Troubleshooting EMC results from a lab

[trevwhite]

Hi all

I have some EMC results back from a lab concerning a current project. It has a couple of places where it has failed radiated emissions. I was hoping I could use a LISN and CDN on cables to track down sources of noise. But I do not really feel confident I can track down all the noise and the EMC lab is asking for a lot of money to retest so I want to be confident before I go back.

One source of noise at 190Mhz I do not actually see on my CDN unit so am thinking the noise might not be on that cable. If I use my near field probes around the circuit I can not find a place where noise at that frequency is much greater than other frequencies. I can find places where I get noise in 100Mhz - 200Mhz but not somewhere where there is a large peak of 190Mhz only I do not currently see.

It seems very tricky.

Anyone got any suggestions, help guides or even know of cheap pre-approval labs in Midlands area of UK that might be able to help?

Thanks

Trev

[tszaboo]

Get in touch with Würth. They will send an application engineer, who will propose (their) part numbers, and solve these issues. I'm very happy with them. The parts might be more expensive, but failing the test again is probably more expensive. And I'm no EMC specialist.

[mikeselectricstuff]

Sending stuff off and hoping really isn't the ideal way to do it.

It's better to go there yourself with a box full of tools and parts, so if any issues are identified in a preliminary scan, you can try fixes there and then. if it doesn't fully fix it, at least you can do experimentation to see what things make a difference and eliminate things.

The only difference between a "pre approval" and a full approval is their level of liability ( and the cost).

It's usually exactly the same kit, facilities and staff. You can typically book a day or half-day and do whatever you can within that time - if you get it all sorted, the test results are just as valid.

This is especially useful if you have multiple variants or products, as you can test as much as you can in the time, and also be fixing one thing while another is being tested.

The only place I've used is RN Electronics in Essex https://www.rnelectronics.com/ but I'm sure there are plenty of others
 

[Tomorokoshi]

What clock rates do you have in your system? Look for an integer sub-multiple of 190 MHz. Can you post a view of that scan spectrum?

[trevwhite]

Thanks for the replies 

I believe I am now booked in for tomorrow at a lab to do all my tests. I was really hoping to be able to do some work before going there. I want to be confident I can solve the problems when I get there.

[fcb]

If you get really stuck trevwhite, and it's not too big - we have an small anechoic chamber, 8591EM, 11940A/11941A probes etc.. in our lab.  Your welcome to noddle about for a day (near Ipswich).

[nctnico]

The trick is to use a near field probe and identify where the frequency band is generated. If I read your text so far think a switching PSU is the cullpritt. If that is the problem area you found with the H-field probes then you can try to put a ferrite bead (with low DC resistance and a peak impedance at around 190MHz) in series with the rectifier diode or add a series resistor (around 5 to 15 Ohm) in series with the gate of the MOSFET. If the MOSFET is internal then you can also put a resistor in series with the boost capacitor. You'd have to check if the PSU still behaves well (doesn't get hot) under full power.

[trevwhite]

fcb: That is really good of you thanks. Did you make your own chamber? Do you have any information on its build?

nctnico/All: The system consists of two units. The lab results I got were the two units together which did not help. I should have really gone with the units but that is another story really...

Anyway the one unit is an STM32f429 running a full colour screen powered from an off the shelf 5V mains input power supply. The uP is runnning up at the 190Mhz frequency but its an internal PLL so not sure about that. It does interface some external memory and this is possibly the route of some noise as I think this might be interface at 90Mhz. Not sure if its down to tracking or not because I have not been able to isolate any peak noise at that level. I have seen a spread of noise from 100Mhz to 200Mhz but the 190Mhz was within that range so I would have thought the other noise would have been picked up as well on the original test results. That's why I say I cant find any 190Mhz noise that is stronger than other noise. There is a switch mode power supply providing 20mA or so to the display backlight but the switching on that is lovely. No ringing and a slope on the switching so I think it should be quiet there.

The second board is powered from an off the shelf, open frame 48V. The board is an LED driver based on a constant current step down buck converter. This is all boxed within an aluminum extrusion. There is mains coming in and leads going out to the LED board it drives. That is why I thought it was noise emitting from the cables. I believe this is the source of the 60Mhz noise. Now I did actually add series resistance to the FET gate and it did make a significant difference and I was quite happy. I also put a common mode choke from the 48V to my board but that seemed to make no difference.

My measurements are all done with a very old CDN unit and also a bit with a 5uH LISN.

Any other suggestions on how to compare emitted noise with that shown on my CDN? Anyone got information on how the two should correlate if at all?

Trev

[fcb]

Nope - we bought it 2nd hand.

I think it's an M&A Instruments "micro anechoic chamber" - it weighs half a tonne and is about 9ft x 4ft x 5ft or thereabouts, it sits on a six wheeled base.  You can get in it...  I'd hate to think what they cost new.

[Andreas]

It seems very tricky.

It usually is.

There are "antenna resonances" which can amplify a harmonic of a fundamental frequency
 in far field (so your near field probes only detect the fundamental).

So I would look for a fundamental frequency fitting to the 190 MHz and any structure
(housing or cable length) that has its lambda/4 or lambda/2 near 190 MHz.
so 0.8 m or 0.4 m * shortening factor (NVP).

with best regards

Andreas

[dmills]

Any slots or painted metal to metal in the case about 35cm long?

If the CPU is clocking at 190MHz, and that is where you are seeing the fail, I would be paying close attention to power and ground around that part, also anything being clocked at around 64MHz would be suspicious (Third harmonic).

A 90Mhz DDR memory will tend to draw current pulses on both clock edges, look to your layout and decoupling, usually obvious when things just dont work, but....

If it is in fact radiated then just making up a dipole or small yagi for 190Mhz and doing the open field site thing can get you hints. Copper tape and posterboard are your friends for making the aerial, and copper tape is your friend for closing up possible leakage paths.
 
Regards, Dan.

[mikeselectricstuff]

Any slots or painted metal to metal in the case about 35cm long?

If the CPU is clocking at 190MHz, and that is where you are seeing the fail, I would be paying close attention to power and ground around that part, also anything being clocked at around 64MHz would be suspicious (Third harmonic).

And decoupling - at 190MHz you may want some small caps as well as the usual 100n decouplers.
Isolating the power to a noisy MCU with ferrite beads can also help

if you can, tweaking clock frequencies and seeing if the peaks move can be a useful way to narrow down the cause

[trevwhite]

This is all really great advice. Thanks

I have done some more testing w.r.t. to the gate resistor of the FET on the LED driver. I used a value of 10R initially and it generated noise. This is all crude stuff by me but I have included a screenshot. Blue trace is the 10R resistor.

I changed the resistor to 47R and you can see the difference. This is from a 5uH LISN between the 48V power supply and the LED driver board. It is not a quiet environment, it's just on my bench, etc but as a comparison of the effect on a conducted cable, I believe it shows something real.

[trevwhite]

This is one of the measurements the EMC lab took

[AndyC_772]

Your peak around 190 MHz isn't a narrow spike, so it's unlikely to be directly caused by your microcontroller or its memory. Those devices are driven off a crystal and will generate noise at very specific frequencies which are exact multiples (or fractions) of the crystal frequency.

My first suspect would be your LCD display; they're notoriously noisy. Does the radiated spectrum change according to the displayed image? (Try an all black or all white image and see what happens).

[mikeselectricstuff]

But make sure the gate resistor doesn't kill efficiency or cause heating/melting of the FET. It's a balance between EMC and still doing anything useful!

[fcb]

That's quite a smudge.  190MHz seems to high for SMPSU noise.  LCD screen drive cable (ribbon)? - copper foil, ferrites and a near field probe might get you somewhere tonight.

Your EMC lab should have an EM investigation kit (usually an older spec/an with a near field probe on it), or perhaps a flat-bed EM scanner (surprisingly useful).

Just remember that EMI is like a squeezing a balloon, you squeeze it out of one area, and it'll pop up somewhere else - so just when you think you've solved it...

[coppice]

Any slots or painted metal to metal in the case about 35cm long?

If the CPU is clocking at 190MHz, and that is where you are seeing the fail, I would be paying close attention to power and ground around that part, also anything being clocked at around 64MHz would be suspicious (Third harmonic).

And decoupling - at 190MHz you may want some small caps as well as the usual 100n decouplers.
Isolating the power to a noisy MCU with ferrite beads can also help

if you can, tweaking clock frequencies and seeing if the peaks move can be a useful way to narrow down the cause

Its not just the value of the decoupling caps that is important. If you really want to see if decoupling is the issue, get a small value SMD cap soldered right across the supply pins of the MCU. Even small lead lengths add enough inductance at 190MHz to spoil the effect of the caps.

[nctnico]

190MHz can be SMPS noise! I've seen SMPS noise in this frequency range before several times. If it is noise from a clock then it will be a single spike.

[dmills]

I have seen insufficiently snubbered flybacks  ring like a bell in the tens of MHz, not sure I have seen one go quite that fast, but it is possible.

That broad mess looks like something data dependent to me combined with an unfortunate but fairly low Q resonance, a graphic LCD is entirely possible as a source, but if you have screwed up badly enough it could be memory data or even an unfortunate power plane resonance.

Make up an H field probe out of some small hardline or handiform and use a pawn shop scanner tuned to 195MHz or so to have a quick sniff around.

I had not considered that such a thing as an E/H field flatbed scanner existed, I **MUST** build one... Pass that old flatbed plotter and some small coax, I think I have a 'project'.

Regards, Dan.

[Andreas]

If it is noise from a clock then it will be a single spike.

Hello,

Modern ECUs partly have spread spectrum oscillators to distribute the peaks over a wider frequency area.
Dont know if it is already active here.
Perhaps the problem could be solved by software.

If its the power supply a small capacitor over the Schottky-Diode could reduce the radiation.

with best regards

Andreas

[nctnico]

I have seen insufficiently snubbered flybacks  ring like a bell in the tens of MHz, not sure I have seen one go quite that fast, but it is possible.

Actually there are three potential sources for generating noise by an SMPS: MOSFET turn on edge, ringing due to the inductor and the rectifier diode's reverse snap off. The latter can easely get in the 100Mhz to 200MHz range especially if it is a fast diode.

[dmills]

And here was me thinking that nobody used snap diodes in that sort of power supply.... But yea, I could see reverse recovery doing that if somebody picked a fast diode without soft recovery.

The fact that all that UHF grass goes down with a little added gate resistance is pointing to something like this for that part of the problem.

Regards, Dan.

[Pitrsek]

Off the shelf power supplies.... have you tested them separately?  Claiming compliance, and being compliant are two way different things. It happens more often than you think. At 190MHz, your regular x5r/x7r capacitors are not capacitors anymore. Would not bother with adding more capacitors, adding multiple values only adds more resonant peaks.
What's your board stackup?
Is everything going to and from the board filtered?
Are your connectors close to each other?
Do you have a solid ground plane?
+1 for the LCD team, this would be first thing I'd test.
How is the LCD connected to rest of the device - especially return/ground?

[mikeselectricstuff]

If you're not supplying the psu with the kit, it makes sense to usr the quietest one you can find. For 48v, telecom rectifiers are awesome.
Most SMPSUs are characterised with a resistive load, so anything else can affect their EMC performance significantly.

If possible, use a linear PSU or batteries!