10-19-2018 03:28 AM
I use ultrascale transceiver generate a 2.5Ghz serdes . If the throughput isn't 100% ,will it be a problem?
For example, if throughput is 1.25Ghz, it will be half of data signal and idle signal combined in phy signal (rxp or rxn signal). If throughput is 2.5Ghz then will be 100% data signal in phy signal.
And now I find out if throughput is above 1.5Ghz, there is no data loss. But if throughput is 500M or 100M , it will be a lot of data loss(88% and 92%). Is there could be a sample problem or clock isn't lock?
10-22-2018 08:10 AM - edited 10-22-2018 09:44 AM
Usually if you are running at 2.5 Gbps you would be using 8b/10B encoding. This would reduce your maximum throughput by 20%. If you are not using an encoding then you would likely run into problems with DC balance on input ac coupling capacitor which would cause data errors. If you are using the 8B10B encoding then you should be able to send any combination of idle characters and data. You should probably be using LPM equalization. Is this the case?
10-22-2018 08:09 PM
I am using 8B10B, and I tried 1G serdes. It won't be a problem for 1G serdes(throughput 1G, 500M, 100M, 10M, 1M)
On the other hand, I changed 2.5G serdes's reference clock from 156.25M to 200M, find out if reference clock is 200M, packet will drop when throughput is 2499M, and then will be worse if throughput going down(1500M loss 1%.500M loss 88% 100M loss 92% ).
When reference clock is 156.25M, only throughput is less then 1500M will have packet loss(1000M 7% 500M 88% 100M 92%).
So it will be the reference clock problem?
p.s. 1G serdes reference clock 156.25M
10-22-2018 11:13 PM
any 8b10b decode error indicated?