Oscillations at a phase margin of 90?

[elmolla]

Hi All,

I'm getting a weird behavior from a negative feedback loop. With a phase margin of 90 degrees I have steady state oscillations around 5% the steady state value although I have a gain of about 60 dB! Any idea?

 

[RF-OM]

Is it simulation or real hardware? Also is it IC or discrete component circuit? Your question will be more clear with these answers.

Best regards,
RF-OM

 

[sachinagg77]

AC simulations dont always present the true picture. First, some of the poles and zeros might not be visisble in the AC simulation (for example, closely placed zero-pole pair) which might have detrimental effect on settling behavior. Secondly, AC analysis are for a particular set of operating condition (operating points). The actual operating conditions might differ. Please make sure that the operating conditions under which you see the oscillations match with that for the AC simulation.

Hope this helps.
Regards
Sachin

 

[svraj]

It will also be good to check the AC simulation settings for phase margin analysis. Are you using the stb function? Are the circuit settings similar to that of the transient analysis settings?

Regards,
svraj

 

[elmolla]

Thanks everybody for the reply.

Concerning the circuit nature, it is an analog IC, I'm using Spectre as a simulator and I'm using the STB analysis for loop stability check and the settings and operating point are the same for both transient and STB analysis.

Any ideas?

 

[LvW]

I understand that both frequency as well as time domain results are not from measurements but from simulations, right ?
In this case, something must be wrong with your STB analysis.
In most of these cases, there are several loops (perhaps hidden ?) - and one has analyzed not the loop which governs the behaviour.
On the other hand, I must confess I am not familiar with the STB analysis since I always use only the ac analysis for stability checks. For my opinion, this is more clear and I always know what I am doing !
Important requirements: You have to choose a point within the loop for doing the loop gain analysis which neither destroys the bias point nor changes the load conditions.
Did you check stability properties at another point within the loop ?

 

[elmolla]

You have a good point LvW; But I've chosen a point in the loop which is not common except with the loop of interest. Also the STB analysis doesn't disturb the bias point.

There is something interesting that I noted, I used to look only at the phase margin result, but when I looked at the loop phase and gain I found weird cut like effect in the phase response of the loop, it is like having two LHP zeros, then the phase drops steeply then gets back on the rise, so I have a shape like the letter N in the phase , but this occurs at a gain or around -25dB!! Am I missing something?

I also tried to modify the circuit, now I'm getting a V like phase response, i.e. a steep fall in phase after the 0 dB point which then begins to rise then stabilizes like two poles then two zeros giving a V-shape like phase response, but still around the -25dB range??

In all the above cases the phase margin is about 90-70 degrees.

Am I missing something?

 

[LvW]

In the meantime I have convinced myself that the STB analysis does not has any influence on bias or load conditions as it is based on the Middlebrook method which is universal.

That means, for my opinion the only cause of strange results can be the place for the probe you have inserted. Try other points within the loop and check if the results are identical. For a single loop system the have to be identical.

 

[elmolla]

Thanks for the fast reply LvW.

The phase response doesn't depend on the position of the probe. I selected the probe position to be the same point I used for the hand analysis, which is a point in a single loop only.

I don't really understand whats going :|

 

[LvW]

There is something interesting that I noted, I used to look only at the phase margin result, but when I looked at the loop phase and gain I found weird cut like effect in the phase response of the loop, it is like having two LHP zeros, then the phase drops steeply then gets back on the rise, so I have a shape like the letter N in the phase , but this occurs at a gain or around -25dB!! Am I missing something?

This phase jump (like an N) indeed is an indication of instabilty. And If the circuit is not stable you cannot trust the gain response. Are you sure to be correct by interpreting the simulated phase margin as positive ?

 

[elmolla]

Do you mean that even if the normal phase margin calculation gives a stable circuit, any phase discontinuity, like the N I've seen, means that the circuit is unstable?

But I've seen this problem even with a continuous like, U shaped phase response?

 

[LvW]

Do you mean that even if the normal phase margin calculation gives a stable circuit, any phase discontinuity, like the N I've seen, means that the circuit is unstable?

No, my point was that such a phase jump could be an indication of instabilty - and perhaps, at the same time, your simulation of the margin is questionable.
I know, that determination of such a margin very often is not done correctly.
For example, the "normal" input has to be grounded during stability simulation.

 

[elmolla]

Thanks for your fast reply LvW

OK. I've considered my setup of phase measurement and its just as given by the book So there isn't really a problem about how the stability margin is measured.

I've chosen the IPROBE point in the loop where I had done the hand analysis; It is a very nice spot, where the loading is very small and the loop gain calculations are just straight forward, so I don't really think there is a problem with the setup.

I agree with you that the phase jump is a bad sign, but why did I encounter the same problem even for a smooth shaped phase response; this same phase response corresponds to a stable transient case at other loading conditions.

Can we say the phase response continuity in a necessary but not sufficient condition to assure the loops stability? If this is the case, then when can we trust the loop analysis, given the behavior of the above case?

I'm not really quite relaxed to hear not many people trust the loop stability technique, its essential in the analysis of any feedback system? How else could we proof our hand analysis?

 

[LvW]

Is it possible to see a block diagram or something similar from your arrangement ?
In this case, it would be much more easy to discuss on a realistic base.

 

[elmolla]

Its a simple LDO; An amplifier followed by a PMOS switch with an RC load

 

[quaternion]

It seems to be like my problem.

 

[LvW]

.....followed by a PMOS switch

A switch ??? No linear control element ?

 

[elmolla]

.....followed by a PMOS switch

A switch ??? No linear control element ?

Sorry.. I meant a PMOS pass device

Added after 4 minutes:

It seems to be like my problem.

Yes quaternion, the only difference is that my oscillations are less sharp, they're more sinusoidal.

I also have one loop which is stable as I said before.

 

[LvW]

I also have one loop which is stable as I said before.

Now I am confused ! In an LDO circuit there is only one loop. And now you state it is stable ? What then is the problem ?

 

[elmolla]

I was talking to quaternion about in this part, because he had the same problem but he had an unstable loop, in a two loop LDO.

My problem is as is. I still get the instability problem in transient although he phase response shows stability

I also have one loop which is stable as I said before.

Now I am confused ! In an LDO circuit there is only one loop. And now you state it is stable ? What then is the problem ?