Differential Pairs and the Consequences of Bad Routing

[DanR]

Hi everyone,

Throughout my career I've tried to always do the best I can while optimising quality and time to market.

Maybe that means taking less time on component selection if I know a board will only be used once, but spending significantly longer on something that is going to be mass produced.

I have recently found myself in a position where differential pairs are frequently compromised in order to save time.
My gut feeling is that this is very bad but I find myself unable to explain why.

Compromised as in not using the correct impedance profile for part of the bus, not using any profile, sometimes separating the pair for a time, bad via entry/exits, etc. Also not adding clearance which results in GND pours getting too close and changing the edge coupled pair to a coplanar edge coupled pair, or allowing different pairs to get too close to one another without a GND shield.

Can someone please help me understand what the consequences are for all of this?

Please keep in mind that we are not talking about high-speed GHz signals as I will never allow those signals to be compromised.
This is about CAN, CANFD, RS485, RS422, and maybe a differential analogue input.


Kind regards,

[robertferanec]

I don't think the peripherals you named are super critical for impedance, especially if the tracks are only a very short length on the PCB. I have had to do a lot of compromises even on USB 2.0 and always worked oki and passed the specification.

Just be careful when you will be routing high speed interface. You may not even see visible problems on badly routed high speed signals and the board often may somehow work, however the performance and reliability may be influenced. For example, there may be some error corrections implemented in the protocol, or interface will automatically slow down etc.

I remember this my 1Gb Ethernet working "oki": https://welldoneblog.fedevel.com/201...xtreme-wiring/

[DanR]

Thanks Robert,

What would you consider to be too long for those signals I listed?

I think we're looking at roughly 600mm through several connectors and boards.


Kind regards,

[qdrives]

If you go through connectors there are already impedance mismatches.
The length depends on the propagation with regards to the rise time.
CANbus is very resilient, but there is also an EMC question.
Then there is the remark from Rick Hartley where he states that most differential pairs on a board are more like two single ended signals as they reference the Gnd and not each other.

[robertferanec]

For the interfaces you listed often the transceiver is placed close to the connector. So it should not be a problem to keep the tracks short.

PS: I can't say the exact distance as I do not know it.