reducing effect of common-mode RF noise from DC-DC converter

[JBeale]

Curious if anyone can suggest a good resource for this? I have a low-level analog sensor system that is reading microvolts (uses opamp with +60 dB gain at 10 Hz - 1 kHz), and the whole system uses +5V at about 100 mA. It's at the end of a long ethernet cable, so I want to power it from a +48V PoE supply.  I'd rather not use a linear regulator which would waste about 5 watts to deliver 0.5 watts.   However my DC-DC converter causes a lot of switching noise around 1 MHz, which affects the dynamic range of the sensor. I can filter the differential-mode signal with multistage LC Pi filters, but in this case it doesn't seem to help much. I believe the problem is a common-mode signal on the power supply.  I've tried a common-mode choke plus bypass caps to case ground, but that has had almost no effect.

An enclosed scope plot shows the kind of thing I'm talking about. When the whole system is powered from a 12V battery, the output is below 20 mVpp of broadband noise, but powered from +48V through the DC-DC converter, the output is over 100 mVpp with clear resonant frequencies present.  The two traces show the output, with the common-mode choke on the power supply (lower trace) and without (upper trace).  Maybe I should just give up the idea of trying to use an efficient converter for this kind of thing.

[DaJMasta]

The basics are covered in the post above, but if you've got a bandwidth limit on the input circuitry, you can also put some low pass filtering on your input amplifier stages, not necessarily just passive filters before the amplifier stage - since your switching frequency is high and the sensor bandwidth is much lower, this can get rid of anything radiated from the switcher or anything picked up on the input lines as well.  A bead on the switcher input with a cap, the switcher module, the output cap (and I would go towards the upper end of the capacitive load the switcher is rated for), and then a secondary regulation stage to get your final voltage, with another reservoir cap is probably going to give you good rejection.  You can also use those same LC filters you're trying on the input lines on the main power rail after the switcher - it's common for sensitive devices to have a final ferrite bead or inductor with a cap right before the bypass for the chip power pins for that extra little bit of power supply noise rejection.

Also, if you're actually transmitting back in digital, you probably want to look into separate power supplies (a second LDO for the digital section) and ground planes for the digital transmission section and the analog amplification and digitizing section.  If you've got this sort of a mixed domain design, I always point towards ADC manufacturer app notes, because there are well documented ways of reducing noise and designing well available specifically regarding ADC layout.

[JBeale]

This is all good advice.   By the way, my circuit uses a 5V opamp. I have a single +5V rail but use the opamp with +/- 2.5 V supplies, by using a TI TLE2426 "rail splitter" to generate the virtual ground.  That part is handy, but it's not infinitely stiff around 1 MHz with only a 10 uF tantalum between output and ground.

I'm embarrassed to admit it, but I just discovered that most of my problem was from putting the scope probe ground on my circuit "virtual ground". I figured that was OK since my 5V power source was nominally floating wrt earth ground (although plenty of stray capacitive coupling). When I instead used my supply V- as the scope ground, the situation looked a whole lot better.

[BravoV]

Use a post-regulator after DC/DC. Consider a combination of a bead (for 100MHz+), an inductor (for MHz filtering), and a high PSRR LDO (TPS7A series, TLV705 series, ADP7112, etc.) for final filtering.

Complementing this post, on the LDO part, also consider LT3042 too as its capable of 79dB at 1 MHz for PSRR.

[jbb]

If the split rails keep giving you trouble, it may be possible to tweak the DC DC converter to spit out +-3.3V. The +3.3V could be filtered for the digital stuff, and LDOs used for the +-2.5V analog.

Texas Instruments call this the Fly Buck and have some app notes.