[time-nuts] 5370A/B measurement floors (was Re: Fixing an HP 5370A...)

[Magnus Danielson]

John Miles wrote:
> It's worth noting that the 5370's accuracy is supposedly never better than 1
> ns (specifically, +/- resolution +/- timebase*interval +/- trigger jitter
> +/- 1 ns systematic).  We tend to rely on it to do much better than that --
> or at least we all count on the "1 ns systematic" to look like random noise
> in the long run -- but it still bears remembering, because that systematic
> error swamps any other effects in the short term.
> 
> Some time ago John A. plotted his 5370's residual floor, about halfway down
> the page at http://www.febo.com/pages/hp5370b/ .  He fed a 1-pps signal into
> a T-connector at the START jack, followed by 35 ns of coax to the STOP jack.
> I ran a similar test recently, but with a TADD-2 1-pps divider driving the
> START jack and a similar amount of coax driving the STOP jack from the 10
> MHz input side of the TADD-2, trying to get an idea of the composite floor
> for a typical measurement using the TADD-2, the 5370B, and a bunch of coax.
> The 5370B was driven by a Thunderbolt for this test.
> 
> My results (attached) were a bit worse than John's, but with a nearly
> identical slope for the duration of his test.  I left it running for a few
> days, and saw some significant flattening on timescales of 3-4 hours that
> matched my air conditioning cycle time (as reported by the shop thermometer,
> aka Lady Heather.)  Based on this graph I don't consider temperature
> variations to be a grave concern above floors of 1E-14, or conversely at
> timescales less than an hour or two, but they should definitely be minimized
> or eliminated for HP 5370A/B measurements below that.
> 
> I still need to try variations of this test with the internal reference,
> with and without the TADD-2 between the START and STOP jacks, and with
> 10-second and 100-second means on the 5370B.  Clearly it would be good to
> achieve better than 1E-10 accuracy at t=1s, since most good crystal
> oscillators are better than that.

It would be interesting if you could plot the tau*ADEV(tau) curve, then 
the long-term deviations would be more apparent. This is a form of 
TDEV-similar plot, but if you only have TDEV that will also help.

tau*ADEV(tau) is about 7E-11 up to tau 1000 and then rises as 
environmental effects kick in.

If I had the time I would ask for the raw-data and fool around with that.

Cheers,
Magnus