doing in house EMC pre-compliance testing - where do I start?

[Simon]

We had a product at work fail EMC in CE marking. So we have bought a spectrum analyser and will be getting the required licence on it for the required filters etc. We have an aerial, and that is all I know.

How do we go about setting this up for the pre-compliance the spectrum analyser claims it can do? The analyser itself has some sort of output as well as input. Is this so that it can self characterise the setup and make the required adjustments to get a roughly good metric against the standard?

I am sorry to say that Telonic have been pretty useless in advising on what we need to the point that I have an aerial but no cable for it, they must be doing so well that dealing with customers in order to make a few more quid on some cable is totally not worth their while.

[fourfathom]

What kind of antenna?  What kind of product?  What was the failure (frequencies, amplitudes)?  I assume this is a free-space test.  And what spectrum analyzer?  What antenna?

I used to do FCC and European (pre-EU) pre-compliance testing for our telecomms gear, and before that for wireless microphones and receivers.  At the simplest, I used a spectrum analyzer, several antennas (a telescoping dipole for the lower frequencies, and a biconical for the higher ones, and sometimes a log-periodic for still-higher frequencies).  Each antenna has a calibration chart.  You may need a preamp -- that and the coax cable also need to be calibrated, or at least have known gain/loss at the frequencies of interest.  My test range was sometimes a grassy field, or at a different company where we did a lot of this we laid down a wire "hardware cloth" screen under a layer of asphalt.

So that's the setup.  This was before the era of modern spectrum analyzers with pre-loaded compliance limits and calibration factors, so I had to do that part by hand: identify the emission peaks apply the correction factors, and compare that to the compliance limits.

I highly doubt that your analyzer can self-calibrate the measurement system using just a measuring antenna.  You're probably going to have to enter the calibration factors yourself.  Again, you need antenna factor calibration data.

If you have the test results for the failed run, you may be able to useful mitigation without a calibrated system.  Just find the specific emissions (frequency, dB over limit) and measure the unmodified system with whatever you have.  Start doing your EMI fixes, and see how much you can drop those specific emissions -- this is a relative measurement, so calibration isn't critical.  If you failed by 10dB and can reduce the emissions by 20dB then you are probably ready to re-test.

The test results can be greatly affected by external cabling to/from your DUT.  Be realistic, don't cheat, and try to maintain the same configuration between tests.

[Simon]

Yes my first aproach would be to look at the system as it failed and use that as a metric against the failure test. then change stuff and see how much better I can make it. First order of business is sniffing around with a near field probe to locate the cause.

I have a tekbox biconical antenna, yes it has a chart of gains at different frequencies so that the analyser can take that into account. What I am confused about is how do I actually set up to do a test with the analyser that will apparently give me the ability to compare to standards. If I am not set up right then what I am doing is meaningless.

[fourfathom]

Yes my first aproach would be to look at the system as it failed and use that as a metric against the failure test. then change stuff and see how much better I can make it. First order of business is sniffing around with a near field probe to locate the cause.

I have a tekbox biconical antenna, yes it has a chart of gains at different frequencies so that the analyser can take that into account. What I am confused about is how do I actually set up to do a test with the analyser that will apparently give me the ability to compare to standards. If I am not set up right then what I am doing is meaningless.

Quick response!  See my additions to my previous post.

There has to be a "How to do an open-field compliance test" tutorial out there somewhere.  It's really pretty simple:  your correction factor is (Antenna factor @ Freq Of Interest + preamp gain @FOI + cable attenuation @FOI).  See what the test range distance is for your particular requirement, and put your antenna there.  Make measurements, probably with vertical and horizontal polarizations (antenna rotation) at the higher frequencies.  Usually your DUT is put on a carousel, but one way or another you need to test at different orientations.  Compare measurements to the standard. 

Your spectrum analyzer probably has compliance limit templates and the ability to enter correction factors, but if that's your question -- sorry, I've never used that stuff.  The last time I did these measurements was perhaps 35 years ago.  But I guarantee that antennas, coax, amplifiers, and radio propagation haven't changed all that much since then.

[Simon]

So I replicate the distances of the standard and use the correction factors of the antenna plus losses/gains of stuff in between and I'm sort of doing the same as the test house.

Interestingly I actually have a 9kHz to 3GHz signal generator so is there a way of testing my setup by generating a known signal and see what I get.

[fourfathom]

So I replicate the distances of the standard and use the correction factors of the antenna plus losses/gains of stuff in between and I'm sort of doing the same as the test house.

Interestingly I actually have a 9kHz to 3GHz signal generator so is there a way of testing my setup by generating a known signal and see what I get.

I've never done it, but you can certainly set up two antennae on your test site, transmit into one and receive on the other.  Of course you will need antenna and coax calibration data for both ends, otherwise you won't be able to separate antenna factors from site loss factors.  I expect that someone has a very small "pseudo-isotropic" antenna with calibration data over a wide frequency range.  It would probably be too lossy for receiving, but it could be a useful test signal source.  Perhaps a design is out there, and if done right, the calibration could be simple math.

But if you are doing outdoor open-space measurements the field (ground under your feet) requirements aren't too difficult.  When I was doing this, the FCC's own site was essentially a grassy field, probably kept well-watered.

I hope someone with more recent experience chimes in here.  No doubt some techniques have changed since I was in the field.

[srb1954]

So I replicate the distances of the standard and use the correction factors of the antenna plus losses/gains of stuff in between and I'm sort of doing the same as the test house.

Interestingly I actually have a 9kHz to 3GHz signal generator so is there a way of testing my setup by generating a known signal and see what I get.

You also need to rotate the UUT to pick out the peak emissions at each frequency. Since the angle for maximum emissions may vary for different frequencies the usual approach is to set the spectrum analyser in peak hold mode and conduct multiple sweeps as the UUT is slowly rotated. Once the frequencies of the highest amplitude emissions are identified from the peak hold scan you can then zoom in for a more detailed look at each emission.

[Simon]

The test house did it at every 90 degrees which makes sense.

[martinr33]

I've been out of this game for longer than everyone, I suspect My experience dates back to the first days of failing EMC compliance...

The short of it is, don't try to replicate the test setup, but use lab tools and your SA to find the porblem emitters.

1) It is tough to replicate the failure conditions unless you have a low-noise space (like a field with limited wireless noise, so away from TV and radio transmitters, cars, and houses.and the ability to rotate the device. But this is less of an issue if you know where the device is failing.

2) Therefore, if you know the failing frequencies you can use near field antennae and your spectrum analyser to figure out where they are coming from.

3) A technique I found effective was to use a scope to probe around the board, and feed its output into the spectrum analyzer. When you find a strong signal on the scope with harmonics in your zone of interest, you have a candidate.

) If you are lucky, your system will have just a few problems. If not, you will be chasing multiple problem frequencies.

[fourfathom]

Martinr33, that is good advice.  But there are often two issues:  Finding the source, and then finding out how that source is escaping to the outside world. 

Sometimes you really can't do anything with the source -- for example a widely-distributed clock in a large system (such as the PBX I had to bring in compliance when the FCC Part 15 rules for this were introduced).  It wasn't practical to filter all the clock signals without radically redesigning all the board/backplane interfaces, so instead I had to put RF gasketing on the enclosure doors and treat all the exiting cables with ferrites (in doing so I believe I invented the "ferrite clamp" core -- I had the ferrite company make me a bunch of E-I cores in an RF mix).

Other times, you have a small device with a few or no external cables, and then you can often fix the source at the root.

But regardless, the first step is to identify the source or sources.

[martinr33]

Not quite on topic, but on my problem unit we tried:
 - conductive paint spray on interior surfaces
 - thick conductive coating on interior surfaces
 - (my favorite) copper plating inside and outside of  the plastic case with metallic copper (ands no, it did not help enough).
 The fix was a tiny capacitor on a long signal line. 

To your point, the problems stem for the sum of smaller parts, and a few single culprits. Design tools fix the worst problems, but the final assembly is another matter.

[KE5FX]

Are you able to share your failing test report?  The compliance lab should have given you some paperwork showing exactly what mask was violated and the frequency(s) at which the violation(s) occurred.  That data is the best starting point.

[Simon]

Are you able to share your failing test report?  The compliance lab should have given you some paperwork showing exactly what mask was violated and the frequency(s) at which the violation(s) occurred.  That data is the best starting point.

We have overshot the line by 20dB at 60MHz, this falls back to the line at 10MHz above and below. I have two candidates, 2 maxon motors and controllers that have no filtering at all and a 3m RS232 coms cable that although screened did not have the screening terminated. Either could be the source.

So yes first steps are near field probes and go around it to find the issue location. But as we fix things it would be nice to have the ability to do a test similar to the one that is being done in the lab.

[miro123]

Hello,
1. Can you perform all CE pre-complaince tests on your device?
   1. radiated - Wideband receive antenas has start price of 5000$. Most of the time you can avoid the farraday cage that add another 200...800K$
   2.Immunity 
     - conducted is easy, it add  500...1000$ rf transformers from e-bay + hobby RF power aplifoers also from e-bay
     - for higher frequency get more expencive.

2. De you get advice from CE lab what get wrong, what are the solutions? - my experience is this labs has good educated test engineers. Normally they guide me how to solve my EMC problems.
==============
Normally I perform the EMC compliance tests in 3 steps
  - pre compliance test on our ems lab ~100K$ in second hand uncalibrated equipment
  - unofficial compliance test - EMC lab near our location - they have calibrated equipment/ antennas and faraday cage.
  - official test - just to get paper

I think that you should post in other forum section, where more EMC experts can advice you.

[kleiner Rainer]

Hello Simon,

buy "Electromagnetic Compatibility Engineering" by Henry W. Ott. Read it - its worth the time and money. Its the EMC bible, period.

Get hold of CISPR14, Part 1 and 2. The Indian version is an older edition, but free:

https://law.resource.org/pub/in/bis/manifest.litd.9.html

CISPR16, Part 1 and 2 describe measuring apparatus and methods. Another important read.

Greetings,

Rainer (doing EMC work for the last 25 years...)

[T3sl4co1l]

I don't feel antennas are all that helpful, outside of a lab setting; you aren't going to have the same gain calibration (due to antenna curve, environment, orientation, etc.), and ambient sources get in the way.  Just a good set of probes (E and H field, current clamp), and decoupling networks (LISN, CDN), covers quite a lot.

The main thing you need to know is, what source is blowing it out by 20dB, and, can you reduce it by at least as much?  If the probe says, whatever say 43dBuV, in your reference orientation, then you make a change and it says 17dBuV -- that's probably a good start.  You still need to know that that reduction was made in a location accessible by ambient fields (coupling to radiation).  Usually this means cables and such.

Note that LISNs aren't useless at 30MHz, that's just where the conducted test happens to end.  You can see quite a lot carried on cables, before it gets into the air.  And a current clamp probe, you can think of as an inline LISN of sorts.  E/H probes help with further pinpointing sources, but mind they tend to be rather low gain, not so effective at tracking noise along cables (where a larger probe, or antenna outright, might indeed be more illuminating).

About the test itself: was this sent off and done by independent techs, no one went along to see what it's doing or to troubleshoot it?  If so, that's rather troubling.  You can only guess what the noise is coming from, and where it's getting out.  Hopefully there's only one thing making a given cluster of peaks; but multiple sources can overlap, and you won't know which.  Nor what connections are likely radiating; hopefully here too the set of suspects is small, but noise can show up everywhere and a little direction goes a long way.

There is at least polarization, assuming you know how the stuff was laid out for the test (got photos from the lab?).  And assuming V/H are listed on the plots, not just merged together(!).  Unless everything's all looped up (circular polarization?) in which case nevermind I guess, heh.

Yes, unshielded cables are a likely culprit.  A cable having screening/braid is meaningless; only if it's tied to local GND at both ends, is there any containment of CM noise within that path.  Other things relatively easy to spot: any connections passing through a bulkhead/shield without grounding or bypassing to it; any PCBs having onboard ground loops between connectors and the main GND plane, or enclosure; etc.

Also, was this just emissions?  Did susceptibility pass?  Is it going back for that later?

I don't know what kind of motors or controllers those are (Maxon is just a brand), but definitely a possibility.  Anything to do with power is an immediate suspect.

RS232 for example, should be very quiet at ~60MHz (unless maybe the driver is shite -- I suppose the charge pump could be badly made or something?), and that's if it's transmitting at all, and at much baud -- it's slew-rate limited, so only the very corners of the waveform correspond to harmonics of any consequence (i.e. in the 10s of MHz), and those corners are very small indeed (i.e. the ~100s of mV during which the signal changes direction between flat and slewing).  RS232 isn't shielded so much for emissions reasons, as immunity: it's a fairly high impedance, so an external electric field couples into the wire at near full scale, corrupting the signal.  With a noise margin of 5V+ with typical interfaces, it should pass commercial limits (say 3V, 3V/m) well enough, but almost certainly demands filtering and/or shielding to pass more stringent, industrial / etc. levels (say 10V+).

Tim

[fourfathom]

Yes, unshielded cables are a likely culprit.  A cable having screening/braid is meaningless; only if it's tied to local GND at both ends, is there any containment of CM noise within that path.

Cables are often the main avenue for emissions to escape a product, and (as T3sl4co1l mentions) even shielding may make no difference.  Often the radiation is common mode, carried on the shielding itself.

You need to consider two things:  The source of a signal, and how this signal is radiated.  If you can keep the source dimensions small enough you may not even need shielding.  A signal source needs an antenna to radiate, and cables are often your main antennae.  So see if unplugging the external cable makes a difference.  Or, use a ferrite clamp on the suspect cable, as close to the DUT as you can get.  Use the proper ferrite mix, and enough turns on the core to be effective -- you aren't trying for a production-ready solution at this point, just searching for the problem.  Later on you can consider cables with built-on ferrites, or common-mode chokes on the circuit board just before the cable connection, or bypassing (but bypassing is often less effective that you would hope.)

I would be happy to point you to good info on suppression ferrites.

[Simon]

Well on the one hand the motor controllers having no filters is quite a red flag, they are 50/8 models or 50V 8A, the person that designed the main board did not think filters were required but spent a long time navigating around ground bounce..... (yes they walk among us...). The RS232 coms happen at 12V over a 3m cable so quite the aerial.

I actually suggested to my boss that we obsolete this product version and come out with a revised one as I have no wish to fix something done so badly that I'd have to start again anyway. Well I have already started again on a bigger brother to this product so once that is working it's just a case of switching the controls out for ours and just having a main board with filters and any power stuff on it being controlled by our new controller. The hard stuff would be done as part of the new project, retrofitting it could be easy.

I still need to work with this product as much as I can and hopefully tell guide the person that designed it to get it right this time and we have some breathing space.

[martinr33]

This does not sound like a motor problem. Motors run at kHz and their inductance fights fast riding pulses. And RS232 is not that fast.

Your problem will be something with a crystal oscillator on board.

One possible fix is to switch to a clock generator with frequency spreading. These devices effectively blunt the peaks of a crystal oscillator by spreading the noise over a wider (but still tiny) portion of the spectrum.

This modulation is key to making emissions limits easier to beat.

At this point, don’t worry about near field probes. Just probe suspect signals for spikes at your failing frequencies.

[T3sl4co1l]

This does not sound like a motor problem. Motors run at kHz and their inductance fights fast riding pulses. And RS232 is not that fast.

Your problem will be something with a crystal oscillator on board.

One possible fix is to switch to a clock generator with frequency spreading. These devices effectively blunt the peaks of a crystal oscillator by spreading the noise over a wider (but still tiny) portion of the spectrum.

This modulation is key to making emissions limits easier to beat.

At this point, don’t worry about near field probes. Just probe suspect signals for spikes at your failing frequencies.

It does sound like a motor problem: if it's a brushed type, the brush noise can extend to 100s of MHz.  Arcing is impressively wideband.  (Some of the earliest microwave experiements, up to 60GHz or so, were done with spark discharge techniques -- back around 1900!)  If it's PWM controlled, and unfiltered as suggested, then 60MHz is a common enough peak in inverter switching loops (corresponding to, on the order of, let's say, 1nF MOSFET capacitance, with a loop of 7nH -- a bit small for a single pair of TO-220s, but representative of a slightly lazy layout with D2PAKs say).  The broad peak suggests modulation; switching noise is impulsive, and the spectrum might be a filtered response, perhaps due to the natural (under)damping of the switching loop, or perhaps due to cabling (say the cable is 2.5m long, so, 1/2-wave resonant).

It can also be digital noise, again due to converters or charge pumps; or subharmonics of the clock frequency, in which case the modulation is due to whatever random data patterns are being emitted.  The peaks should be visible if this is the case, but it depends: the signals might repeat at quite a low rate (MHz, kHz even), thus giving quite a dense forest, and subsequent filtering (due to poor PCB layout, cables, etc.) gives the overall figure of the spectrum.  The modulation might not even be very strong; I've seen MCUs emit a highly pure F_CLK before, even at fairly high frequencies (harmonics) where you'd expect it to be dropping off instead; and selected harmonics at that (e.g., 336MHz standing out prominently, but not 168 or 504MHz, nor sidebands thereof).

Given the broadening, spread-spectrum is unlikely to help the former (the data patterns are already spreading it out!), but the latter can be reduced quite effectively (since the pilot tone is F_CPU itself (and harmonics), apparently unmodulated by data).

But yeh, that's all of EMC: "it depends".

Tim

[Simon]

The motors are brushless. I'm not sure what frequency the main micro runs at, it's an STM32 with an M3 or M4, it may have an 8MHz crystal and use the internal PLL. The cabling in the box is not the best.

[Simon]

So the bit I am wondering about, well several bits.

The tests are done in dBµV/m, the spectrum analyser has dBµV, does this mean that the test is measuring the V/m based on the known length of the aerial and the analyzer is just giving the voltage it picks up (referenced to 1µV)?

The antenna comes with a table of antenna factors. Some dB/m, so if I put these in it adjusts the trace to take these into account. Nom the analyser only has a dB metric for these not dB/m.

Really my biggest problem with all of this is that every item seems to use a different metric or that some metrics may be the same but they are called differently. I know that dBm is actually dBmW but we all call it dBm. So when it comes to dBµV is this another unit of measure that just misses the last bit off and is actually dBµV/m?

Turning the attenuator on on the analyser lowers the noise floor. Turning on the preamp does too. I am guessing that this is a data acquisition thing, the attenuation is being compensated for by some amplification but the attenuation reduces the noise to a point where the ADC will not resolve it, same logic for turning the preamp on.

[T3sl4co1l]

Yes, calibrated to dBuV/m in the air; antenna factor accounts for frequency response as well as converts from uV/m to uV.  And yes, dB relative to 1uV.

Calibration ultimately comes from an antenna of standard geometry, like an electrically short dipole, which might not be very efficient or well-matched, but the electric field from it is well-defined.

Tim

[Simon]

Well I did some playing around this afternoon. I set up a signal generator with a dipole that is about 600mm and just put about 50dBµV on it. I worked out how to set the analyser up to get a readout that allowed me to see the signal against the background noise.

I increased and decreased the output by 10dB and saw a corresponding change in the peak on the screen. So I now know that if I take some measurements on my failed equipment, whatever that is I need to take 20+dBµV off.

So if I put 600µV into this 600mm dipole antenna, is that 1mV/m? or given the aerial characteristics I have entered does that mean that when I read off 50dBµV that is 50dBµV/m?

I tried putting steel mesh around the antenna with holes that are 50mm, this did not reduce the noise floor as unplugging the aerial did not. So I assume that I am in a quiet part of the building and that the only way to further drop the noise floor if needed is to reduce the RBW, the test house runs at 100kHz, at 9kHz we can drop the noise floor, at 200Hz it's nice but takes some time. But changing the RBW does not alter the actual measurement of my sample signal so I am guessing that I can use the RBW I need and still get relatively comparable measurements.

[T3sl4co1l]

The whole story of it is:

- You have two antennas of given gain factors.
- You position them in some orientation.
- You apply some voltage to one.  It generates... whatever field, given by its antenna factor.
- That field propagates (or perhaps doesn't, if in near field), with attenuation given by Friis antenna equation, plus reflections from nearby geometry, multipath, etc.  (The superposition of which interferes and may reinforce or reduce the actual reading at the far end.)  Geometry can be ignored in an anechoic chamber, all those modes are absorbed and direct transmission remains.
- The other antenna picks up that field at whatever distance it's at.
- You read some voltage at it, given by its antenna factor.

Also, polarization, if oriented differently to each other, or by going through enough oddly-angled reflections.

Transmission losses are quite substantial, so you may need quite some transmission power (well, ~mW say), and a preamp on the receiver, to see anything on the spec -- an instrument with fairly high noise figure, generally (but whatever it is, it's in the manual).

If the antennas are identical, then you basically square (as V, or double as dB) the antenna factor (or put one into the signal generator, but if it's fixed voltage, then, this must be done at the receiver).  You should then measure reasonably flat gain.  But again, assuming freedom from reflections, and adequate distance between (to not be in the near field).

Lower RBW gets better selectivity for narrow-band signals: random noise has a density behavior (V/rtHz), so lower RBW (effectively, the pixels on the display represent narrower bands) gives lower noise floor, while narrow signals remain full amplitude.  Modulated signals are spread out by nature of the modulation, and so generally read lower, as with noise (because they don't fit within RBW).

There is still a statistical bias, depending on detector type (peak / avg / etc.).  At very low RBW, many sampling periods might go into one pixel, and you are simply displayed the peak or average or whatever of those readings.  The spec's manual will show how these trade off.

Offhand, I would expect RBW of 10 or 100kHz is fine here, unless your noise floor sucks and you need the edge from lower RBW, just to see things.

Tim