[time-nuts] Racal-Dana 1992 switches

[Bruce Griffiths]

Jeff Mock wrote:
> How do you pick the optimal difference frequency?  I see that 1kHz has a 
> nice numerical property where you can read the frequency directly off 
> the counter, you just need to mentally prepend the first 4 digits. With 
> computers it's not that important, the difference can easily be a 
> strange number if it optimizes performance. I'm wondering what 
> difference frequency optimizes the performance of the mixer thing or if 
> it really matters?
>
> Do you worry about the phase-noise contribution of the 10.001MHz source? 
> As I do the math, it seems that the phase noise of the mixing signal is 
> subtracted out after the mixing, so it shouldn't mater that the 
> 10.001Mhz source comes from a frac-N synthesizer and has a few random spurs.
>
> You say this isn't state of the art.  Why not?  Can't you run the timing 
> collection for longer runs and get higher resolution results?
>
> jeff
>
>
>   
Jeff

Whilst a naive analysis indicates that the phase noise of the offset 
oscillator isnt critical, in practice it is.
The dependence on the offset oscillator phase arises because zero 
crossings of the 2 beat frequencies do not occur at the same time.
The greater the time difference between the 2 zero crossings the more 
critical the phase noise of the offset oscillator becomes.
Random spurs do matter, as JPL has found with their nominal 100Hz beat 
frequency system, performance is much better when the synthesizer is set 
for a 123Hz beat frequency.

The optimum beat frequency is that which minimises the system noise.
As the beat frequency is lowered the resolution, for a fixed accuracy in 
measuring the zero crossing times, increases.
However as the beat frequency decreases the mixer output noise density 
increases.
Thus a balance between increasing noise at low offset frequencies and 
increasing resolution at low offset frequencies has to be found.
JPL currently uses 100MHz mixer input frequencies and a nominally 100Hz 
beat frequency.
NIST have also moved from 10MHz mixer input frequencies with a 10Hz beat 
frequency to 100MHz mixer input frequencies and a 100Hz beat frequency.

The system isnt state of the art because the zero crossing detector 
performance is far from state of the art.
A diplexer isnt the optimum IF port termination for low noise at low (< 
100kHz) beat frequencies.
State of the art systems have a system noise level of around 1E-14/Tau 
or less.
The use of tuned circuits in the zero crossing detector increases its 
phase shift tempco over that of a state of the art design.
Thus as the averaging time increases ambient temperature variations will 
ultimately limit the system noise.

In fact a similar performance to the proposed system is possible using 
an analog phase comparator if one takes a little care in the design.

Bruce