Ethernet design guidelines

[Sylvain Munaut]

Hello,

I'm working on a PCB where there is an ethern 10/100 RJ45 port.
But I'm having a hard time finding guidelines to route thos signal.
I can put the magnetics close to the RJ45 (traces ~0.5 inch) but I need
to carry the data signals further (about 3 inches). They are
differential signals so routing them as a differential pair sounds
obvious but any other recommandation ? Particular differential impedance
to match ?


Thanks for any hint.


Sylvain

 

[Mac]

On Sat, 02 Jul 2005 18:57:58 +0200, Sylvain Munaut wrote:

Hello,

I'm working on a PCB where there is an ethern 10/100 RJ45 port.
But I'm having a hard time finding guidelines to route thos signal.
I can put the magnetics close to the RJ45 (traces ~0.5 inch) but I need
to carry the data signals further (about 3 inches). They are
differential signals so routing them as a differential pair sounds
obvious but any other recommandation ? Particular differential impedance
to match ?


Thanks for any hint.


Sylvain

I used to have a differential pair impedance calculator. I don't have
it anymore, but for the most part, this rule will work: route the two
signals which make up the pair immediately adjacent to each other on the
same layer and make them the same length. Then, most importantly, use
a trace width which makes the individual tracks have a characteristic
impedance of around 55 Ohms to ground (or the nearest power plane). Then
the differential impedance will be about right.

Oh, I guess I should explain my methodology. You are trying for 100 Ohms
differential impedance. This is from the Ethernet spec. The 55 Ohm figure
comes from this highly technical formula:

Zo = Zdiff / 2 + epsilon.

Zdiff is the desired differential impedance, epsilon is uh, or rather has
been experimentally determined to be around 5 Ohms. ;-)

--Mac

 

[Sylvain Munaut]

Mac wrote:

On Sat, 02 Jul 2005 18:57:58 +0200, Sylvain Munaut wrote:


Hello,

I'm working on a PCB where there is an ethern 10/100 RJ45 port.
But I'm having a hard time finding guidelines to route thos signal.
I can put the magnetics close to the RJ45 (traces ~0.5 inch) but I need
to carry the data signals further (about 3 inches). They are
differential signals so routing them as a differential pair sounds
obvious but any other recommandation ? Particular differential impedance
to match ?


Thanks for any hint.


Sylvain


I used to have a differential pair impedance calculator. I don't have
it anymore, but for the most part, this rule will work: route the two
signals which make up the pair immediately adjacent to each other on the
same layer and make them the same length. Then, most importantly, use
a trace width which makes the individual tracks have a characteristic
impedance of around 55 Ohms to ground (or the nearest power plane). Then
the differential impedance will be about right.

Oh, I guess I should explain my methodology. You are trying for 100 Ohms
differential impedance. This is from the Ethernet spec. The 55 Ohm figure
comes from this highly technical formula:

Zo = Zdiff / 2 + epsilon.

Zdiff is the desired differential impedance, epsilon is uh, or rather has
been experimentally determined to be around 5 Ohms. ;-)

--Mac

Thanks ! Exactly what I was looking for.


Sylvain

 

[Sylvain Munaut]

PeteS wrote:

I use the following perl code for diff pairs (when I need to calculate
stuff - mostly I tell the board fabricator what I need and let them
tell me what the track widths and inter layer thicknesses are).

Note that a diff pair should be length matched to a certain degree (to
prevent differential to common mode conversion; i.e. to make sure your
differential signal is still differential). In your case, that's not
necessarily too important. The numbers of consequence to you are
t(prop) ~ 160psec/inch on FR4

Thanks for the script !


Sylvain