Isolated DC-DC converter output filtering

[Harvs]

For my current project I needed an isolated supply for part of the circuit. So I got a couple of RECOM RBE-0505S 5V to 5V 200mA isolated DC to DC converters, same sort of thing in the Dave’s USB supply.

However I was concerned about noise for this application, and my investigation is the purpose of this post.  All scope grabs are done a 100% output load.
So at full load with no output filtering at all, the following was what I got:


Having a look at the datasheet, that’s easily within spec, but not good enough for my application.
So I added some output capacitance, in this case a Nichicon 330µF 130m? 35V electro that I had plenty of. Also they were brand new from Digikey, so I knew they were the real thing. This decrease the ripple down to about 1/5th what it was:


At this stage the amplitude is pretty reasonable for a switchmode supply, about 0.22% of output.  Also, what’s not evident from just the single scope grab is the ripple now gradually reduces as load decreases.  Previously it has some weird things going on, particularly around 50mA.  Further, the voltage regulation to load changes is much better as would be expected.
So next thing I did was to add a pair of ferrite beads in both the positive and negative lines, with 100nF caps either side.  This greatly reduced the high freq components.


At this point, I thought I had a quite reasonable outcome, and it meet my requirements.  However, I saw a damaged motherboard (written off due to physical damage in a move) in the corner that was loaded with polymer caps, so I pulled out the heat gun and stripped a few off it.


The ones I chose were 560µF 6.3V, and as you can see, they’re tiny! I set up my sig gen and scope to try and measure the ESR.  I couldn’t do it accurately with this setup, but it was around 10m?.  So I replace the electro with the polymer cap and removed the ferrite bead/ceramic cap filter. 


As expected the output ripple is pretty low but there’s some high freq components they look pretty nasty.  Adding the filter back in with the poly cap and this was the result…


Speaks for itself really, anything there is just noise in the measurement because I get the same result with the supply on or off.  Down side is the poly caps cost almost $1ea, but it’s pretty awesome. 

[ptricks]

You can remove just about all the noise with a pi filter.
If I am reading the scope output correctly, still early, looks like you could use 2x 470uf caps and 1x25uh inductor .
one cap parallel with output, then series the inductor, then parallel the second cap.

[Harvs]

You can remove just about all the noise with a pi filter.
If I am reading the scope output correctly, still early, looks like you could use 2x 470uf caps and 1x25uh inductor .
one cap parallel with output, then series the inductor, then parallel the second cap.

That's basically what I'm doing, only using ferrite beads instead of an inductor.  The dc-dc converter has some cap in it, which I've paralleled with a 100nF cer chip, then the ferrite bead in series, then another 100nF and the 560uF cap in parallel.  Ferrite beads are cheap and have quite low DC resistance compared to some small inductors.

[SeanB]

If you rip out the DC-DC converter ferrites in the motherboard you will find they will work a lot better.  low resistance windings too.

[Harvs]

If you rip out the DC-DC converter ferrites in the motherboard you will find they will work a lot better.  low resistance windings too.

I agree that they'll have much lower DC resistance, after all in a motherboard they're carrying high currents.  But in this case, with the converter only capable of 200mA, the ferrite beads with their 150mOhm DC resistance only cause a 30mV drop.  This converter is going to be followed by a 3.3V LDO anyway, so small losses don't cause any problem.

Given that with 2x100nF (0603) caps, 2xFerrite beads (0805) and the single polymer cap I have a noise level undetectable by any equipment I have, I don't see how they can work much better?   

[baoshi]

I have a 2w 0505 converter similar to Dave's made by Mornsun. The datasheet is here http://www.mornsun-power.com/UploadFiles/pdf/IB_S-2W_EN.pdf
In the datasheet they specifically mentioned ripple control. And they claim that the capacitance should not be too large.

[jeroen74]

Add a cap of ~1nF between the isolated grounds with a voltage rating equal or better then the isolation voltage of the converter. Ceramic caps on the outputs, close to the pins, work very well too.

[Harvs]

Hi Baoshi, that converter you've got is output line regulated, the one I'm using certainly isn't.  So there's some differences, and I'd say the one you're using probably has a lot better ripple performance to start with than my cheapskate version.

jeroen, what does the cap across the isolation barrier do?  RF suppression?

Cheers

[jeroen74]

Yes, you often see these caps in isolated topologies for EMI control. IIRC the official term is stitching capacitor.

[baoshi]

Hi Baoshi, that converter you've got is output line regulated, the one I'm using certainly isn't.  So there's some differences, and I'd say the one you're using probably has a lot better ripple performance to start with than my cheapskate version.

jeroen, what does the cap across the isolation barrier do?  RF suppression?

Cheers

Yup Harvs you're right. Mine is regulated output.
I just measured its output ripple, at 300mA it is only 12mVp-p. After one 60Ohm ferrite bead it drops to almost nothing.