XAUI crosstalk assumption
I accepted the assignment of considering connector crosstalk in the last
XAUI conference call. We have been assuming 4% crosstalk, and reasoned that
most of it will come from connectors. Here is my initial result for maximum
connector crosstalk that is consistent with these assumptions. Constructive
critique is welcome! (Why do I stick my head out like this?)
-Dawson
Here is a worst-case crosstalk scenario for the lane C receiver (lanes
choices are arbitrary):
Lane A driver >--------------------------------------------> Lane A
receiver
||
||
\/
Lane C driver >--------------------------------------------> Lane C
receiver
/\
||
||
Lane E receiver <--------------------------------------------< Lane E
driver
Lane A crosstalks onto lane C at the far end due mainly to use of a common
connector. (There is relatively little assumed crosstalk in the PCB due to
good layout practices. Crosstalk in the connector is more difficult to
control even with excellent design due to lack of ground planes, pin
proximity, etc.) The crosstalk is attenuated about 6 dB at 1.56 GHz as it
propagates down 20" of the lane C traces. The worst position for the
connector is at the very far end where the lane A signal edges are fastest.
The edge rates are degraded as the signal propagates down the trace, so
moving the connector away from the lane a driver results in less coupling.
Lane E crosstalks onto lane C at the near end, due again mainly to the
connector. The worst position is at the very near end where there is no
signal attenuation over the PCB. Moving the connector away from the near end
adds PCB trace attenuation and reduced crosstalk seen at the lane C
receiver.
Using the above scenario and an assumed 4% maximum allowable total
crosstalk, this allows 4%/(1+0.5) = 2.67% crosstalk per connector. In other
words, there can be 2.67% NEXT and -6 dB as much FEXT, adding to a total of
4.0% total crosstalk. (The 4% assumption has been used by the XAUI group to
check signal margin with minimum received signal levels, or 20" interconnect
length. Shorter lengths allow larger crosstalk.)
This 2.67% connector budget is larger than the worst case pair-to-pair
crosstalk result reported by one connector supplier. There is margin for
additional sources of crosstalk.
-Dawson