Ramblings about feedback control loops

On Sunday, October 6, 2019 at 9:04:16 AM UTC+11, snipped-for-privacy@columbus.rr.com wro te:

etty much did not understand any of it at the time. I just remember one gr aph that seems to get drilled into my brain. The ole 3dB of overshoot and

45 degree phase margin. Like , man , you may not know anything about contr ol loops but that is the absolute nirvana of all control theory....Gotta ha ve that 3 dB overshoot.....

rld problems that require control loops. The idea of oscillations and BW s tart taking hold. Seeing how to close a loop using a bode diagram and phas e margin takes hold.

w you do it! I am even understanding it now!

me that I have been duped. The best control loop is the absolute most over damped loop that you can possibly build that is still responsive enough to get the job done. Wow, they never taught me that in school

Depends on the problem. There is actually a second order loop that always g ets closest to the target value at any time after the impulse.

The damping factor is 0.7, and the response overshoots by about 5%, but the re's no ringing.

If you don't want any overshoot, you have to set the damping factor a bit h igher - at 1.0 or above - but you are farther way from the target value for a bit longer, though there's not a lot in it.

It sounds as if you lecturer didn't implant quite the right message.

--
Bill Sloman, Sydney

PLLs used for repeating network clocks (e.g. for SONET, synchronous Ethernet, etc.) are required to have at most +0.10dB gain peaking. This is to stop jitter buildup when a lot of PLLs (i.e. network nodes) are cascaded.

It's been a while since I designed one from scratch (since in modern products this is done by a single off-the-shelf IC (e.g. Si5347A, etc.)), but when I did, I used to aim for +0.05dB gain peaking. This results in a loop that's massively overdamped. It's also very tolerant to parameter variations.

Regards, Allan

On Saturday, October 5, 2019 at 6:04:16 PM UTC-4, snipped-for-privacy@columbus.rr.com wr ote:

etty much did not understand any of it at the time. I just remember one gr aph that seems to get drilled into my brain. The ole 3dB of overshoot and

45 degree phase margin. Like , man , you may not know anything about contr ol loops but that is the absolute nirvana of all control theory....Gotta ha ve that 3 dB overshoot.....

rld problems that require control loops. The idea of oscillations and BW s tart taking hold. Seeing how to close a loop using a bode diagram and phas e margin takes hold.

w you do it! I am even understanding it now!

me that I have been duped. The best control loop is the absolute most over damped loop that you can possibly build that is still responsive enough to get the job done. Wow, they never taught me that in school

I never took a control class, (though I've read some books), so I'm not sur e exactly what you are referring to. Is the 3dB overshoot the response to a change in the control parameter or a change in the load on the plant?

I use to make my control loops snappy.. quickest response with some over shoot as you say. These days I make 'em slower. But TBH it's hard to tell the difference.

I'd like to read a good control book. Any suggestions? George H.

Pretty much did not understand any of it at the time. I just remember one graph that seems to get drilled into my brain. The ole 3dB of overshoot an d 45 degree phase margin. Like , man , you may not know anything about con trol loops but that is the absolute nirvana of all control theory....Gotta have that 3 dB overshoot.....

world problems that require control loops. The idea of oscillations and BW start taking hold. Seeing how to close a loop using a bode diagram and ph ase margin takes hold.

How you do it! I am even understanding it now!

n me that I have been duped. The best control loop is the absolute most ov erdamped loop that you can possibly build that is still responsive enough t o get the job done. Wow, they never taught me that in school

ure

l

Overshoot is a time-domain idea. A critically-damped loop has no overshoot. Both underdamped and overdamped loops have longer settling time. Peaking in the frequency domain is a related but distinct issue. IIRC critical da mping in a second-order system occurs at a damping factor delta = sqrt(0.

5), and there's no frequency response peaking below delta = 1.0.

Cheers

Phil Hobbs

Pretty much did not understand any of it at the time. I just remember on e graph that seems to get drilled into my brain. The ole 3dB of overshoot and 45 degree phase margin. Like , man , you may not know anything about c ontrol loops but that is the absolute nirvana of all control theory....Gott a have that 3 dB overshoot.....

l world problems that require control loops. The idea of oscillations and BW start taking hold. Seeing how to close a loop using a bode diagram and phase margin takes hold.

s How you do it! I am even understanding it now!

on me that I have been duped. The best control loop is the absolute most overdamped loop that you can possibly build that is still responsive enough to get the job done. Wow, they never taught me that in school

sure

o

r

ell

t.

Right, snappy in that it responds fast, not that it settles faster.

Peaking in the frequency domain is a related but distinct issue. IIRC crit ical damping in a second-order system occurs at a damping factor delta = sqrt(0.5), and there's no frequency response peaking below delta = 1.0. Hmm OK... (delta, (damping) = 1/(2*Q) ?) Is this different than 2nd order filters? Where delta (or Q) = 0.707 (sqrt(0.5)), is flat in frequency but has some bumpies in the time domain?

George H.

snipped-for-privacy@columbus.rr.com wrote

That is certainly the safest way to design control systems.

Just got to make sure there is enough margin.

For example I have been making a product for 25 years which has a switching PSU, outputting 100mA.

Now I need to get 500mA out of it. It does it easily, but over 200mA it goes unstable. The more load, the worse it gets - as expected because the output caps (which form the dominant pole) become less effective with a lower load resistance in parallel with them.

So I had to increase the main feedback cap from 10nF to 100nF. Actually 20nF was enough, so I went for 100 :)

Exactly what you suggest.