cancel
Showing results for 
Show  only  | Search instead for 
Did you mean: 
hpoetzl
Voyager
Voyager
2,153 Views
Registered: ‎06-24-2013

Glitch in eye scan?

Looking at those beautiful eye scans generated by Vivado via the IBERT core ...

MGT Eye ScanMGT Eye Scan... I'm wondering what causes the glitch in the middle where looks like either the bottom or the top part is shifted down by about 10%.

I'm pretty confident that the actual signal eye doesn't look like this but the effect is very consistent, i.e. it shows up in all eye scan diagrams I've done with Vivado/IBERT yet.

Is this a one-off error in the software or some weird effect in the way the sample point voltage is configured or maybe something completely different I haven't even considered yet?

 

Thanks in advance,

Herbert

 

-------------- Yes, I do this for fun!
0 Kudos
13 Replies
roym
Moderator
Moderator
2,093 Views
Registered: ‎07-30-2007

I've never seen this sort of problem.  I can only guess that the equalization is adapting during the scan causing this.  I would suggest holding the various CDR and equalization attributes and seeing if holding one of them makes this go away.  Also does changing from LPM to DFE or vice versa affect the scan.

See tables 4-9 and 4-10 of UG578 to see the hold attributes for GTY.  Other transceivers should have similar tables.




----------------------------------------------------------------------------
Don't forget to reply, kudo, and accept as solution
Be sure to visit the Resources post periodically to keep up with the latest
https://forums.xilinx.com/t5/Serial-Transceivers/Serial-Transceiver-Forum-Guidelines-and-Useful-Resources/td-p/1173590
----------------------------------------------------------------------------


hpoetzl
Voyager
Voyager
2,085 Views
Registered: ‎06-24-2013

Hello Roy!

I've never seen this sort of problem. 

That is quite interesting, because I see it all the time not just occasionally

I can only guess that the equalization is adapting during the scan causing this. 

I would suggest holding the various CDR and equalization attributes and seeing if holding one of them makes this go away.  Also does changing from LPM to DFE or vice versa affect the scan.

I set all the *HOLD port properties to 1 but the effect didn't change.

I also verified that those settings are actually active by changing from pattern to fast clock which really gives bad results until I un-hold those settings ...

See tables 4-9 and 4-10 of UG578 to see the hold attributes for GTY.  Other transceivers should have similar tables.

As far as I can tell, there is no DFE for GTP tranceivers (UG482)

 

eye_scan_hold

 

 

-------------- Yes, I do this for fun!
0 Kudos
hpoetzl
Voyager
Voyager
2,079 Views
Registered: ‎06-24-2013

After artificially closing the eye, by playing around with the transmitter settings, I was able to obtain some really nice scans showing the problem in all detail ...

xc7a50t_6G_eyescan_1

 

xc7a50t_6G_eyescan_2

 

xc7a50t_6G_eyescan_3

 

-------------- Yes, I do this for fun!
0 Kudos
gguasti
Xilinx Employee
Xilinx Employee
2,066 Views
Registered: ‎11-29-2007

What you are measuring is probably due to two different RX configurations. The eye scan loop is much slower than the RX adaptation loop. I would agree with Roy.

have you tried with setting the RXOSHOLD=1 as well?

hpoetzl
Voyager
Voyager
2,050 Views
Registered: ‎06-24-2013

What you are measuring is probably due to two different RX configurations. The eye scan loop is much slower than the RX adaptation loop. I would agree with Roy.

Not sure why the ES_VERT_OFFSET should affect the RX adaptation or equalization, maybe you can elaborate on that ...

have you tried with setting the RXOSHOLD=1 as well?

Yes, as can be seen in the Vivado Screenshot I posted earlier.

But because the glitches are so nice, here is a new one again with all *HOLD port properties set for both tranceivers involved in the links.

xc7a50t_6G_eyescan_hold

 


 

 

-------------- Yes, I do this for fun!
jlantz
Xilinx Employee
Xilinx Employee
1,974 Views
Registered: ‎12-10-2009

If you set the eyescan to do 100% of the horizontal scale (UI) and 100% of the vertical scale (codes) does it do the same thing?  For this test you can do increments of 4 on both to speed up the scan. 

0 Kudos
hpoetzl
Voyager
Voyager
1,946 Views
Registered: ‎06-24-2013

Yup, the same (only smaller :) with a full scan ...

 

 

xc7a50t_6G_eyescan_full

 

-------------- Yes, I do this for fun!
0 Kudos
jlantz
Xilinx Employee
Xilinx Employee
1,934 Views
Registered: ‎12-10-2009

Is this an optical or electrical interconnect?

0 Kudos
hpoetzl
Voyager
Voyager
1,916 Views
Registered: ‎06-24-2013

electrical.

-------------- Yes, I do this for fun!
0 Kudos
gguasti
Xilinx Employee
Xilinx Employee
1,901 Views
Registered: ‎11-29-2007

the eye scan loop is not "affecting" the adaptation loop. But they happen at the same time.

It is possible that while the eye scan is running, and this operation takes a very long time, the equalizer adapts to a better setup. As result, you have the overlap of two half eyes: the first half is with equalization A, and the second half is with equalization B.

I mentioned RXOSHOLD because it is not listed in GTP UG, UG482 table 4-9 (equalizer), but it appears only at table 4-11 (CDR ports). Thus is possible that in testing we forget about this port.

However in your eyescans I see the glitch always happening at the passage by vertical offset  = 0. I cannot explain this. A change in the equalization would affect the eye at random vertical level.

could you please repeat the scan, only for that subregion where the glitch happens?

thanks

Image 1553244227.png

 

 

0 Kudos
hpoetzl
Voyager
Voyager
1,886 Views
Registered: ‎06-24-2013

could you please repeat the scan, only for that subregion where the glitch happens?

Not sure how to run the scan only for an offcenter subregion but I certainly can zoom in with Vivado ...

Here some closeups of the glitch:

xc7a50t_6G_eyescan_zoom

 

-------------- Yes, I do this for fun!
0 Kudos
travisc
Moderator
Moderator
1,855 Views
Registered: ‎10-19-2011

Can you do an export on the raw data on one of these strange eyes? Just right click on the eye and select export, then hopefully you can attach it here. It might be usefull to see both ends of the spectrum, a plot with many samples, and one with very few samples.

-------------------------------------------------------------------------
Don’t forget to reply, kudo, and accept as solution.
-------------------------------------------------------------------------
0 Kudos
hpoetzl
Voyager
Voyager
1,830 Views
Registered: ‎06-24-2013

Sure, here is a new scan (default and heatmap) together with the CSV data ...

 

xc7a50t_6G_eyescan_csv

 

xc7a50t_6G_eyescan_heatmap

 

-------------- Yes, I do this for fun!
0 Kudos