Non internally compensated op amp question.

[Optikon]

Hi all, this question is general and conceptual in nature.

Non internally compensated op amps do not contain the comp-cap internal to the device and require external compensation to be stable.
These devices are inherently unstable. If this is true, what is the mechanism that is providing 180 deg phase shift?

For example, say I wire up closed loop gain of 2 non-inverting op amp with no comp, I expect instability but what is responsible for the phase shift? Are the the gain stages of the internal design responsible for the phase shift?

Now suppose I wire for CLG = 0.5. No comp. Is it unconditionally stable?

Comments, insights?

Thanks!

 

[Roff]

For stability in a closed loop system, the phase shift around the loop must be less than 360 degrees at all frequencies where the gain is greater than one. In an uncompensated op amp, each gain stage has an associated bandwidth due to parasitic capacitance and/or transistor Ft considerations, so each will have one or more poles (RC networks). An RC network approaches 90 degrees phase shift asymptotically as the frequency increases (it is 45 degrees at the -3dB point). So, if you have an op amp with 2 or more gain stages (true of almost all of them), they will be unstable in a feedback loop if the loop gain is greater than one when the phase shift is 360 degrees (180 degrees due to inversion, 180 degrees due to phase shift).
Some op amps are compensated by making the bandwidth of one stage significantly lower than the rest, which causes the overall gain to go below 1 while the phase shift is less than 360 degrees. Others add a lag-lead network to one stage, which accomplishes the same purpose, but allows overall bandwidth to be higher. The lead portion of the network basically stops the rolloff of the compensated stage at the frequency where the rolloff of the second stage begins.
Higher closed-loop gain configurations generally don't require such heavy compensation, because the loop gain is less and therefore will go below one (0dB) before the phase shift is excessive. In fact, that is the main reason for the existence of uncompensated op amps. They allow you to optimize the bandwidth for your particular closed loop gain.
A CLG of 0.5 is only possible in the inverting mode, and the loop gain is 2/3 *(open loop gain). This is pretty close to full open loop gain, so I wouldn't expect the configuration to be stable.


Whew! :roll:

 

[Optikon]

For stability in a closed loop system, the phase shift around the loop must be less than 360 degrees at all frequencies where the gain is greater than one. In an uncompensated op amp, each gain stage has an associated bandwidth due to parasitic capacitance and/or transistor Ft considerations, so each will have one or more poles (RC networks). An RC network approaches 90 degrees phase shift asymptotically as the frequency increases (it is 45 degrees at the -3dB point). So, if you have an op amp with 2 or more gain stages (true of almost all of them), they will be unstable in a feedback loop if the loop gain is greater than one when the phase shift is 360 degrees (180 degrees due to inversion, 180 degrees due to phase shift).
Some op amps are compensated by making the bandwidth of one stage significantly lower than the rest, which causes the overall gain to go below 1 while the phase shift is less than 360 degrees. Others add a lag-lead network to one stage, which accomplishes the same purpose, but allows overall bandwidth to be higher. The lead portion of the network basically stops the rolloff of the compensated stage at the frequency where the rolloff of the second stage begins.
Higher closed-loop gain configurations generally don't require such heavy compensation, because the loop gain is less and therefore will go below one (0dB) before the phase shift is excessive. In fact, that is the main reason for the existence of uncompensated op amps. They allow you to optimize the bandwidth for your particular closed loop gain.
A CLG of 0.5 is only possible in the inverting mode, and the loop gain is 2/3 *(open loop gain). This is pretty close to full open loop gain, so I wouldn't expect the configuration to be stable.


Whew! :roll:

Thanks for the good explanation! A few points of confusion still exist though.. Consider what is used as an opamp model (like a spice macromodel for example). There, a voltage dependent voltage source driving an RC (comp section) that drives a buffer output stage. In this model, no 180 inversion exists and so phase shift is at most 90 degrees. Am I correct to say that this model is no good for evaluating stability?
Have you seen what I'm talking about?

Now in a real opamp.. lets say something mundane such as a 741. The 180 degrees of inversion is due to the input stage differential pair with collector output (common emitter config?) is this right? Because the following stage incorporates the comp-cap and output is just something similar to an emitter follower where only Ft phase shift takes place.
I suppose Ft phase shift takes place in _ALL_ transistors but I'm looking to understand where the dominant areas are. Let's not consider the ones that are "way out there."

On a different but related note,

If I remember correctly the Barkhausen criteria is the condition that must be satisfied for oscillation which, is closely related to stability. The conditions are stated as you mention, Gain = 1, phase shift = 180 deg. in feedback network for the canonical form. Gains greater than one will usually be reduced to one by circuit non-linearities. But can a more general statement be made in terms of loop gain:

Gain >= 1 and n*360 degrees where n is positive integer >= 1? Intuitively, it seems that this too would satisfy the condition for oscillation (instability). From the feedback point of view, the circuit does not know about time delay really, 360 degrees, 720 degrees, 1440 degrees should all look the same from this standpoint. Can this more general statement be made? I have not seen mention of this in the literature.

Your comments are most appreciated!
Thanks!

 

[Roff]

Consider what is used as an opamp model (like a spice macromodel for example). There, a voltage dependent voltage source driving an RC (comp section) that drives a buffer output stage. In this model, no 180 inversion exists and so phase shift is at most 90 degrees. Am I correct to say that this model is no good for evaluating stability?
Have you seen what I'm talking about?

Well, any macromodel will have inverting (180º) and noninverting inputs, but if there is only one pole, it should be unconditionally stable unless you add external poles, or if the combination of the feedback resistor and input capacitance is excessive.

Now in a real opamp.. lets say something mundane such as a 741. The 180 degrees of inversion is due to the input stage differential pair with collector output (common emitter config?) is this right? Because the following stage incorporates the comp-cap and output is just something similar to an emitter follower where only Ft phase shift takes place.
I suppose Ft phase shift takes place in _ALL_ transistors but I'm looking to understand where the dominant areas are. Let's not consider the ones that are "way out there."

I think the 741, even though it has 2 gain stages and a follower output stage, has one dominant pole whose corner frequency is set by the product of the output impedance of the 1st stage (many meghoms) and the 30pF feedback capacitor of the 2nd stage. This keeps the phase shift at 90 degrees pretty much throughout the useful bandwidth of the part, making it very stable in the absence of lots of additional phase shift in the feedback loop.

Gain >= 1 and n*360 degrees where n is positive integer >= 1? Intuitively, it seems that this too would satisfy the condition for oscillation (instability). From the feedback point of view, the circuit does not know about time delay really, 360 degrees, 720 degrees, 1440 degrees should all look the same from this standpoint. Can this more general statement be made? I have not seen mention of this in the literature.

Well, time delay=phase/frequency. The higher the phase shift at a given frequency, the higher the delay. I don't know what happens if you still have loop gain at integer multiples of 360º phase shift.
Optikon, I'm not a font of knowledge on this subject. I've just been doing analog design for 40 years, and many years ago (before 741s), I designed and compensated a few discrete transistor op amps. I've forgotten some of the stuff I knew then, and have certainly learned a lot of stuff that I didn't know then.

 

[Optikon]

Consider what is used as an opamp model (like a spice macromodel for example). There, a voltage dependent voltage source driving an RC (comp section) that drives a buffer output stage. In this model, no 180 inversion exists and so phase shift is at most 90 degrees. Am I correct to say that this model is no good for evaluating stability?
Have you seen what I'm talking about?

Well, any macromodel will have inverting (180º) and noninverting inputs, but if there is only one pole, it should be unconditionally stable unless you add external poles, or if the combination of the feedback resistor and input capacitance is excessive.

Now in a real opamp.. lets say something mundane such as a 741. The 180 degrees of inversion is due to the input stage differential pair with collector output (common emitter config?) is this right? Because the following stage incorporates the comp-cap and output is just something similar to an emitter follower where only Ft phase shift takes place.
I suppose Ft phase shift takes place in _ALL_ transistors but I'm looking to understand where the dominant areas are. Let's not consider the ones that are "way out there."

I think the 741, even though it has 2 gain stages and a follower output stage, has one dominant pole whose corner frequency is set by the product of the output impedance of the 1st stage (many meghoms) and the 30pF feedback capacitor of the 2nd stage. This keeps the phase shift at 90 degrees pretty much throughout the useful bandwidth of the part, making it very stable in the absence of lots of additional phase shift in the feedback loop.

Gain >= 1 and n*360 degrees where n is positive integer >= 1? Intuitively, it seems that this too would satisfy the condition for oscillation (instability). From the feedback point of view, the circuit does not know about time delay really, 360 degrees, 720 degrees, 1440 degrees should all look the same from this standpoint. Can this more general statement be made? I have not seen mention of this in the literature.

Well, time delay=phase/frequency. The higher the phase shift at a given frequency, the higher the delay. I don't know what happens if you still have loop gain at integer multiples of 360º phase shift.
Optikon, I'm not a font of knowledge on this subject. I've just been doing analog design for 40 years, and many years ago (before 741s), I designed and compensated a few discrete transistor op amps. I've forgotten some of the stuff I knew then, and have certainly learned a lot of stuff that I didn't know then.

Thanks for all your comments..They are very much appreciated!

I've been trying to wrap my head around the origins of phase shift in opamps and their models. Sometimes, it seems it is an uphill battle. You've got vendors that give you crude models, textbooks and app notes with anywhere from one to 3 dominant pole models all claiming to model stability.

Can you describe the meaning behind 3rd order intercept? I'm confused about what this really is and means. I havnt found an explanation that I can intuitively grasp yet.

Thanks again!