Here's a 24uH CMC (made fom an 82uH toroid choke) with a low-ESR 680uF after it. Peaking is minimal, and attenuation at the 65kHz switching spike is -44dB.
Looks good. So let's hook it up to the RPS30-15 and see if that is true. If it is, the switching spike should be down around -96dB with it.
And it's not. It is still up around -84dB just like it is with normal (non-common mode) filters. These isn't ANY common mode cancellation going on.
So far the only CM configuration that has reduced the switching noise on ground is the 3.1uH CMC with 1000uF cap.
Looks good. So let's hook it up to the RPS30-15 and see if that is true. If it is, the switching spike should be down around -96dB with it.
And it's not. It is still up around -84dB just like it is with normal (non-common mode) filters. These isn't ANY common mode cancellation going on.
So far the only CM configuration that has reduced the switching noise on ground is the 3.1uH CMC with 1000uF cap.
Last edited:
Not mounted, it's just the fairly-chunky choke with very short (1") leads, sitting on the bench with the capacitor aznd resistive load soldered to the output leads.
Perhaps an alternative approach for DM attenuation, that may benefit CM as well, is to identify a poly cap with an impedance minima at the nominal switching frequency. Granted most cap datasheets indicate a poly cap would need to be a physically large part, with a large capacitance to get its Z minima down towards 100kHz. The lowest datasheet curve I quickly identified was about 350kHz for a 6u8 Wima MKP4, but their product range does go up to 68uF for 100Vdc.
Does that mean you firstly identified a Z minima at some frequency (at or close to the smps noise signal) using say an RC filter and sweeping the frequency range ?
I used a network analyzer to find a film cap with impedance minima @ around 65kHz. In-circuit, it did nothing to stop the switching freq currents from contaminating ground.
Correct.
Only live and neutral input connections on the RPS. It didn't make any difference when I was working with an XP Power unit.
Would you be ok to post a photo of the DUT and connections to the spectrum analyser that makes the plots ? Is the analyser the only other connected device powered from the mains ac ? What noise floor do you get if the active to the RPS is disconnected (but the neutral remains connected - or is it designed to have both connections switched)?
@trobbins I see what you're getting at. There are no such noise sources coming from the mains or other equipment connected to it. I have three separate mains circuits at the bench, the computer is on one of them by itself (shared only with the room lighting.) I have a large iso xfmr I can use if needed but it rarely is. You can see the noise floor for this setup in the measurements, with 16 averages it's down around -124dBV. There's nothing wrong with this test setup, been using it for years. It's high-impedance circuits that occasionally present challenges, not low-Z ones like this.
I haven't, the live and neutral connections are marked on the RPS.
Something I want to look into later is the how well balanced the impedance of both sides of my homebrew CM chokes are. I noticed the PE-62912 data sheet specifies a 1% max imbalance. That's pretty tight for a cored inductor.
Something I want to look into later is the how well balanced the impedance of both sides of my homebrew CM chokes are. I noticed the PE-62912 data sheet specifies a 1% max imbalance. That's pretty tight for a cored inductor.
Last edited:
You're thinking about it the wrong way. The inductance is only high for currents that are common-mode, which is to say going in the same direction in both windings, which is not how current being supplied to the load normally flows. The inductance is low in differential mode, which is what the load sees. So, the output impedance is nowhere near as high as you would think from measuring the inductance of only one winding.
Even if the output impedance really was too high, just use a bigger capacitor. If the bigger capacitor reduces the effectiveness of the filter because it has too much ESL, then use multiple capacitors in parallel.
Are you absolutely sure it's common mode noise? Try turning the power supply off and measuring the noise again. You can also try shorting an oscilloscope probe (just clip the ground lead to the tip so the ground wire forms a loop), and waving it around the area to see what radiated noise it picks up.
I checked the inductance of both windings on the CMC's I've used thus far. The 3.1uH was the best matched, but the 24uH choke was the worst. So I tweaked turns and windings dress to get it within 0.1uH at 22uH. The result is a 4dB improvement. The switching noise peak came down from -84.7 to -88.7.
So balance matters. But the difference is not huge.
So balance matters. But the difference is not huge.
What load are you using? Here are a couple of articles, one mentions impedance matching.
https://www.murata.com/en-us/products/emc/emifil/products-search/selectionguide/highspeed
https://www.murata.com/en-us/products/emc/emifil/products-search/selectionguide/highspeed
Here's the noise with the RPS off, everything else unchanged. I left the marker where the switching freq spike was. Other than the -103dBV spike at 88kHz, it's quite clean.
Last edited:
Three 10W resistors in parallel to get 15 Ohms.
The impedance matching they're talking about is maintaining the integrity of high-speed digital signals.