Conversation

Notices

1. Embed this notice
   Andrew Zonenberg (azonenberg@ioc.exchange)'s status on Tuesday, 24-Dec-2024 21:21:23 JST Andrew Zonenberg

   Managed to tear myself away from holiday stuff long enough to look at the sim run for the QSGMII TX pair.

   It's decidedly less well optimized than the RX but, honestly, I want to get this thing out the door and it's probably good enough (S11 of -16 dB at 5 GHz, -13.6 dB at 7.5 GHz).

   TDR response shows the main thru line around 105Ω which is probably a meshing artifact, then a big high impedance spike (to 115Ω) early on with some ringing, then a big low-impedance dip (to 85Ω) at the connector launch.

   • Embed this notice
     Andrew Zonenberg (azonenberg@ioc.exchange)'s status on Tuesday, 24-Dec-2024 21:21:21 JST Andrew Zonenberg

     in reply to

     And just for kicks, let's repeat with some real data.

     Here's 5.15625 Gbps (close enough) PRBS-15 from a Kintex-7 GTX (no UltraScale+ test boards handy but should be close enough) through 24+ of coax then into my 16 GHz scope.

     Rich Felker repeated this.
   • Embed this notice
     Andrew Zonenberg (azonenberg@ioc.exchange)'s status on Tuesday, 24-Dec-2024 21:21:21 JST Andrew Zonenberg

     in reply to

     And here's what we model.

     Top is the actual waveform seen by the scope, bottom is channel emulation through the EM model of the QSGMII launch on the board. I can barely see the difference.

     Going to call this "good enough".

   • Embed this notice
     Andrew Zonenberg (azonenberg@ioc.exchange)'s status on Tuesday, 24-Dec-2024 21:21:22 JST Andrew Zonenberg

     in reply to

     S21 response does show some ~1.5 dB dips from reflections, but not until 12.3 GHz.

     I don't *like* it but I'd probably get away with it for QSGMII.

   • Embed this notice
     Andrew Zonenberg (azonenberg@ioc.exchange)'s status on Tuesday, 24-Dec-2024 21:21:22 JST Andrew Zonenberg

     in reply to

     Time domain simulation says yes: it's completely fine. The eye is big enough you could drive a truck through it.

     Using +/- 300 mV swing at the TX it passes the QSGMII RX eye mask with a massive amount of margin and zero mask hits.

     Obviously the real eye won't be this pretty as there will be losses in the cable, but those will probably dominate whatever my board layout did to the signal.

     I could repeat the model doing channel emulation with actual waveforms coming off a Kintex-7 GTX at 5 Gbps through a half meter or so of coax, but with results this good it's not going to change anything.

     Might do it anyway just for fun.