[time-nuts] Rubidium Performance: DDS noise effect on 10 MHz

[Rex]

On 2/4/2012 12:17 AM, John Miles wrote:
>> In the current 5680A units, the 10 MHz output comes from the 60 MHz VCXO
>> (divided by 6 in CPLD) and not direct from a DDS. If my architecture
>> understanding is right, the DDS signal output is mixed with the VCXO
> output
>> only at the 114th harmonic of 60 MHz, and it's the PLL (looking at the
> optical
>> signal from the Rb) that drives the VCXO to keep it lined up.  As I
> understand
>> it, DDS phase noise should be divided by a factor of 6*114 by the time it
>> appears at the 10 MHz output, and at larger frequency offsets the
> amplitude
>> should also be (significantly) reduced by the PLL loop filter.
>>
>> I don't have any phase noise measurement tools myself, so this is just an
>> academic argument, but if there is significantly more noise on the 10 MHz
>> than expected for a 60 MHz VCXO, I wonder if it's just inadequate
> filtering of
>> an internal power rail.  Is the unit under test being driven by a linear,
> or
>> switching supply?
> What made me suspect the AD9832 is that the PN/spur behavior is so close to
> what I'd expect from it.  Analog Devices' own plots contain a lot of strong
> spurs, and the chip's actual SFDR spec is in the -70 dBc range that I'm
> seeing (blue trace attached).  But yes, you're right, it would make a lot
> more sense to use the DDS as an offset generator for the control loop,
> rather than as an output device.
>
> I tried it with a couple of different supplies, linear and otherwise, and
> the noise didn't change meaningfully.  Since it's so far out of line with
> FEI's specs, I have to believe that I either got a couple of bad examples --
> although they lock very quickly and look essentially unused inside -- or I
> don't have the pinout right, and am using the wrong ground or something like
> that.  Seems pretty unlikely that this is normal operation.   Let's see what
> JohnA comes up with.
>
> -- john
>

John,

 From your comments, it sounds like you may be measuring one of the 
earlier 5680A's that could be tuned over a large range with a DDS output.

These earlier ones had a 50.25 MHz internal osc which was locked after 
multiplication to the rubidium frequency. Then there was a DDS on the 
output that could be programmed over a wide frequency range.

The newer ones have a 60 MHz internal oscillator. This is multiplied up 
to 6840 MHz and subtractively mixed with ~5.3125 MHz from a DDS to get 
to the rubidium frequency. The feedback from the rubidium cell locks the 
60 Hz which is divided by 6 for the 10 MHz output. No DDS on the output 
generation. The DDS in the loop can digitally adjust the rubidium lock 
frequency that tunes the 60 MHz, to fine tune the output only around 10 
MHz.

Do you know which type of 5680A you are measuring?