DP83848C: Random Link-Downs for ~12-15s

Hello Mathias,

It does look like your RBIAS is slightly off from the recommended resistor value, are you able to test this with the 4.87 k-ohm resistor?

Thanks,

Vibhu

Hi Vibhu

I'm still waiting for the ordered 4.87kOhm 0.1% Resistors, but I did measure the current values:

Vbias=1.1857V

Rbias=4.85kOhm (which would be within the 1% range, but when you do the math it could be slightly out of range with 1% resistors, I agree)

:/

So you say a "not perfect" Rbias could lead to random Linkdowns?

In this case one of my next steps would be the opposite way: go to 4.7kOhms for example and check if the same behaviour happens more often...

Hello Mathias,

It would be one of the first things to narrow down and eliminate as causing the problem as the device is not characterized for other values of Rbias.

Thanks,

Vibhu

Hello Vibhu

Today did a bit of testing and tried to force the error to reappear again by changing the Rbias value.
1st test with Rbias = 5.1kOhm: successful
2nd test with Rbias = 4.78kOhm: successful

Both tests lasted about 2h.

So I'm really starting to think, that the Rbias isn't the problem here.
I certainly will change the resistors to one single 4.87kOhm 0.1% resistor, but still I doubt that this was the reason...
Testing this propperly would take a few weeks, since the error appears so seldom...

Do you know of any other reason why the DP83848CVV could lead to a short linkdown?

Here the rest of the schematics:

Other information:

- The link partner of the DP83848CVV is a switch. We tested different switches and always had the same error.

- Just the link between the DP83848CVV and the switch's port to which it's connected with breaks down

Hello Mathias,

Thanks for sharing those results, I agree it looks like the RBIAS is likely not the issue.

Is there a reason why you believe the link down is due to the PHY and not a higher layer? I think you may need to monitor the link for the next link down and read registers to see what is happening during this period. Another thing to try is to see if the PHY sends out the FLPs once it sees there is no link.

You mentioned you've tried different switches, were you by chance able to test different DP83848 parts as well?

What is your ppm error of your clock?

Thanks,

Vibhu

Hello Vibhu

We just know that we receive a linkdown notification from the DP83848CVV when the error happens. The reason / source of this linkdown is unknown.
=> What do you have in mind?
I will look into this with our embeddedSW team. So we might be able to trace all the MDIO register values during the error.
Unfortunately I have no idea how to check if the FLPs are sent, because we still aren't able to reproduce the error, which is why me have to wait several days for it's reappearance...

No, we didn't try different DP83848 yet, because we have several designs in our company with a working DP83848CVV and never experienced a similar behavior before...

The clock has a frequency tolerance of +/-50ppm, which meets the requirements in the DP83848's datasheet.

Tomorrow I will test two more things:
- Influence of the temperature on all the components
- Measurement of the signal quality of the 100BASE-TX signals according to ANSI 9.1.x (eye pattern, jitter, levels, etc.)

Thanks a lot,
Mathias

Hello Mathias,

Thanks for the update, please keep me posted on the testing.

Thanks,

Vibhu

Hello Mathias,

I was wondering if you had any updates here. Please let me know.

Thanks,

Vibhu

Hello Mathias,

I am closing this thread for now please let start a new thread if you have new results to share.

Thanks,

Vibhu

Hi Vibhu

I just measured the eye pattern, jitter, voltage levels and so on and it looks totally fine.

To whom it may concern: To generate a test pattern on the DP83848C
I used a MSP-EXP430F5529LP to write the MDIO registers:
Register 0000 = 8000 (=Reset)
Register 0000 = 2100 (=100Mb/s, Full Duplex)

My colleague from the embeddedSW is currently on a business trip,
so I expect some news from him by the end of next week.
I'll keep you posted, so please don't close it yet ;)

Regards,
Mathias