When in doubt, consult prior work. I found the power supply that I had lifted the PE common mode choke from. It was on the output of a 120V supply, from an old HP 54110D DSO. In fact, all of the rails on that board had CM chokes on the outputs, some followed by CLC filters. And I got to see how they were using the CMC'c.
And it turns out I was confusing myself when posting the attenuation curves for common and diff mode for the CMC's I had made. I ID'ed them backwards. Common mode connection originates from the windings on the same end of the toroid, not the opposite end as I was doing.
So the noise spectra in post #116 (and prior ones) is incorrect. Here is the correct one for the 22uH CMC. Compare it to post #116.
The switching freq peak is down to almost -95dBV. And the 3.5kHz peak is at -108.
Now we're getting some common mode rejection. More a little later.
And it turns out I was confusing myself when posting the attenuation curves for common and diff mode for the CMC's I had made. I ID'ed them backwards. Common mode connection originates from the windings on the same end of the toroid, not the opposite end as I was doing.
So the noise spectra in post #116 (and prior ones) is incorrect. Here is the correct one for the 22uH CMC. Compare it to post #116.
The switching freq peak is down to almost -95dBV. And the 3.5kHz peak is at -108.
Now we're getting some common mode rejection. More a little later.
After seeing the PE-62912 on the output of the HP supply, I decided to try it with the RPS30. HP used 56uF caps on both sides of the CMC, which seemed small to me. I first tried 68uF, and the results were about the same as my 22uH CMC, only noisier at low freqs. So I put 100uF on the RPS output side and 1500uF on the load side. And it looked pretty decent.
The 3.5kHz spike is down to -116dbV, and the main switching spike is down to -100.
Next I'm going to experiment with cap values on either side of the choke, and then take a look at the output impedance. But this is looking much better. Switching ground noise has been reduced by 16-20dB.
The 3.5kHz spike is down to -116dbV, and the main switching spike is down to -100.
Next I'm going to experiment with cap values on either side of the choke, and then take a look at the output impedance. But this is looking much better. Switching ground noise has been reduced by 16-20dB.
There wasn't much more to be had with the lytic on the RPS output. From 330uF and up the switching spike started increasing, but anything in the 47uF-220uF range with low ESR was good and measured pretty similar. A couple more dB reduction in the switching spike was gained by moving that cap to the RPS pcb. It's output cap is unpopulated. I added a 100uF low-ESR lytic across the 1500uF on the load side to counteract the bigger cap's ESL.
So here's what it looked like with 220uF on the output, PE-62912 CMC, and 1600uF on the load side.
The switching peak is down to -102dB. The 3.5kHz peak was a little higher today no matter what, I have no idea why. I also verified that the 88kHz spike is in the environment and being picked up by the probe lead.
So the next question is, what does the output impedance look like? I normally use an HP 4276A for quick looks at power supply impedance, but it was not at all happy with the variable noise at low freqs coming from the RPS. It wouldn't lock below 200Hz. So I used a network analyzer to sweep it.
It's still a noisy-looking curve down there, but it worked. Impedance is in the 200-300mΩ range up to 900Hz, then the large output cap dominates, until about 12kHz when it starts heading back up to near 100mΩ. On it's own, the output capacitors don't have this rise, so this is a consequence of using the CMC.
With a large Z and Zphase transition right in the upper midrange, I wouldn't want to directly power a piece of high-quality audio gear with this. (Though, the potential for "midrange magic" is quite high... 😊 ) But for my application, which uses the SMPS as a DC converter and pre-regulator for a linear regulator powering a low-power Class A amp , it will probably be OK.
This is good enough to give a listen to, so this weekend I'll compare it to a Power One linear PS as a pre-reg (which sounds excellent), using a very good LM3886-based amp for reference.
So here's what it looked like with 220uF on the output, PE-62912 CMC, and 1600uF on the load side.
The switching peak is down to -102dB. The 3.5kHz peak was a little higher today no matter what, I have no idea why. I also verified that the 88kHz spike is in the environment and being picked up by the probe lead.
So the next question is, what does the output impedance look like? I normally use an HP 4276A for quick looks at power supply impedance, but it was not at all happy with the variable noise at low freqs coming from the RPS. It wouldn't lock below 200Hz. So I used a network analyzer to sweep it.
It's still a noisy-looking curve down there, but it worked. Impedance is in the 200-300mΩ range up to 900Hz, then the large output cap dominates, until about 12kHz when it starts heading back up to near 100mΩ. On it's own, the output capacitors don't have this rise, so this is a consequence of using the CMC.
With a large Z and Zphase transition right in the upper midrange, I wouldn't want to directly power a piece of high-quality audio gear with this. (Though, the potential for "midrange magic" is quite high... 😊 ) But for my application, which uses the SMPS as a DC converter and pre-regulator for a linear regulator powering a low-power Class A amp , it will probably be OK.
This is good enough to give a listen to, so this weekend I'll compare it to a Power One linear PS as a pre-reg (which sounds excellent), using a very good LM3886-based amp for reference.
Last edited:
I don't like how it sounds. It's become a little thinner, more edgy, electronic-sounding. Less smooth, effortless, and solid than with the analog prereg. I doubled and then tripled the load-side capacitance (without measuring to see the effects) and it's little changed.
Is it terrible? No. But it is a step back quality-wise.
I'll stick with the analog prereg for this project.
If anyone else wants to further optimize using the CMC's to lower the noise and get the output impedance lower too, I'll watch watch interest. But I'm done wih this for now.
It's a pity, because it is so lightweight and runs nice and cool.
Is it terrible? No. But it is a step back quality-wise.
I'll stick with the analog prereg for this project.
If anyone else wants to further optimize using the CMC's to lower the noise and get the output impedance lower too, I'll watch watch interest. But I'm done wih this for now.
It's a pity, because it is so lightweight and runs nice and cool.
OK, one more kiss before we part. I had to see, what if anything has changed since increasing the load-side capacitance.
: The good: The output impedance has come down above ~300Hz and its bump at ~900Hz has been damped.
: The bad: The noise floor in the audio range has come up by 4-5dB, to -120dbV.
: More bad: The switching spike, which we worked so hard to get down to -102dB, is back up to -96. It may well have just come along for the ride with the rest of the noise increase.
Also weird is that the peak freq moved up to 67kHz.
The raised noise floor above 1kHz correlates well with the "thinness" I heard.
Using these CMC's to lower switching noise appears to be a delicate dance with many interactions and tradeoffs.
: The good: The output impedance has come down above ~300Hz and its bump at ~900Hz has been damped.
: The bad: The noise floor in the audio range has come up by 4-5dB, to -120dbV.
: More bad: The switching spike, which we worked so hard to get down to -102dB, is back up to -96. It may well have just come along for the ride with the rest of the noise increase.
Also weird is that the peak freq moved up to 67kHz.
The raised noise floor above 1kHz correlates well with the "thinness" I heard.
Using these CMC's to lower switching noise appears to be a delicate dance with many interactions and tradeoffs.
Last edited:
Would it be worth with experimenting with R-C values in series across the output, to provide some damping - not just more&distributed capacitance.
I suspect effects will be load-current-dependent - but do keep going, it is interesting to see the graphs so far!
I suspect effects will be load-current-dependent - but do keep going, it is interesting to see the graphs so far!
I don't have a schematic for the 54110D. I have an op/pgm manual, but it appears they never pubished a service manual for it. These things are true boat-anchors. I was very happy to get off of the last good one I had! The others have been a good source of parts, though.
@martin clark Methings a parallel RC in series with the RPS output would be counterproductive, since the output impedance is highest at low freqs as it is.
@martin clark Methings a parallel RC in series with the RPS output would be counterproductive, since the output impedance is highest at low freqs as it is.
Thanks @rayma it's a good explanation on EMC filtering for mains inputs. Does not translate directly to outputs, from what I can see. And in my case, the converter is floating, so chassis can't be the return path. I'll keep poking around and see if I can find somethign relevant.
@Mr Evil was right about the load-side capacitor needing to be large. But not the one on the SMPS side.
@Mr Evil was right about the load-side capacitor needing to be large. But not the one on the SMPS side.
It's simple enuf to sketch easily.
It is labeled "Primary Power" on the pcb. This supplies raw DC to local inverters and linear regs on each pcb. The 54110D is one of the noisiest DSO's I've ever seen; this may be why, in part at least.
There's also a 15k R in series with a zener across the 1st 56uF, zener value unknown.
Worth noting, the 56uF caps have three terminals, with grounded cases.
LOL. Looks like most SMPS's I've seen, from the pulse xfmr on out. Some have another LC on the output.
It's probably a flyback converter, in which case it can only be half-wave since the pulses coming out of the transformer are only in one polarity.
In terms of lowering CM noise, is there anything to grounding the cases of the 'lytics? In the RPS30, all of them are crammed right up against the pulse xfmr. I notice that the larger industrial switchers either have grounded-case lytics or space them well away from other parts.
I have spent 2-3 hours searching for details on using CM chokes as output filters, and finally found ONE reference to it, in a Maxim / Analog Devices paper on proper layout and component selection for EMI reduction:
https://www.analog.com/en/technical-articles/proper-layout-and-component-selection-controls-emi.html
Contrary to the title, no details are given on component selection, and the post-CMC filter caps shunt noise to earth ground, which doesn't exist in the RPS switchers.
https://www.analog.com/en/technical-articles/proper-layout-and-component-selection-controls-emi.html
Contrary to the title, no details are given on component selection, and the post-CMC filter caps shunt noise to earth ground, which doesn't exist in the RPS switchers.
I don't have any grounded-case lytics of appropriate size/quality, so that will have to wait for another time. I experimented more with the capacitors on either side of the CM choke, and didn't really learn anything new. The cap between the RPS30 and CMC is best in the 47-220uF range. The load-side lytic has to be large. I saw no benefit above 4700uF.
So for now this is as good as it gets. The 4700uF has 100uF lytic and 10nF ceramic across it to tame the big lytic's ESL.
(There's a typo in graph title, it should be 220uF, not 2200...)
Sound-quality-wise, this was a step closer to the analog prereg. To put it in context, if I hadn't compared it to the analog prereg, I would have considered this to be pretty good.
It bugs me that the larger CMC isn't working better than this at reducing the switching spike. With a large output cap, it should be 50dB better at 65kHz than with the 22uH CMC. So I put together a test jig to measure attenuation through the CMC's, using a power amp as a low-impedance source. With a large output cap, the 22uH CMC is behaving pretty close to spec. The PE-62912 CMC should be atttenuating 65kHz by nearly 90dB, but in-circuit it is only 6-8dB better than the 22uH CMC. So something else is going on. Perhaps a bifilar-wound CMC (less leakage) would do better?
So for now this is as good as it gets. The 4700uF has 100uF lytic and 10nF ceramic across it to tame the big lytic's ESL.
(There's a typo in graph title, it should be 220uF, not 2200...)
Sound-quality-wise, this was a step closer to the analog prereg. To put it in context, if I hadn't compared it to the analog prereg, I would have considered this to be pretty good.
It bugs me that the larger CMC isn't working better than this at reducing the switching spike. With a large output cap, it should be 50dB better at 65kHz than with the 22uH CMC. So I put together a test jig to measure attenuation through the CMC's, using a power amp as a low-impedance source. With a large output cap, the 22uH CMC is behaving pretty close to spec. The PE-62912 CMC should be atttenuating 65kHz by nearly 90dB, but in-circuit it is only 6-8dB better than the 22uH CMC. So something else is going on. Perhaps a bifilar-wound CMC (less leakage) would do better?