Directional Antenna for Detecting Radiated EM Emissions 1

[Henneshoe]

I am testing a design for a device with a built in Ethernet switch for CE certification. We failed the EM Emissions portion of the test due to noise spikes at 125, 250, and 375 MHz. I immediately suspected the Ethernet communication because the actual clock rate of 100Mbit/s Ethernet is 125Mhz. When I forced the connection to 10Mbit/s the noise spikes went away (this reinforced my suspicion of the Ethernet communication).

I have some ideas for routing modifications that might solve the problem, but I would like to test my modifications in-house before going back to the external lab for certification. We have a spectrum analyzer so I thought I might be able to connect some sort of directional antenna to it and probe the circuit board for noise at the problem frequencies before and after the modifications.

I am having problems trying to buy/make the antenna. When I look online for directional antennas in this frequency range, all I find are roof-top TV antennas (VHF band). I also tried making a small loop from coax cable, but I wasn't able to pick up spikes at the frequencies of interest.

Any thoughts on what I could try next?

 

[MacGyverS2000]

Maybe not the most scientific, but I've always used a straightened paperclip jammed into the end of the meter's cable, trimmed to a length of interest. Maybe not the ideal solution for 125MHz due to the 22" length for a 1/4 wave, but who says always need to pick up the main freq.

If you want me to look at it, email me the PCB for just that section and I'll give it a once over (though I will say most transceiver chip datasheets do a good job at giving you a well-shielded layout).

Dan - Owner

http://www.Hi-TecDesigns.com

 

[zappedagain]

Shoe,

If you have the room I've had good luck with both of the solutions you mentioned.

I have an old rooftop TV antenna (you can get them fairly inexpensively from the big box stores) that I connect directly via coax to a 50 ohm spectrum analyzer; the 75 ohm/50 ohm mismatch isn't a problem for seeing signals that are strong enough to fail EMC tests.

I often come back into the lab and 'sniff' for emissions with a loop on the end of coax; 0.5" diameter has been good for the frequency range you mention. The loop will sense near-field emissions, so if you aren't picking anything up your source may be in a location other than where you are looking. Check the power supply leads for starters.

John D

 

[VEBill]

In the commercial world, how do you distinguish what's emitted from the box under test and being leaked from the external Ethernet wiring? In the avionics world, we used shielded wiring with 360-degree shield termination at the box (for both test and actual installation). Thus the only thing being tested is the box itself.

If you're seeing Ethernet frequencies being leaked, then maybe they're coming out on the (commerical, unshielded) Ethernet wiring. Cat5/6 is twisted pairs, but that only gets you so far.

 

[TurbineGen]

I have made quick and simple shields by wrapping aluminum foil around the device of interest and grounding the aluminum foil. I'd start by wrapping the ethernet cable.

Another place to check is if you have a switching voltage regulator. These give off a lot of RF noise. Sometimes just adding a capacitor can greatly reduce this.

------------------------------------------------------------------------
If it is broken, fix it. If it isn't broken, I'll soon fix that.

 

[VEBill]

For certification testing, you obviously can't wrap the UUT in foil. And you can't even shield the cables unless the system is to be installed with similar shielded cables.

 

[Henneshoe]

Thank you for all the suggestions. Macgyvers2000 thank you for the generous offer to look over the layout. For now I still feel confident I can hunt down the problem (with the help of a few internet tips); if I am no closer in a couple of weeks maybe I will have to call in more support.

The device is for an industrial environment and we use shielded twisted pair for the Ethernet media. We started the test with the shield un-terminated on the laptop side and ac terminated to earth via a 1nf capacitor on the EUT side. We also tried terminating the shield to earth on the laptop side by stripping the insulation and connecting a piece of copper tape around the shield and onto the ground plate. Next, in the same manner we terminated the shield directly to ground on the EUT side. The emissions did not change much (1 or 2 db) with the different termination techniques.

One concern with the termination on the EUT side is due to the test constraints the EUT must be 1m above the ground plane. We try to reduce the impedance in the ground connection of the device by making our ground connection through a 2cm wide copper braid, but I think the impedance of the braid is still pretty high at 375Mhz.

For my in-house testing I made a new loop antenna following zappedagain's specifications. For the loop I used a coil from an old RFID tag that was laying around the lab. The new antenna works great. When wave the loop over parts of the PCB I get noise spikes at the frequencies of interest that are 20dB over the noise floor.

Good suggestion to check the switching supply. However, the switching supply we use on this device is a MC33063. It switches pretty slowly. In our past experience, if there is an emissions problem with this switcher it will be at frequencies below 100MHz.

My plan is to use the loop antenna to find what parts of the PCB are noisy then try to quiet them down. So far it looks like the noise comes from the vicinity of the Ethernet switch.

 

[VEBill]

"...1m above the ground plane....2cm wide copper braid....I think the impedance of the braid is still pretty high at 375MHz..."

That length at that frequency is close to a complete tour around the edge of the Smith chart.

 

[zappedagain]

This could be as simple as poor capacitor bypassing at the IC driving the Ethernet cable. At these frequencies you probably need 0.01uF or smaller, and right at the IC pins. Any lead inductance can negate any improvements pretty quickly at these frequencies (if you tack it on with 1" leads it might not do any good at all, but with proper mounting may fix the problem).

Whenever anybody comes on with EMC issues I always give a plug for where I learned how to solve EMC/EMI issues:

http://www.emiguru.com/books.htm#EDN

Remember, cables make great antennas! Any common-mode currents will radiate. Most Ethernet circuits have a choke right before they leave the board; yours didn't get cost-reduced out, did it?

John D

 

[MacGyverS2000]

I prefer Howard Johnson's "High-Speed Digital Design: A Handbook of Black Magic"... can't count the number of times I've read that one, and I still go back for more. I need to pick up Part II...

Dan - Owner

http://www.Hi-TecDesigns.com

 

[VEBill]

EMC 101: If you know what makes a good antenna, do the opposite.

 

[Higgler]

How about removing a VHF rabbit ear adjustable length antenna from an old TV and using that for your sense antenna.

 

[biff44]

Well, FCC certs are hard to meet! The most likely problem is that your cables attaching to the unit are radiating like big broadband antennas. Can you clamp some big honkin' ferrite beads across them during the test?

You can not dither the clock frequency, as you are probably trying to meet some ethernet clock speed spec, but spread spectruming the clock would make the heights of the frequency spikes go down a lot.

There are shielding and absorbing materials that you could use. Tell us how exactly this unit is to be used in the final product.

http://www.MaguffinMicrowave.com

Maguffin Microwave wireless design consulting

 

[Henneshoe]

So using my RFID coil antenna I was able to determine that adding a 1nf capacitor between the Analog Ground Plane and Earth Ground on the device caused the noise level to decrease. We went back to the EMC lab and passed the "official" FCC test.

The idea of adding a capacitor between the Ground Plane and Earth Ground came from an app-note. Before I start the re-layout I would like to have a better understanding of what is happening. I think noise is coupling from the electronics side of the transformer to the Cat5 side. The capacitor is giving this noise a return path so it does not need to travel down the the ground braid or Ethernet shield. In addition to the capacitor I also plan on increasing our isolation zone between the 2 sides of the transformer from 1mm to 1.5mm.

I am just looking for a sanity check here. Does it sound like I am on the right path? Does anyone have any other suggestions?

Thanks!

P.S. I also linked the referenced app-note.

 

[zappedagain]

Radiation is caused by common-mode 'signals'. Your capacitor 'bypassed' the cable and gave the common-mode noise a shorter (and thus quieter) path to follow.

The transformer in the Ethernet interface (layer) is supposed to block the common mode signals; maybe at 125 MHz they still leak through? Or there might be some distortion that is getting 'rectified' by non-linear effects and radiating? I'm thinking out loud now because I have a similar problem that I need to fix eventually when I get back in for emissions testing. With an unshielded Ethernet cable I fail immunity testing because lots of noise is introduced into my system; a shielded cable passes. The Ethernet keeps communicating just fine with either cable during the immunity testing. So not to sidetrack your thread, you also might see if things are different with a shielded vs unshielded Ethernet cable.

I'm glad to hear you passed.

John D

 

[Henneshoe]

So like you said, if the capacitor helps, then it provides a shorter return path for the CM Noise. The next questions are, "What is the CM noise source?" and "How does the CM noise couple to the cable side of the Ethernet transformer?"

For the first question, if the return path involves the analog ground plane, then I assume the CM noise source is either the Ethernet Phy or something coupling on the traces between the Phy and Transformer. For fix this, I will try to increase the separation between the diff pairs and other traces. Also, I read an app-note that suggested adding a small (22pf) capacitor in parallel with the termination resistors on these traces. The capacitors will remove the fast edges on the Ethernet traces, not necessarily the CM noise. I did try this modification during our FCC test and it did help some in my situation.

For the second question, the small Ethernet transformers have some non-negligible capacitive coupling between the primary and secondary windings (15pf in my case). This capacitive coupling could be a path for CM noise. Maybe a different transformer would help?? For my design, a CM choke is integrated in the transformer, I would think the choke would remove most of this noise. The other path I can imagine is capacitive coupling across the isolation zone between the primary and secondary of the transformer. This is the reason I plan to increase this zone from 1mm to 1.5mm.

These are just my thoughts from my investigation so far. Hopefully they will help with your issue.