AC-coupled on-board Ethernet

[buenos]

I have designed a board where (among many other things) I have a 1Gig Ethernet controller chip directly connected to a 1Gig Ethernet switch chip. The connection is AC-coupled, there is no transformer in between, and pullup or bias resistors.
The problem is that the link only comes up in 100Mbps mode. 100M runs on 2 diffpairs, 1G mode runs on 4 diffpairs.
The MDI connections are straight 0--0, 1--1, 2--2, 3--3. I have even tried to swap them with very short wires (0--1, 1--0, 2--3, 3--2), but still it comes up as 100Mbit only.
So the question is:
-Does it need bias resistors, if yes in what configuration?
-If termination/biasing is the issue then should it also not work in 100Mb mode either?

 

[FvM]

The ports ususally require external termination and DC bias. It's particularly unlikely that the 1000BASE-T hybrid circuit can work without correct termination.

 

[buenos]

For 100Meg only controller they always have external termination resistors.
For 1Gig controllers I have never seen it.
Bot the controller and the switch have reference design schematics, and on those there are no termination resistors, but all Ethernet ports are directly wired to magjacks. The eth co datasheet actually says that there are on-chip 50R terminations both input and output, the switch chip datasheet does not mention anything as it is very short.

 

[FvM]

For 1Gig controllers I have never seen it.

But some definitely have like SMSC LAN7500. All in all it's sligtly absurd to give no information about the involved chips and expect an answer why your solution doesn't work. I'm convinced that you'll find out, by carefully reviewing the datasheets and reference designs.

 

[buenos]

I probably cannot mention part numbers as these are new chips under nda.
The reference designs only show chip to magjack configurations. In those ref designs there are no resistors between the magjack and the chip.

 

[FvM]

- Does the chip use the magjack center tap for biasing?
- Does the chip need a low impedance DC path between + and - pins?
- It might be that hybrid operation depends on common mode isolation.

 

[buenos]

Both chips' reference designs have magjacks an all ports.
The switch chip datasheet does not mention anything about the IO circuits on the Ethernet ports.

"It might be that hybrid operation depends on common mode isolation. "
-I don't understand this.

 

[FvM]

I understand from your answer that the chip documentation specifies 1:1 ethernet transformers with unconnected center tap and no additional components.

But you don't know if either the low impedance DC path or the common mode isolation of the transformer is required by the chip.

I wouöld connect a transformer for test to check if 1000 MBit link is established then. After that you can further narrow down the cause.

 

[buenos]

OK, so the test results:
-AC coupling: 100Mb half duplex
-DC coupling: no link
-magnetics: 1G Full duplex.
For the next board revision I am using 2 magnetics between the 2 chips. But I should also make the existing prototypes work.

By the way, the ref designs describe magjack connections like this:
-controller: all center taps together, decoupled to ground and connected to 1.8V
-switch: all center taps decoupled to gnd separately.
I have replicated each with separate transformers then connected the line sides.

 

[FvM]

-controller: all center taps together, decoupled to ground and connected to 1.8V

Thanks for finally referring to the bias question.
Did you verify if the 1.8V bias is required by the controller? If so, the transformer could be possibly replaced by two chokes or high Z ferrite beads at each pair.

 

[buenos]

OK, I will try this on Monday:

 

[buenos]

I tried it with 600R ferrite beads but it didn't work at all. We didn't have any chip inductors in the lab. Then I tried it with 50R resistors and that worked.

 

[FvM]

Interesting! But no idea why.