DP83TD510E: DP83TD510E: More robust Link in scenario with millisecond connection losses on physical connection

Hello,

Thank you for your reply. Please allow me until Wednesday 2/15 to get back to you.

Sincerely,

Gerome

Hello,

I have relayed this information to our internal team for comment. I hope to get a response back next week. Please note TI is on holiday 2/20/23 in observance of President's day.

Sincerely,

Gerome

Hello,

Some questions from internal team:

1. How are the disruptions created? Shorting MDI lines or something?
2. The above observations are counter intuitive i.e. if the disruption is present for shorter time that gives worse result?
3. How did you conclude the phy did a retraining or not? There is no link drop in both cases, correct?

Sincerely,

Gerome

Hello Gerome,

Thank you for your reply. To the questions from the team
1. The disruptions were created by quickly discnnecting and reconnecting both MDI lines with a solid state switch (TMUX7211).
3. We observed the MDI lines before and after the solid state switch with a differential probe on an oscilloscope (blue and red trace). We also monitored the ethernet packages through the phy with the trigger output of a capturing device (yellow trace) and the state of the solid state switch (green trace). The following image shows a disruption of 1ms.
 

[https://imgur.com/a/SfOZkv7, 2562x1396]
 

A retrain is clearly visible because both first PHYs go silent, then send link pulses and then send training sequences at a lower amplitude before sending at full amplitude.
 

The following image shows a disruption of 50ms:

[https://imgur.com/a/DO4A7Um, 2562x1396]

The phy recovers the link without a full retrain.
 

2. We also think that the observations are counter intuitive but the following could be happening : Short disruptions (<500us) do not lead to a scrambler unlock and therefor not to a link drop. During long disruptions (>10ms) the PHY correctly transitions to the SEND_IDLE state and is able to regain synchronization to recover the link. During disruptions between (500us and 10ms) some register or part of the logic has not transitioned to the correct state and a link recovery is therefor not possible.

Sincerely,

Kristian

Hello Kristian,

20th Feb is a US public holiday. I'll let Gerome respond as soon as he's back to office.

--
Regards,
Gokul.

HI Kristian,

Please allow me time to bring this information to the team. At worst, I expect an update by EoW.

Sincerely,

Gerome

Hi Kristian,

I apologize for the delay. I will ping our internal team regarding your query and get back to you next week.

Sincerely,

Gerome

Hi Kristian,

Can you confirm the concern is that for disruptions between 500us-10ms yousee the link drop and a retraining whereas for longer disruptions it recovers without complete re-training?

In no case phy gets completely stuck, correct?

Sincerely,

Gerome

Hi Gerome,

that is absolutely correct, the PHY doesn´t get stuck.

Sincerely,

Kristian

Hi Kristian,

I have relayed this to our internal team. Please note that I am out of office until next Tuesday, and I hope to reply with an update upon my return from the team.

Sincerely,

Gerome

Hi Kristian,

To understand the context of your query, you only want to understand why you are seeing such behavior, for 500us to 10ms where link drops and retraining is needed but for longer disruptions, it recovers without re-training?

Sincerely,

Gerome

Hi Gerome,

are there any updates from the team? Have they been able to replicate the issue?

For more context:
During field test we had issues with link drops due to vibrations at the connectors. In measurements we determined that the channel is only ever disturbed for a few milliseconds. This had us puzzled, because according to the 802.3cg standard the PHY should be able to recover the link with disruptions up to 200ms. Therefore we did some further investigation and lab tests, the results which we have posted earlier.

For our application it is important that the connection is not interrupted for more than about 100ms maximum. A retrain takes much longer than that (up to 3s) and will lead to a full stop of the system, since the communication is critical. We therefor want to know, if there is a way for the PHY to recover the link with the 500us to 10ms disruptions as well. Since the PHY is able to recover longer disruptions it should be possible from our perspective.

Single Pair Ethernet is a great technology and solves a lot of problems for us but with the issue it is currently unusable for us.

Sincerely,

Kristian

Hi Kristian,

Thank you for the context. I have relayed this to the team, and I do think this will help clarify. Please expect a response by mid-week.

Sincerely,

Gerome

Hi Kristian,

After discussing with team, we understand the concern regarding retraining and your system. We wanted to understand a few things:

- What is the timeline you are looking for a solution? 

- Can you confirm you are not using any register configuration and are running default settings, correct? If not, please share configuration

- How repeatable is your observations that you are seeing regarding re-training for 500us-10ms? Is this consistent over many IC's?

Sincerely,

Gerome