Two opposing currents result in a mutual impedance; they obstruct one another. The PCB calculator takes this into account when calculating the trace geometries, although you specify the impedance and it indirectly solves for the geometry. But the opposing currents is the source of this impedance.
When a signal current passes a 0A ground plane it looks like an opposing current. The higher the signal level, the more impedance it sees. Hence, you get an attenuation of the signal, or a loss, or dissipation - and this dissipation goes somewhere: onto the ground plane. Here it forms on opposing current for other differential pairs, and in effect it's a crosstalk antenna.
A ground pour is fine as it keeps the board from warping due to the large copper-free areas otherwise. But I'd remove it altogether from the differential pairs, and space those out a little more for good measure since you have plenty of board area to work with. They can get in close around the magjack pins, that's fine.
Do you have a scope? If so scope out the supply ripple at the magjack. You may find decoupling it slightly will reduce this. I don't think it's a huge deal, but it's so easy to check and add a few caps as needed when you're going to spin another board rev anyway.
But fix the ground pour and maybe tweak the spacing first. There's little reason to worry about other problems until the obvious ones are fixed.
Ideally, you'd use a high-speed differential probe to look at the pairs, trigger on one of the (R)MII error signals, and confirm crosstalk. Then you can confirm it's fixed in your rev 2 boards.
Just my $.02...
Home
Help
Search
Login
Register
