Conducted emissions issue with multiple Mean Well AC/DC converters

[prasimix]

I will definitely try this when I finish building LISN and get a spectrum analyzer.

[jkostb]

Besides splitting the caps i.e. making a pi filter I would also advise to look at you input bypass caps of preregulator module. You use 3 ceramic caps of 4.7uF. Please note that the capacity of ceramic caps drop under DC voltage. You can easily loose > 50% depending on package size. For this reason you see combination of ceramic caps and polyester caps. Polyester caps don't suffer from capacity drop under DC voltage but have higher ESR, but much lower than an electrolytical cap. As you probably know a Buck converter draws a discontinuous input current. The input ripple current is for this reason supplied from the input bypass caps. Insufficient input bypass caps or using wrong type can increase differential mode noise currents and can be a root cause for conducted emission fail! I think that the peaks at 220Khz are really caused by differential moide noise currents. You can easily do the math. Common mode noise currents are usually caused by parasitic capacitance e.g. parasitic capacitance between heat sink and metal housing. But these parasitic capacitances are usually in order of 100pF. The impedance of 100pF at 220Khz is 7.2Kohm, which is too high for common mode noise currents. So my advise is try also to use a polyester cap at input when experimenting and look whether this helps.

[trobbins]

If you are using the DPC405 power module, then a quick view of the schematic for the pre-reg shows a buck regulator with I suggest extremely low bulk input supply capacitance.  You may want to clarify for yourself how much input supply ripple is being generated (I couldn't easily identify that from the github info, and note that ripple and noise of the output seemed to be the only priority being designed for), and look closely at how that could be modified.  One thought is that you have chosen a poor product to integrate for your application.

[prasimix]

Yes, ripple and noise of the output was not just the only priority but the only thing I thought was important. When I finally managed to get to the lab I see that taking care of the input ripple is also important, something I’ve only read about before. So I need to get a spectrum analyzer as soon as possible and build a LISN that I can proceed with real corrections and not just speculation.

[prasimix]

I return to the topic. In the meantime I got the Siglent SVA1015X and built LISN thanks to @Jay_Diddy_B who gave me the schematic and referred me to its design discussed here. LISN has 10 dB attenuator on its outputs. I used a ready made extruded box for which I made a front and rear panel to make construction easier. This is actually the second and wrong revision of the PCB which is 10mm shorter because I misinterpreted the dimensions of the box. If these 10 mm are added, a filter up to 6 A as well as a PCB AC power switch can be installed. Maybe I'll do it once in the future. In this case I had to shorten the box by 10 mm







EDIT: Please note that LISN shown above is not open source.

[prasimix]

The first thing I tried was to get measurements comparable to those from the lab where the Keysight EMI receiver and other LISN are used. I can’t say I’ve made it, but what I’ve achieved is to see the difference for different configurations. For the beginning I lowered the limit lines by 10 dB because like I said LISN has attenuator at the output.

I also discovered too late that I hadn’t collected enough measurements while I was in the lab, as well as that some had insufficiently clear names so I’m no longer sure what they referred to.

I decided to start with the measurements from scratch by adding component by component. On Siglent I have the EMI filter turned on, used Quasi Peak detection with dwell time at 108.1 us (because I see it was set that way on Keysight), RBW is 9 kHz.

Measurement 1: AUX-PS module with two Mean Well IRM converters, no load (no MCU module or Mean Well LRS connected):



Measurement 2: AUX-PS module with two Mean Well IRM converters, no load (no MCU module) with 2 x Mean Well LRS connected without load (i.e. no power modules):



This combination already seems to be causing problems. If I wrote an accurate comment on the measurement in the lab then it matches, i.e. my first problem is that the combination of the two types of Mean Well converter generates the problem. Lab measurement is shown in post #16.

Measurement #3: AUX-PS module with two Mean Well IRM converters, no load (no MCU module) with 2 x Mean Well LRS connected with resistive load. Situation is not better:

[prasimix]

Measurement #4: AUX-PS module with two Mean Well IRM converters without 2 x Mean Well LRS but with MCU module connected:



Measurement #5: AUX-PS module with two Mean Well IRM converters without 2 x Mean Well LRS with MCU module connected this time with added filter on the AC input (Xin=330n, 2x5 mH, Xout=100n, Yout=2x4n7):



Measurement #6: AUX-PS module with two Mean Well IRM converters with MCU module connected with added filter on the AC input (Xin=330n, 2x5 mH, Xout=100n, Yout=2x4n7) and 2 x Mean Well LRS without load. I'm not sure if that is good enough. It is below quasi-peak limit (blue line), but still above average limit (cyan):

[prasimix]

It seems that real difference can be measured only when filter is added on the AC input, i.e. before filters built in Mean well LRS converters. More measurements follows.

Measurement #7: AUX-PS module with two Mean Well IRM converters, with MCU module connected, with 2 x Mean Well LRS, added filter on the AC input (Xin=330n, 2x5 mH, Xout=100n, Yout=2x4n7), DCP405 without load connected:



Measurement #8: same as above but with DCP405, Iout=max (5 A):



I currently only have a common mode filter with twice the inductance at hand and with it the situation looks better.

Measurement #9: AUX-PS module with two Mean Well IRM converters, with MCU module connected, with 2 x Mean Well LRS connected, added filter on the AC input (2x10 mH, Xout=100n, Yout=2x2n2), DCP405 without load connected:



Measurement #10: same as above but with DCP405, Iout=max (5 A):



I plan to get and test a few more common mode filters with an inductance of 15 to 20 mH before going to the lab. I believe that should help.

[poorchava]

I recall a similar situation with RSP-2000 power supply (1200W, 12V). It had perfect results with ohmic load, and then all hell was breaking loose when we loaded it with basically a huge 8-phase step down converter (which was well engineered and was generating relatively little EMI when powered from a battery). The noise source was impossible to track down with NF probes.

In the end what worked was the exactly right magical combination of Y caps between different grounds in the system and a custom CM choke that passed all 3 mains lines (L,N,PE).

What we suspect was happening was that noise was being radiated from the internals of the supply and coupled to some other traces/planes that acted as antennas. It took several months to squish emissions enough that it passed the industrial limits.

[prasimix]

Relatively speaking I made a real progress some time ago after experimenting with different additional filter connected at the AC input. Once again this is situation with default setup without filter:



... and now with filter:



Now I have to confirm this results in the lab.

[prasimix]

It seems to me that I will not be able to continue testing with the selected filter for the simple reason that the CM choke overheats too much. I overlooked that it was rated for just 1 A
The CM choke in question is WE 750314978 and works almost magically, nothing like it in that inductance range (up to 20 mH) filters so well. I don’t know what the trick is, maybe in its high leakage inductance? This is its specification and how it is connected to get results from previous post:

[jkostb]

Wurth does has common mode chokes for higher currents. Have you looked at WE-CMBNC series. For example  7448030333 2.5A rated current. You can find a diagram with insertion loss. Frequency range matches with your measurements above. I don't know whether damping is sufficient for your application

https://www.we-online.de/katalog/en/WE-CMBNC?#744803033

[TimNJ]

The Wurth choke sems to have a pretty typical leakage inductance for a common-mode choke. 300uH (min) /10mH (typ) = 3% leakage inductance ratio...I think most CMCs are in the range of ~5%, though it can vary considerably with bifilar vs sectional wound. In this case, it's clearly sectional wound with large separation between the windings, so very well may have higher leakage inductance than 300uH.

If that hypothesis is true (that a high leakage inductance is helpful), then that indicates a largely differential-mode issue. It's very convenient to use the leakage inductance of a CMC to achieve adequate DM filtering, but in some cases, you might need a dedicated DM choke. Powdered iron, Sendust, or High-Flux core types is probably what's needed to maintain decent inductance at high current.

[prasimix]

Wurth does has common mode chokes for higher currents. Have you looked at WE-CMBNC series. For example  7448030333 2.5A rated current. You can find a diagram with insertion loss. Frequency range matches with your measurements above. I don't know whether damping is sufficient for your application

https://www.we-online.de/katalog/en/WE-CMBNC?#744803033

Actually I did, the following one: 7448030417 that has a much higher inductance (17 mH) and rated for 4 A, but still has a minor effect. I tried few others with inductance about 10 mH but again without notable effect, that is strange.

[prasimix]

The Wurth choke sems to have a pretty typical leakage inductance for a common-mode choke. 300uH (min) /10mH (typ) = 3% leakage inductance ratio...I think most CMCs are in the range of ~5%, though it can vary considerably with bifilar vs sectional wound. In this case, it's clearly sectional wound with large separation between the windings, so very well may have higher leakage inductance than 300uH.

If that hypothesis is true (that a high leakage inductance is helpful), then that indicates a largely differential-mode issue. It's very convenient to use the leakage inductance of a CMC to achieve adequate DM filtering, but in some cases, you might need a dedicated DM choke. Powdered iron, Sendust, or High-Flux core types is probably what's needed to maintain decent inductance at high current.

Yes, it’s one of two planned steps: to add a dedicated DM choke or to take the Mean Well IRM-10 modules out and try something else e.g. RECOM. Cui Inc. has interesting and price attractive modules but it is stated on Mouser that products are available only to OEM/EMS Customers and are not shipped to consumers in the EU.

[prasimix]

I continued experimenting today by introducing DM choke. First usable case is shown below.



Without load connected on the power modules I got the following:



... and with max. load on two channels it looks like this (I'm on the edge of average limit, but still below quasi-peak limit):



Second usable case include even larger CM choke, 2x11 mH and smaller DM choke (220 uH):



Without connected load:



... and with max. load on two channels:



It seems that DM of about 200 uH and CM of 2x3.3 mH or more is needed. Also X2 on the input has to be 330 nF or more.
I need to find now suitable CM choke rated for 4 A or more since used one is for 2 A only. Another thing that is possible worth trying is two CM filters in series with different characteristics that their leakage inductance is big enough for DM suppression.
A new challenge is to put two chokes (DM+CM or maybe two CM) on the PCB without increasing its dimensions a lot.

[TimNJ]

Try putting the choke between two X-caps as Pi filter. Generally, for the same total capacitance (L-C vs. C-L-C) you can get much more attenuation with a Pi filter.

See the below simulation. I've added modest parasitic series inductance and resistance to each element. Using a Pi filter can get you more attenuation or you can get the "same" attenuation as a L-C with smaller components

[prasimix]

Excellent suggestion, many thanks! I tried to split existing 430n into two 220n as shown below:



New measurement without load:



... and with max. load:



The final result is comparable with 3 times larger CM choke!

[Jay_Diddy_B]

Hi,

You might want to build a DM/CM Separator as I described in this thread:


https://www.eevblog.com/forum/projects/diy-dm-cm-seperator-for-emc-lisn-mate/msg3117396/#msg3117396

If you know if the emissions are CM or DM that you can add the appropriate filtering.

I went to alot of trouble to optimize the performance. But if you just built one, you should get most of the benefits even if the result isn't 100% accurate.

Regards,
Jay_Diddy_B

[prasimix]

Hi,

You might want to build a DM/CM Separator as I described in this thread:


https://www.eevblog.com/forum/projects/diy-dm-cm-seperator-for-emc-lisn-mate/msg3117396/#msg3117396

If you know if the emissions are CM or DM that you can add the appropriate filtering.

I went to alot of trouble to optimize the performance. But if you just built one, you should get most of the benefits even if the result isn't 100% accurate.

Regards,
Jay_Diddy_B

Thanks Jay_Diddy_B, I followed that topic, indeed. However, I haven't done anything about it yet because what is left open is what with the cores. Where to get them?
BTW, I tried to treat this issue as CM only and DM only, applying chokes that address one or another mode. But it seems that both of them are very prominent and need to be handled.

[Mr. Scram]

I return to the topic. In the meantime I got the Siglent SVA1015X and built LISN thanks to @Jay_Diddy_B who gave me the schematic and referred me to its design discussed here. LISN has 10 dB attenuator on its outputs. I used a ready made extruded box for which I made a front and rear panel to make construction easier. This is actually the second and wrong revision of the PCB which is 10mm shorter because I misinterpreted the dimensions of the box. If these 10 mm are added, a filter up to 6 A as well as a PCB AC power switch can be installed. Maybe I'll do it once in the future. In this case I had to shorten the box by 10 mm

That LISN looks great! I may have to build one myself at some point.

[Jay_Diddy_B]

Thanks Jay_Diddy_B, I followed that topic, indeed. However, I haven't done anything about it yet because what is left open is what with the cores. Where to get them?
BTW, I tried to treat this issue as CM only and DM only, applying chokes that address one or another mode. But it seems that both of them are very prominent and need to be handled.

Hi,

The cores I used the prototype are hard to get. There is a minimum order quantity.

I have tried a core that I obtained from a P0354NL common mode choke in one of the two locations.  This core is too small for one of the locations.

I have also received some ferrite cores that are sold by Digikey. I will test these in the near future.

Regards,
Jay_Diddy_B

[prasimix]

Hi all, just to summarize this topic, because a few days ago I passed the conducted emission testing. This of course required adding an AC input filter in front of all Mean Well modules so the scheme now looks like this:



Worst case scenario with the max. load in the lab looks like this:



Probably the peak around 20 MHz could be ironed out, however this is also passing and I am currently having other problems since I found out that my MCU module is doing a serious problem with conducted emission for which I will open a new topic.

Thanks to everyone for the assistance, especially @Jay_Diddy_B who gave me the LISN schematics.

[Kean]

Good work!
Any way to modify the original power board on the Crowd Supply units?
Maybe with a daughter board holding the filter components and some track cutting.
Not sure if there is enough space to safely fit it...

[prasimix]

Hm, highly unlikely. That was my first approach for test lab, but when I realized that both CM and DM choke is required I gave up and make another revision. The new challenge was to put everything together in the existing PCB dimensions and possibly the same budget. In the end, I succeeded in that by throwing out the triacs and their drivers and heat sink and putting relays, which are used on the BP3C backplane as well. So the audit that passed the conducted emission test now looks like this:



Even this revision isn’t necessarily the last one, because I don’t know how the people in charge of LVD testing will like it. For now, the latest AUX-PS module allows a passing of the EMI conducted emission for the entire BB3 chassis.

I will follow up on CE testing in the main BB3 thread soon.