4
\$\begingroup\$

We are having trouble with a CISPR-11 (EN55011) Radiated Emissions Class B compliance test in one of our solutions. The troubling part of it is basically an industrial motherboard (with a commercial SSD and commercial RAM) within a well grounded metal enclousure, sharing space with an industrial switch, a switching 12V industrial power supply and a industrial touch panel and monitor. Every single one of these industrial components have its corresponding compliance certificate for CISPR-11. The antenna picks up the following waveform (Vertical orientation, 3mts distance, 30 to 200 MHz bandwidth). Crosses above the threshold are non-compliant peaks. non-compliant peaks

The main offender is a 60MHz peak that looks like a strong square wave in frequency domain. That is new to me in these tests; I am more used to specific harmonics in the spectrum. If we turn off the motherboard (and thus also the RAM and SSD), the 60MHz and also those crests at ~80MHz dissapear. Turning of the rest of the components (except the 12V PSU) has no effect.

I have the motherboard's compliance test and procedures, and their peaks are nowhere near in power or frequency. They also used commercial RAM and a mechanical commercial hard drive.

Has anyone seen this waveform before for radiated emissions and/or has any idea, given the description of our system, what could be generating it?

EDIT: Based on comments and answers, I will clarify some things. First, I have 1 Ethernet cable (unshielded) inside the case, connected to the switch, also inside, but then 3 Ethernet SSTP cables going out from the switch to 3 ethernet Basler cameras. These 3 do have a ferrite clamp on each, but I haven't fiddled with the one inside. I'll note that down.

The SMPS was our main target for the whole ordeal but I still need to try ferrite beads on its outputs and maybe enhance the shielding around it.

Unfortunately we do not have our own spectrum analyser and our scopes are up to 60MHz. We do not develop high frequency custom electronics, so they were not neccesary. Given the circumstances, we may adquire one and do our own pre-compliance testings, as many of you suggested.

I will keep this updated.

EDIT2: So I tried making a makeshift antenna out of the scope probe, and we identified a single spot in the LCD screen using FFT, which is directly in front of the motherboard, with a peak exactly at 60Mhz. I am quite sure we found the place where the signal is originating, though still couldn't find the culprit. I tried everything suggested here, and altough they didn't quite fix this particular emission, all suggestions are greatly appreciated as the will surely enhance compatibility.

We identified, however that the screen enclousure is plastic and in some spots (including the one with the 60MHz peak), there's no conductor material between the interior and exterior of the whole enclousure, so next step is to place a metal sheet between back of the screen and everything else. I will post the results for anyone interested.

\$\endgroup\$
8
  • \$\begingroup\$ switching edges associated with a SMPS. to give you an idea, risetime around 20-50ns will produce components around the 20-40MHz \$\endgroup\$
    – JonRB
    Mar 19 at 20:53
  • 4
    \$\begingroup\$ Do you have your own spectrum analyzer with E-field probes? While it is impossible to test for compliance with such equipment it should be fairly straightforward to pick up the culprit frequency and then "sniff" around the board for peak amplitudes. \$\endgroup\$
    – Arcatus
    Mar 19 at 20:57
  • 1
    \$\begingroup\$ Was there a data cable externally connected to the DUT? This is sometimes the biggest offender \$\endgroup\$ Mar 19 at 21:30
  • 1
    \$\begingroup\$ Try disconnecting some ethernet cables (or disabling interfaces). They're a good source of radiation, especially if the jack isn't well connected to the chassis, and (speculation) may not have been fully active when the device in question was tested. Not that anyone would cheat like that on purpose. I've seen passing equipment fail at hundreds of MHz, only when a cable was plugged in * and looped back * \$\endgroup\$
    – tomnexus
    Mar 20 at 1:24
  • 3
    \$\begingroup\$ your progress update is much appreciated. that way others can learn from this question. \$\endgroup\$
    – tobalt
    Jun 24 at 18:19
3
\$\begingroup\$

It looks like roughly 5MHz impulses with a resonance at 60MHz to me judging by the sidebands (20 per 10 MHz) but could be a 60MHz clock with roughly 5MHz recursive sidebands.

The fastest way to find the offending signal is to use a small loop antenna probe to DSO or scope (Coil or wire in a small loop) to sniff out the near-field noise. If you have 2 sized loops you can get a quick look then a smaller loop to get near the chip, trace etc. 60 MHz is about the size of a small room resonant frequency.

If indeed it is the data on the external cable then a Balun on the cable is a valid fix for that until you put a CM choke or SMT PHY on the data stream on the MOBO.

This is also the best practice for pre-test scanning the signature emissions before paying for the unintended radiation tests at 3m at an outside lab. I would use a current loop probe for near-field inside the box and an E-field probe or just a wire on a coax end into the Spectrum analyzer for a broadband spectrum. Antenna efficiency drops for longer than 1/4 wavelengths but it is still broad spectrum.

\$\endgroup\$
1
  • \$\begingroup\$ Thanks, I will try fiddling, removing or shielding the data cables. \$\endgroup\$ Mar 20 at 23:13
2
\$\begingroup\$

It's not that easy to spot the source of that radiation.

If the test program running on top of the OS is not reading or writing files from/to the SSD than the problem might be the RAM module.

Try a shielded RAM module.

I understand that in production you will have to use the unshielded RAM module but the problem now is spotting the source of radiation.

If you have access to the BIOS you may try to slow down RAM accesses.

Try using RF gaskets to cover the case enclosures.

You are not far from compliance.

\$\endgroup\$
1
  • 2
    \$\begingroup\$ Actually with an SA , its really easy to find these unintended radiators up close outside and inside the box. The Friis loss and radiator shape changes the spectrum from far field at 3m to near field and 3cm but finding the source is not hard. Once it's found fundamentals determine the desired fix with bandaid foil and ferrites until a design correction can be made. \$\endgroup\$ Mar 20 at 1:00
1
\$\begingroup\$

I would suspect the power supply - specifically, ringing of the SMPS switching waveform and / or harmonics of that sneaking back into the conducted path.

Try using CM filters (clamp-on beads) on the PSU's input and output feeds. Also, try applying an R-C snubber to the PSU switch side.

Also, try shielding the PSU's inductor.

There's a poor-person's way to sniff out these kinds of problems: if you have a scope with FFT capability, make a near-field antenna by attaching the ground lead to the scope probe. You now have a cheesy RF probe. Then, go hunting for that mystery 60MHz. My guess is it'll be coming from the PSU's inductor.

Want something fancier? Make your own probe: https://www.eng.mu.edu/~richiej/seminar/aidi.pdf

Or buy a set: https://beehive-electronics.com/probes.html

Considering what lab time costs, this is a prudent investment.

\$\endgroup\$
7
  • \$\begingroup\$ 60 MHz seems a tad high for a SMPS with low Q emission and 5MHz sidebands but perhaps loads on the supply. \$\endgroup\$ Mar 19 at 21:23
  • \$\begingroup\$ The pulse width for the PWM signal can be as small as ~100ns for a high switching rate and stepping ratio. 60MHz would be a sixth harmonic. I've seen ringing/oscillation stuff like this on some SMPS, especially when they go into DCM mode. \$\endgroup\$ Mar 19 at 22:51
  • 1
    \$\begingroup\$ Fourier spectrum doesn't work like that. With any pulse width measured at 50% peak, or PW50 , it is the 10% to 90% rise time that creates all the harmonics and there are nulls for every harmonic f that fits a full cycle within your pulse width or PW50. This means a PW50= 100ns has harmonic nulls at 100ns . Let's say the fundamental was 1MHz wit h10% duty cycle then there are NULLs @ 10MHz, 20 MHz,30, 40,50 and 60Mhz. etc. Then the harmonic with half power is 0.35/Tr rise time from 10 to 90%. So a 50% duty cycle or 500ns has nulls at all the even harmonics \$\endgroup\$ Mar 20 at 0:32
  • \$\begingroup\$ I like those poor-man options, thanks. I will try those with what I have at hand. \$\endgroup\$ Mar 20 at 23:13
  • \$\begingroup\$ At 60MHz (radiated emissions) the problem can't be the Power Supply that has a clock of 200 kHz. The harmonics of 200 kHz have virtually zero energy at 60 MHz. \$\endgroup\$ Mar 21 at 8:03
1
\$\begingroup\$

The 60 MHz is fairly low-Q. That suggests to me that it is a self-resonance of something like a filter component or a looped PCB track or something.

I'd suggest you remove the RAM and SSD and see if the MoBo still has the same 60 MHz spot - the clocks will still buzz. If it does, contact the manufacturer and push on them that their product does not meet its spec, they might have something to say in return.

\$\endgroup\$
2
  • \$\begingroup\$ I did contact the manufacturer and replied with a copy of their own succeful compliance test. There's not much I can say against that. I think its self resonance too, I will update when we have the metal shield for the lcd screen. \$\endgroup\$ Jun 25 at 19:23
  • \$\begingroup\$ @JavierCambiasso You could send them your own scan and say "Look, your product is defective. The scan you sent is different from this board." But I'd suggest removing the RAM and SSD first. \$\endgroup\$ Jun 25 at 19:28

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.