Common mode noise mystery

[Phoenix]

- Common-mode choke : Bourns DR331-474BE

Bi-filar, so negligible DM leakage inductance (caleld LL in the datasheet).

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

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...

I could conceed that there exists an observable degree of 1MHz (switcher bandband frequency) common mode noise. I would love to understand the path it is taking (radiated or capacitively coupled) as the path is not obvious... However this is consistent with the CM choke attenuating the 1MHz spike a few dB in the spectrum.

The million dollar question now (the originial intention of the filter) - what's the minimum design to meet CISPR25.

[uski]

I could conceed that there exists an observable degree of 1MHz (switcher bandband frequency) common mode noise. I would love to understand the path it is taking (radiated or capacitively coupled) as the path is not obvious... However this is consistent with the CM choke attenuating the 1MHz spike a few dB in the spectrum.

I'm still not sure if it's CM or DM noise. Measurements are conflicting in this regard : the capacitor alone filters the noise somehow so that's mean DM noise, but the traces clearly show there is some CM noise. Maybe the noise I have has both CM and DM components.

I need to make a filter or something to distinguish between CM and DM noise for sure.

[Phoenix]

I'm still not sure if it's CM or DM noise. Measurements are conflicting in this regard : the capacitor alone filters the noise somehow so that's mean DM noise, but the traces clearly show there is some CM noise. Maybe the noise I have has both CM and DM components.

The majority of the noise in your application is definately DM - it fits the all the results as well as the theoretical circuit behaviour (discontinuous current pulses from a buck converter). You're right, however, the overall EMI spectrum is always a combination of both DM and CM. Circuit/topology, construction/layout etc. will determine what is dominant at which frequency. That 1MHz component having quite a bit of energy is probably CM and DM.

Maybe there is capacitive coupling from the extra long supply cable over the ground surface leaking 1MHz - try a shorter supply lead (unless of course this is defined by the standard).
There are also some large loop areas in the cable where you've seperated the twists - try tightening these up, they can inductively couple.

A splitter/combiner is a useful tool, but you need to appreciate its imperfections or it can lead you astray too.

[T3sl4co1l]

200us/div is *way* too low to resolve harmonics in the MHz+ range.

Tim

[uski]

For people still reading the thread, you're not going to regret it
My scope has a fancy "DPO" capability. So I thought, why not use it ? It may help see more things.

Setup :
- Positive side LISN on Ch1
- Negative side LISN on Ch2

I reproduced the current pulses from my radio transmitter using my active load.
You can see them here, because they generate more noise on the power supply :


Zoomed in :


As you can see, the traces are so noisy that it's very hard to see what's going on.
Yea we see there is some DM noise outside the pulses, but inside the pulses it's just too noisy to see anything.

And then I had an idea...

Perfect differential mode noise means that if Ch1 goes high, Ch2 should go low. If Ch2 goes low, Ch1 should go high.
So if I set the scope to XY mode, I should see a perfect diagonal line, if the noise was 100% differential.

If the line is not diagonal, it means some noise current does not flow normally, and that's probably common mode noise.

So...

Here is the control trace, with the UUT unpowered :


Here is the UUT powered but idle :


Here is the UUT powered and the active load simulating the pulses from the radio transmitter :


See that weird looking spot at the top-left of the center ?

I was able to make the phenomenon clearer by setting the active load to draw current permanently :


This is clearly not a straight line.

So, unless I'm mistaken, it means I have some non-differential noise.

The "line" is also a bit slanted and offset but it may come from the oscilloscope not being calibrated (cal due 2010) so I'm not drawing any conclusion from this, especially because we're dealing with milliVolts so that'd be sensitive to calibration being off. It could also be the LISNs being slightly different. Green would be the correct line, red is the actual line :

[T3sl4co1l]

Any imbalance in CM/D impedances will also push one into the other, and such.  Much more dramatic: any mismatch in probes, cables, etc.  Phase shift or amplitude differences (at whatever wideband frequency) obviously will lead to huge measurement errors.

Those spikes look pretty solidly diff though.  Even though the impedance might be low, that just means it takes a low impedance to get things started... which supply bypasses do a fine job of providing.  Best solution of course is adding a high impedance, like an inductor.

Tim

[Phoenix]

Any imbalance in CM/D impedances will also push one into the other, and such.

Ah the "whack a mole" effect; whack one down, another one pops up.

uski - don't get too tied up in your measurements. Poor measurement technique will lead you up the garden path.

[uski]

Hi guys

Ah the "whack a mole" effect; whack one down, another one pops up.

uski - don't get too tied up in your measurements. Poor measurement technique will lead you up the garden path.

Any suggestion about improving the measurements ?
I really want to explore this issue 100%, it gets me a better understanding of what's happening and I see it as a learning experience.

Thanks

[Phoenix]

Power splitters/combiners (0deg, 180deg) are useful, but you need to undertand what they actually give you (especially knowing if your CM is symmetric/unsymetric or balanced/unbalanced) - I'll see what references we have at work next week. As T3sl4co1l explained any mismatch in LISN/cable etc. will result in imperfect cancellations so results from splitters/combiners (and oscilloscope with math) should be taken as indicative only, not absolute.

Also with regards to using the 'scope, you need to consider how the signal is processed for the frequency domain result (peak, QP detector, bandwidth etc). So time domain measurements may not reflect how bad/negligible something is in the frequency domain.

[uski]

Hi there,

I just wanted to close the discussion with the final measurements.

22) With the original cable, unfiltered


20) With the filter badly assembled as described previously


21) With the filter correctly assembled


Conclusion analysis : the filter badly assembled does filter out some noise but not all, it probably filters the DM noise.
The correctly assembled filter makes the current goes through the common-mode choke and probably removes the remaining CM noise.
The filter does its job properly so I'm happy.

Thanks all for your help !