the CL circuit will do then bully for you.
I am recommending a more prudent method to lesser mortals that involves measuring the CL system to make sure it does what it is supposed to.
80% accurate.
This is a great question and probably best answered by Andyc. There is a great thread about simulators somewhere around here.
A real circuit contains many, many more "components" than you tend to see in a typical LT Spice model and the "typical" device model parameters will not be identical to the parts you actually use, so it's a garbage-in, garbage-out thing.
So my opinion is that simulation is hugely useful but it is no substitute for measuring the real circuit.
A agree completely. Having good transistor models is especially important, and often difficult. This is especially the case for output transistors. Once again, Andy_c is a good reference on this. It is also the case that there are numerous parasitic elements in a real amplifier than can strongly influence things, especially when it comes to local instabilities that often occur at higher frequencies.
Stability of the output Triple is a good example. If the collectors of the pre-drivers, drivers and output transistors are all connected to the same rail bus, any common inductance in that bus may cause instability by way of feedback at high frequencies on that rail bus. Some modest HF filtering of the rail bus as it goes from output stage to driver to pre-driver can help a lot. Even 1 ohm and 0.1 uF to ground between the output transistor rail and the driver rail can help. These are the sorts of things that will not show up in an ordinary SPICE simulation unless great care and good guessing is incorporated.
Cheers,
Bob
Is it rutting season? If you want to assert that in your omniscience you can measure an OL circuit and then predict exactly what the CL circuit will do then bully for you.
I am recommending a more prudent method to lesser mortals that involves measuring the CL system to make sure it does what it is supposed to.
Having a bad day, Brian?
Set aside the limitations in determining the important parameters (gain and phase margins, etc...) closed loop measurements are not going to give much of a clue about how to improve the design, for anything that goes beyond the cdom standard design.
One could of course argue that circuits are reciprocal, so the loop gain can be calculated from the closed loop response, well, speaking about practical issues, you should try to do this.
I can agree that closed loop measurements are more convenient and accesible to DIYers, but to claim that this should be the method of choice, that's more than a stretch.
We end up disagreeing, be it nit-picking or not. We are just not on the same wavelength (oops, you would say we are not on the same frequency).
I think most of the rest of the readers here know what I mean. I also suspect that some of them are saying that we are both right.
Do the simulation I suggested and you will see what I mean.
Cheers,
Bob
I agree with all this. I don't think I said that CL is the only measurement you should do, I thought I said that you should always do it and not rely only on OL measurements. I was talking in terms of how well the circuit is performing. Your are saying that to investigate how to improve a design you need to decompose the system, if you will, and that's very true.
thanks to traderbam and bob for their answer about sim s accuracy....
apart from this, there s a debate that seems to stick to definitions
of what is that and what is this, although there s attempts to clear the
misunderstandings..
i will add more confusion perhaps, but t seems to me that
the open loop transfer function define completely the closed
loop responses of a system...
also, about bob s exemple of a somewhat equivalence between a
phase shift and a delay, i think it s right in that case..
since the phase shift increase with frequency, this could be perfectly viewed as
a constant delay in the amp s response, at least in a restricted frequency
range , as pointed by bob...
a closer study would surely show that the phase shift growth vs frequency
is higher than the rate growth vs frequency of a constant time delay equivalent
phase shift...
hope this last one will not bring more havocs on the thread..
regards,
wahab
apart from this, there s a debate that seems to stick to definitions
of what is that and what is this, although there s attempts to clear the
misunderstandings..
i will add more confusion perhaps, but t seems to me that
the open loop transfer function define completely the closed
loop responses of a system...
also, about bob s exemple of a somewhat equivalence between a
phase shift and a delay, i think it s right in that case..
since the phase shift increase with frequency, this could be perfectly viewed as
a constant delay in the amp s response, at least in a restricted frequency
range , as pointed by bob...
a closer study would surely show that the phase shift growth vs frequency
is higher than the rate growth vs frequency of a constant time delay equivalent
phase shift...
hope this last one will not bring more havocs on the thread..
regards,
wahab
you comment the first half of the statement,
where i m saying that a curve can be confused
with its tangent in a small interval...
here the full statement :
quote
"since the phase shift increase with frequency, this could be perfectly viewed as a constant delay in the amp s response, at least in a restricted frequency
range , as pointed by bob...
a closer study would surely show that the phase shift growth vs frequency
is higher than the rate growth vs frequency of a constant time delay equivalent phase shift..."