1
\$\begingroup\$

Maybe this question is perhaps too broad for the site: however, after looking on Google, searching the literature and asked to a friend who's a specialist in the design of SMPS without getting any useful answer, I decided ask it here, apologizing in advance if it is "Off-topic".


Question: do there exists switching power supply topologies which do not rely on the standard PWM paradigm but work, for example, by producing a variable, voltage controlled, frequency sine oscillation? And in case the answer is yes, what are their most used structures?

A little bit of background

  • I got interested in the topic after seeing the timeline of the switching power conversion as given in the wikipedia: it spans at least 50 years, so it seems plausible that a great deal of different ideas have been developed. So why focusing on designing fixed frequency PWM modulators/demodulators?
  • In general manuals and application notes and about switching power supplies, it is occasionally said that "the frequency is kept fixed in order to improve the EMC compliance". However, it seems more that this is a design trend and not a design requirement imposed by deep physical limitation, as it is seen for example in [1], so why always choose the PWM way of design?

References

[1] Mikko Kuisma, "Variable Frequency Switching in Power Supply EMI-Control: An Overview" January 2004IEEE Aerospace and Electronic Systems Magazine 18(12):18 - 22, DOI: 10.1109/MAES.2003.1259021.

\$\endgroup\$
10
  • 3
    \$\begingroup\$ Look up hysteresis control. The problem with variable frequency is that it is much harder to filter the output ripple. \$\endgroup\$ – user110971 Sep 5 '20 at 19:24
  • 3
    \$\begingroup\$ Related is constant on-time control. \$\endgroup\$ – DKNguyen Sep 5 '20 at 19:32
  • 2
    \$\begingroup\$ @DanieleTampieri with variable frequency you can have low as well as high frequency switching. The frequency is also dependent on the transients. So it depends what the load is. Digital loads tend to draw current on the clock edge while analog loads draw continuous currents most of the time. Hence you are not sure where on the filter curve you are going to end up. They have their uses though. Bang-bang control has faster settling times and can have lower ripple, if you know what the load is. However, since most SMPS are sold as separate units, this is not applicable in general. \$\endgroup\$ – user110971 Sep 5 '20 at 19:50
  • 2
    \$\begingroup\$ Self-oscillating topologies are also worth looking into. \$\endgroup\$ – a concerned citizen Sep 5 '20 at 19:54
  • 2
    \$\begingroup\$ Absolutely theres such a thing. Its called PFM pulse frequency modulation. techweb.rohm.com/knowledge/dcdc/dcdc_sr/dcdc_sr01/897 \$\endgroup\$ – Kyle B Sep 5 '20 at 21:14
3
\$\begingroup\$

To regulate to a certain voltage with DC-DC switching, you generally vary the duty cycle of the driver waveform. There’s several techniques to do that, including these two well-known methods:

  • fixed frequency, variable duty cycle (PWM)
  • fixed on-time, variable off-time (sometimes called COT, constant on-time)

In reality regulators may use more than one method depending on the load condition. And between each type, there are trade-offs for ripple vs. efficiency vs. transient response.

There’s also finer details about the type of control loop (e.g., current mode, voltage mode, etc.) that influences the regulator choice.

Lately, I’m tending to use COT for high-current supplies with fast transients, since holding the voltage above a minimum is more important than accuracy (logic core supply.)

For more noise-sensitive supplies fixed frequency PWM is preferred which, while it responds more slowly to transients, will have less (or more predictable) ripple.

MORE: (shout out to John D for the comment) There’s a newer, more-efficient topology that in fact uses frequency: LLC resonance. In broad strokes, this type adds a tuned L-C circuit inline with the switch. The power delivered to the load is adjusted by varying the PWM frequency to above or below the L-C resonance point to create a regulator loop.

LLC resonance supplies are more complex than PWM but have the advantage of lower switching losses. More here: https://www.electronicdesign.com/power-management/article/21805811/llc-resonant-converters-raise-the-powerefficiency-bar

\$\endgroup\$
2
  • 1
    \$\begingroup\$ There are also resonant topologies like LLC which use the resonant curve of a tank circuit for regulation. \$\endgroup\$ – John D Sep 5 '20 at 21:03
  • 1
    \$\begingroup\$ Very clever these LLC resonance supplies. \$\endgroup\$ – Fredled Sep 5 '20 at 23:15
1
\$\begingroup\$

exists switching power supply topologies which do not rely on the standard PWM paradigm?

Yes.

Pulse density modulation is a thing, and that's basically frequency-modulation.

Also, there's "spread-spectrum" SMPSes, where the clocking is intentionally non-uniform to spread the switching noise in frequency domain instead of producing sharp harmonics.

Also, many low-power SMPS controllers can switch between a PWM and a pulse-density mode in least-load situations.

In general manuals and application notes and about switching power supplies, it is occasionally said that "the frequency is kept fixed in order to improve the EMC compliance".

So, if you want to keep the switching noise constrained to discrete frequencies, you'll make damn sure to keep the frequency constant. I mean, that's the math of what "spectrum" is.

However, if a strong periodicity interferes with your application (and it often does), then you'd want to avoid that. That's why things like the PCIe bus with the option for spread-spectrum clocking and the aformentioned spread-spectrum SMPSes exist.

So, that statement is to be taken with a grain of "... depending on what you consider to be bad noise for your application".

However, it seems more that this is a design trend and not a design requirement imposed by deep physical limitation,

What? That doesn't make sense. The math is pretty clear: you do something periodically, you get that periodicity as component in your spectrum. That's not a "trend". That is the underlying math and physics in their purest form.

\$\endgroup\$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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