SMPS feedback compensation

[kubeek]

I am working on my prototype, which doesn´t use traditional feedback with optocoupler but rather optocoupled mosfet drivers.
Could anyone hazard a guess on the values of the R17-R23 and C31-C33?

Do I even need compensation, or is the driver path fast enough?
Thanks

[sawreyrw]

You need to post a schematic.

[kubeek]

F*****ck, thank you sawreyrw. I was wondering why no one was replying
Feedback goes straight to the output node after the LC filter of the halfbridge.

[hesener]

Hi, I would always use a defined feedback loop behaviour, otherwise you are relying on component variations that may spoil the party.... opto-isolated gate drivers usually are a tad slower than monolithic gate drivers, so there is a "natural" roll-off in there but it may not be what you are looking for in transient performance.

In a control loop, the controller should be the slowest element in the chain, else you run the risk of parasitic oscillations. Can you share further specs of your PS?

[kubeek]

It is supposed to be a 750W PSU for audio amp working at 75kHz, producing cca 2x45V,14A. I know the circuit can work without any regulation quite fine, but I wanted to maybet try some experiments with the feedback, so that is why I am asking which of the components I guess-placed there are the right ones for compensating.

Both the optocoupler and driver have propagation time of max 100ns, which seems ok to me for the 13us period?

Do you have any info on how to compensate based on the step response of the PSU? Something like If you see overshoot, reduce C33.

Also I would like to know how can I model the AC response of the circuit, can I just remove the PWM and rectifiers, and feed the AC source straight into the transformer?

[hesener]

Hi kubeek,

try http://focus.ti.com/lit/ml/slup098/slup098.pdf
or http://www.eetasia.com/ARTICLES/2000...URCES=DOWNLOAD

as first sources of information. With these circuits you should be able to calculate the components according to the step response you want (its really about the corner frequencies).

On the prop delay, that sounds about right for these components but compared to the total period it is quite long. That will show up as a "dead time" in your control loop, and will bring transient response donw, but thats explained in these articles.

On the modeling, simulating switching power circuits is a little tricky. There is excellent literature (a very good book from Christophe Basso). Try replacing as many parts of the circuit as possible with SPICE-built-in voltage or current sources, to reduce simulation time and improve convergence. Difficult to explain it in a nutshell..... You may want to try and simulate the power supply as it is, and then connect a sinusoidal current source to the output and sweep that frequency, at something like 1mA or so (with some DC current to resemble a typical operating point).

just my two cents......

[kubeek]

Originally Posted by hesener

You may want to try and simulate the power supply as it is, and then connect a sinusoidal current source to the output and sweep that frequency, at something like 1mA or so (with some DC current to resemble a typical operating point).

You mean I connect AC current source across the output, measure the output voltage and that should give me the bode plot so that I can see gain and phase margin?

[hesener]

yes. first, put a DC current to "load" the power supply, then in parallel a AC current source with +/- 1mA, and sweep the frequency, and then plot the output voltage. You will not see gain, but output impedance.

If you HAVE the control loop up and running in simulation, you can simply add a voltage source of e.g. 1mV AC into the loop, and sweep it. It will show you gain and phase no problem.

Sorry if I was unclear in the previous post, but "AC response" can mean different things

[kubeek]

I didn´t really have the guts to go throug the whole Basso´s book, but I know that he uses some averaged models of the pwm in the simulations. I spice capable of doing the AC analysis on such switched and non-linear circuit without the averaged model?

[darkfenriz]

The main reason for compensation is the phase lag due to LC filter. The optocoupler pole(s) are usually at least an octave higher than that, so this is no different to compensate. The only difference is you can have less margin, because there is no CTR uncertainity, which normally tanslates to open loop gin uncertainity.

All in all you have to tailor the error amp to have phase lead at the LC resonance. This is a straightforward stability criterion which you use everywhere.

My favourite approach is to use type III compensation with the values out from AC simulation where I use an L and C and not the ESR, which i often believed to form a transfer zero, but I don't assume that.
Then I take L/2 to simulate inductance drop at high current to check if everything stays fine.