We have tried to measure the AD9912 output frequency resolution, but our measurements always show a DDS output frequency corresponding to an FTW with the LSB set to zero.
From the datasheet we would expect a step of 1e9/2**48 = 3.55 uHz, but since the FTW LSB seems to be always zero, we get the same frequency measurement result for e.g. FTW_even and FTW_even+1. We then see the expected 7.11 uHz step at FTW_even+2, but no change in frequency at FTW_even+3. The measured frequencies at even FTW agree with what we expect to better than 0.1 uHz (depending on the measurement time).
We have tried this with two different AD9912 boards, the evaluation-board AD9912A/PCBZ, and an ARTIQ-board called Urukul: https://github.com/sinara-hw/Urukul/wiki
On the Urukul-board we tried 'readback' of the FTW once it has been set. We do see the correct programmed odd FTW ending in a '1' - but the output frequency corresponds to an even FTW with a '0' LSB.
Our results look like this:
The measurements are performed in two different ways. One is with a Microsemi 3120A phase-meter, and to verify we divided the output frequency down to a 1 Hz signal and made the measurement with a Keysight 53230A time-interval counter. Our frequency reference is a Hydrogen-maser output at either 10 or 100 MHz which is used to generate a 1 GHz SYSCLK for both boards. The same Hydrogen maser is used as the reference for the measurement instruments. In both cases we changed the FTW with 30 or 60 minute intervals, and computed the average frequency during the time the FTW was constant. With a Spectrum Analyzer the output signal looks reasonably clean, with expected levels of harmonic spurs.
Additionally we saw this same effect around 400 MHz output frequency (not shown in the figures).
Any ideas on what could be causing this? Is there a documented measurement/setup for how to verify the 3.55 uHz resolution? Is there some mode/setting in the AD9912 that could be causing this?
Upon investigating the issue in your post, we have discovered an unintended design flaw that invalidates the least significant bit of the frequency tuning word. As a result, the effective frequency tuning resolution of the AD9912 is, indeed, 47 bits rather than 48 bits.
The next revision of the data sheet will be corrected accordingly.
We appreciate you posting your findings and bringing this issue to our attention.
Thanks for confirming this!
A feature - not a bug ;) meaning we won't search any further for faults in how we program/use the DDS or in our measurement system.