09-13-2009 10:52 AM
I'm interested in using GTX/GTP but I need to have the TX serial stream at the exact same rate as the
RX stream. I see how to use the CDR features to recover the serial stream clock (not sure how high
a rate can be moved around the chip), but I don't see any way to use the recovered clock to use
the TX serial clock. The only way I see to have TX=RX is to do external CDR and feed the recovered clock
into the shared PLL. It seems like such an obvious thing to want to do so what am I missing?
09-14-2009 08:11 AM
The recovered clock from the received datastream has more jitter than is acceptable to be used as the reference clock for the transmitter. For protocols that need a locked RX/TX, such as Sonet, you need to use an external clean up PLL that starts off near the target frequency and then switches over to just the recovered clock for the exact frequency.
The SiLabs SI5321 is an example of one device that can do this for SONET rates.
09-16-2009 08:56 AM
09-16-2009 03:26 PM
The MMCM/PLLs are not great at reducing jitter so this has a low chance of success in generating a quality REFCLK for the MGTs.
07-09-2014 06:29 AM
I managed to use a PLL to generate a GTGREFCLK at ~148.5MHz for a HD-SDI passthrough.
Input the gtx_rxoutclk to a PLL with mult 16, div 8 to achieve a 2x clock multiplication from 74.250MHz.
Works fine on a Kintex7 :)
07-09-2014 08:08 AM
The performance degradation with a simple low utilization design under nominal conditions is not that great. However, this is not the same as having a complex high utilization design that is operating under worse case PVT conditions.
07-11-2014 12:02 AM
Definitely true. I should say we are only using it for testing/proof of concept purposes with HD-SDI. So yes, in some cases it can provide a reasonable solution, just don't expect too much from it :)