Silencing a MeanWell SMPS

[helge]

There are a few catches when first trying to use a switch-mode power supply (SMPS) in a low noise application (involving sensitive amplifiers, radio frequency and broadband analogue circuitry and the like), most of which can be overcome with the proper choice of passive filter components in the right spots. Apologies for not making the text easier to understand with diagrams and drawings. They can mostly be found in the references given below and I'll add them if there is demand for it.


What came as a surprise to me is that the conducted electro-magnetic interference (EMI) specs of a power supply are only concerned with the INPUT conducted emissions. Usually one or two common mode filters with a few percent of stray inductance will cover most of the single and differential mode noise isolation needs (At least when you want to pass EMI compliance tests).



But then the converter itself has only limited intput-to-output isolation (minor issue unless you're plugged in next to a high power VFD or AC motor) and introduces new common and differential mode noise with a rich set of harmonics up to some tens of megahertz and broadband features caused by ringing induced by rectifier reverse recovery etc. (major issue).

Consider a flyback inverter: the switch node sees up to 2-3x the rectified mains because during the demagnetization phase the output voltage scaled by the winding ratio is reflected to the primary and adds to the input capacitor voltage. So if you have 230V mains it is not uncomon to see 600-800V MOSFETs in these converters, even 1kV types are used. Now the primary and secondary windings are inevitably capacitively coupled and form a capacitive divider with all the capacitances that return to case ground or the input negative rail.
That is one of the major sources of common-mode output noise and the reason why some switch-mode transformers have a Faraday shield (winding or foil in an open turn).
Whether or not such an E-field shielding is an option can depend upon the targeted price, power density and appliance class you're designing for. If a Faraday shield winding is not feasible, ceramic capacitors with sufficient voltage / safety rating (Class X1, X2, Y1, ... proper voltage rating and failure mode) can be placed to return the common-mode noise from the output to the input negative rail.
Typical values: for 12V output place 10µF/16V or 25V across the immediate output of the rectified secondary side and use 2.2-22nF across OUT+ or OUT- back to input negative (this capacitor is afaik not permitted for Class II devices) . The capacitive divider for the common-mode output noise now has a few pF due to inter-winding capacitance in series to, say, 10nF. 500Vpp on the switch node will show up as 0.5Vpp common mode noise.
Follow that with a high current common-mode filter coke and your common-mode noise is below the magnitude of your differential-mode noise whereas even tens of volts would be possible without the primary-secondary bypass cap.

The differential mode noise and all the harmonic goodness it brings also needs to be managed. Depending on the type of ferrite used in the common-mode filter on the output you may observe that the filter becomes useless in the single-digit megahertz region. These filter components can be selected parametrically and can be complemented with higher frequency toroidal chokes. You can also use shielded flat wire inductors with really high series resonant frequencies (SRF). I've picked a 2.2µH 6A shielded inductor with >100MHz SRF.
A bit of de-Q-ing and decrease in symmetry of the common-mode filters with a single resistor across one of the windings can also be tested.
Most of the times you'll only see an "EMI cannon" ferrite filter rod. These are cheap and cover a wide frequency range but they radiate to and also pick up much more EMI from the surrounding than toroidal or shielded SMD inductors.

Let me finish with a few links and pictures of the SMPS I modified to clean up the 1-20 MHz region. The board-level SMPS needs to be screwed to a metal chassis which is connected to PE. In the last picture I highlight the modifications explicitly.

http://www.interpoint.com/product_documents/DC_DC_Converters_Output_Noise.pdf (highly recommended)
http://www.interpoint.com/product_documents/DC_DC_Converters_EMI_Conducted_Interference.pdf
http://www.justradios.com/safetytips.html
https://en.wikipedia.org/wiki/Appliance_classes
https://www.maximintegrated.com/en/app-notes/index.mvp/id/716
http://www.ti.com/lit/an/snva489c/snva489c.pdf
https://www.tdkinnoveta.com/pdf/iDQ_fulldatasheet_072313.pdf (properly designed EMC output filter for POL DCDC applications from TDK)

ps. the SMPS in question is the MeanWell PS15-12.

[NiHaoMike]

At work, a Meanwell PSU is the reference for a "noisy charger"!

[Richard Head]

Helge

I agree with everything you mentioned. I would also like to add that a metal enclosure around the PSU with bypassing (ideally feedthrough caps) on all leads entering and exiting the chassis makes a huge difference. Also, don't forget about the SRF of the filter capacitors. The bigger X-caps have a pathetically low SRF and cease to be capacitors above a few Mhz. Same applies to 275VAC metalised paper Y-caps. They top out at about 30Mhz for 4.7nF. Check the Wima datasheets for SRF. From an RF point of view the EMI caps are terrible, but we have to live with them. Trying to get an open frame PSU to be electrically quiet is an exercise in futility.

[helge]

Richard,

thank you for the detailed information. I've since put the whole project back together and tested its performance. The main PCB also has filtered DCDC converters on it (and provisions for shielding cans) so there exists an intrinsic lower bound to what a test tells about the performance but we now seem to be on par with cheap linear lab power supplies.
I'll also try to fabricate a shielding cover for the SMPS board for further investigations.
I opted against using the shielded power supplies we had bought because they too were built down to a price and at 50W rated output I suspected more light-load-ness generating broadband noise and more severe current spikes.

As an excuse I might add that the supply and a two-stage common-mode filter share their own compartment formed by several thick metal plates - made me think I might get away with it for now.

Regarding shielding cans and ferrite adhesive foil: it appeared as if modifying the surroundings to control the stray fields would back-fire in a way that noise would show up where it was previously just of minor magnitude.
The main DCDC converter @ 600kHz is placed on the opposite side of the PCB onto which an RF hybrid module is mounted and these two are only separated by a 4-layer PCB with two solid power planes. With that we just got lucky as we did with the larger ceramic bypass caps that tend to cancel out dI/dt effects on critical spots on the PCB.

Side note: Vishay has some nice e-field shielded power magnetics: http://www.vishay.com/docs/49762/_ihle-4040dc-5a_vmn-pt0433-1502.pdf definitely an option for new designs.

Not so lucky was the choice to use a TDK-Lambda Z+320 programmable power supply right next to the aforementioned assembly. Our previous attempts to control input and output noise of that unit were less successful. Any idea if directly bypassing the output leads to the chassis has merit?

[Richard Head]

Helge
The link took me to Mouser but I got lost in the site!
There's no doubt that the EMI of a switchmode can be reduced to a negligible level with appropriate shielding and filtering. It will increase the manufactured cost though and the magnetics required to control the EMI start to rival the main switching magnetics! A VF resonant topology (such as LLC) starts to look attractive as they are inherently quieter. LLC has already displaced PWM topologies in telecom switchmode rectifiers with power densities greater than 23W/cubic inch including active PFC. The fact is though that most applications don't require really low level EMI so the PSU manufacturers don't cater for that market segment.

[helge]

Richard,
sorry for the link problem, I've updated it with a link to the manufacturer.

While the LLC topology has quite and advantage with respect to efficiency and noise spectrum it is a rare find as a low powered option for 85-265V AC systems.

Vicor has some quite interesting resonant switching products that might be an option:
V300C12C75B : 300V DC nom. (180-375V) in, 12V 6.5A out
http://cdn.vicorpower.com/documents/datasheets/ds_300vin-micro-family.pdf
www.vicorpower.com/documents/applications_manual/fas_trak_apps_manual.pdf

but then I saw the graphs ...

[T3sl4co1l]

I suspect most PSU modules are certified (precertified really, as they aren't really finished products) based on a physically short load wires/connections, if any.  Therefore, when put into an actual system, with long wires on the output side, or additional ground connections, the output CM voltage gets transmitted into all sorts of antenna opportunities.

Thus, one should be cautious of the output conducted, and overall radiated, emissions, from any power supply module, really.

I recently worked on a project that basically investigated these parts:
http://www.digikey.com/product-detail/en/ECE05US12/1470-1110-ND/4487491
http://www.digikey.com/product-detail/en/ECL10US12-P/ECL10US12-P-ND/4476530
http://www.digikey.com/product-detail/en/EML15US12-P/1470-2618-ND/4809088

Started with the tiny square one, failed by a fair margin.  Additional line-side filtering was tried, with little effect.  I suspect they don't even take the space to add an internal EMI filter on those things; I haven't seen a look inside one to be sure.

Some more aggressive prototyping/testing, with the middle and last parts.  The middle one is almost good enough on its own, but needed ~50uH CMC and a Y cap on the output side to secure an excellent pass margin.

The last one is fine all its own, and passed with considerable margin when placed in the "excellent margin" circuit.

It may be an added benefit that the medical version also has extremely low isolation capacitance, too.

Tim

[Richard Head]

Tim
Most commercial PSU's are tested to EN61000 (if in Europe)which calls up the original CISPR 22 specification. CISPR 22B is more stringent than CISPR 22A by about 10dB. Most commercial equipment is expected to comply with CISPR 22B. The spec is very specific about how the test is performed with the idea of ensuring repeatability. The test has quite a few phases but the most important (to me) are the conducted test (on both input and output leads) and the radiated test. If you fail the conducted test don't even bother with the radiated test as you'll fail if your conducted emissions aren't low.
Experience has shown me that  a common mode choke is generally required on the DC output as well as the AC input. The output diodes generate a huge amount of noise when they recover.
The military standards such as Mil Std 461 are similar but include more comprehensive susceptibility (CS01 and CS02) testing and test up to 1Ghz for conducted emissions (compared to 30Mhz) if I remember correctly.
The EUT is placed on a wooden table to create the worst case common mode situation. For radiated emissions the test antenna (bi-conical) is adjusted in polarisation to produce the strongest signal and is moved around the chamber until max signal is obtained. Or that's how it is meant to be done!
 

[helge]

found some test related documents to the Z320-1.3, they have these lovely setup drawings you talked about:
http://www.lambda-germany.com/KB/Zplus200-HV-IEC61000-DATA.pdf
http://www.lambda-germany.com/KB/Zplus400-HV-RELIABILITY-DATA.pdf

I've attached the schematic of the filter that now sits between the MeanWell power supply and the load. Perhaps it would have been a better idea to bring the 2.2µH inductor closer to the PSU terminals - but the PSU should also have a DM choke in place. The CM choke values are a bit low but they are tiny and come in potted SMD packages. The filter was initially designed to filter a PH50A280 module output.

http://www.tdk-lambda.com/products/sps/ps_pm/ph-a280/pdf/ph50a280_iec.pdf

Apparently no RF filter on the output is prescribed by the typical application circuit so I suppose it is included in the module. The input side is a totally different story. The EMI conducted emissions test foresees a two-stage CM filter assembly, quite an undertaking.

[chris_leyson]

Hi Helge

That was a great post, thankyou. I was testing a galvo driver on the bench one day and all of a sudden this noise appears at a few hundred kHz, where did that come from, turned out to be common mode noise from a cheap and very nasty Maplin bench supply that a colleague had just switched on, he was sitting a few feet away. Bloody Chinese crap.

I think far too many engineers forget about primary to secondary capacitance, "It's got a transformer in it, therefore it's isolated"  That's only DC isolation not high frequency AC. A screen between primary and secondary helps a little bit but not much.

Resonant converters are the way to go, you only have filter at a few spot frequencies, and no nasty high dV/dt edges to deal with.

I've been thinking about active common mode filters whereby you sense the common mode current at the output and then actively null it out. The same could be done on the input but nothing simulated or prototyped yet.

[T3sl4co1l]

Most commercial PSU's are tested to EN61000 (if in Europe)which calls up the original CISPR 22 specification. CISPR 22B is more stringent than CISPR 22A by about 10dB. Most commercial equipment is expected to comply with CISPR 22B

In my example, we were testing to 22B.

Note that this test only covers what goes up the AC line, and what's radiated into space.  It doesn't care about internal connections in your system, which is why output noise from a PSU module can be higher.

I'm not aware of any tests that specify how PSU modules need to be tested.  I didn't look at the standard, and don't know if it covers that sort of thing.  I'm sure it covers much more than merely the product we were testing that day...

The spec is very specific about how the test is performed with the idea of ensuring repeatability.

Oddly, the lab tech said he doesn't like the test very much: that it's quite inconsistent on setup methods.

For example, the power and auxiliary cord positions, lengths, and how much is folded up.  You're allowed to use whatever stock cable you bring, but the chances are, it won't be the exact length and position and angle specified in the standard.

Especially the ones where you need to stack ferrite beads on cables, and use voltage and current probes, to ensure controlled impedances.  Yech, that bothers me just thinking about it.  Though I think those pertain to susceptibility standards, not CISPR 22 specifically.

If you fail the conducted test don't even bother with the radiated test as you'll fail if your conducted emissions aren't low.

I don't think this is true.  You can very easily add a small value CM filter that controls radiated without affecting the fundamental and first few harmonics.  Conversely, you can add a huge CMC that takes care of conducted, but which has so much capacitance, the radiated remains strong.  Radiation can also come from the PCB itself, or between wires, without regard to conducted.

Experience has shown me that  a common mode choke is generally required on the DC output as well as the AC input. The output diodes generate a huge amount of noise when they recover.

Well, that can be part of it.  But schottky diodes are very popular, as are synchronous rectifiers.  Of course a poorly timed sync rect will exhibit something like recovery (or do even worse, if it's leading or lagging the ideal switching points), but it's certainly not limited to PN junction diodes.

I've built one flyback power supply, that makes a metric shitload of output-side noise.  It appears to be the primary switching edge, pushed through interwinding capacitance and winding resonances, showing up as ground voltage.  To the tune of a few hundred volts, if not for the Y caps.  (It's so noisy, it's unstable and squeals without the correct selection of Y caps!)

This is peculiar, because with an essentially identical circuit, and transformer windup, but made for 6.3V instead, the circuit is quiet without needing secondary CM filtering!

So it's not just primary coupling, but it's also complicated induction in the secondary winding, too.  High voltage outputs seem to be more difficult, at least for the windups I've been using with this series of supplies.

(For reference, http://seventransistorlabs.com/Images/DistAmp2.jpg shows a 100V model of the PSU I'm talking about.  The "high voltage" one is not pictured, but is made for 150-300V output.  This one is also noisy, but not quite as bad; you can see I used a nice CMC on this one's output, to deal with that.)

Tim

[Richard Head]

I've been thinking about active common mode filters whereby you sense the common mode current at the output and then actively null it out. The same could be done on the input but nothing simulated or prototyped yet.
This area is definitely worth investigating. It will never replace the filter magnetics but it can reduce the size of the first CM choke massively. The first CM choke is sized to provide sufficient attenuation at the first spike above 150Khz.(Conducted test only starts at 150Khz so many designs have the converter operating just below this).  If you inject an anti-phase signal into the gnd I believe you can achieve at least 20dB reduction of the 2nd harmonic. That should allow you to significantly reduce the size of the first CM choke core.
However, a resonant topology neatly sidesteps most of these issues.
By the way Tim, what on earth is the device in your picture. It looks like an RF amplifier.     

[chris_leyson]

Thanks Richard, a quick Google for "Active Ripple Filtering" and I found these papers, which is exactly what I had in mind.
http://dspace.mit.edu/openaccess-disseminate/1721.1/86968
http://www.hamill.co.uk/pdfs/aearffui.pdf

By the way Tim, what on earth is the device in your picture. It looks like an RF amplifier.

I was wondering too, HF distributed amplifier ?

[T3sl4co1l]

Indeed, that's exactly what it is.  It should be 40W (CW / PEP) output capacity, 30MHz BW (rise time around 15ns), and ridiculously inefficient (class A, unless you want to run it at higher voltage and get unidirectional pulses out of it).  At the moment, I have the frequency response tuned fairly well (at somewhat less BW than I was expecting, but tuning 9 inductors and 7 capacitors is a stone cold bitch in case you were wondering!).  Haven't tested it at full power yet, because I don't have a preamp to drive it!

I'll post about it on the HAM section when it's done.

Tim

[helge]

I've just assembled a list of filter components for the next round of testing. Is it just me or is the classical approach to 2-stage common-mode filtering to use a NiZn CM choke first for the 10-100 MHz components, followed by lower frequency (micro-/nanocrystalline metal powder?) CM filter.


I'll also add the reference to an article which gives a brief introduction to filter magnetics selection and optimization (*sigh*, I'm not alone):

The spectrum is usually spread from 10kHz to 100MHz. Filtering the transmission wires is a challenge, which often leads to redesigned filters, tested in a “trial and error" procedure and optimised only rarely in terms of weight, size or costs (Mayer, 1998).

Common mode chokes and optimisation aspects:
http://www.adv-radio-sci.net/12/143/2014/ars-12-143-2014.pdf

[helge]

Different SMPS, same story - and the quest continues!

I'm back with another rounding of SMPS silencing and I've actually put a VICOR V300C24T75 to the test (180-370V in, 24V out, 75W). Apparently the job ain't done with a resonant topology. I'm particularly seeing transients around the input side of things which are impropely handled.
http://cdn.vicorpower.com/documents/datasheets/ds_300vin-micro-family.pdf

As it stands the module still has some problems with the amount of capacitance provided behind the filter (330µF/400V crops up in some places, in my setup there are 6x 47µF before a CM/DM filter and another 2x 47µF behind that following implementations I've seen in miscellaneous 120W - 240W PSU designs, maybe that filter in between is not such a great idea afterall.

Since I'll stay below 0.4A in my application a slightly wasteful single-transistor gyrator appears tempting (see attached image, n001 across C6, n002 across C7). FZT653 or FZT849 might do the trick, since the circuit is mostly floating a high beta low voltage transistor is used.

Vicor also provides secondary side active hybrid EMI filters the likes of which are described here:
https://www.pes.ee.ethz.ch/uploads/tx_ethpublications/biela_APEC06.pdf

futher reading exposes different approaches which are more or less efficient. For the high frequency stuff dissipation seems to be acceptable (most of the power should by recycled by non-dissipative snubbers anyway).
http://www.ecti-thailand.org/assets/papers/504_pub_25.pdf
https://www.prbx.com/wp-content/uploads/pimfiles/899/QPI.pdf
http://www.rle.mit.edu/per/ConferencePapers/cpWPET02_achow.pdf

[VK3DRB]

"Leave them alone! They mean well."

Another silly business name, created by someone without doing his research into the English language connotations. Their power supplies are not that great from my experience, but they mean well.

But it goes the other way too.

IBM tried to introduce a machine into France called a "computer". The word was very close to something very rude in French. The name had to be changed to L'ordinateur.

IBM created a TV advertisement about 20 years ago trying to flog their global solutions. They had a poor Chinese girl in rural China speaking in a perfect American accent dressed looking like an American kid. It was an insult to all the impoverished rural Chinese who at the time had no hope of ever owning a computer. But I guess the average American viewer would not have known any better.

In the 1980's I developed the very first first computer manufacturing line in Tianjin, China. The American manager of the joint venture formally named the Chinese workers after US presidents, complete with business cards. I remember a Reagan, a Roosevelt, a Franklin. Its a similar arrogance and cultural ignorance on part of the Americans that caused so many innocent Pakistanis and Afghanis to be sent the Guantanamo Bay to be tortured byt eh US government. Why? The culturally ignorant programme of offering between $3,000 and $25,000 reward to Afghanis to turn over "enemy combatants".

[mmagin]

What came as a surprise to me is that the conducted electro-magnetic interference (EMI) specs of a power supply are only concerned with the INPUT conducted emissions. Usually one or two common mode filters with a few percent of stray inductance will cover most of the single and differential mode noise isolation needs (At least when you want to pass EMI compliance tests).

If that surprises you, you'll faint when you see how they measure output ripple.  It's like, "run twisted pair to the load, put some extra external capacitance across the load, limit the scope bandwidth to 20 MHz."  Well, of course it looks a lot better with some low-pass LC filtering!

At least it makes me pretty mad.  About as mad as I get thinking about "spread-spectrum clocks" to make EMI problems go away.

[TimFox]

I fought a rear-guard action against switching power supplies in low-noise analog systems before retiring.  I insisted that the schedule include engineering time to eliminate the power supply noise if it were truly necessary not to use a linear supply.

[helge]

TimFox that's always a good thing if there is room for it. Me, I have to fit a >= 50W SMPS (V300C24T75 on a custom board) and a HomePlug AV interface into a 38x38mm box and the interface requires some very silent 2-30 MHz.
I'm finding it hard to locate the culprit without a proper spectrum analyzer. Right now I'm only seeing a very smooth switching waveform on the secondary side with 20-50mVpp. The residual primary waveform is not visible on the 2mV/div scale on the scope (<400µVpp) when I'm coupling to the combined data / power cable.

What's very weird is that the impact of the noise is lessened by floating the PSU + communication board. It feels like a CM noise issue but attempts to bypass CM noise show no effect.

[T3sl4co1l]

Identifying noise can be tricky.  You should have a current transformer, that you can clip around a cable (or pass a cable through) to measure the CM current flowing on it.

The signal level will be quite small, so you may need a preamp as well.  (The noise floor, for a 10-100MHz bandwidth into 50 ohms, is 3~10uV, and still less into a narrower bandwidth, such as what a spec would read.)

Since the wires will act as antennas, you may get ~mV of ambient RF on them, which needs to be subtracted out somehow (mentally or otherwise).  Hopefully, the offending signals won't be at the same frequencies, so they can be identified.

Bypassing or grounding shields, enclosures, circuit grounds, etc. to a ground plane working surface can help.

First of all, note that grounding is orders of magnitude more successful than trying to pile ferrite beads on a cable.  A ground strap can have an impedance ratio of 10-100, i.e., a cable has a characteristic impedance above the ground plane in the 100~300 ohm range, while the strap can have a parallel impedance of 1~30 ohms to ground.  Whereas ferrite beads are rarely above a factor of 10, even for several turns through a large core, measured at the frequency where impedance peaks -- a series impedance of ~kohms. 

Of course, combining both gives you geometrically greater attenuation, which is what's so awesome about shielding combined with chokes.

Secondly, once you can isolate and shield independent sections of the system, you can reduce the network of cables and boxes to something more manageable.  An extreme example would be each box having a stack of ferrite beads at each connector, and the middle sections of the cables having their shields all shunted to the ground plane, for some distance.  So each interconnection between boxes has an L-C-L filter topology, where "C" is a direct ground strap, or at least some RF bypass caps, and the "L"s are ferrite beads.  Thus, each box is isolated from its neighbors (common-mode-wise), and the CM voltage is easily measured by adding a "sense wire" loop through the stack of ferrite beads on that side.  Furthermore, if the circuit ground inside each box can be bypassed to the ground plane, then the CM voltage, from each connector, can be measured independently -- no need to resolve loops and track down the one or more offenders!

Since that's a total PITA... keep in mind that EM waves obey the sausage effect.  More ferrite beads on one cable, can make another cable worse, because that CM voltage or current has to go somewhere.

Tim

[helge]

Here's a quick snapshot that shows a transient coupled to the shorted grounding strap with two planar E cores clamped around twisted input leads to the Vicor SMPS board.
The second channel (for trigger, hidden) has been picked up around the input filter to ensure the switching action is the actual cause of the transient. There are other stray fields that couple into the wire loop but they are not synchronized.

regarding the sudden end around 50 MHz... didn't have a proper scope around and had to use a 40 MHz DSO.

[helge]

Realizing that maybe the PH50A280 performance is not so bad at all I looked for CM noise on the PSU direct output and the effects of a two stage CM filter leading to a PTN78000.
While the on-board output filter of the PH50A280 board is quite good it is now dedicated to the sensitive RF section. All other components must be connected directly to the output filter capacitor, prompting the need for more filters downstream.

A 5V rail is generated with a PTN78000 from the PH50A280-24 output.

Here is some common-mode mess before the "external" CM filter (In the initial post I was talking about 0.5Vpp, haha):

No wonder the RF component had some "minor issues",

here's the filter:

component specs:


and here's some combined junk from behind the filter and some transients picked up on top of the higher PTN78000 switching frequency.

[Richard Head]

Helge
I'm glad you are getting some improvements with your filtering attempts.
But, you committed a cardinal sin! You passed the input and output wires of the filter right next to each other. That allows the noise to couple from the noisy side back onto the clean side bypassing the filter. Always keep the the input and output physically separated unless they are screened. Also, for really stellar filter performance place the filter components in a shielded enclosure ideally with each section in its own sub enclosure. Basically, use tried and tested RF techniques. With these methods one is able to get the conducted and radiated noise levels down to the noise floor of the spectrum analyser/measuring receiver.

[helge]

I hope there is still room for amends  It's a quick and flawed test but it already shows

a) that i forgot to attach the center 1210 DM cap.... and that the long ground straps are rubbish
b) even a crude CM filter eliminates most of the 1-10 MHz ringing with formerly extreme amplitude
c) the high pass filtered version of the transients that are situated directly at the switching pulse edge are still present and also couple into the leads -> better probing technique is also required in addition to a sane and RF compliant arrangement with proper shielding around cables and power converters.

I'd rather put that filter pretty close to the power supply where it belongs and maybe select simpler CM filter chokes for the far side around the DCDC converters.

And then it would be sweet to get a new scope from Keysight, scopemonth, *sigh*



ps.
I also have one of these on my desk, "EN 133200"

they're not even trying, are they?