I tried to design a step up SMPS to create ~200-300V from a 24V input with an expected power of ~12W max. Since the project is for a hobby, I chose to attempt a hybrid boost-flyback converter to get the needed voltage gain with a reasonable duty cycle. I think i have a non-stable system at some output voltages, there is a low frequency ~7Hz triangle wave of ~10V on my output of 250V. Other then that it seems to work fine with ~85% efficiency. When I tune it to a stable voltage (~50% of random voltages are stable), there is no low frequency ripple, only the expected ~1V output ripple at the switching frequency.

For compensation I made some assumptions (transformed the load / output voltage to primary side) and followed a typical boost compensation procedure. I attempted to follow suggested design steps in the PMIC (LM3478) datasheet. I added a low frequency pole but, apparently that did not work so well.

Basically I built a compensation network for ~48V output, with a ~1.5mH inductor, ignoring the fact it is a flyback. Any easy ideas of how to stabilize the output? I'm at the point to blindly play with the compensation network and see if it changes anything.

I used the following app notes for the basis of my design: AN-5081, AN-IPS-08

Design Info:

  • \$ f_s=120kHz \$ ; \$ V_i = 24V \$ ; \$ V_o = 200V-300V \$
  • Load is steady, resistive.
  • Transformer is 2 mil air gaped, 52uH Primary, 1.8mH Secondary, 1:5.6 ratio.

Any ideas to better stabilize the system would be helpful... Unfortunately Controls was not my forte in collage.

Image showing event I am talking about: Image showing event I am talking about

Steps I have Tried:

  • The controller uses current mode feedback as well as voltage, I overcompensated the current feedback to almost disable it per the data sheet, it had no effect. This is not sub harmonic anyways...
  • I have tested with multiple loads ranging from 3W to 15W, low loads exhibit this effect slightly more.
  • I tested this approximate design and transformer using a TL495 on a breadboard and it was successful and stable with no load, and no compensation network.

Schematic: Schematic

Bode Plots: Bode Plot

  • \$\begingroup\$ Does it work ok if you remove the link from the dotted secondary and connect that part of the winding to ground instead? Have you tried testing it on load? \$\endgroup\$
    – Andy aka
    Commented Dec 30, 2018 at 11:06
  • \$\begingroup\$ @Andy aka Yep I have done a pile of load tests to look at efficiency. Light loads seem to be a bit worse. The transformer connection you speak of is made with a plane, so it is not super easy to cut cleanly but I could try it. I also tried adding a pile of slope compensation to make it opperate only in voltage control mode instead of current mode and it did not fix the issue. \$\endgroup\$
    – MadHatter
    Commented Dec 30, 2018 at 14:57
  • \$\begingroup\$ I'm also just guessing that this is what unstable looks like? If it helps I tried this basic architecture and my transformer with a TL494 on a bread board and it worked ok unloaded and uncompensated. I did not give it loads to test. Which tells me it is something with the driver chip I am currently using, so it is probably the controller compensation? \$\endgroup\$
    – MadHatter
    Commented Dec 30, 2018 at 15:06
  • \$\begingroup\$ Maybe the controller is going into burst mode i.e. every 1/7 seconds it wakes up and pushes a few cycles through to the secondary then goes siltnt for a while and repeats. What does SW do? \$\endgroup\$
    – Andy aka
    Commented Dec 30, 2018 at 15:11
  • \$\begingroup\$ The gate drive waveform just has a constantly varying duty cycle from ~30-70% during these events. It does this under 3W loads and as high as 15W loads while the system is limiting the duty cycle to limit primary current thus dropping the output voltage. \$\endgroup\$
    – MadHatter
    Commented Dec 30, 2018 at 15:13

1 Answer 1


For anyone interested, the issue turned out to be layout related. It seems the feedback signal was such a high impedance, with the most sensitive part running near the transformer, noise was coupling into the system and being amplified.

There is an effective ~100x amplification of noise after the high voltage divider created by R9,10 & TM1 on the output.

The high voltage divider was made with lower value components and moved closer to the LM3478 IC, and the issue has disappeared as well as there is overall better stability on load shifts. At the cost of dissipating way more power.

enter image description here


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.