Thread Links Date Links
Thread Prev Thread Next Thread Index Date Prev Date Next Date Index

[802.3_ISAAC] Return Loss - Adhoc

Hi Thomas,

Thanks again for making the presentation.

 

Here is a summary of what I brought up during the call. If you get some time to address these questions, I would really appreciate it.

The simulation presented suggests impedance mismatches, if accurate, they would imply a massive recall of existing SERDES systems—particularly with GMSL and FPD-Link—which together account for billions of links deployed in the field today. To my knowledge, there have been no recalls related to cable return loss, and this includes during my experience building systems around these specifications.

 

SERDES systems operate with a variety of cable lengths and interconnects, and yet they perform reliably in production. Your measured results are far better than the simulated ones—showing up to 20-30 dB improvement in low-frequency RL. This raises the question of why relaxing the spec by +6 dB at low frequencies is necessary. SERDES has established itself as a gold standard, and changing the RL requirement to accommodate unverified simulation risks undermines proven industry practice.

 

As Ragnar has already noted, the fundamental question remains: how often do real-world systems experience this level of impedance variation?

 

On slides 4–7, the very different S11 vs S22 isn’t evidence that “four in-line connectors” inherently blow up low-frequency return loss. It’s the segmentation and where the first mismatch sits relative to the port. From Port 1 the first mismatch is only ~0.3 m/1 m away, while from Port 2 it’s ~4.56 m, so (a) the phase of the first reflection is different, and (b) the two-way path loss to that reflection is larger from Port 2, which reduces the apparent RL. That’s why S22 sits lower than S11 in your plots. This aligns with your own conclusion that low-frequency RL is dominated by cable segment properties/impedance deviation, while connectors dominate at higher frequencies—not the other way around. 

 

When referring to your slides about this exact behavior, you might have an error with S11 and S22 on the symmetric cable segments on slides 9 and 10, which should have nearly identical S11 and S22.

 

See Ragnar’s presentation about this: https://www.ieee802.org/3/ISAAC/email/pdfyf5P9oeh4z.pdf

 

Below, you can find my questions, as well as one additional question I added regarding the in-line connectors.

 

Questions for Clarification

  1. Why should 802.3dm require a +6 dB increase in low-frequency RL compared to GMSL?
  2. Can you show a physical cable construction that replicates the simulated RL profile?
  3. What is the measured standard deviation of impedance for cables in current production?
  4. If the simulated RL is truly critical, why have there been no SERDES recalls despite billions of deployed links operating 6 dB below the proposed spec?
  5. Why did you have such short cable segments back to back?
    1. What is the reasoning behind this, and why is it not implemented in current SERDES products, as it would be in breach of their specifications?  

 

Thanks,

TJ Houck

 

Infineon Technologies Americas Corp. – Detroit

System Architect

Mobile: +1 (765) 426-9832

tj.houck@xxxxxxxxxxxx

 

To unsubscribe from the STDS-802-3-ISAAC list, click the following link: https://listserv.ieee.org/cgi-bin/wa?SUBED1=STDS-802-3-ISAAC&A=1