Author: Pawel Plucinski
Date: 2016/10/17


Aurora 8b10b using GTY on the VCU108
====================================

Introduction

Officially, the Aurora8b10b protocol does not support the GTY transceivers (only GTX and GTH).
The GTY transceivers are meant to be targeted for higher line rates, Xilinx recommend
to use Aurora 64B66B protocol for better throughput. As the consquence the Aurora 8b10b IP
core does not offer anything for the GTY transceivers.
There is not really any technical reason behind Aurora 8B10B protocol not
supporting GTY transceivers. 

In order to test the Aurora8b10b protocol on the GTYs, the specific designs (configs 1-4) are
implemented using the custom method but based on the files generated by Vivado (as for example
GT wizard).

Setup

1. The configuration_1: VCU108(GTY) -> VCU108(GTY)

Aurora 8b10b protocol is implemented on the VCU108 using GTY@5Gbps. 
This test is peformed on the single QUAD (4-Aurora lines). The data
are being transmitted on the TX lines and received on the RX lines.
The GTY transceiver is bonded to QSFP module (the QSFP loopback module).
The design is equipped with the pseudo random data generator and data analyzer.
Additional features: Sideband channels, Diagnostic: VIO and ILA component (data monitoring),
Duplex mode.

2. The configuration_2: VCU108(GTY) -> VCU108(GTY)

Aurora 8b10b protocol is implemented on the VCU108 using GTY@6.4Gbps. 
This test is peformed on the single QUAD (4-Aurora lines). The data
are being transmitted on the TX lines and received on the RX lines.
The GTY transceiver is bonded to QSFP module (the QSFP loopback module).
The design is equipped with the pseudo random data generator and data analyzer.
Additional features: Sideband channels, the VIO and ILA component (data monitoring),
Duplex mode.


3. The configuration_3: VCU108(GTY) -> ZYNQ(GTX)

Aurora 8b10b protocol is implemented on the VCU108 using GTY@5Gbps, while the
ZYNQ board using GTX@5Gbps.
A single Aurora line is tested. The data are being transmitted on the TX line from
VCU108 board and received on the RX line (ZYNQ board).
The VCU108 module is equipped with the pseudo random data generator, while the ZYNQ
board with the data analyzer to check the data integrity.
Additional features: Timer, the VIO and ILA component (data monitoring), 
Simplex TX and RX.


4. The configuration_4: ZYNQ(GTX) -> VCU108(GTY)

Aurora 8b10b protocol is implemented on the VCU108 using GTY@5Gbps, and on the
ZYNQ board using GTX@5Gbps.
A single Aurora line is tested. The data are being transmitted on the TX line from
ZYNQ board and received on the RX line (VCU108).
The ZYNQ module is equipped with the pseudo random data generator, while the VCU108
board with the data analyzer to check the data integrity.
Additional features: Timer, the VIO and ILA component (data monitoring), 
Simplex TX and RX.


Implementation method

The ZYNQ board owns the GTX transceivers, so the standard Aurora8b10b IP core and
the standard implementation method can be used to implement the test design.

Taking into account that the Aurora8b10b IP core does not support the GTYs transceivers, 
so, the custom method needs to be used. And, here is the procedure which is needed
to accomplish the goal:

1. The standard Aurora8b10b IP core is used to generate the Aurora8b10b protocol files
(link layer), and design example files.
The Aurora8b10b GUI uses the GT wizard to configure (automatically) the physical layer,
but in that mode only the GTH or GTX transceiver can be configured.

2. Thus, in order to configure the GTY transceivers, the GT wizard needs to be used in the
standalone mode. The user is in charge to setup the physical layer manually (line rate, 
speed, ref clock,  user clock, quads, encoding, etc). The generated files are needed
to build the complete desing.

3. The final step is needed to merge the link and physical layer manually. In order
to form the project, the user is obliged to bond the relevant components. 

4. It's recommended to revise the design using the standard Vivado tool
(code schematic view, logfiles, etc). A substantial number of project files needs
to be visually inspected (including the xdc files).

Summary

The performed tests indicate that (in all cases), there is no problem to initialize
the link. The link is up the stable. The data analyzer inidicate no transmission error.
In addition, the chipscope (TX and RX, other signal monitoring) does not report any
problems.



Conversation with the Xilinx experts on this aspect:

=========================================================================================================================

Date: 2016, July 13

Hello Pawel,
 
Hope you are doing fine.
 
Thanks for filing a service request with Xilinx Technical Support. I am Ashish and will be working with you on this SR.
This is in relation to your issue "Aurora IP Queries" filed with us.
 
Please check my answers for your query inline:
 
1)      Do you know if there is any plan to upgrade the Aurora 8b10b core to be able to implement this protocol for the GTY?
        For the moment only GTH is supported.
 
> I checked in for the internal resources to see if there is any plan in near future to support Aurora 8B10B protocol
in Ultrascale GTY transceivers but right now it is not yet planned. If I come across any plan/update regarding this,
I will make sure to inform you about it.
 
2)      I can see that GT wizard also offers the Aurora 8b10b protocol for the GTY . Have you tried to use the GT wizard
        with Aurora protocol 8b10b for the GTY transceivers. If yes, have you experienced any problems?
 
> Even if GT wizard offers Aurora 8B10B protocol for GTY transceivers, it will just customize the GT part of the protocol.
> It will not provide the complete protocol as the Aurora IP provides. So you can simply check the GT attributes and other
> settings which it configures but not the protocol. Even if it is implemented, it will only implement GT part of Aurora
> 8b10b IP and not the complete protocol.
 
Feel free to let me know in case of any further concerns regarding this issue.
 
Thanks and regards,
Ashish
 
Ashish Dilip Dohale,
Worldwide Technical Support,
Xilinx
 
Technical Support: http://www.xilinx.com/support/
Xilinx User Community: http://forums.xilinx.com/

=========================================================================================================================
Date: 2016, July 19

Hello Pawel,
 
Thanks for waiting on this.
 
Please check answers for your questions inline:
 
 I understand that only the physical layer (transceivers) will be configured but the rest of protocol can be
 probably emulated (in VHDL/Verilog), right? 
 
>Yes this is correct.
 
Have you tried to implement the Aurora 8b10b protocol (based on the Aurora IP core) for the UltraScale family?
 
>Though I have not tried it, I can confirm that validation of this IP is already performed on the development side
>before releasing updated version of IP. So there should not be any problem regarding constraints generated by the tool.
>You can even cross-verify them by comparing with ones specified in board user guide (UG1066 from VCU108 board).
 
How to get these specific line rates (I mean what is the best procedure to configure the hardware to avoid problems): 
4.8 and 9.6 Gbps? (based on the reference clock 320.64 MHz).
 
>As we had discussion regarding this particular concern, best approach would be programming the oscillator
>frequency so that it matches with one of the values recommended by Aurora IP customization wizard for
> particular line rate.
 
Feel free to let me know in case of any further concerns regarding this.
 
Thanks and regards,
Ashish


=========================================================================================================================
Date: 2016, September 14

Hello Pawel,
 
Thanks for your response and glad to know the root cause of issue is figured out.
 
I checked this internally and came to know that there is not really any technical reason behind Aurora 8B10B protocol not
supporting GTY transceivers. But since GTY transceivers are meant to be targeted for higher line rates, we recommend
to use Aurora 64B66B protocol for better throughput.
 
So I would strongly recommend to use Aurora 64B66B with GTY transceivers because if you land into some issues while
using Aurora 8B10B, it would be difficult to find out root cause since this combination is not tested and verified
at our end.
 
Regarding image processing, feel free to raise a service request for issue which you are facing. It will be directed
to appropriate engineer.
 
Thanks and regards,
Ashish

=========================================================================================================================
Date: 2016, October 05


Hello Pawel,
 
Thanks for the updates.
I have downloaded the video but I am yet to analyze it since I am quite involved this week with some tasks
(yeah I need to complete them by Friday since I will be travelling next week).
 
But I will definitely go through the video and let you know my remarks.
I would really appreciate if you can send me the ready-to-use design for VCU108 which tests Aurora 8B10B
using GTY transceivers as mentioned in your response.
 
I will test it at my end and we can surely include it in our internal database and it will be useful
for anyone who wants to implement similar application in future.
Thank you so much for all the efforts and keeping me updated with the progress of this project.
 
I would give you a buzz if I have some queries before including this information in our database.
Hope you speak with you soon.
 
Thanks and regards,
Ashish
=========================================================================================================================