I plan on designing a variable power supply (1-50V, isolated), able to provide a power of about 50W. For efficiency's sake, linear regulation is out of the question. The natural thing would be to design a flyback AC-DC supply, but there are a number of problems raised:

  • The fact that it is variable complicates things: for example, the aux winding will not provide a fixed voltage. Also, will the flyback topology be appropriate for a large range of duty cycles ?
  • I did not find any off-the-shelf flyback transformer that would be appropriate. Either the output voltage is not right, or the power, or the aux voltage, or they are not in stock anywhere. So I'd certainly have to design my own transformer, but I don't feel confident about this part (neither regarding the core choice, the design, or the DIY construction of this).
  • Even the other parts of the flyback supply design seems risky: the snubber across the primary winding (which seems to consume a lot of power, by the way), the optocoupler feedback (which seems to require some tuning), etc ...

On the other hand, I feel confident about designing an appropriate DC-DC buck converter. There are a lot of tools from various manufacturers to help in designing them (TI, Linear Technology, ...), and it seems there are a lot less parameters that could lead to a non-functioning design.

So, I plan on using a simple toroidal transformer (much more easily available) to do the isolation and an initial stepping-down of the AC voltage, then the usual diode bridge + filtering capacitor, and then the appropriate DC-DC conversion step.

Apart from the fact that the transformer will be a lot bigger (and more expensive) than for a flyback design, are there some drawbacks of using this topology ? It seems nobody really does this, but I can't think of any reason that would make this a bad idea. The efficiency should still be pretty high, right ?

  • \$\begingroup\$ 1st the buck can't work in wide range, 2nd why someone would use a buck to have variable voltage supply, because if you need a bench PSU the switching isn't suitable, too noisy, go for linear. \$\endgroup\$ Commented May 23, 2016 at 9:23
  • 2
    \$\begingroup\$ I planned to make a pre-regulator with a buck, and then a linear regulation. The linear part simulates very well, but in the worst case, the main fet has to dissipate about 70W. That's way too much. There has to be some efficient pre-regulation. \$\endgroup\$
    – dim
    Commented May 23, 2016 at 9:30
  • \$\begingroup\$ You could get more efficiency with the toroidal transformer variation if you switched between taps on the secondary. \$\endgroup\$ Commented May 23, 2016 at 12:33
  • 1
    \$\begingroup\$ @AndrewMorton I know. I even designed a circuit to dynamically configure the secondaries either in series (18V@4A), parallel (36V@2A) or a kind of x3 configuration ([email protected]) to make the most out of the transformer. But I don't want to use multiple transformers, or an impossible to source multitap one, so in the worst case I still have 72W to dissipate. \$\endgroup\$
    – dim
    Commented May 23, 2016 at 17:59

2 Answers 2


Your design is good. It is OK to use 50 Hz isolating transformer and then buck DC-DC stage. However: you should take care of interference and isolation. This may be not so easy, if you do not have good enough PCB technology at hand.

You can not make your buck stage on one IC because of high voltage / power requirements. You can use ready-made buck controller (for example, LTC3810) and two small power MOSFETs rated for 80 V operation. LTC3810 can regulate its output in range from 0.8 V to 50 V. You have to select the proper feedback correction to run the switching regulator in such a huge output voltage range.

Additional benefits of this design are: built-in over-current protection and soft start.

The problem is: you need a good PCB technology to make this design work fine.

As for the switching noise - it is only a matter of proper filtering. I have designed a lot of very sensitive analog and RF circuits, all containing switching regulators. It is enough to filter the output by one or two stage LC filter with MLC capacitors and low quality factor inductors. I suggest to add one common mode choke on output of your DC-DC stage and another one - on its input.

The linear regulator after the switching one does not help against a switching noise, it is a myth. You can check specs for linear regulators for AC suppression: it is pretty low above 0.5 MHz. And we are talking about noise / interference from the switching frequency up to 100 MHz, at least. So cheap passive components (inductors, ferrite beads and MLCC) solve this problem.

The only drawback of using 50 Hz transformer compared to fly-back switching design is: the capacitance between Mains and your DC output is ~100 times larger with 50 Hz transformer. It may cause problems in RF applications.

Once again: your major problem in this design is the PCB technology (2 layers is absolute minimum, 4 is good) and proper PCB layout design.

  • \$\begingroup\$ Thanks. Actually, the purpose of the secondary linear regulator is more to hide the "approximations" of the initial dc-dc conversion phase (which will be hard to tune perfectly due to the wide output range, as you said) rather than smoothing the switching noise, you're right. I'll use some ferrite and caps anyway, but the switching noise is not my main concern. I don't design medical devices (hopefully). The PCB will be 2 layers. I'll try do lay it out carefully. \$\endgroup\$
    – dim
    Commented May 23, 2016 at 18:58
  • \$\begingroup\$ Good luck. Take care of thermal design: on-board power devices must have good path to dissipate the heat. \$\endgroup\$
    – Master
    Commented May 23, 2016 at 19:42

Off-line flyback design is certainly not the simplest power supply challenge. Whatever you do, you also need to control the inrush current and the resistors doing the limiting are subject to a LOT of power during that phase. So you pretty much need purpose-built through hole resistors. Adding to that you need to consider X and Y capacitors as well as a thrysistor to bypass the inrush current limiter after the precharging is done.

With regards to "not suitable voltage", flyback controller can certainly adjust the voltage to a different level than the "advertised" voltage, it may not just be quite as efficient. Aux winding voltage is indeed not regulated, that's why you need a linear regulator AND a voltage clamp for it. "Tuning" optocoupler feedback is no different from tuning any other SMPS feedback circuit. Certainly the equations are a bit different but the principles are the same.

You may run into trouble with the "1 volt" business, duty cycle aside having 50x variance in output voltage is likely to cause various funkyness. It might not be the worst idea to use the flyback to provide 12V and then a Sepic circuit to adjust that to the desired voltage which keeps the duty cycle somewhat sane.

  • \$\begingroup\$ An inrush limiter for a 50W design ? The transformer is not more than 100VA. Honestly, I doubt it is necessary. Regarding the funkiness resulting in the wide output range, I expect it. As long as the secondary linear regulator can hide it, it is fine. \$\endgroup\$
    – dim
    Commented May 23, 2016 at 18:49
  • \$\begingroup\$ @dim Yeah, limiter. 50W is already a fair amount of power and you will have a bulk capacitor at the primary capable of sinking north of 100A when you flick the switch. If you need special purpose resistor to withstand the precharging, it's not very healthy for the rectifier, common mode choke etc either. And your fuse is liable to trip. For one off hobbyist PSU it may not matter but for something used commercially it's a no-no. One infomercial about it bearpwr.com/pdfs/article_EP_0306_inrush.pdf \$\endgroup\$
    – Barleyman
    Commented May 24, 2016 at 1:27
  • \$\begingroup\$ An EDN article about another basic technique edn.com/design/analog/4349191/… 100A is likely being optimistic considering the bulk capacitor ESR can be pretty small. \$\endgroup\$
    – Barleyman
    Commented May 24, 2016 at 1:33

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.