[time-nuts] Why using HP5370 ext-ref is (maybe) a bad idea

[Magnus Danielson]

On 28/02/14 22:51, Poul-Henning Kamp wrote:
>
> A long time ago, I found out that the HP5370 is quite sensitive to
> qualities of the external reference signal and after playing around
> with it a bit, I decided to run my HP5370 from its own OCXO since
> that was both reproducible and eliminated what I suspected was the
> root cause.
>
> While playing around with John's new CPU board, and now having a bit
> more kit in my lab, I decided to revisit this detail.
>
> The setup I created is the following:
>
> 10 MHz GPS locked "lab-standard" feeds ext-ref on the HP3336.
>
> The HP3336 generates 10MHz/0dBm which feeds ext-ref on the HP5370
>
> The same lab-standard also feeds the start input of the HP5370
> which is setup to start-common, TI, 1k samples, output stddev.
>
> And then I step the phase of the HP3336 generated 10MHz through
> 0...360 degrees relative to the lab-standard.
>
> The result is the attached plot, where for every 18 degrees
> the stddev increases by 8-10ps, roughly 40%.
>
> This is evidently because the ext-ref on the HP5370 is multiplied
> to 200MHz, which is what drives the counter circuits.
>
> Another way to run this experiment, is to set the HP3336 to
> 10.0000001 MHz and log the stddev's over time while the
> two clocks sweep each other by in phase.  Doing it this way
> can give you a plot of much higher resolution.
>
> And that scenario is where the trouble starts:
>
> If the HP5370 ref-in clock synchronous to the experimental
> signals, you will most likely be lucky, but sometimes you will not
> and the noise will be much larger.
>
> If the HP5370 ref is not synchronous, for instance running of its
> own OCXO, the two phases will sometimes conspire briefly and you
> get a few noisy samples, but the average will almost always be good.
>
> I have not tried to calibrate/trim the HP5370 to see what that
> does to these spikes, but it would be an interesting experiment.

I'm not a bit surprised. I tried using the normal HP5370 trimming 
routines, but I found that using my SIA3000 helped a lot on the 200 MHz 
synthesis chain.

Also, as I have told before the board doing the 10 MHz logic spews out a 
lot of 5 MHz with overtones, which is a simple mod away.

Would be interesting to see if you could trim these systematics down by 
tweaking the syntesis chain. Maybe some peaks is good indicators for a 
particular stage offset, which would be expected from the x5 followed by 
x4 multipliers with tons of filters. The 50 MHz tank would be a good 
prime suspect I would think.

Cheers,
Magnus