3
\$\begingroup\$

In my design I want to connect a PHY (88E1512) to a switch which contains another PHY (BCM53115M). The devices are separated by a few inches of PCB and will be impedance matched. Both have the same ground reference but different power supplies. Ethernet also seems to have a 1m distance requirement between ports.

I'm able to place magnetics in between the two but I wanted to know the logic behind requiring magnetics between two on board PHYs. I have also heard of using capacitive coupling instead of magnetics all together and was wondering for such short distances are magnetics better or would simply capacitive coupling be better like in this question: Connecting two Ethernet PHY without magnetics?

Note: Both devices seem to have termination inside them.

\$\endgroup\$
5
  • \$\begingroup\$ Magnetics in this situation are clearly unnecessary. However, the BCM53115M seems to have two MAC ports available. Why don't you connect directly the MAC behind the 88E1512 to a MAC port of the BCM53115M, bypassing the two PHYs? \$\endgroup\$
    – dim
    Commented Jun 6, 2018 at 20:11
  • \$\begingroup\$ @dim thanks for the answer! We already have devices occupying those ports unfortunately. Since both devices I listed have termination inside them do you think I could safely connect them directly without any external termination and without series caps? \$\endgroup\$
    – Mackenzie Goodwin
    Commented Jun 6, 2018 at 20:33
  • \$\begingroup\$ You may still need a single capacitor for each line (two in series are unnecessary), if the PHYs have different voltage levels. I am not sure whether the PHY specs indicate voltage levels for each +/- line according to ground, or simply differential levels. If what is specified is just the differential levels, and both PHYs have different reference voltages, you may run into problems without a DC-blocking cap. I'm not quite sure I'm saying here so I won't make a full answer. For the termination, if it is integrated in the chip, you must not add it externally. \$\endgroup\$
    – dim
    Commented Jun 6, 2018 at 20:48
  • \$\begingroup\$ Also, I believe the 1m distance between nodes for ethernet comes from requirements for the collision detection. So, if you just have two MACs talking to each other (no risk of collision), I think you can get rid of this constraint. Again, I'm not sure what I'm saying. \$\endgroup\$
    – dim
    Commented Jun 6, 2018 at 20:51
  • \$\begingroup\$ In conclusion, I think connecting those directly with short traces and a capacitor should work fine... But don't blame me if it doesn't. \$\endgroup\$
    – dim
    Commented Jun 6, 2018 at 20:55

1 Answer 1

Reset to default
0
\$\begingroup\$

Magnetics is always better because it is the standard way and it gives the expected result: the connection is established and kept stable.

Capacitive coupling, especially in 1000BASE-T, is always a lottery very often resulting in a long trial-and-error-error-error unfanny process having no visible end and confidence in the result's stability.

Therefore, if your design/project time is finite, choose the well-known, old-school, transformer-based method of coupling/decoupling. In your case you need only one transformer that is comparable in occupied area with a bunch of capacitors + other discrete parts in help of them.

\$\endgroup\$
1
  • \$\begingroup\$ Thanks for the answer. I talked directly to Marvell on the topic and they also said capacitive coupling is unreliable. That is exactly why I don't have it as an example design. \$\endgroup\$
    – Mackenzie Goodwin
    Commented Jul 9, 2018 at 16:23

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.