> > The time to write an 8 bit word to the DDS is 1+3.8+2.1+3.8+10.5 ns =
> > 21.2 ns. From Table 13 of the AD9914 Data Sheet.
> 
> It seems there are further limitations when using the register programming
> mode. Page 29 says:
> "The parallel mode allows the user to write to the device registers at rates 
> of
> up to 200 MBps using 16-bit data (or 100 MBps using 8-bit data)."

I had not noticed this discrepancy between p 29 and p 33 of the AD9914 Data 
Sheet. Indeed the 100 MBps rate is slower than one would guess just adding up 
the required delays in Table 12. Daniel, do you understand this? 

> > Why do you anticipate this taking so long (>1000 ns)? It looks like
> > the
> > AD9914 can be written far faster.
> 
> The other operations. As explained in a previous email thread, a RTIO
> transition takes ~1 microsecond to program, and a DDS programming takes
> two such transitions for FUD, plus all the other steps. The DDS bus bandwidth
> limit is not responsible for a large part of the delay here.

I had overlooked that the DDS programming would be performed using the RTIO  
core. Indeed this has been discussed in the past.
https://ssl.serverraum.org/lists-archive/artiq/2014-October/000113.html

I'd like to discuss options for making this faster when you visit Boulder next 
week. I gather options include
a) streaming from DRAM
b) offloading DDS programming to a MIGEN state machine that's faster than RTIO

I'm also don't well understand what the uP is doing for each RTIO transition 
(and why it takes 100's of clock cycles). It would be helpful to discuss this 
too at greater length. 

-Joe 
_______________________________________________
ARTIQ mailing list
https://ssl.serverraum.org/lists/listinfo/artiq

Reply via email to