Conducted emissions issue with multiple Mean Well AC/DC converters

[prasimix]

I started the EMI testing procedure of the EEZ Bench Box 3 which is carried out in an accredited lab. I was informed that the results of the conducted emissions were negative and that I should try to correct it.
I was in the lab today and got a slightly better picture of the situation, and the situation is complex. Here's what it's about. On mains voltage I have the following Mean Well AC/DC converters: 2 x IRM-10 (for +5 V and +12 V for 10 W each) and up to 3 x LRS-150F (for + 48V / 155 W) but I tested with only one. Mean Wells have some input filters, and the simplified wiring diagram and "conducted emissions" results look like this:



It is important to note that the MCU module is not a simple ohmic load. The picture is similar if the LRS converter is removed from the scheme and only IRMs remain. The first suspect is a 5 V IRM that powers the MCU module. For this reason, I tried to insert a filter in front of the IRMs, temporarily removed the LRS and it gave positive results:



However, the luck was short-lived because as soon as I added the LRS converter even without any load the situation got worse and the test did not pass:



Since the LRS has filter at the AC input, I also tried to connect it directly to 230 Vac and leave the IRMs behind the filter, but that didn’t help. At one point I doubted the CE certifications of the Mean Well module, so to dispel that doubt I connected them directly to 230 Vac, and loaded both IRMs with ohmic loads. The results are then good:



My conclusion is that an IRM loaded with a non-ohmic load combined with an LRS even without a load generates the problem. All the filters I prepared (6 of them) failed to fix this issue. I don’t know if it’s because they all had CM chokes with insufficiently high inductance (below 2.2 mH). I also had a seventh filter that had a CM choke of 2 x 40 mH but unfortunately it turned out that one was not working (one of its lines is broken).

Your input and suggestions are welcome. Many thanks in advance.

[RoGeorge]

I'm not very sure this will work, but I'll try to put some resistors at the output of each Meanwell power source (as a constant load).  My assumption is the current either varies too wildly, or it's too small.  Those are switching power supply, they try to keep the required voltage by altering the switching frequency or the on/off ratio, it depends on their type.

If they already have a constant load, the real load won't affect the switching frequency/period so much.  Also, a switching power supply will struggle with no load, sometimes it might not even work without a minimum load.  Either way, an extra resistor or a small light bulb to ensure a minimum current should help.

No idea if this is the main cause, but at least it's easy to test.  Put some load on all the Meanwell sources and see if the noises decrease enough to pass the tests.

[jbb]

Like RoGeorfe said, some minimum load might help.

If Mean Well supplies are base on flyback (very traditional for an off line supply, nothing wrong with it) they will be in Discontinuous Conduction Mode (DCM) at low loads. In DCM the switch nodes will likely exhibit some ringing. Try loading things up to 50% load and see what happens...

A possible extra complication; when the assorted supplies are paralleled up, their EMC filter caps are somewhat in parallel, which may shift filter resonances around.

[jkostb]

I suspect that the conducted emission issues are caused by common mode currents (L/N -> MCU -> PE). This is confirmed by your experiment where you added a net filter consisting of a common mode choke.

The problem with adding an unloaded ACDC adapter is that the combination of netfilter and unloaded AC DC adapter can cause stability issues. You always need to ensure that the output impedance of the netfilter is lower than the input impedance of the regulator (which in your case is unloaded). When the power supply gets instable you can get conducted emission issues!

Have you tried to add a common mode choke just before the MCU module (and remove net filter) before the meanwell power supplies? I would also add some ceramnic caps over the V+/V- lines to the MCU (100nF) for additional diiferential filtering.

Lastly Meanwell power supplies are very cheap, but you get what you pay for. I know that some Meanwell power supplies have inadequate netfiltering especially for reactive load. If your MCU module is properly filtered you can always try ACDC modules from other manufacturer (TDK-Lambda)

[trobbins]

The MCU is shown with an internal PE link to negative supply rail.  Can that link be modified to provide a lossy high-frequency impedance, but still provide an acceptably low galvanic resistance and protective current conduction capability (which I presume is a requirement)?

[prasimix]

Many thanks to all of you. I missed to add into initial post results of all three Mean Well directly connected to the AC mains and without any load. Mean Well seems to be doing its job well for that money:



I believe @jkostb guessed what the problem might be: connected GND to PE on the MCU module side. I suspected this at the end of the test, but I was not able to disconnect the GND and see what the result would be. It’s definitely something I’ll try right away the next time I go to the lab.

I will also prepare a configuration where it will be possible to insert a CM filter between the Mean Well and the MCU module.

@trobbins I'm not sure I really need low galvanic resistance from GND to PE, but I could put some Ferrite bead for the test, or just separate it with HV caps (e.g. 1 to 4n7) to collect spikes when connecting a USB cable from a PC.

[jbb]

Glad you’re narrowing it down!

At my previous workplace the serious EMI researchers had specially modified LISNs (3 phase too) where you could select the per phase, differential mode (3 modes for 3 phase pairs) or common mode.

You know, ribbon cables are infamous for EMI issues. Could you clip on a ferrite choke around the whole ribbon between the Aux board and MCU board?

[prasimix]

You know, ribbon cables are infamous for EMI issues. Could you clip on a ferrite choke around the whole ribbon between the Aux board and MCU board?

Good point. I will put this on the TODO list for the next lab visit.

[Dave]

I'd consider buying/renting/making a LISN box to do some preliminary testing on the bench. Frequent lab visits tend to get quite expensive when you're basically kicking around in the dark, hopelessly trying to please the EMC gods.

[prasimix]

I totally agree with that. My next question was just supposed to be: is there any DIY LISN schematics to recommend? They have a little monster in the lab NSLK 8128 (see below) although I also believe that a more modest version can serve the purpose quite fine.
Together with some entry level spectrum analyzer he would not be in total darkness believing that everything in the lab would go as it should.

[Jay_Diddy_B]

prasimix,

I have sent you a PM regarding the DIY LISN.

Regards,
Jay_Diddy_B

[prasimix]

Great, already replied

[TimNJ]

Rules-of-thumb for conducted emissions:

150-500KHz -> Differential mode noise
500KHz-30MHz -> Common-mode noise

It's just a rule-of-thumb. It's possible to have DM noise in the CM range and vice versa.

I would try a MnZn ferrite common-mode choke on the inputs/output of your DIB power module, maybe something from Coilcraft's CMT line. You can try a lossy ferrite cable core, maybe try a core made from Fair-Rite 31 material.

The LRS-150F looks like a pretty old school design. I'm not sure if it actually is, but newer designs using quasi-resonant flyback or resonant LLC may have lower EMI. Obviously, it's not guaranteed, but may be worth thinking about if you are still having a hard time.

LRS-150F certainly takes the cake for cost. It's pretty massive as far as 150W open frame power supplies go at about 6" x 4". EPP-200 is 140W, 4" x 2" and uses an LLC topology. Personally, I think topology is probably not your biggest concern, but just a thought.

[prasimix]

Thanks @TimNJ for you input and mentioned Rules-of-thumb. Currently my main concern is not supplying DIB power modules but IRM-10 and powering of the MCU module. But, yes from other side, trying to attenuate noise generated by IRM-10, added AC filter affects IRM-10 + LRS-150F combination even without DIB power module connected. I need to see first what happens when I disconnect GND from PE on the MCU module.

You are completely right that LRS-150F is not silent, and for that reason I added a CM choke (RN112-4-02-0M7) to the input of the power module to reduce the input noise to some reasonable extent.

[jkostb]

Regarding:

@trobbins I'm not sure I really need low galvanic resistance from GND to PE, but I could put some Ferrite bead for the test, or just separate it with HV caps (e.g. 1 to 4n7) to collect spikes when connecting a USB cable from a PC.

Please note that a PE connection to chassis is usually done for safety (e.g. if your product must comply with low voltage directive).  From EMC point of view it is totally ineffective, because the wiring has too much inductance. If you make a connection between PE and GND (=chassis) because of safety then I would certainly not recommend inserting a ferrite bead. Another thing which you have take into account is that shield of the USB cable is probably also connected to PE (unless you use a laptop). So the metal housing of the MCU can be indirectly connected to PE through a USB cable attached to a desktop computer. If I look at you diagram you have several modules. If these modules all have the same reference (=GND in your diagram) then you must ensure that all these modules are properly bonded (=low impedance path). Safety grounding should be bonded only at one point to your system reference.

For debugging of your conducted emission issue you need to understand how the currents flow and if the problem is caused by common mode or differential currents. I suspect that the problem in your case is caused by combination of both. The high frequency components are probably common mode. You can verify this with a current clamp and a spectrum analyzer. Note that you are looking for currents in microampere range!

[prasimix]

We have not yet reached the LVD test, as they want me to pass EMC (conductive and radiated emissions) first.
I don't know if they will require that the GND of the MCU module must be connected to the PE or it can remain floating just as is the case with the GND on all power modules. This is intentionally done so that their power outputs can be connected in series or parallel as needed.

[prasimix]

I was in the lab again today and we made new measurements.

1. For start I tried the AUX-PS which has both Mean Well IRM-10 converters for +5 V and +12 V installed and without a connected MCU module:



2. The next thing I tried was to connect two Mean Well LRS converters to which BB3 power modules (DCP and DCM) are connected (without load attached). MCU module was still disconnected:



3. For the next measurement I replaced the AUX-PS module with another on which the IRM-10 +12 V module was removed. LRS converters were not connected, nor the MCU module. I also added a filter (100 nF X2 + 2 x 0.7 mH + 2 x 2.2 nF Y2). That looked great:



4. Next I connected the MCU module but on which the PE is not connected to the GND like last time. This time it also looks very good:



5. Unfortunately adding back LRS converters with connected BB3 power modules creates a problem again and we have breaking the limit in a wide range from 150 kHz to almost 1 MHz:



6. The situation is no better with the load connected to the outputs of the power modules:



It turns out that the problem is with LRS converters and complex loads such as BB3 modules that have switching power pre-regulators at the input stage. I think now I will really have to get a spectrum analyzer and make a LISN (for which I have already got a schematic!) to continue testing outside the lab.

The question is whether I need to deal with the filter at the AC input or maybe better the filter at the DC input of the power module switching pre-regulator.

[jkostb]

You always need to insert the filter close to the source. From your experiments it is clear that the problem is  caused by the load of the MeanWell LRS module ( DIB module).  During a conducted emission test you measure the noise voltage between Line and PE and Phase and PE. Both voltages have to be below the limits. You can derive the currents easily from these voltages by divding the peak valus by 50Ohm. So you can investigate this problem further by use of a current probe (e.g. Fisher F-33-1) and LISN. First you need to figure out whether the problem is caused by common mode or differential mode currents. Note that the Meanwell LRS module has a functional earth, which means that the PE line is internally connected to the DC- output.

In your measurements I observe peaks between 150-200Khz. These are most probably direct from the switching power supply. The peaks between 300-400Khz are probably harmonics of the switching frequency of the SMPS. So the noise currents are probably differential noise currents (to be confirmed with current clamp) and I would start adding capacitors between DC+ and DC- just after the Meanwell LRS module. Since module  has 48V output, I would use polyesther caps.  You can increase the filtering by inserting inductors in the DC+ and DC- lines(only if required). Be careful with inductors with ferrite cores, because these can saturate.
Then you can look with current probe whether it reduces the noise currents. Given that the frequency is <1Mhz adding bypass caps combined with inductors (if required) is probably sufficient. Common mode currents are usually much higher in frequency.
You also need to investigate the routing of wiring. Don't run input and output wiring close to each other. Hope this helps.

[trobbins]

prasimix, what 'is' the DIB power module - both looking in to its power input terminals, and if it has any other external connections, and its loading profile?

[prasimix]

Many thanks @jkostb and @trobbins. Please find below simplified wiring diagram of DIB DCP405 module connected to LRS converter.

Mean well has PE input but it is not connected to the DCout- for sure, e.g. output is "floating" otherwise I would not be able to combine the power module outputs in series.

[jkostb]

I see in the block diagram of DIB module a common mode choke. I have looked at specs of this choke and this choke has maximum attenuation around 150Khz (depending on load current). You  have a peak in conducted emission diagram also around 150Khz. So this confirms that your problem is not caused by common mode currents. So I don't think that you will solve the problem by adding more common mode chokes.  Problem is most likely  caused by differential noise currents from the switching power pre-regulator. What is the switching frequency of this module? If you have a near field magnetic probe you can connect it to a spectrum analyzer and look whether you also observe peaks around 150Khz. If this is the case I would increase the input bypass capacitor to this module.Note that C1 (10nF) is not effective around 150Khz.  because it has an impedance of 100Ohm at 150Khz.  So try adding bigger capacitor (e.g. 10uF)  over the  + and - input lines after the common mode choke.  Do you know what the input capacitor value is of the switcher? You also have to examine the board layout. This input capacitor needs to be located in such a way that the critical current loop is small!

[prasimix]

Please let me know how did you realize that max. attenuation for the used RN-112 is at 150 kHz? This is its frequency characteristic:



Switching frequency in this case was about 220 kHz. Currently I have the following capacitors on pre-regulator input connected in parallel: 3 x 4u7 X7R + EEEFK1H470P (47u elco).

[Jay_Diddy_B]

Hi,

Let me add a few things to the discussion.

Reference:

https://www.schaffner.com/fileadmin/media/downloads/application_note/Schaffner_AN_CISPR17_Measurements_E8.pdf

This talks about the issue with filters measured in a 50 / 50 environment and then used in a 0.1 100 environment.

0.1 100 is correct for Differential Mode emissions. The LISN inputs are in series.

0.1 25 is correct for Common Mode emissions because the LISNs are in parallel for CM inputs.

Inductor Model

Here is the curve from the datasheet:



From the datasheet we know:

The nominal inductance is 700uH
The Rdc = 24m

From the curve we know that the SRF is around 3MHz
We can calculate parallel capacitance as 10pF

We need to add a parallel resistance to set the maximum attenuation, at resonance, to be 32dB in the 50 test circuit:




If I run the model and plot V(vin)/V(vout) I can get the attenuation of the CM choke per CISPR 17 in a 50 50 test circuit:



Given the simplicity of the model, this is pretty close to the datasheet.

Application Circuit

The CM choke forms a filter with the Y caps. The X cap connects the two lines in parallel, so the Y capacitors are considered in parallel.

If we believe Schaffner, and accept that noise source is 0.1 and you trust me that a pair of LISNs presents 25 to CM signals.

Then the application circuit can be modeled:



The result from running the model shows how the CM choke works with the Y-caps to create an effective filter:



Y-cap location

The location of the Y-caps depends on if you are stopping emissions or improving immunity.

In a power supply, normally the plan is to reduce emissions, so the Y-caps should be located on the input of the power supply, the mains side of the CM choke.

Here are the modelling results showing the impact of having the Y-caps on the wrong side:



In the diagram of the DCP405 the Y-caps are on the wrong side of the CM choke.

Regards,
Jay_Diddy_B

[prasimix]

You're completely right. Usually in a power supply, normally the plan is to reduce emissions. However I put that filter in the first place to prevent the emission in the other direction: from Mean Well (that is power supply, too) to the power module. Of course back then I didn’t think about the consequences of why I now have trouble passing conductive emissions tests.

This is what the output of a power module powered from Mean Well without a filter looks like:



... or with filter that has Y-caps on the "wrong side":

[Jay_Diddy_B]

Hi,

You can split the Y caps, to make a PI filter and put some on each side of the CM choke.

Jay_Diddy_B