Hello,
I've spent the last few years working on multiple designs that incorporate a variety of 8 Gb/s+ SerDes interfaces (PCIe, 10GBase-KR, 25GBase-KR, etc) and I think I have a pretty good understanding of the various trade offs that are involved in SerDes channel design from a PCB perspective. However one issue that has come up a few times where I work that I've never had a good answer for involves impedance discontinuities:

I get that in a normal world, the ideal goal is to have the impedance of the channel match the driver and receiver termination values (100 ohm diff, 85 ohm diff, etc). But in the case where an Impedance discontinuity does exist, what matters more: the length of the violating regions or the number of violating regions?

As an example if I have a channel that looks like this:
----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
100 ohm transmitter > 100 ohm diff PCB trace > 100 ohm diff connector > 100 ohm diff PCB trace > 100 ohm diff connector > 100 ohm receiver
----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------

Everything is good, but what happens if the system mandates connectors that are 85 ohm nominal impedance but the transmitter and receiver devices are fixed at 100 ohm:
----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
100 ohm transmitter > 100 ohm diff PCB trace > 85 ohm diff connector > 100 ohm diff PCB trace > 85 ohm diff connector > 100 ohm receiver
----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------

In the channel above their are two impedance discontinuities at each connector and four transition points (if that matters?) between 100/85 and 85/100 ohm within the channel.

Would it be the best (of the not so best options) to still route all PCB traces at 100 ohm nominal and just deal with the discontinuities at each connector or would it be better to to do something like this:
----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
100 ohm transmitter > 85 ohm diff PCB trace > 85 ohm diff connector > 85 ohm diff PCB trace > 85 ohm diff connector > 100 ohm receiver
----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------

In this case the channel between the transmitter and receiver is at least homogeneous or does having the whole channel mismatch with the transmit and receive terminations cause more issues?

Any thoughts appreciated,

Thanks!
Jack

 

The case of single discontinuity is fairly well understood in terms of accepting some mismatch versus taking steps to mitigate the mismatch in real time. I'm thinking here of a transmitter, an antenna tuner, a feedline and an antenna. In this case the mismatch can be monitored and adjusted away as the carrier frequency of the transmitter changes. In the case of point to point connections, each discontinuity represents a point that can be modeled as some amount of transmission and some amount of reflection. The physical length of the transmission line between the two discontinuities, or between the source and a discontinuity, or between the discontinuity and the receiver will dictate the number of times a reflection will bounce back and forth between them and with what amplitude on each succeeding reflection. What you care about is those multiple reflections bouncing back and forth with enough amplitude to affect the receiver thresholds turning 1's to 0's or 0's to 1' or code groups to other code groups.

For example, a return loss of 20 dB is equivalent to a fairly benign SWR of 1.22:1 Not worth worrying about for voice communication. The mismatch between 100 Ω and 85 Ω is in this ballpark, but I have no idea if this would work on one of your examples or not. It would depend on the physical dimensions and and the actual physical details of encoding and decoding.

It is possible that a TDR device might be helpful in understanding and analyzing this behavior.
https://en.wikipedia.org/wiki/Time-domain_reflectometer