Feedback in Isolated Power Converters

[palpurul]

Hello,

I am getting into power electronics, I had some knowledge in designing simple buck and boost converters, but no experience in isolated stuff.

I have been looking at some simple reference designs in TI's website, I am still not sure how feedback signal is utilized across the isolation.

Let me show you one example.


Referece Design

This design uses LM3481, it seems like there isn't even a feedback loop, but only over voltage protection using COMP pin (my initial guess, probably wrong)
I don't know how COMP is utilized normally, but I always thought some kind of frequency compensation is implement using some passive components at the output of the errors amplifier.

Here are the questions that I have about this:

• In this design feedback is pulled to ground. Does that mean that the designer is forcing LM3481 to max. duty cycle without any feedback?
• I don't know how optocoupler is used, but it's not providing any analog feedback and it just trips when there is an over voltage
• What's the point of compensation network at COMP pin if there is no analog feedback, can we just leave it open let the optocoupler pull it low with the internal pull up?
• What's all of the circuitry on the left that is all connected to optocoupler's LED? If you don't want to explain I'd understand 

As I said before I don't have much experience in power electronics, but I'd like to work with isolation and similar ICs like UCC2808 (and its variants) and LM3050 (and similar stuff). What are your recommendations for me, please let me know

PS:
Here is another design that does not use any feedback across isolation.

[Marco]

The output of the feedback Error Amplifier is not low impedance, otherwise the COMP pin would never do anything. The compensation input does affect the duty cycle even with FB grounded.

No idea about the rest.

[Someone]

I don't know how optocoupler is used, but it's not providing any analog feedback and it just trips when there is an over voltage

Power electronics are almost entirely analogue in nature, this seems to be a misunderstanding that optoisolators are binary on/off devices (very very few have hysteresis on the input and are usually separately classified as digital isolators).

What's all of the circuitry on the left that is all connected to optocoupler's LED? If you don't want to explain I'd understand

There are application notes on those parts: SNVA005B, SNVA761A, SLVAF37

[Faringdon]

The error amplifier is in the TL431....it does the regulation, as the vout goes above regulation more current flows in the opto, and the COMP pin gets pulled down...the COMP pin connects directly to the  PWM comparator...so thats how it works.
I will atttach LTspice of same kind of cct...it is free simulator

Also try to find "Designing with the TL431" by Dr Ray Ridley...here..
https://www.ridleyengineering.com/design-center-ridley-engineering/49-circuit-designs/203-a21-designing-with-the-tl431-2.html

Dr Basso in his books covers TL431 feedback in best detail
Your opto is acting in analog, also you can put in a faster opto with eg a 5v supply pin on it...this gives faster feedback for you, if you so want....eg you dont want so much vout overshoot etc etc...then again, you can always put RC across the 127k for overshoot reduction.
That pri side feedback network looks a bit overdone...i would just have a single small cap on the comp pin. say 10nf to 100nf.

I must admit i dont like to use your R15...i instead prefer say an 820R across the opto diode. With R15 it kind of makes it harder to really reduce current in the opto diode.

I am sure you know the TL431 draws about 1mA of bias current all the time....it draws this into its cathode, but cannot draw it any more if v(cath) <2v.
This is why i prefer a normal opamp to a tl431...because opamp has a separate vcc pin.

[Geoff-AU]

This design uses LM3481, it seems like there isn't even a feedback loop, but only over voltage protection using COMP pin (my initial guess, probably wrong)
I don't know how COMP is utilized normally, but I always thought some kind of frequency compensation is implement using some passive components at the output of the errors amplifier.

The trick is to look at the functional block diagram (see attachment, not letting me inline for some reason)

EA is the error amplifier, and it compares the FB pin to the 1.275V reference.  When FB goes higher than the reference, COMP gets pulled low by the error amplifier .  The EA also has limited output drive, and so you can add a compensation network to its output to fine tune the regulator's behaviour.

So by grounding FB and just pulling COMP low whenever the regulation point is reached, you're just taking a shortcut.  Whether the shortcut does what you want or blows up in your face depends on the specifics of the situation, but if the manufacturer has given a reference design then they have hopefully thought about it a bit and tested that it works properly.

In this case, it's much easier to pull COMP low with an opto-isolator than it is to pull comp "slightly higher than 1.275V".

[Wolfram]

As mentioned by other posters, the TL431 is the error amplifier in this case, and the LM3481 error amplifier is bypassed. Having the error amplifier on the output side minimizes the impact of optocoupler gain (which is usually very badly controlled) on the output voltage regulation.

[palpurul]

The error amplifier is in the TL431....it does the regulation, as the vout goes above regulation more current flows in the opto, and the COMP pin gets pulled down...the COMP pin connects directly to the  PWM comparator...so thats how it works.
I will atttach LTspice of same kind of cct...it is free simulator

Also try to find "Designing with the TL431" by Dr Ray Ridley...here..
https://www.ridleyengineering.com/design-center-ridley-engineering/49-circuit-designs/203-a21-designing-with-the-tl431-2.html

Dr Basso in his books covers TL431 feedback in best detail
Your opto is acting in analog, also you can put in a faster opto with eg a 5v supply pin on it...this gives faster feedback for you, if you so want....eg you dont want so much vout overshoot etc etc...then again, you can always put RC across the 127k for overshoot reduction.
That pri side feedback network looks a bit overdone...i would just have a single small cap on the comp pin. say 10nf to 100nf.

I must admit i dont like to use your R15...i instead prefer say an 820R across the opto diode. With R15 it kind of makes it harder to really reduce current in the opto diode.

I am sure you know the TL431 draws about 1mA of bias current all the time....it draws this into its cathode, but cannot draw it any more if v(cath) <2v.
This is why i prefer a normal opamp to a tl431...because opamp has a separate vcc pin.

Thank you for your response!

I am still learning about all of these stuff, so forgive if I ask very obvious or dumb questions.

I probably have a lot to learn, but let me ask one question. How do they make the regulation? Is optocoupler used in the linear region to make things work? I was thinking opto is used almost in a digital, but it seems like it's not the case as @Someone pointed out.

[palpurul]

This design uses LM3481, it seems like there isn't even a feedback loop, but only over voltage protection using COMP pin (my initial guess, probably wrong)
I don't know how COMP is utilized normally, but I always thought some kind of frequency compensation is implement using some passive components at the output of the errors amplifier.

The trick is to look at the functional block diagram (see attachment, not letting me inline for some reason)

EA is the error amplifier, and it compares the FB pin to the 1.275V reference.  When FB goes higher than the reference, COMP gets pulled low by the error amplifier .  The EA also has limited output drive, and so you can add a compensation network to its output to fine tune the regulator's behaviour.

So by grounding FB and just pulling COMP low whenever the regulation point is reached, you're just taking a shortcut.  Whether the shortcut does what you want or blows up in your face depends on the specifics of the situation, but if the manufacturer has given a reference design then they have hopefully thought about it a bit and tested that it works properly.

In this case, it's much easier to pull COMP low with an opto-isolator than it is to pull comp "slightly higher than 1.275V".

So in this case we are basically bypassing the EA amplifier and drive its output directly with the optocoupler I guess.?

[palpurul]

As mentioned by other posters, the TL431 is the error amplifier in this case, and the LM3481 error amplifier is bypassed. Having the error amplifier on the output side minimizes the impact of optocoupler gain (which is usually very badly controlled) on the output voltage regulation.

I think I get the general idea, but I have seen designs that uses the optocoupler as a digital element (just a simple voltage divider at the input of the optocoupler and connect the output of the optocoupler to CS pin). I don't if this is common practice for cheap stuff, and I might actually not making correct assumptions. 

Here is the desing:


I think it just trips after a certain point with the heavy filtering on the HVDC side (100nF in paralell with 3.6k), I am pretty sure, but I might be wrong.

[Geoff-AU]

I think it's easiest to understand these things qualitatively..  yes it is in the linear mode of operation, but I often think about it more in terms of causal X and Y relationships..  If X happens then the circuit does Y.

In the first example, the TL431 really wants to regulate quite precisely at a particular voltage so if the output voltage overshoots, the TL431 pulls really hard through the opto, a lot of light shines across and the COMP pin gets decked into the ground.  If the voltage then undershoots, the TL431 switches the LED off and the switcher goes crazy again.  This is loop instability, and your output voltage oscillates around and below the setpoint.  You need some "analog" in the system to calm things down (filtering, and various knees in response curves).

Ideally what you want is that as the output voltage reaches its desired point, the TL431 juuust starts pulling enough current to switch the opto on, which juuuust pulls the COMP down enough that the power supply cools off enough to supply the output load at the desired voltage.  The trick is, this bias point needs to allow for a range of output loads, temperatures, component tolerances etc.  But having succeeded in picking a good bias point, then you are into the realm of small effects where the output voltage gets a tiny bit higher, the TL431 pulls on the LED a tiny bit more, and the COMP bin just whispers down a few millivolts and the system stays at its regulation point.  The COMP pin feedback network has a capacitor for "fast" filtering and an RC for "slow", so it can tame a couple of types of instability.
 
The second example is the same thing in a cheap and nasty way, as voltage goes up TL431 turns on Q5 which yanks on the COMP pin.  Since there's no filter network on the primary side this will be pretty brutal, and I expect HVDC is fairly loose with its regulation requirements.