On 09/19/2014 11:06 AM, Slichter, Daniel H. wrote: > I heartily agree on all counts. However, for back compatibility with > current DDS and TTL breakout boards I thought we should at least have > the option. We can use commericial SCSI cables to improve the > reliability over the current homemade ribbon cable method. > Furthermore, I have done some looking into various chips to do > serialization/deserialization to enable us to use high speed serial > for the FPGA-DDS and FPGA-TTL connections. The issue is that almost > all the chips I found have nondeterministic latencies from chip to > chip, and I worry about all the effort we are putting in to making a > nice RTIO core if in the end we are limited by the performance of the > deserialization at the end (jitter of +/- 1 slow (parallel) clock > cycle is a typical spec as well).
I fear you have your terminology confused up here. A jitter of one parallel UI is unlikely. Anyway. Latencies are already non-deterministic in our setups. Worse, they are unknown and uncontrolled. We need a return channel anyway to pin them down. And once you have a return channel you can use any physical layer. > If we are able to find a suitable technical solution for the SERDES, > my thought for high speed serial would be to use DisplayPort > connectors and cables, which have 4 differential pairs and can be > used for long (3 m) runs at full bandwidth (~8 Gbit/s for DisplayPort > 1.0, more than we likely need). DisplayPort is pretty much one-way. If you really need that many lanes and that data rate, something like SNAP12 would do it. Optical layers have the advantage of much longer lengths and galvanic isolation. > There exist chipsets which implement specific SERDES protocols (e.g. > HDMI, DisplayPort) but the latencies and jitter associated with these > are even worse. For the applications for which they are used, there > just isn't much incentive to care about absolute timing to the degree > we need. Robert. _______________________________________________ ARTIQ mailing list https://ssl.serverraum.org/lists/listinfo/artiq