03-03-2017 02:18 AM
We had implemented Eye Scan using 7 Series FPGAs Transceiver (Kintex-7 GTX), For a "prescale value" of "5" we are getting the Eye Scan "Error Count" as "zero" and "Sample Count" as "65535", and the "Internal Data Width" is "40". For these results, if we calculate the BER using the formula "BER = Error Count/(Sample count * (2^(prescale + 1)) * Internal Data Width)", we are getting "BER" as "Zero".
Is any default value has to be considered as BER for an Error Count of Zero?
03-03-2017 08:31 AM
Hi @prasanthvthycaud ,
There is a reference design for doing eye scan, described in XAPP743. It runs on Microblaze and contains full source code for different transceiver types. Did you take a look at it and compared with your implementation? That would help determine the source of problem.
03-05-2017 10:07 PM
Thanks for your reply.
Sorry that I had missed to mention my objective in measuring the Eye Scan. Using the BER values obtained during each iteration for different offsets of EYE Scan, want to plot a Statistical Eye Plot.
I had referred the XAPP743, and it had mentioned about an algorithm called "Gear Shifting", as per the Gear Shifting if we get the "zero error count", the measured values is not to be considered and the iteration has to be repeated by changing the prescale value.
I had implemented the "Gear Shifting" based on XAPP743, and the threshold values were also set based on the XAPP743 reference design. Even after implementing the Gear Shifting algorithm the error count was same, was getting zero even with the change in prescale value.
The threshold values used are, 1, ERROR COUNT TOO SMALL = 30, 2, ERROR COUNT TOO LARGE = 3000, based on XAPP743 reference design.
I am trying the Eye Scan Analysis for SRIO Protocol, but IBERT core doesn't support this protocol.
03-06-2017 06:03 AM
I mentioned IBERT because I found it useful to verify results with my "My EyeScan" implementation for JESD204B links.
One think to check is GTX QPLL/CPLL and line rate you use. If you look in DS191 Table 91 you will see there is a "hole" for 5.0Gbps line rate range if using QPLL (you should use CPLL for 5.0GBps). This could also produce strange EyeScan results.