cancel
Showing results for 
Show  only  | Search instead for 
Did you mean: 
philippcw-s
Visitor
Visitor
741 Views
Registered: ‎01-09-2020

Nonlinearity of eye scan function (Kintex-7 GTX transceiver)

Jump to solution

Hi,

I am currently working on a function to automatically check eye mask compliance of a given serial interface using the eye scan feature of a GTX transceiver in a Kintex-7 FPGA. For this function I need the eye scan data to represent the signal at the input pins of the FPGA in the best possible way.

To avoid manipulation of scan results by the automatic adaption functions of the transceivers, I've set the parameters of the linear equalizer in LPM to fixed values using the override ports and the DRP.

Using this fixed configuration I've made measurements at different input amplitudes. For comparison I measured the height of the eye in the middle of the eye with an oscilloscope at the input of the FPGA and with the eyescan function of the FPGA. The results of these measurements are shown in the attached picture.

As you can see in that picture, there is no linear connection between the actual voltage at the input of the FPGA and the results of the eye scan function. I guess this non-linearity is caused (mainly) by the AGC.

My question is: Has this AGC simply a non linear characteristic amplifying lower voltages stronger or does the AGC include an active adaption compressing the whole eye at higher voltages?

In case there is an active adaption: Is there any way to turn this of?

I can't find any detailed information about the AGC in the transceiver user guide. I know DFE mode offers some more options here, but unfortunately DFE has some other disadvantages for me.

Thank you for your answers!

Tags (4)
Diagram.PNG
0 Kudos
Reply
1 Solution

Accepted Solutions
roym
Moderator
Moderator
631 Views
Registered: ‎07-30-2007

I'm afraid we don't test these settings for linearity.  We are more interested in having enough margin.  It is interesting to note that the majority, if not all, of your plot is above the "eye mask minimum" where we guarantee you will have enough margin over process, voltage and temperature.  See https://xilinx.sharepoint.com/sites/xkb/SitePages/ArticleViewer.aspx?articleNumber=52529




----------------------------------------------------------------------------
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
----------------------------------------------------------------------------


View solution in original post

5 Replies
roym
Moderator
Moderator
696 Views
Registered: ‎07-30-2007

I'm not sure you should expect a linear relationship even without the AGC.  In your plot you appear to reaching saturation where the eye isn't going to get any bigger.  

You can't turn off the equalization completely.  The best you can do is to override all the LPM parameters and set them low or neutral.  This is talked about in the NIDRU app note   Set:

RXLPMEN=1
• RXLPMHFOVRDEN=1
• RXLPMLFKLOVRDEN=1
• RXOSOVRDEN=1

See UG476 for the registers.  Set the RXLPM_*F_CFG registers to 0.  Set RX_OS_CFG to mid range - try both 13'h1000 and the default 13'h0080 and see which works best.




----------------------------------------------------------------------------
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
----------------------------------------------------------------------------


philippcw-s
Visitor
Visitor
649 Views
Registered: ‎01-09-2020

Hi, 

thank you for your answer!

I have to admit I'm rather hoping for a linear relationship than really expecting it, since it would make my task much easier. I'm aware that linearity is probably not the thing those transceivers are optimized for.

I already used the configuration you're suggesting while capturing the data I attached to the initial post. As those settings should reduce the effect of the linear equalizer as far as possible, it seemed likely to me, that the nonlinearity is caused by the AGC, which I can not control at all.

In the end it does not really matter, where the nonlinearity comes from. If there is no way to reduce it, I have to include the nonlinearity into my analysis. The really important question for me is: Does the transceiver behave different at signals with higher amplitudes (amplifying the whole signal - including low voltages - less) or is the effect I see a simple saturation? (Obviously there could also be a combination of both, but that would mess things up for me completely.)

If the transceiver adapts its gain I have to use different scales depending on the amplitude of the signal. If there is "only" a saturation, I can create one scale for all amplitudes out of my data.

0 Kudos
Reply
roym
Moderator
Moderator
632 Views
Registered: ‎07-30-2007

I'm afraid we don't test these settings for linearity.  We are more interested in having enough margin.  It is interesting to note that the majority, if not all, of your plot is above the "eye mask minimum" where we guarantee you will have enough margin over process, voltage and temperature.  See https://xilinx.sharepoint.com/sites/xkb/SitePages/ArticleViewer.aspx?articleNumber=52529




----------------------------------------------------------------------------
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
----------------------------------------------------------------------------


View solution in original post

philippcw-s
Visitor
Visitor
599 Views
Registered: ‎01-09-2020

Ok, then I'll have to work with what my measurements tell me.

Unfortunately, I can't access the link you've sent.

0 Kudos
Reply
roym
Moderator
Moderator
578 Views
Registered: ‎07-30-2007

The link just says that if you have 45 vertical codes of clearance and maintain the other GT specs, like voltage noise levels and refclk jitter then you have enough margin to cover PVT variations.




----------------------------------------------------------------------------
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
----------------------------------------------------------------------------