0
\$\begingroup\$

I am feeding the output of an isolated ACDC converter (this one) into a DCDC converter (that one), and the ripple out of the ACDC supply jumps by a factor of 5 when the DCDC converter is connected - which apparently is nearly entirely passed through to the output of the DCDC.

Ripple out of the ACDC without any load connected but a 10uF output cap: 140mVpp enter image description here

Ripple out of the DCDC supplied from a benchtop power supply: 30mVpp enter image description here

Ripples when the ACDC supplies the DCDC:

  • 650mVpp on the output of the ACDC (with or without 10uF cap)
  • 550mVpp on the output of the DCDC enter image description here enter image description here

(blue is ACDC's output and DCDC's input, yellow is DCDC's output)

The fact that both increase when they are connected together means to me that they are interacting.

Both oscillations seem to be at the same frequency as the DCDC switching (about 360kHz), so my current theory is that as the DCDC downstream switches the rail ON, which discharges the ACDC's output cap voltage, the ACDC's controller starts to compensate for that and passes some current through - but the DCDC controller downstream switches off and the rail voltage starts to increase before the ACDC acts on it at the next loop cycle.

It's the only explanation I have for the fact that both waveforms are so in-sync and close in amplitude whereas I'd normally expect the DCDC to show on the waveform clearer signs of attempting (in vain, by limited bandwidth) to regulate out some of the ripple.

If it is true, heavy filtering could improve this. To calculate what's necessary though, I need to confirm - or at least increase confidence - that this is what's happening. What's your take? Any other ideas?

Note: When connected together, 10x4.7uF are already present at the input of the DCDC. I'd be surprised if this is too high a capacitance for the stability of the ACDC.

\$\endgroup\$
8
  • \$\begingroup\$ It is normal that a regulator, even a linear regulator, cannot pertubations from passing either way between it's inputs and outputs. The control loop bandwidth just isn't high enough to respond to high frequency transients. Converters can interact with each other via beat frequencies. \$\endgroup\$
    – DKNguyen
    Commented Sep 28, 2022 at 16:04
  • 1
    \$\begingroup\$ The first question is, what is your measurement setup? Will the measurements be wrong because ground loops or simply magetic pulses coupling to scope probe? Standard ripple measurements tend to be done with a 10uF cap maybe and 20 MHz bandwidth limit enabled on scope. \$\endgroup\$
    – Justme
    Commented Sep 28, 2022 at 16:09
  • \$\begingroup\$ Noise no, because the bandwidth of the controller required would have to be extremely high, but these oscillations are 350kHz. It seems like conducted EMC from switching \$\endgroup\$ Commented Sep 28, 2022 at 16:24
  • \$\begingroup\$ More and better capacitors? \$\endgroup\$
    – winny
    Commented Sep 28, 2022 at 16:26
  • \$\begingroup\$ @Justme: since the frequency of these oscillations are 360kHz I doubt radiated emissions would cause that, and unless I'm mistaken chances of a ground loop causing that are low since the considered PSU seems isolated (except for 40nF capacitive coupling to Earth and 4nF coupling to neutral). I'm already limiting the bandwidth to 20MHz \$\endgroup\$ Commented Sep 28, 2022 at 16:29

1 Answer 1

2
\$\begingroup\$

If a downstream DC/DC converter can 'see' sufficient ripple from an upstream power supply, yes it can affect it's control loop. Find which frequencies are becoming a problem and use an filtering to knock out those frequencies.

You can do this in a number of ways. LC filtering is better then RC because the power isn't being dissipated in a resistor, you then have to deal with ringing or other effects from an LC filter. Another way is to use a ferrite filter, but whatever you do the ripple should be reduced to the input of the DC DC converter.

I'd start by looking at simply adding capacitance and then trying other filters.

\$\endgroup\$
5
  • \$\begingroup\$ Thanks for confirming. I've added 10uF (because that's what I had at hand) between the two units and it didn't change either signal at all. Do you think it's still worth designing a CLC filter centred around the DCDC's frequency? \$\endgroup\$ Commented Sep 28, 2022 at 17:42
  • 1
    \$\begingroup\$ I would try that, I've used LC filters on the outputs of DC DCs before with great success to reduce ripple. If the AC/DC's loop is in the kHz to MHz range then you might have to use a big LC filter with something in the 10uH range for the L. If you are in the MHz range you might be able to just use a ferrite filter that has a high impedance in the frequency of interest. \$\endgroup\$
    – Voltage Spike
    Commented Sep 28, 2022 at 17:53
  • \$\begingroup\$ Just looking at the datasheet of that TDK converter, the typical application circuit shows a 100 uF input capacitor and 0.1+22 uF output capacitors. So that 10 uF seems far too small. Also, don't forget that cables have their own inductance. \$\endgroup\$
    – jaskij
    Commented Sep 28, 2022 at 18:33
  • \$\begingroup\$ @jaskij Thats 10uH for the L \$\endgroup\$
    – Voltage Spike
    Commented Sep 28, 2022 at 18:54
  • \$\begingroup\$ I currently have 10x4.7uF as the input cap of the DCDC, the 10uF is on top of that right across the ACDC PSU to see if that changes anything to decouple right at the ACDC PSU. I wonder if half of the recommended cap really makes that much of a difference compared to the full - some manufacturers just slap a nice round number without much need for this high a value. \$\endgroup\$ Commented Sep 28, 2022 at 20:21

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.