Feedback loop nonsense?

[cupoftea]

Hi,
We all know that in an SMPS's, the output LC filter resonant frequency should be at least 3x more than the SMPS's crossover frequency.
Theory tells us that if it isnt then this filter will interfere with the feedback loop of the SMPS.
However,
In real life....ie shedloads of reverse engineered SMPS's......
This is never the case...........to do so would usually mean a ridiculously small inductor in the output LC filter...not to mention a ridiculously small capacitor getting used.
Not to mention that often the loads that get connected to the SMPS almost always comprise a good few 100's of uF at their input....so by the time the SMPS is powered up...the "actual", real LC output filter resonant frequency is way, way under the crossover frequency of the SMPS.

So what's behind all this?

 

[dick_freebird]

Product optimization begins with rules-of-thumb. Never ends there.
If somebody can "get away with something" and reduce BOM cost,
they will.

Of course sometimes you only -think- you got away with it....

 

[Easy peasy]

" We all know that in an SMPS's, the output LC filter resonant frequency should be at least 3x more than the SMPS's crossover frequency. "

What a load of malarky .... back to polytech

 

[cupoftea]

What a load of malarky .... back to polytech

...Page 32 of the below...

https://www.ti.com/seclit/ml/slup340/slup340.pdf

..aswell as many many other refs saying same thing...i can send here if you wish.

..but i see your point entirely though.....after all, nobody obeys the "3x" rule.

--- Updated Jul 24, 2023 ---

..Will consult the two great texts of Dr Basso later, to see if he has opinion on the "3x" rule

--- Updated Jul 24, 2023 ---

Page 17 of this also supports the "3x" rule...(or at least...res freq more than the xover freq)

https://www.power.com/sites/default/files/documents/an47.pdf

--- Updated Jul 24, 2023 ---

Page 17 of this also supports the "3x" rule...(or at least...res freq more than the xover freq)

https://www.power.com/sites/default/files/documents/an47.pdf

 

[cupoftea]

…though I confess, once when at a telco, the designer put a tiny ceramic cap bank at the output of a 500W half bridge, then a filter inductor, then a huge electro cap bank…and that went unstable….and caused 5 million quids worth of 4G test results to be scrapped….so there is something behind it.

 

[Easy peasy]

" the designer put a tiny ceramic cap bank at the output of a 500W half bridge, then a filter inductor, then a huge electro cap bank "

this does not sound like a designer who really knows about feedback loops

--- Updated Jul 25, 2023 ---

Caps are there to:

1, soak up ripple current - which would other wise propagate to the load,

2, soak up current and limit voltage rise if the load is suddenly disconnected - i.e. give the volt loop time to work,

3, provide peak pulse current to loads - again to allow the f/b loop to ramp up to a high load state,

4, to provide a pole to the power stage in association with series ( sometimes just leakage ) inductance that the control loop can work around - having large o/p caps defines this pole and allows stable operation if the control loop is properly designed.

 

[dick_freebird]

If it behaves as a transconductor then there's no such thing as too
much C.

But that's the question.

A current mode control is a transconductor, a voltage mode control
is not.

 

[cupoftea]

Thanks, and you agree, that if the output LC filter resonance frequency is << xover frequency, then it needs accounting for in the power stage transfer function.?
Its interesting if your divider comes off upstream of the output filter inductor.
Basso doesnt go through feedback loops with huge LC output filters at the output of the SMPS....other than to comment on the resonant frequency of same.

 

[mtwieg]

You should clarify that this rule of thumb is meant to apply to second stage LC filters, not the "main" choke and filter capacitors which see most of the ripple current. For the main LC, the rule of thumb is typically the opposite, with the crossover frequency being above the natural resonant frequency. Still a crude rule of thumb though with lots of room for if's and but's.

 

[cupoftea]

Thanks, yes indeed its second stage LC filters we speak of.

What a load of malarky .... back to polytech

And.....Dr Basso speaks on this subject in his great book "Switch Mode Power Supplies..."
Basso, on page 685 states......QUOTE>> ...the cutoff frequency of this filter must be well above the crossover frequency (at least 2 to 3 times) to avoid further stressing the phase at this point" >>UNQUOTE.

...And its this that doesnt seem to be the way most people do things.....typically they are well under 2 or 3x the crossover frequency....specially when you consider added loads which have a big 'lytic in front of them.

In the above, Basso is speaking about reducing the vout ripple on the vout of flybacks.

 

[Easy peasy]

This point needed clarifying in your earlier post:

" Thanks, yes indeed its second stage LC filters we speak of. "

but the " we " should have been replaced with "I", as you are not speaking for others

 

[cupoftea]

Thanks, yes i must admit the term "second stage LC filter" isnt something ive heard of, as opposed to the "output LC filter" though. The output inductor of an half bridge say, and the output cap, i usually hear being called the "power stage LC ".

this does not sound like a designer who really knows about feedback loops

..Thanks, Yes, Anyhow, the point about the unstable Full Bridge, was that it was seen to be stable at max load and at 50-100% load transients......but when it was connected to the RF amplifier, which had a large electrolytic in front of it, it then went unstable.

The power supply on its own though, was stable, in spite of very heavily violating the 3x rule by itself.

I have seen a lot of PSU's, which have a second stage filter, and dont go unstable until the "second stage LC filter" resonant frequency gets way below the "3x xover frequency" level.

 

[dick_freebird]

Electrolytics cannot be considered "capacitors" at high
frequencies like your usual loop BW. The ESL buggers
them well before that. "Instability" could be provoked by their
self-resonance?

 

[cupoftea]

Thanks, though in this case, its the lowness of the frequency of this 'lytic bank with the second stage output filter inductor thats in question, and indeed the problem.........where the capacitance of this lytic bank is relevant.
Ayk, we are speaking of the lowness of the second stage LC resonance in comparison with xover frequency.

I think Dr Ridley covers this in his small white book on control..but mine is locked up 100 miles away.

 

[cupoftea]

This paper on page 10, shows how the transfer function gets adjusted for the addition of the second stage LC filter......
Practical Feedback loop design considerations for switch mode power supplies, by Hangseok Choi

Hangseok Choi

 

[cupoftea]

Second stage LC filter sizing:
As an example, here are the bode plots for a flyback SMPS with xover at 600Hz.
...And also for the same flyback, but this time with a second stage LC filter with F(resonance) = 919Hz

As you can see, the Bode plots still show stability , even after the addition of the LC filter (10uH and 3mF).

This shows that the " rule of thumb" discussed, is not correct.

The factor seen in the equation on page 10 of Dr Hangseok Choi's doc "practical feedback loop...." was used to multiply by the original transfer function to get the Transfer function with the second stage LC.

This shows why only extreme cases of external LC filter cause instability.

In the shown plots, the phase margin is certainly reduced, but not giving total instability.....the phase is still above 180degrees at the crossover.