EMI: I'm in or out the directive? (edit: poor man's testing method)

[mcinque]

I'm about to sell a project essentially based on atmega328 @16MHz, 74 logic, linear and switching regulators, keyboard, display, some relays and mosfets.

Since I haven't much money to spend with proper EMI tests, I'm trying to understand how much radiated emissions my project generates, and basically, I need to know only if I'm into the EU directive, doing a sort of pre-compliance test.

With a 15-2700MHz RF explorer, a 50MHz DSO and a couple of beehive magnetic and e-field probes (with no preamplifier in between), I've analyzed the emissions of my project.

Only with the probes really, really close (near contact) to a specific zone of the MCU, I've seen a 31.071MHz, 59dBm main peak and some harmonics up to 150MHz (with lower energy) using the magnetic field probe, and a 150KHz electric field near in contact with the switching regulator using the e-field probe and DSO.

All the rest of the PCB areas shows basically no considerable activity.

The problem is: how can I say that I'm in the directive or not for radiated emissions? I've NOT a RF test chamber nor a proper antenna, neither the proper instruments.

Well, I guess that I can compare my project's emission values with similar projects that has already been CE approved: Arduino UNO boards, LiPo chargers that uses atmega328 and so on...

Doing this, I discovered that my project radiates less or the same emissions compared to that devices.

I know that this is really a poor man's test, but I'm confident that this method has it's own logic and that it does have sense: if they are EMI ok with those values, me too.

What do you think about this testing? Can I trust my pre-compliance method for radiated emissions?

[qno]

It is hard to say if you do not have a reference.
Check the levels of a CE compliant device.

You also have to comply to suseptability for external fields and esd.

[mcinque]

It is hard to say if you do not have a reference.
Check the levels of a CE compliant device.

It's exactly what I did, and on more than one device. I wonder if this raw method it's reliable. Of course I want only to see if I'm into the limits, not to measure exactly emissions values with uncalibrated and cheap hardware 

You also have to comply to suseptability for external fields and esd.

Oh yes, this is relatively easily achievable with a proper shielding, guards and esd discharge paths.

[Precipice]

It's exactly what I did, and on more than one device. I wonder if this raw method it's reliable.

Short answer: no. Not to within, say, +-30dB, to pick a figure by (somewhat educated, experienced) guesswork. Even chasing a single frequency that failed in testing, so you can really concentrate on making things as repeatable as possible, knocking 10dB (say) off with a fix on the bench, and measuring it before and after, is almost entirely unrelated to what it'll measure back at the test site. 
a 60dB fix, though, is probably real!

Nearfield measurements are also almost entirely decoupled from the other half of the game, which is 'what gets coupled onto the wires leaving my product'.

You also have to comply to suseptability for external fields and esd.

Oh yes, this is relatively easily achievable with a proper shielding, guards and esd discharge paths.
[/quote]

If it was easy, people wouldn't fail test. I'm bitter, having just come off an extremely difficult few months getting an automotive product approved, but even on simple products, I'm never as confident as you sound.

Of course, it depends how much you care / how much you've got riding on it.

[mcinque]

If it was easy, people wouldn't fail tests.
 I'm never as confident as you sound.

Oh well, probably I'm too confident due to my inexperience... 

[Precipice]

Oh well, probably I'm too confident due to my inexperience... 

Yeah -  a few rounds of EMC testing will beat that out of you
Have you asked any local EMC houses what a precompliance run would cost? It might not be as bad as you think?

(It's why I'm building an EMC chamber at the moment. 2500 more screws to go, and I'll see whether it's RF-tight. I hate screws - but not quite as much as I hate failing EMC!)

[Neilm]

Only with the probes really, really close (near contact) to a specific zone of the MCU, I've seen a 31.071MHz, 59dBm main peak and some harmonics up to 150MHz (with lower energy) using the magnetic field probe, and a 150KHz electric field near in contact with the switching regulator using the e-field probe and DSO.

You are measuring near field noise - not far field noise. EMI is usually measured at 3m or 10m with the antenna in vertical and horizontal planes. The unit is then rotated until the worst combination is found. This is done in either an Open Air Test Site (OATS) or a properly set up EMC chamber, the results of which are then correlated to the OATS. When you go into a chamber you will see it covered in ferrite tiles which are there to remove reflections. Dave did a tour of one a few years ago - look up the video. [EDIT - found it - Episode 202]

EMC compliance is not easy - there are things you can do to improve your chances of complying, but unless you put your unit in a well sealed metal box (ie with no openings) you are liable to have potential problems. The only thing I have every tested that passed first time was basically in a solid metal box with only a very small opening. At that we had issues making sure the paint was not put in critical areas causing fails.

I would have thought an emissions test would take no more than a day on a unit as simple as you have described, probably less.

Neil

[mcinque]

You are measuring near field noise - not far field noise. EMI is usually measured at 3m or 10m

True, depending on device class (A/B).

But I probably miss something:

say that I go on a basement (zero RF from outside) and I measure near field emissions (or even "far" using a DIY log periodic antenna) of a couple of devices similar to mine (almost same components, MCU, quartz etc) that I'm sure to be full EMC compliant (note the "devices similar to mine" and "almost same components": I'm not comparing the emissions of an VHF transmitter to an ArduinoUNO board; say that I've designed an alarm clock and I'm using others CE alarm clocks as reference).

Using an uncalibrated spectrum analyzer, probes and antenna, I get some values about their radiated emissions. Say that I find the clock frequency and its harmonics together with some other filth. Say that I find that all this values never go over 70dB. I store this number and I use it as reference limit.

Using precisely the same measurement setup in the same location, now I power up my device and measure its radiated emissions, moving and rotating it until I find the worst scenario. Also here I find the clock and its harmonics and all the measurements values reach at max 60dB.

Why my device shouldn't be into the directive?

[Neilm]

I will make no pretense at being an expert in this field but how do you know your basement has zero RF? Is there an electric light in there? I have seen mains cables re-radiate EM frequencies at significant levels. What is behind the walls?

Second, how do you know what the emission spectra of your reference devices are? If you want a transfer standard, there are noise sources that transmit known levels of noise, but they are also expensive.

Thirdly, I have seen a demonstration with 2 PCBs. They had exactly the same components in exactly the same circuit. The only difference was the layout of the PCBs. There was about 20dB between the emission spectra of the two.

Fourth, I once spent 2 months trying to track down a signal with spectrum analyser and calibrated probe kit. I could see it but there were other frequencies that looked far worse. Putting the unit into the GTEM at our in house facility showed the issue very clearly.

The tests you have done could satisfy you that you have a very good chance to pass, but the equipment you are using is not calibrated so creating a technical file for CE compliance from this data would be laughed out of court if you were prosecuted for selling non-compliant equipment (that is the extreme case - I have never heard of anything more than fines or forced rework for EMI related issues).  Showing a result from a recognised test house would probably be accepted as your due diligence.

As my experience has been challenged on previous threads when discussing EMI and safety - I have spent the last 10 years designing test equipment and part of my job was to do EM testing at our in house facility. When the EMI standard our instruments have to meet changed, I ended up reworking the instruments that failed.

[mcinque]

Thank you Neilm to reply to my post.

how do you know your basement has zero RF?

You're right, probably I should have say "lowest RF possible". Using the RF explorer in that ambient I can see only noise floor.

Is there an electric light in there?

Only a led flashlight.

I have seen mains cables re-radiate EM frequencies at significant levels.

It's easy to believe that.

Thirdly, I have seen a demonstration with 2 PCBs. They had exactly the same components in exactly the same circuit. The only difference was the layout of the PCBs. There was about 20dB between the emission spectra of the two.

Absolutely. Any change to the design affect the emissions. But the point is that you CAN effectively notice the difference in excess. That's my only goal.

The tests you have done could satisfy you that you have a very good chance to pass, but the equipment you are using is not calibrated so creating a technical file for CE compliance from this data would be laughed out of court

Oh, you can bet it! It is not my intention to produce any technical file or report. I've not the experience/competence for doing it and, like you say, nor the proper equipment. I'm looking only for something that could see me if I could pass or not.

What about using a mini DIY RF test shielded chamber? A medium size metal box properly shielded with no holes (I mean waterproof welded except for the RF out) to use it as test box?

[Precipice]

You're right, probably I should have say "lowest RF possible". Using the RF explorer in that ambient I can see only noise floor.

You can't see FM radio (88-108MHz) or cellphones or UHF TV? Hmm. Does an FM radio or cellphone work in the basement?
Those are generally around or well above the limit line, round here. You might be in the middle of nowhere, of course - but if they're not showing at all, I think that shows how unrelated your near-field study is to an actual EMC sweep...
A sweep with a DIY Log periodic (and a preamp!) would be a lot more convincing (but be warned that Log Periodic antennas need calibration, or at least confirmation that they're working properly) -
www.ets-lindgren.com/manuals/3147.pdf shows some curves for a (presumably) well made one.

You're going down a well trodden path, by the way. Welcome to the happy world of EMC on the cheap!

(I still think you should at least talk to a local test house. Round here, they're very helpful, and may be able to offer you advice or more practical help. They definitely understand the difficulties tiny manufacturers face - not least because they want you to become a bigger manufacturer and use their proper services)

[Neilm]

The tests you have done could satisfy you that you have a very good chance to pass, but the equipment you are using is not calibrated so creating a technical file for CE compliance from this data would be laughed out of court

Oh, you can bet it! It is not my intention to produce any technical file or report. I've not the experience/competence for doing it and, like you say, nor the proper equipment. I'm looking only for something that could see me if I could pass or not.

Before placing any equipment on the market, you should have prepared a technical file containing the arguments why it passes the relevant EU legislation, both EMC and safety requirements. These are used as evidence for the signing the Deceleration of Conformity. This should list all the standards it meets. Theoretically, these could be asked for by anyone who wants to sell the equipment. Practically, they are not usually asked for.

[sandy]

Hi,  I dont want to demoralise you ,but sooner or later you will realise how much time you have wasted with these tools.  You must at least try to buy a second hand  spectrum analyzer and get it calibrated, this will help you rectify at least 50% of your problems. You simply cannot rely on DIY tools. If you want to rely than imagine your self batteling with the norms and guidelines of  'The International Electrotechnical Commission' with and arsenal of DIY and uncalibrated probes . Sorry but no hope here as they measure in nano watts with great accuracy to make a pass or a fail.

[mcinque]

Wait wait... I never stated that I want to skip the report. I want only to do some kind of a pre-compliance. This could save money when I'll send the device to the test site.

[PA0PBZ]

Only with the probes really, really close (near contact) to a specific zone of the MCU, I've seen a 31.071MHz, 59dBm main peak...

Isn't that like about 800 Watts? 

[AndyC_772]

Just to back up what others have said, the work you've done is all good and worthwhile if it convinces YOU that your product would pass radiated emissions if you were to submit it for testing.

It will not, however, convince anyone else, because it's not complete, quantitative, or representative of the test setup that's mandated by the standards.

If you're confident that it would pass testing first time, then now is the time to go ahead and send your kit to a lab for a formal test.

As well as radiated emissions, you likely also need to test for ESD and both radiated and conducted immunity. If it plugs into other equipment, or the mains, then you may well need to do conducted emissions too.

I know it's not what you want to hear, but besides the 'negligence + hope' approach, there's really no valid alternative to doing at least some testing at a calibrated test facility - sorry.

[mcinque]

Only with the probes really, really close (near contact) to a specific zone of the MCU, I've seen a 31.071MHz, 59dBm main peak...

Isn't that like about 800 Watts? 

Oops! minus -59dBm

[mcinque]

if it convinces YOU that your product would pass radiated emissions if you were to submit it for testing.

If you're confident that it would pass testing first time, then now is the time to go ahead and send your kit to a lab for a formal test.

exactly.

It will not, however, convince anyone else, because it's not complete, quantitative, or representative of the test setup that's mandated by the standards.

Absolutely agree.

As well as radiated emissions, you likely also need to test for ESD and both radiated and conducted immunity. If it plugs into other equipment, or the mains, then you may well need to do conducted emissions too.

It is floating since it runs only on battery and does not have any plugs or connections with other devices/equipment. I'm confident that I should not comply conducted immunity, only radiated. Am I wrong?

[AndyC_772]

If it doesn't plug into anything else, then that implies no cables (correct)?

If so, then there's no conducted testing to do. It also dramatically improves your chances of getting through radiated emissions and immunity first time, because in my experience something like 80% of all noise problems are related to the cables. No cable == no efficient aerial.

[qno]

What is your battery voltage?

In a distant past I worked for an EMC house and tested a baby phone that run on a 9V battery.

The carrier was so weak that it did not extended above the limit at 3 meters.
The harmonics where hardly visible.
So if you are running on batteries and you did not build a specific transmitter with an big end stage
you do not have to worry.

A shielded box is not necessary, look at your SPSU for your laptop. It is plastic.

What you need to do is keep the current loops as small as possible.
Easies way to do this is use a 4 layer PCB with one layer dedicated to ground.
So every track runs over or under its return.
This keeps the loops very small so your track will not radiate.
It also makes it very difficult for tracks to pick-up unwanted signals.

[Precipice]

If it doesn't plug into anything else, then that implies no cables (correct)?

No charger either? If so, then yes, everything's a whole lot easier. I still think that nearfield sniffing is slightly too unrelated to 'real' radiated to count very much, but a 1m sniff with an amplified antenna in a quiet room might correlate reasonably well. Don't forget to rotate the product, and raise / lower the antenna.

If so, then there's no conducted testing to do. It also dramatically improves your chances of getting through radiated emissions and immunity first time, because in my experience something like 80% of all noise problems are related to the cables. No cable == no efficient aerial.

Yep. How come everything I ever make looks like an electric octopus? Cables everywhere...

[mcinque]

If it doesn't plug into anything els, then that implies no cables (correct)?

Correct for the PCB. For the entire project there are a couple of cables that goes from the PCB to the front panel of the case, connected to 5V pushbuttons, reed switch and on/off switch. Should I use shielding? The return path should zero any currents.

What is your battery voltage?

14.8V.

So if you are running on batteries and you did not build a specific transmitter with an big end stage
you do not have to worry.

No specifics transmitters were built on the PCB. There is nothing that can radiate intentionally anything: every IC and the switching regulator are properly (at least I think!) decoupled/filtered, there aren't floating pins or traces... in some way I've tried to apply all the suggestions and rules I've read (but maybe I haven't read - or understand? - enough )

What you need to do is keep the current loops as small as possible.

This is a good advice that I've followed. All the (small) currents should be zeroed by the near return path/cable GND.

Easies way to do this is use a 4 layer PCB with one layer dedicated to ground.

mmm... this would raise dramatically the costs of the pcb manufacturing.

No charger either? If so, then yes, everything's a whole lot easier.

Yes, no charger.

but a 1m sniff with an amplified antenna in a quiet room might correlate reasonably well. Don't forget to rotate the product, and raise / lower the antenna.

I dont' have a proper amplifier (I should need something with a very wide band)... what do you suggest?

[AndyC_772]

Ah... so, you have a 2 layer board and some internal cabling.

Without a solid, unbroken ground plane, you're just asking for trouble.

It's all about loop area, which governs how efficiently the board works as an antenna. If you have a track carrying a current on a layer immediately above an unbroken ground plane, then the return current can and will flow in the plane, and will approximately follow the route of the track above it. The loop area depends on the trace length and the layer-to-layer spacing in your PCB, which is usually small. This means the PCB is an inefficient antenna, which is exactly what you want.

Without a ground plane, the loop area still depends on the trace length, but now also depends on the length of the (routed) ground and the spacing between each current-carrying trace and the corresponding ground trace. This area is typically *much* larger, so the PCB acts as a much more efficient antenna. A lot more energy is radiated or absorbed by your circuit.

Also, without power and ground planes, you have very poor decoupling above 100 MHz or thereabouts, and any current spikes drawn by your circuit will cause corresponding high frequency voltage spikes due to the inductance of your power and ground traces. These can radiate.

For all but the most completely trivial circuits, I don't use anything less than a 4 layer board. It's not worth it. Unless you're making thousands of units, the extra cost of a couple of PCB layers will be small compared to the time you spend testing and reworking the product when it has EMC or signal integrity problems.

[Precipice]

I dont' have a proper amplifier (I should need something with a very wide band)... what do you suggest?

There's a link in here
http://www.edn.com/electronics-blogs/emc-emi-rfi-esd/4378157/An-EMC-Troubleshooting-Kit--Part-1b-Emissions-

 to "This $50 module covers 20 to 3000 MHz, has a gain of 18 to 23 dB, noise figure of 2.7 dB and
can handle up to +10 dBm power input. It runs off 12 V DC."
Basically just a minicircuits ZX60-3018GS+ module on a bit of board with connectors and a battery.
Curves are here www.minicircuits.com/pdfs/ZX60-3018G+.pdf - it's not flat, but neither will your antenna be!
You can get better noise performance elsewhere, but it's not _bad_.

That entire EDN series is worth a read, if you hadn't found it before.

[mcinque]

There's a link in here

Ordered!