search

sinara-hw / Urukul

4 channel 1GS/s DDS (AD9910 or AD9912 variant)
13 stars 7 forks source link

L9 #7

Closed hartytp closed 5 years ago

hartytp commented 5 years ago

@gkasprow is L9 populated by default. IIRC, it's not, but that's not marked on the schematic.

sbourdeauducq commented 5 years ago

Using the built-in XO does kill reliable/predictable IO_UPDATE timing, sync, and phase modes, which I foresee will be implemented at some point.

There is differential stability between Urukul channels is unaffected which further reduces the need for absolute external references.

If you are unlucky, with no sync between DDS and FPGA, some channels can register IO_UPDATE in one cycle and others in the next.

jordens commented 5 years ago

Yes: surprisingly the XO is a stupid idea if you need an absolute phase or/and absolute frequency. I am not saying that the XO should be the only solution! Currently none of that is implemented and exactly nothing is killed. And it only matters for very specific use cases (only AD9910, only if the SYNC clock and the RTIO clock are derived from the same source, only if the FPGA drives SYNC, only if absolute phase determinism is needed). The default relative phase mode works fine. And using the XO won't even kill synchronization between DDS. And finally there is the SYNC clk on the EEM connector that can be used to retime IO_UPDATE to meet S/H at all DDS. Still all with the XO or an external clock.

sbourdeauducq commented 5 years ago

And using the XO won't even kill synchronization between DDS.

It does between Urukul cards.

hartytp commented 5 years ago

@jordens okay, maybe I misunderstood you. I'm certainly not trying to imply that the MMCX is the best solution for all use cases, or that it should be the only option provided. I simply feel that it's a useful enough option that it should be enabled in the "default" Urukul configuration without having to get a soldering iron out.

I can see the utility of all three clocking options (XO, SMA, MMCX) and expect all to be used regularly. That's why I like having the two muxes to make that simple.

hartytp commented 5 years ago

More generally, we've previously had in depth discussions about the features you suggested removing. Someone has made a case for each of them, and we agreed that they add value to the base design. As @gkasprow pointed out, the marginal cost of these features is low compared with volume pricing, so I still think it makes sense to keep them in the default design.

If you feel strongly that we should reopen these discussions then we can do. I just don't think that I have much to say on top of all the previous discussions on these matters.

gkasprow commented 5 years ago

One can always order previous release...

hartytp commented 5 years ago

@gkasprow yes, but:

Anyway, since it's all open source, any group can order whatever version they like. If they're ordering enough units then a custom version with some features removed might well make sense.