Ethernet CM choke - Which side?

[ricko_uk]

Hi,
I always thought the Ethernet's CM choke is placed on the cable side but I just noticed that sometimes it is placed PHY side. And is also on several RJ45 with integrated magnetics as well as in transformers used externally from the connector (see attached screenshots from different datasheets).

Which one is best or standard and why?

Thank you

[PGPG]

If (because of EMI) you need to add CM choke and your PHY output stage is powered through trafo you can't add it at driver side as CM core will saturate by DC supply current so you add it at cable side. It is most common solution, I think.
But the cable has some common mode impedance that can resonate at some frequencies making EMI problems. So called "Bob Smith termination" is needed to avoid it. The center tap on cable side should be connected to "GND" through 75Ω. But the CM choke in serie with it is ruining this termination. So to have good Bob Smith termination you have to move CM-choke on the PHY side.
If receiver is not powered through trafo you can do it at least at receiver side.
If PHY output is not powered through trafo you can also at output move CM choke to PHY side and you get your second situation.
But I think PHY powered through trafo is simpler to design so:
- you can have CM choke at PHY side provided you run third wire through CM-choke to compensate DC current, or
- you can left CM choke at cable side and add extra L (at cable side of CM choke) with center tap just to have a pin for Bob Smith.
Adding third wire - bigger core = costs and space.
Extra L = costs and space.
So, because of costs there are solutions with 'semi' Bob Smith (ruined by CM choke at cable side).

If it seems being complicated then consider PoE

[T3sl4co1l]

Normally, CMC goes on the media side. It's isolated, only the ESD capacitor loading it, two wires, easy job.

There are three cases for PHY side:
1. Push-pull driver, two-line CMC: both pins sink (or not) only, thus there is CM current at this point and wild voltages result. Disallowed!
2. Push-pull driver, three-line CMC: equivalent to the two-line media-side case.  Popular for PoE, as choking the power flow (large DC bias) would be onerous.
3. Bridged driver, either CMC: both pins sink and source symmetrically, thus a 2-line is permissible here.  (Note that the center tap doesn't need a connection to AVDD; it is usually still bypassed to GND with a small capacitor.)

Bridge drivers are common for GbE+ but sometimes seen on 10/100 PHYs, perhaps compact/integrated ones; signal transmission amplitude is usually reduced as a result (or at least as compared to 10BASE-T definitions), but this is rarely a problem with the sensitivity and compensation used in modern receivers.

If building (2) from discrete components, consider two data-line chokes in parallel (one winding common (CT), remaining windings to TX+,-), as multiline chokes are kind of nonexistent.  But, you may have a 3-line available as part of the "chipset"* you're using; the manufacturer may have related/relevant components available.

*In reference to your other thread https://www.eevblog.com/forum/projects/descrete-transformers-pairs-for-100-mpbs-ethernet/msg5680235/#msg5680235 where you're looking at "chip" style transformers, see what I did there.

Tim