Low common mode noise DC/DC converters?

[Martinn]

I am working (for fun mainly) on an isolated USB oscilloscope. As I aim for 0.5 mV/div or better sensitivity, I am aiming for low noise also from the power supply, which would be isolated DC/DC from USB 5 V (output 5V, better +-5 V, 1-2 W)
While trying out output filtering options, I fought lots of common mode noise as well. Dave Erickson (DIY-SMU) has an article on the problem: https://www.djerickson.com/dc-dc/

So I am wondering what my options are.

- finding an off-the-shelf unit with low CM noise: Recom just launched the K series (planar transformer 1 MHz) https://recom-power.com/en/products/dc-dc-converters/isolated-dc-dc-converters-k-series/k-series.html?1
with the RKK having 10 pF coupling capacitance only. Of course CM noise is never specified, but low coupling would hint to low CM injection as well, I suppose.

- DIY? Besides LT1533&such  https://www.analog.com/en/parametricsearch/11497#/p5362=Push-Pull, TI has some transformer drivers https://www.ti.com/isolation/power-signal-isolators/transformer-drivers/overview.html with the interesting UCC25800 (LLC resonant, 9 V min so no USB) or the newer SN6507 (adjustable rise times).
What transformer core types would be best? Low coupling/high leakage would favour resonant, which is the point of the UCC25800 as I understand. Not sure about push/pull forward with high leakage.

Honorable mention of course goes to the DMM7510 coax transformer, but I think this is excessive unless you want to build a 7.5 digit DMM. https://www.eevblog.com/forum/metrology/dmm7510-coax-transformer/
Again, this is about common mode noise, which is far more difficult to get rid off than output voltage noise.

Nice test setup (last page): https://calex.com/pdf/8common_noise.pdf

[Weston]

Do you just need DC isolation, or do you also need low coupling capacitance because you want to float the oscilloscope to probe floating nodes with a high dV/dt (like high-side gate drive in a SMPS)?

Any CM noise from the DC/DC converter can be addressed with filtering, but the filtering and coupling capacitance also impacts the rejection of external noise, which may be a bigger issue than the noise generated by the internal DC/DC converter.

Common noise is going to couple into the oscilloscope measurements through voltage drop on the oscilloscope leads. The ground and signal connection have different impedances which leads to common mode to differential mode conversion.

Any common mode noise generated by the isolated DC/DC can be shunted with a sufficient capacitor across the isolation barrier and adding common mode chokes on the input and output. However, the larger this capacitor gets the lower the impedance for CM noise between the USB and the oscilloscope circuit connections (noise can be induced in either direction). USB (or what you are probing) can be quite noisy, so its possible that these sources will be a bigger noise contribution than the internal DC/DC converter.

I was working on a USB powered rogowski coil recently and used one of those cheap little 2W isolated DC/DC modules which seem to have ~20pF of coupling capacitance. With a cheap USB power supply separate from the scope a few mV of spikes were induced on the input to the oscilloscope due to common mode noise from the USB adapter. When the probe was powered from the scope there was very little noise as there was no CM noise between the USB power input and BNC output.

https://www.digikey.com/en/products/detail/mornsun-america-llc/A0509S-2WR3/16348392

Just following the manufacture recommendations for the filtering of the DC/DC and a small SMD common mode choke was enough to push the noise from the DC/DC itself down to the noise floor.


I got adequate performance for my application with the cheap DC/DC module. If you want lower noise the bigger issue might be rejection of external CM noise rather than CM noise generated by the DC/DC itself. A low coupling capacitance would help with that.

It also might somewhat become a matter of how big a common mode choke you are willing to use.

[jbb]

“To a man with a nail, everything looks like a hammer.”

That UCC25800 (LLC resonant) looks like a really cool hammer. There is a TI ref design with multiple isolated outputs from one driver chip, which is nice.

The ref design also includes a companion Wurth transformer (see) with very low capacitive coupling (<1pF I think).  In addition, an LLC driver should have fairly gentle switching edges. So I’m principle this could be a strong contender.

But you would need a boost converter to feed it from USB.

I would note that the incoming USB supply has quite a bit of variation, like from 4.25V - 5.25V. So that a boost converter would also be a good pre-regulator, which is nice.

[David Hess]

The low noise DC to DC converters I am most familiar with in multimeters and oscilloscopes all use small high frequency inverters instead of switching regulators.

I would start by discarding designs which do not provide balanced drive to a transformer; reducing common mode swing at the primary will reduce it at the secondary.

[floobydust]

With off the shelf DC-DC converter moduels, they are terrible for CM noise and I'll add a CM choke to their output and if necessary a Y-cap.

[Martinn]

With off the shelf DC-DC converter moduels, they are terrible for CM noise and I'll add a CM choke to their output and if necessary a Y-cap.

Not sure about the significance of a Y capacitor. The whole point of isolating a measurement front end is... to isolate it! A Y capacitor bypasses this isolation; starting from a 10 pF (like gate drive) DC/DC converter adding a 10 nF Y cap will increase coupling 1000x. This will increase the AC current you see on ground line of the mesurement (isolated) side. Of course if it's about DC only any Y capacitance will be fine.

I would start by discarding designs which do not provide balanced drive to a transformer; reducing common mode swing at the primary will reduce it at the secondary.

Fully agree, not sure how much hard-switching push-pull or more sinusoidal drive (like in the UCC25800 (LLC resonant)) is important. Probably as the LT1333 is hard-switching (strictly rise-time controlled, so more like trapezoidal) this can't be too bad.

Unfortunately the UCC25800 won't run from 5 V USB (9 V min) and

But you would need a boost converter to feed it from USB.

given the huge peaks in the boost currents I don't particularly like the idea.

Just found another thread where this has been discussed: https://www.eevblog.com/forum/metrology/power-supply-for-voltage-references/

[floobydust]

Your problem is the DC-DC transformer winding capacitance between sides. Measure it and note datasheets do not specify it.
So you will get common-mode fundamental switching freq. of say 150kHz, 300kHz, 1MHz etc. and harmonic spurs at the output of the DC-DC converter. This is normal.

[Martinn]

With some Recom converters isolation capacitance is specified. For example their new K series, this one https://recom-power.com/en/products/dc-dc-converters/rec-s-RKK.html?1 has 10 pF.
Also their gate driver isolators https://recom-power.com/en/applications/gate-drivers/gate-drivers.html?2 have 10 pF specified isolation capacitance.

[Kleinstein]

For the CM background the capacitance is one factor and the other is the effective voltage, that depends on the symmetry / shilding and also the voltage. Here it can help to start from a relatively low voltage. So a converter from mains voltage has a much more difficult start than starting from 5 V or so.  A high frequency can help in 2 ways: it makes the CM mode choice more effective and the a higher frequency the transformer can be smaller size, which from the general trend gives less capacitance (there were times when they measured capacitance in cm ).

One sometimes finds data on the capacitance and one can measure it relatively easy. What is usually not found is how much of the voltage is effective acting on that capacitance, so how good the symmetry is - this also depends on the drive type - full wave bridge drive may be more effective than half bridge or push pull with a split winding.


A CM choke alone has limited effect - it mainly works together with some y capacitance to ground or a shield / guard.

[floobydust]

Thanks I did not notice the RKK datasheet spec. of 10pF

What will you do with 10pF @1MHz plus harmonics coupled to your secondary? The 1W modules are flyback converters, so it is an extra ugly pulse.

Again, I add a CM choke to the module output, which greatly reduces the interference, but there is still some.
Many isolated instrumentation products use Y-caps to contain RFI/EMI like 1,000pF+47Ω to PE. It's not a problem to do this, it swamps the (transformer) coupling capacitance and limits EMI problems by giving it a path to ground. You can put provisions for this on your PCB but unpopulated... unless you find it useful.
Otherwise you need a DC-DC converter with an electrostatic shielded transformer, which is a specialty item.

[spostma]

DC-DC converters with higher voltage rating tend to have a lower capacitive coupling.

Also ones that have reinforced insulation tend to have lower capacitive coupling (search for IEC60601)

For example, Mornsun G_S-1WR2 & H_S-1WR2 Series have 5pF typical between input and output.

[spostma]

... and Murata NXJ2 Series has just 2.0 - 2.5pF coupling!

[Martinn]

With off the shelf DC-DC converter moduels, they are terrible for CM noise and I'll add a CM choke to their output and if necessary a Y-cap.

Some update here. As suggested by floobydust and also Recom https://recom-power.com/en/rec-n-very-low-noise-filter-for-isolated-dc!sdc-converters-46.html?1 I tried CM chokes on input and output. For reliably measuring CM noise I made a PCB, which measures the ground voltage over a 50 ohms shunt (into terminated coax, center connector).
Adding two Würth Elektronik SL-1 330 uH chokes at input and output together with four 22 uF MLCCs (before/after each choke) got rid of most CM noise to a degree where measurements were dominated by nearby radio transmissions in the 140-220 MHz region.
Finding the right chokes needs some experimentation, I got the small CM choke kit from Würth, which helped a lot.
After this, I added a 470 uH/1000 uF output filter, which pushed the output noise below the input noise limit (2 nV/rtHz, 9 uV RMS at 20 MHz BW) of my LNA (I'd say noise is easily < 1 uV RMS).
Remembering Jim Williams' idea of a trigger probe, I triggered the scope from the primary side (where there was enough noise) and set scope to average, revealing a remaining 17 uV p-p switching noise (20 MHz BW).

Bottom line if you need now noise (both CM and output), it looks like it can be had even with commercial DC/DC modules. The RFM-0505S was not too bad to start with, definitely buy a variety of different models and do measurements. Finding the right CM choke was trial+error for me, optimizing the output filter of course depends on your requirements.

[jonpaul]

Use shielded low capacity transformer.

Shield to pri side rtn

Jon

[tszaboo]

Bottom line if you need now noise (both CM and output), it looks like it can be had even with commercial DC/DC modules. The RFM-0505S was not too bad to start with, definitely buy a variety of different models and do measurements. Finding the right CM choke was trial+error for me, optimizing the output filter of course depends on your requirements.

You should test that converter under various loads. They are often driven into saturation, where the noise figures are very different than under light load. The transformer in those is tiny.

[branadic]

I'd like to draw your attention to

https://xdevs.com/article/vref_psu_p1/
https://www.eevblog.com/forum/metrology/power-supply-for-voltage-references/

Maybe that gives you an idea or two.

-branadic-

[Martinn]

I'd like to draw your attention to

-branadic-

Pickering, yes, have seen this at some time. My first thought when I saw that construction was to use conductive ESD filament and 3D print the shields. It seems you came to the same conclusion!
Another thought would be the DMM7510 coax transformer https://www.eevblog.com/forum/metrology/dmm7510-coax-transformer/

Did you measure common mode injection using your setup - and under what conditions? I posted an image of my test setup, which uses a 50 ohms shunt and a 1 nF Y capacitor. Both values are somewhat arbitrary, but I kept them constant for my setups to have some fixed reference.
I am particularly not sure about the "Y-cap". In power supply design, this is the primary measure to keep conducted emissions down, but for an isolated measurement frontend a Y capacitor would effectively cause error currents when probing a circuit that has a floating (wrt earth) ground. I quickly measured the capacitance between the input LO terminal and the front panel USB shield of my DMM6500, which came out at 230 pF. Maybe with an 7.5 or 8.5 digit DMM one has to be more careful about frontend to Earth capacitance.

All said for now I am looking for an isolation for a USB scope, +5 V USB in, +-5 V out, 1 W should be sufficient. So more in the order of sugarcube size; tiny compared to the pickering setup. But thanks for sharing - very interesting indeed!

- Martin