Author Topic: DIY EMC test cage  (Read 5309 times)

0 Members and 1 Guest are viewing this topic.

Offline MartinL

  • Regular Contributor
  • *
  • Posts: 51
Re: DIY EMC test cage
« Reply #25 on: September 26, 2022, 10:29:19 am »
Bob Smith terminations may be helpful, I used a Wurth jack that had them built in.

You didn't mention whether you have a common mode choke on the power side - i.e. L2 in the SLUA469 schematic?

Even if your power converter is now clean by itself I think this may be helpful, as without it there is a path for noise to escape from the board onto the cable by going through the converter onto the power pairs, bypassing the common mode chokes in the TX and RX pairs.
 

Offline MartinL

  • Regular Contributor
  • *
  • Posts: 51
Re: DIY EMC test cage
« Reply #26 on: September 26, 2022, 12:10:14 pm »
In fact, there's a test you could do to confirm that hypothesis by checking which pairs are radiating - cut off the outer insulation of the ethernet cable, separate the TX/RX pairs from the power ones, and put your current probe around each group in turn.
 

Offline temperance

  • Frequent Contributor
  • **
  • Posts: 420
  • Country: 00
Re: DIY EMC test cage
« Reply #27 on: September 26, 2022, 12:33:16 pm »
What did you use instead of bob smith terminations? Because without those, common mode noise will be all over the place. Much like operating a switching power supply without a capacitor between sec 0V and primary 0V.




 

Offline nctnicoTopic starter

  • Super Contributor
  • ***
  • Posts: 26755
  • Country: nl
    • NCT Developments
Re: DIY EMC test cage
« Reply #28 on: September 26, 2022, 02:57:56 pm »
Bob Smith terminations may be helpful, I used a Wurth jack that had them built in.

You didn't mention whether you have a common mode choke on the power side - i.e. L2 in the SLUA469 schematic?
There is. And extra series inductors as well.

What did you use instead of bob smith terminations?
Well... uhhh.. nothing !

Quote
Because without those, common mode noise will be all over the place. Much like operating a switching power supply without a capacitor between sec 0V and primary 0V.
That sounds very plausible. I'm going to add the Bob Smith termination and see what that does.
There are small lies, big lies and then there is what is on the screen of your oscilloscope.
 

Offline temperance

  • Frequent Contributor
  • **
  • Posts: 420
  • Country: 00
Re: DIY EMC test cage
« Reply #29 on: September 26, 2022, 03:30:02 pm »
An extra inductor might create problems if not sufficiently damped.
 

Offline nctnicoTopic starter

  • Super Contributor
  • ***
  • Posts: 26755
  • Country: nl
    • NCT Developments
Re: DIY EMC test cage
« Reply #30 on: September 26, 2022, 05:00:07 pm »
That has been addressed as well by adding shunt capacitors. The design is a bit tricky because there is a limited amount of capacitance that may be directly at the connector of a PoE port. The inductors also have quite a bit of series resistance to have extra dampening.
« Last Edit: September 26, 2022, 05:04:08 pm by nctnico »
There are small lies, big lies and then there is what is on the screen of your oscilloscope.
 

Offline nctnicoTopic starter

  • Super Contributor
  • ***
  • Posts: 26755
  • Country: nl
    • NCT Developments
Re: DIY EMC test cage
« Reply #31 on: September 27, 2022, 11:35:17 pm »
Meanwhile I have tried the Bob Smith termination on the center taps of the transformer at the cable side but that makes no difference at all. I also tried different ethernet transformers with no real improvement. I messed something up in the ethernet circuitry. There are some series resistors and TVS diodes as well. I'm going to take those out tomorrow (it is late already) in order to strip the ethernet circuitry back to the bare minimum and go from there. I'm seeing a hard spike at a fixed frequency. I doubt it is the PoE power supply but I'm going to verify that again. Also, adding ferrite clamps over the ethernet cable does nothing at all.
« Last Edit: September 28, 2022, 12:12:29 am by nctnico »
There are small lies, big lies and then there is what is on the screen of your oscilloscope.
 

Offline MartinL

  • Regular Contributor
  • *
  • Posts: 51
Re: DIY EMC test cage
« Reply #32 on: September 28, 2022, 09:02:17 am »
I'm seeing a hard spike at a fixed frequency.

What frequency?

Assuming 100Mbps ethernet, the symbol rate on on the wire is 125MHz so you'll see peaks at multiples of this, in my experience 250MHz is the main one. If that's what you see then look at the circuitry between the PHY and the jack.

If you see 25, 50 or 100MHz peaks then those are coming from the MII interface on the other side of the PHY, so focus your efforts there.

Anything that's not a multiple of 25MHz, it's coming from somewhere else on the board.
 
The following users thanked this post: 2N3055

Offline nctnicoTopic starter

  • Super Contributor
  • ***
  • Posts: 26755
  • Country: nl
    • NCT Developments
Re: DIY EMC test cage
« Reply #33 on: September 28, 2022, 11:51:30 am »
The spike I'm using as a 'reference' is a strong one at 43.52 MHz (according to the spectrum analyser which may have a frequency error) but there is a whole band of noise popping up. And it is only there when the ethernet link is up so it is not coming from the PoE power supply. At the worst peak the difference between ethernet active / inactive is about 25dBuV (where the inactive state hits the noise floor of the EMC lab's test setup) so it is not just a small spike. I'm going to look at the TVS diodes as a next step. It could be these are rectifying part of the signal causing intermodulation products. I'm not ruling out other power supplies but all rails towards the PHY have filtering (ferrite bead + >10uf decoupling in various sizes & values).

I know about the high frequency products that come from some ethernet PHYs which is why I specifically choose a PHY chip which has low EMI by design (at a slightly higher price). This was a problem with the previous version of the product that used a 'cheap' PHY from SMSC / Microchip (the extra filter components needed made it expensive).

Update: it is not the TVS diodes, signal filtering or PoE supply. There is something else going on.
« Last Edit: September 28, 2022, 08:19:22 pm by nctnico »
There are small lies, big lies and then there is what is on the screen of your oscilloscope.
 

Offline MartinL

  • Regular Contributor
  • *
  • Posts: 51
Re: DIY EMC test cage
« Reply #34 on: September 29, 2022, 11:07:49 pm »
I doubt any spectrum analyzer will be off by more than a few tens of ppm, so I expect that 43.52MHz is accurate.

Interesting number. Exactly 17 times 2.56MHz, which is useful since it's 2^8 x 10kHz. I doubt that's a coincidence.

I don't know what else you have on your board, and what it else it does when the ethernet link is up, but I feel like there's a clue there.
 
The following users thanked this post: 2N3055

Offline nctnicoTopic starter

  • Super Contributor
  • ***
  • Posts: 26755
  • Country: nl
    • NCT Developments
Re: DIY EMC test cage
« Reply #35 on: September 29, 2022, 11:49:55 pm »
Meanwhile I think I have found the problem. There seems to be nothing wrong with the ethernet interface on the device itself. I measured the signals and impedance (TDR) and that looks just fine. Also a second board measures identical.

Though, the device is connected to only two pairs out of four so two pairs are open (it is not possible to connect the open pairs to the device; a design choice). With the unused pairs terminated at one end, the signal at the offending frequencies drops significantly. Ballpark 10 dB according to what I have been able to measure with a current probe (a few turns on a ferrite clamp) and a DIY ISN using a relatively short UTP cable (3 meters). Cable length does seem to matter with two pairs open. During the EMC test I used a 15m cable. During the most recent EMC test I left the unused pairs open as part of trying to rid of the noise from the PoE injector. All in all it looks like an error in the test setup.

Addition: nice catch on the frequency. It probably has to do with the symbol rate in the ethernet idle pattern.
« Last Edit: September 30, 2022, 03:48:59 pm by nctnico »
There are small lies, big lies and then there is what is on the screen of your oscilloscope.
 
The following users thanked this post: 2N3055

Offline nctnicoTopic starter

  • Super Contributor
  • ***
  • Posts: 26755
  • Country: nl
    • NCT Developments
Re: DIY EMC test cage
« Reply #36 on: October 21, 2022, 01:52:27 pm »
Update: with the other end of the ethernet cable terminated on all pairs, it passed but not with a large margin. It is surprising how much of the ethernet signal leaks out despite the use of twisted pairs.
There are small lies, big lies and then there is what is on the screen of your oscilloscope.
 
The following users thanked this post: Kean, 2N3055, Sighound36


Share me

Digg  Facebook  SlashDot  Delicious  Technorati  Twitter  Google  Yahoo
Smf