I am designing a forward converter that is supposed to work as a pre-regulator for a linear supply. The forward converter is supposed to be able to give output from 4V to 34V and 0 to 5A. While compensating the feedback loop I found out that the worst-case scenario is when the forward converter is supposed to provide 5A at 4V (low voltage and high current) and in this case my phase margin is just 26 degrees as shown below in the bode plot:

enter image description here

I have read it many times on internet that we should have at least 45 degrees of phase margin. But my transient response still seems pretty good to me as shown below:

Transient Response at 4V, 5A:

enter image description here

Step Response from 5A to 4A and from 4A to 5A:

enter image description here

At full load, i.e 34, 5A I have around 20kHz bandwidth and 70 degrees of phase margin.

This is my schematic:

enter image description here

Switching frequency of the converter is 100kHz.

So I want to know whether is it okay to have this low phase margin and still have stable power supply when implemented on actual hardware?

  • \$\begingroup\$ I'm not altogether convinced that your open loop gain response would produce the closed loop response when you closed the loop. Did you do open loop measurements correctly? I also don't see much of a difference in the PMs either around 100 kHz - are you sure you got this correct? \$\endgroup\$
    – Andy aka
    Commented Dec 11, 2021 at 18:10
  • \$\begingroup\$ You also seem to have a high Q resonance around 100 kHz. Need take more data points to resolve what's happening near there. \$\endgroup\$
    – jp314
    Commented Dec 11, 2021 at 18:19
  • \$\begingroup\$ It looks like you have SIMPLIS set up correctly for the loop response, but I agree with Andy that the closed-loop transient response doesn't seem like it follows from your open-loop response. Also, if you have 27 degrees of phase margin with nominal component values, what happens over tolerance, temperature and aging? You could easily get in trouble in production. \$\endgroup\$
    – John D
    Commented Dec 11, 2021 at 20:18
  • 2
    \$\begingroup\$ How did you compensate your circuit? Usually, you compensate for the worst-case power stage response which seems to be 4 V in your case. For your information, the compensation circuit around the TL431 is not that of a type 2: a single cap. is enough and the emitter should have a cap. for the high-frequency pole. Have a look at my free 60+ SIMPLIS templates where all the converters are automatically compensated. \$\endgroup\$ Commented Dec 11, 2021 at 20:40
  • \$\begingroup\$ Yes you guys are correct. I was measuring the open loop bode plot the wrong way and explains why I was not getting the expected compensation results. I went through the UC3845 datasheet and saw its internal circuit and corrected my mistake. \$\endgroup\$
    – tinkerer
    Commented Dec 13, 2021 at 9:08

1 Answer 1


The control loop in your example uses a TL431 which is a self-contained op-amp plus reference. This part needs at least 2.5 V to operate properly. When you reduce your output voltage to 4 V, you are forced to reduce the LED series resistance to a low value (100 ohms in your example) to ensure regulation considering the 1-V drop of the optocoupler LED (see my book on loop control). This LED series resistance sets the mid-band gain of your compensator. And with such a low value, you have a very large crossover at a 4-V output compared to the one you have at 34 V. And in this high-frequency region, the phase lag is already well pronounced hence the poor phase margin.

One way to overcome this is to decouple the fast lane from the slow lane. Please read this seminar that I built some years ago which explains what are the slow and fast lanes but that is where the issue is I believe. You have several options to decouple the fast lane but the only one that can work in your case is an auxiliary \$V_{cc}\$ coming from an auxiliary winding. Considering the ratio of 34/4 which is 8.5 you may have to think of a clever solution to produce this extra stable voltage. I have used one of my free SIMPLIS platforms to simulate a 34-V forward converter whose output can be folded back to 4 V:

enter image description here

The calculations of the type 2 featuring a disconnected fast lane are given in the right-side sheet and they automate the process in SIMPLIS. I have shot for a 1-kHz crossover and, as expected, the power stage response does not change significantly in relationship with the output voltage setpoint. I have a decent phase margin of nearly 60° in both cases, 34 V and 4 V output. The low-side resistance of the divider is changed for doing so: don't touch the upper resistance as it would affect the compensator response while the low-side does not (only the dc setpoint changes). When you follow these recommendations, the response at 34 V out is ok:

enter image description here

and so it is when \$V_{out}\$ is set to 4 V:

enter image description here

Finally, the transient response shows a good stability at both output voltages:

enter image description here

  • \$\begingroup\$ I was going through the seminar you have linked for the last few days and ended up using a type 3 compensation to get enough phase margin at all conditions. Now my worst-case PM is 48 degrees. To increase Rled max limit I am using TLV431 instead of TL431 which is very hard to get at a reasonable rate in my area so not much practical. I will be using 12V supply for the opamp in linear stage so will use that to decouple the fast lane. I will try to get it working with the TL431 and auxiliary winding you suggested. I used your book to design the power stage. Thank you so much for all the help. \$\endgroup\$
    – tinkerer
    Commented Dec 13, 2021 at 8:59
  • \$\begingroup\$ Glad to read this and good luck with this design! \$\endgroup\$ Commented Dec 13, 2021 at 11:59

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.