Common mode noise mystery

[uski]

Hi,

Following some problems with common-mode noise in one of the products I was involved with, I had to design a filter to remove that noise.
That filter has been put inside the power connector of the system.

The factory making the cables screwed up, and did not connect the wires at the right place.

So, here's what I should have been getting, and what I actually got (uskiCAD 1.0 )


And to my surprise... the noise is filtered out.

Here's the measurements (purple trace: noise on the negative wire, cyan trace: noise on the positive wire - see this thread for measurement setup)

Cable with no filter :


Cable with the wrongly connected filter :


Pretty good for a wrongly connected filter huh ? How's that possible ? Can someone explain ?

What I'm seeing here might be an artifact caused by the LISNs maybe, or something else. But if it's actually working then it's really good as it's possible to filter out common mode noise irrelevant of the DC current, as the DC current doesn't go through the common mode noise. But I doubt it. Any explanation ?

uski

[EMC]

No, the 'least impedance' theory would apply to differential mode; doesn't explain this.   Good food for thought through, I am thinking that a common mode choke "absorbs/cancels" rather than the usual "filter/alternative ac path" solution.   A third plot of how it should work would help.   e.g. Is it only 1/2 or 1/3 as effective as it should be?

And what is the small square component in the positive line at the input?   

[MagicSmoker]

...
And what is the small square component in the positive line at the input?

Probably a ferrite bead.

To the OP: try measuring the common mode *current* on the DC supply; I suspect that will be more realistic looking vis-a-vis a bypassed filter not filtering.

[dom0]

I am thinking that a common mode choke "absorbs/cancels" rather than the usual "filter/alternative ac path" solution

An ideal common mode choke is just a large inductance to common mode signals. It's the usual LC filter (as a differential filter) with the cap(s) behind it. But I don't really see how this should give you that much attenuation when that network isn't actually in the current path. So either it's the first cap that shunts it, or something else was changed, too (e.g. grounding arrangement, added shielding somewhere, different setup).

[uski]

Hi,

The parts are :
- PTC fuse : Bel fuse 0ZCF0075AF2C (0.75A), it's there to protect the windings of the coil in case of short circuit at the output of the filter
- Common-mode choke : Bourns DR331-474BE
- Caps : 1uF 100V (TDK CGA4J3X7S2A105K125AE)

The cable is a design update from a previous design we had.

There is a connector on both end, one side connects to the car (to take power + signals), other side goes to the device, with wires in between. Nothing spectacular.

On the initial version, the factory simply soldered the wires on the pins of the connectors, on each side.
On the new version, there is a small PCB added to the connector on the car-side, where they should solder the power wires (this is a 3D inside view of that connector) :



New+ and New- is the supposed connection point of the wires. This is the output of the filter.
Old+ and Old- is where they actually soldered the wires, like they did before when there was no filter. The PCB takes these points as the input of the filter.

So it's the exact same connector, with the exact same wiring location, exact same wire length... it's just that we added a PCB with a filter, that is not used... yet it's still filtering somehow.

I do have some data of how it "should" look like, I got this plot doing some tests with the exact same component (minus the PTC) but it was soldered "dead bug style" on wires (without the whole input connector you see on the 3D picture), and the filter was located right next to the device instead of inside the connector at the other side of the cable, so it's not perfect for a comparison, but it can give an idea (as the picture says, add 10dB so the reading is 41dBuV) :



I'll be able to take some measurements with the filter wired correctly in roughly 2 weeks. But it looks like the "bypassed" filter isn't that bad !

The more I look at it, the less I understand how it's working. I'll check the measurements tomorrow.

[Phoenix]

Why do you say it's a common mode problem/noise?

The filter capacitors you have drawn are differential mode filtering only, that common mode choke on it's own probably isn't going to do a whole lot for common mode noise. I'd suggest your noise is more likely differential mode and just the capacitor at the output is doing a damn good job of filtering. Just because you see the noise on both wires does not make it common mode.

[T3sl4co1l]

Impossible to say without pictures or circuits of the surrounding equipment, or..., but I would agree, you have a mystery on your hands.

So do what any good engineer would, and gather more data.  There is no mystery, only insufficient data.

Don't suppose you have a proper ground plane, LISN(s) and stuff to set up a sort of demonstration?  Then some voltage or current probes, or CM/Diff coupling networks, would be helpful for testing it.  Determine which end the noise is coming from, and where it's coming and going out of (it has a ground return somewhere -- at low frequencies, there's no radiation current to "ground" one end of the cable!).  Then determine what kind of filter (including characteristic impedance! it matters!) is needed to address it.

Tim

[Phoenix]

I agree, analysis of the circuit and construction would go a long way to determining the noise source.

It looks like an OBD connector (CISPR 25 being automotive); it also isn't mounted in an earthed/ground chassis (photo shows it sitting on a block of foam). This would lead me to hypothesis that it is VERY UNLIKELY that the noise is common mode, almost certainly differential. CM is usually capacitively coupled from the device to earth/ground/chassis through an unintended path.

Does the circuit contain a small switching power supply at 100's of kHz? The baseband frequency is probably too low for it to be a digital clock.

A bigger cermaic capacitor at the terminals or even on the PCB is likely to help dramatically (or smaller, depending on the self resonant frequency). Perhaps 2 small discrete inductors (or chip inductors) in each of the supply lines instead of the common mode would also help - be careful to choose a high bandwidth (probably dissipative) magnetic material.

(Please correct me if any of my observations are wrong, so I can adjust my hypothesis.)

[uski]

Hello,

Four reasons why, in my opinion, it must be common mode noise and not differential mode noise :
- Noise level is the same on each LISN (see the two curves that match on the plots)
- Noise levels I see are the same than the noise level measured by a test house with a R&S LISN
- Adding a capacitor (1000uF 25V) at the LISN does not change the readings
- Increasing the value of the capacitors at the input of the power supply on the device itself does not change the readings

Yes the big noise peak is at the frequency of a switching buck converter in the device, running at ~1MHz.
It used to run at 650kHz but I increased it to ~1MHz, it reduced the noise level a little bit and makes it easier to filter.

I should move it to 1.9MHz to get away from the AM radio spectrum, as per EMC's wise recommendation in the other thread.

uski

[Jay_Diddy_B]

Uski,

Try these experiments and report the results:

1) remove the CM choke:



2) remove the CM choke and connect the two capacitors in parallel with very short wires:




One further test.

In the original measurement, no filter, you have a signal level of 63dBuV. This is 1.4mV rms which is big enough to measure with an oscilloscope. Connect the output of both LISNs with the same length of cable to an oscilloscope. Use the 50 Ohm setting on the scope or use 50 terminations. You should be able to see if the signals are common mode, in phase, or differential mode anti-phase.



Regards,

Jay_Diddy_B

[uski]

Hello,

I did 6 measurements :
1) Filter badly connected, like in the cable
2) No filter
3) Capacitor 2200uF//1uF
4) Filter badly connected, common mode choke removed (as Jay_Diddy_B suggested)
5) Filter badly connected, common mode choke bypassed by wires (as Jay_Diddy_B suggested)
6) Filter correctly connected

Measurement setup picture :


Capacitor as a filter picture (test number 3) :


Bad filter connection picture (test number 1) :


Common-mode choke removed picture (test number 4 - test number 5 is same with wires instead of CMMC coils) :


Correct filter connection picture (test number 6) :


Results (purple trace (trace 2 on the DSA) is noise from the negative terminal LISN, cyan trace (trace 3 on the DSA) is noise from the positive terminal)

1) Filter badly connected, like in the cable


2) No filter


3) Capacitor 2200uF//1uF


4) Filter badly connected, common mode choke removed (as Jay_Diddy_B suggested above)


5) Filter badly connected, common mode choke bypassed by wires (as Jay_Diddy_B suggested above)


6) Filter correctly connected

[3roomlab]

sorry i need to ask a noob Qn

what is the purpose of the CISPR 25 boxes?
and is there a shielding function for the large piece of copper desktop?

[Phoenix]

- Noise level is the same on each LISN (see the two curves that match on the plots)

This is exactly characteristic of both differential and common noise. It doesn't actually help suggest one way or the other.

The only difference between CM and DM is the phase relationship. The EMI reciever can not measure phase, thus CM will appear exactly the same as DM without further testing equipment (splitters and combiners). The use of splitters and combiners gets a bit complex, so I wouldn't recommend going down that path just yet.

- Noise levels I see are the same than the noise level measured by a test house with a R&S LISN

This validates your LISN/receiver which is really good because you can experiment in your lab with confidence , but does not validate of your interpretation.

- Adding a capacitor (1000uF 25V) at the LISN does not change the readings

Electrolytic - probably not going to help for your problem. You have noise at 1MHz++ you should to look at the self resonant frequency of your components. You should consider going smaller, not bigger.
For example
- 100nF MLCC capacitor i randomly chose has a self resonance of about 15MHz, above this value it acts as an inductor. You need smaller than 100nF even... Put in 100pF even. Better yet parallel up 100pF and 100nF, let me know how that goes.

- Increasing the value of the capacitors at the input of the power supply on the device itself does not change the readings

See above - it's not the size that matters, but the self resonant frequency. The bigger the capacitor the lower it's effective frequency range.

Yes the big noise peak is at the frequency of a switching buck converter in the device, running at ~1MHz.
It used to run at 650kHz but I increased it to ~1MHz, it reduced the noise level a little bit and makes it easier to filter.

I thought I saw it change frequency,
That 1MHz might be easier to filter with a differential LC, as long as the parts you have selected are actually effective at those frequencies. E.g.

100nF MLCC
http://ds.yuden.co.jp/TYCOMPAS/ut/detail.do?productNo=LMK105BJ104KV-F&fileName=LMK105BJ104KV-F_SS&mode=specSheetDownload
The graph on the bottom right shows its impedance with frequency. The impedance drops linearly to about 10MHz, hits the self resonant frequency and then starts going up.
Below 10MHz this capacitor is a good capacitor
In the middle this is the self resonant frequency
Above 100MHz this capacitor is a good inductor (eww) - not going to help your noise.

10mF Electrolytic
http://www.vishay.com/docs/28332/138aml.pdf
Figure 16 shows the impedance with frequency (this is quite hard to find for electrolytics)
Interpolating a 2200uF capacitor stops being a good capacitor around 200kHz! No wonder it had little effect...

(edited now for less frustration, sorry)

[Phoenix]

what is the purpose of the CISPR 25 boxes?
and is there a shielding function for the large piece of copper desktop?

Those boxes are called a LISN (line impedance stabilisation network), they perform 3 functions:
1. To provide a stable constant and fair source impedance for the device under test. The supply of the DUT could be anything (electrical grid, car battery etc.) the LISN ensures that the impedance characteristic of the supply does not interfere with the measurement.
2. An EMI filter of its own, to remove noise coming from the source.
3. It provides the take off point for the EMI reciever measurement. I.e. the BNC/coax going to the the EMI reciever.

The copper desk is a very good ground/earth - importantly, it means everyone has the same ground reference (small amounts of ground bounce between test apparatus might show up in this sensitive testing). It is where the LISN shunts high frequency external noise before it gets to the EMI reciever and intereferes with the measurement. It also provides the easiest return path for any common mode leakage in the DUT.

[T3sl4co1l]

Electrolytic - next to useless for your problem!? You have noise at 1MHz++ you NEED to look at the self resonant frequency of your components to determine IF THEY ARE DOING ANYTHING. You need to go smaller, not bigger.

Calm down: the electrolytic will still be a negligible impedance (<< 50 ohms) well into the MHz.  It might even be beneficial to dampen the 1uF.  Though the impedance is likely so low in all cases (presuming there are bypass caps elsewhere in the circuit, too) that it doesn't make much difference.

Tim

[T3sl4co1l]

Hello,

I did 6 measurements :
1) Filter badly connected, like in the cable
2) No filter
3) Capacitor 2200uF//1uF
4) Filter badly connected, common mode choke removed (as Jay_Diddy_B suggested)
5) Filter badly connected, common mode choke bypassed by wires (as Jay_Diddy_B suggested)
6) Filter correctly connected

Thank you for the testing.

So, it looks like you were unable to reproduce the supposed "like in cable" measurement taken earlier.  Seems like the cable was connected correctly after all, perhaps you misread the connections..?

Always good to do a sanity check.  If you can't reproduce the condition, it ain't science!

The correct connection looks to do a fine job, though, as one would expect!

Tim

[Phoenix]

Calm down

Sorry, I have now.

presuming there are bypass caps elsewhere in the circuit, too

It appears the noise is being generate by the main onboard switching power supply. The board probably has sufficient decoupling on the switcher output, but that isn't going to help the input side. Especially if it's a buck converter (most likely 12V down to 5V/3.3V?) with discontinuous supply current pulses...

[uski]

Thank you for the testing.

And thanks for helping

So, it looks like you were unable to reproduce the supposed "like in cable" measurement taken earlier.  Seems like the cable was connected correctly after all, perhaps you misread the connections..?

The conditions are slightly different, because the purposes of the measurements were different.

The very first measurements I did were with the incorrectly assembled cable itself, compared with a cable with not filter. The idea was to compare the "badly assembled" cable with a cable with no filter, under the same conditions. And we see that the filter, despite being assembled incorrectly, does remove some noise. To do that test, I had to use matching connectors, so the parasitic components (wire length...) were slightly different and may have some effect.

The 6 measurements I did in a row were done in a slightly different setup where I used simple wires between the device and the LISNs, and I then connected different "filtering" stuff exactly the same way, exactly at the LISN. The purpose was to compare the different filters when used with the exact same wiring, not to reproduce anything I did before. It's a different experiment. It's OK to me that the results are not exactly the same and it does not mean anything.

The main difference between the two test setups is that for the first tests with the assembled cables, there was some wire between the filters and the LISNs, whereas for the second test the filter was put directly at the LISN.

I want to emphasize that the test with the 2200uF capacitor was done with a 1uF capacitor in parallel with the 2200uF capacitor (see the picture) and I was careful to match (as closely as possible) the wire length with the filter PCB. I also used the exact same 1uF capacitor that is used on the filter PCB. And we clearly see that this filter, by itself, doesn't do anything. So I doubt it's only DM noise, as it would have been damped by the 1uF ceramic cap.

But I need to be 100% sure that this is not DM noise. I would be surprised, but I need to check.
I see 3 options :
- Jay_Diddy_B time-domain method with an oscilloscope (but I'm not sure my scope is sensitive enough, and I may have problems identifying the noise signals in the noise time-domain traces)
- LISN MATE as described here (resistive signal adder). But I'm having trouble sourcing the 16.7ohms 0.1% resistor - I submitted a sample request, we'll see if it goes through
- Magnetic current probe, with both power wires going through, that should remove DM noise and only output CM noise. But it'll be an uncalibrated measurements and I never did it before so I won't trust the results 100% which is not great

[uski]

And also, something interesting :

During the series of 6 tests I did, those two yield the same results :
1) Filter badly connected, like in the cable
4) Filter badly connected, common mode choke removed

This means that filtering effect seen with the filter in parallel with the power supply wires is not related to the CMMC.
So that leaves only two components in the circuit : one capacitor and the PTC fuse.

However, those tests are similar :
2) No filter
3) Capacitor 2200uF//1uF

Which means that the capacitance alone is NOT the source of the filtering effect.

So it's either that the capacitors alone do nothing, but with a resistor in series it does something (I assume the PTC is equivalent to a resistor in its untriggered state), or that the filter PCB acts like an antenna or something and absorbs the noise.

Thoughts ?

[EMC]

Hello,

Four reasons why, in my opinion, it must be common mode noise and not differential mode noise :
- Noise level is the same on each LISN (see the two curves that match on the plots)
- Noise levels I see are the same than the noise level measured by a test house with a R&S LISN
- Adding a capacitor (1000uF 25V) at the LISN does not change the readings
- Increasing the value of the capacitors at the input of the power supply on the device itself does not change the readings

Yes the big noise peak is at the frequency of a switching buck converter in the device, running at ~1MHz.
It used to run at 650kHz but I increased it to ~1MHz, it reduced the noise level a little bit and makes it easier to filter.

I should move it to 1.9MHz to get away from the AM radio spectrum, as per EMC's wise recommendation in the other thread.

uski

Very difficult to prove common mode diff' mode without phase info.     The only suppression that can work in the 'unintended' config is the first cap; and it will only be effective for diff' mode.  The noise must be about 180 out of phase and cancelling across the first cap.  Verification: 1. Scope to see phase on +ve & -ve lines.  2. Take the rest of the circuit out by lifting the choke input pins. If verified then the choke can be removed from the production BOM; I will go you halves in the saving!

Sent from my SM-N9005 using Tapatalk

[Phoenix]

I looked again at the photo of your 1000uF (and any others with the piggyback style) and I believe the test is flawed because you have the DUT going into the LISN and the filter board located on the otherside piggybacking. Layout/placement matters a lot. I wouldn't try to draw any significant conclusions from this setup.

If you could set something like this up (see attached diagram) and try 100nF // 100pF MLCC or 1uF // 1uF or 2.2uF // 100nF; a few combinations around those sizes, that would be a valid test demonstrating a DM only filter. But it must be setup inline without fly leads (i.e on the PCB or well made protoboard).

[uski]

Hi,

I looked again at the photo of your 1000uF (and any others with the piggyback style) and I believe the test is flawed because you have the DUT going into the LISN and the filter board located on the otherside piggybacking. Layout/placement matters a lot. I wouldn't try to draw any significant conclusions from this setup.

OK so I modified the test setup.

This time, this is what I did (tests 10 11 12) :


Without the filter PCB (tests 13 14 15 16 17) :


Capacitors at the device connector (tests 15 16 17) :


You can see the filter under test is at the bottom left, and is inline between the device (on the right), and the LISNs (at the top).
The trace from the spectrum analyzer is from the positive-side LISN. This time I didn't bother taking the readings from the negative side LISNs as all the previous tests showed it was matching in amplitude.

Also this time I used a log scale for the frequency as allowed by the new firmware of the DSA832.

So, today I did 8 measurements with this setup (numbers start from 10 to distinguish from the previous measurements - I want each test to have a unique number within this thread):

10) Filter correctly connected
11) Filter correctly connected, common-mode choke shorted with 2 wires
12) Filter badly connected (cables to the device connected at the input of the filter, output of the filter floating)
13) No filter (filter removed and input/output wires connected together)
14) 1uF capacitor instead of the filter
15) No filter (like 13) + 1uF capacitor soldered onto the device connector
16) No filter (like 13) + 1uF//100nF capacitor soldered onto the device connector
17) No filter (like 13) + 1uF//100nF//10nF capacitor soldered onto the device connector

And here are the results :

10) Filter correctly connected


11) Filter correctly connected, common-mode choke shorted with 2 wires


12) Filter badly connected (cables to the device connected at the input of the filter, output of the filter floating)


13) No filter (filter removed and input/output wires connected together)


14) 1uF capacitor instead of the filter


15) No filter (like 13) + 1uF capacitor soldered onto the device connector


16) No filter (like 13) + 1uF//100nF capacitor soldered onto the device connector


17) No filter (like 13) + 1uF//100nF//10nF capacitor soldered onto the device connector

[Phoenix]

14) 1uF capacitor instead of the filter
15) No filter (like 13) + 1uF capacitor soldered onto the device connector
16) No filter (like 13) + 1uF//100nF capacitor soldered onto the device connector

What's the physical configuration difference between 14 and 15? That's a huge 10dB difference at 1MHz for moving the location of that 1uF. I guess it shows just how important placement location is. The results of 16 show addtional high frequency attenuation by the 100nF, but may not even be required to meet CISPR25.

Do you have the datasheet for the common mode choke (or a plot of its leakage inductance)? I'm curious how much DM inductance it's contributing, some CM chokes can have a few percent leakage and make significant contribution to DM - others have a lot less. Maybe that 1MHz is radiating somewhere, can you shorten the leads or is their length dictated by the standard? Can you place that small wire that's dangling back onto of the foam block and twist the black/red all the way to the end?

[uski]

What's the physical configuration difference between 14 and 15? That's a huge 10dB difference at 1MHz for moving the location of that 1uF. I guess it shows just how important placement location is. The results of 16 show addtional high frequency attenuation by the 100nF, but may not even be required to meet CISPR25.

14) 1uF capacitor instead of the filter : the filter pcb has been removed and a 1uF cap was soldered accross the wires at the location where the filter was.
15) No filter (like 13) + 1uF capacitor soldered onto the device connector : the filter PCB was removed and a 1uF capacitor was soldered at the connector (see the photo where you see the caps soldered on the black connector, this picture was for 17, for 15 it's the same with just 1 cap)

I assume the parasitic inductance and or resistance of the wires between the device and the capacitor, in 14, makes for the difference, probably creating a RC filter ?
And the common-mode choke might be actually helping in this regard, maybe the parasitic resistance of its coils recreate that effect and create the same type of filter. Maybe the magnetic effect of teh common-mode choke does not help at all... which would mean DM noise... YET... see below.

Do you have the datasheet for the common mode choke (or a plot of its leakage inductance)?

Yup, see :

The parts are :
- PTC fuse : Bel fuse 0ZCF0075AF2C (0.75A), it's there to protect the windings of the coil in case of short circuit at the output of the filter
- Common-mode choke : Bourns DR331-474BE
- Caps : 1uF 100V (TDK CGA4J3X7S2A105K125AE)

==> http://www.bourns.com/docs/Product-Datasheets/dr331.pdf

And now what you've all been asking, time-domain stuff :




Measurements taken with a Tektronix TDS 754D scope, goes down to 1mV/div, with the "no filter" setup like test 13. Inputs set at 50 ohms impedance, AC coupling (this creates a high pass filter of 200kHz inside the scope), 1X. Trigger Ch1 normal single-shot.

Black : trace from the positive LISN
Green : trace from the negative LISN
Purple : trace defined as Ch1+Ch2

Because the purple trace is not 0, I think it means we have common-mode noise. What do you think ?
I'm not sure this test is relevant... there are so many frequency components in the noise, that it's hard to know what we're looking at.

The cursors are aligned to some sharp peaks which are spaced 960ns apart which is 1.04MHz...

To do that measurement I had to change the setup slightly and connect an active load at the output of the switcher, because intended RF emissions were fuc*** up the results (my device has a RF transmitter and it's its current pulses that load the switcher - it's possible the oscilloscope doesn't have any RF filtering at its input so it was seeing the HF EM field from the RF emissions).

Sanity check :


Noise is still here, similar amplitude, same pattern, so using the active load instead of the built in RF transmitter to load the DC/DC does not change the test conditions.

[uski]

And just for fun, so you understand why I moved to an active load when using the oscilloscope, here's a video showing what happens when using the RF transmitter on the PCB and waving my hand on top of the device :


So this is why I had to use an active load. The amplitude on the scope with this test was around 250mV... (Ch1 on top comes from the positive-side LISN and Ch2 on the bottom comes from the negative-side LISN), this shows one pulse from the radio transmitter :


Definitely not our noise !!! Amplitude is far too high. It's inducted voltage from the RF emissions !

And the exact same setup, transmitter off, using the active load instead (so the noise from the PSU is still there - but much much smaller, as we no longer see the parasitic EM field from the radio transmitter) :


Now it's better. We know the noise is still there from the sanity test above done with the spectrum analyzer. It's normal we see nothing under the 100mV/div setting.

Zommed in, Ch1+Ch2 trace added, now we see the true PSU noise (the noise is now continuous because I can't create the pulses with the active load - the RF transmitter does pulses):


This is slightly off topic but I thought it was interesting to show you how easily you can fuck up your measurements if you're not careful. It took me a few minutes to find out where this 200mV noise was coming from (the spectrum analyzer didn't see it due to filtering so it puzzled me a little bit - now it's obvious, but at first, it's puzzling)