[time-nuts] Data collection cycle -- multiple sources to onecounter?
Tom Van Baak
tvb at leapsecond.com
Sun Jun 19 11:10:15 EDT 2005
There are quite a few coax switches you can use:
HP 59307A VHF switch
HP 34907A with HP 34905A RF multiplexer module(s)
HP 3488A with 44472A VHF switch plugin(s)
HP 54300 oscilloscope probe mux
VXI system with HP E1472A 50 Ohm Multiplexers
The dwell time will depend on the quality of your
GPS source. I mean, if you're doing really long
term collection, one data point per day is all you
need. But if you have white receiver noise or
sawtooth jitter to remove then make sure your
measurement is long enough to cover that. An
M12 needs less averaging time than a VP, for
But things to watch out for:
I think 15 minutes sounds fine. If you're looking
for exactly 4 readings per hour, on the hour, you
have to take into account the second(s) it will
take to activate the switch and let the line settle.
Maybe you can activate the switch in between
1 PPS pulses and just make it in time. But in
general don't take 900 second readings and
expect you'll get exactly 96 readings over the
86400 s day.
If you are measuring 1 PPS your switch set up
will probably work fine. But if you are measuring
5 MHz RF then watch out for impedance changes.
The other 3 lines will be at no load for 45 minutes
while the one you're measuring is at 50R. Most
frequency standards are affected by changes in
load. One solution is the (more expensive, usually)
VHF switches with internal termination for all inputs
regardless of the state of the switch. The other
solution is to let the frequency settle after a load
change; but this can take minutes. Constant load,
regardless of switch position, is the ideal. Another
idea is to 50R terminate each input with a BNC tee
at the switch and make your measurements at
1M. Check this with a 'scope before you trust it.
Again, all this concern is for RF signals; not to
worry about this with GPS 1 PPS signals.
To preserve relay contact life when using switches
like the 59307A, think gray code instead of binary
code; that is; make your measurements so that
you have only one relay contact change per
measurement. Look at the schematic, or listen
with your ears. You want a switch order like
00 01 11 10 instead of the normal 00 01 10 11.
Gray order is one relay change per measurement
while binary order is 1.5 (average) relay changes
per measurement; so gray gives you 50% more
relay contact life.
Expect some of your surplus switches to have
trashed contacts due to age. I don't know if you
can detect contact wear with a micro ohmmeter
but you might want to just run them through a
self test, using the same ref signal through the
relay mux and see if the 5370 standard deviation
is 25 ps (good contacts) or something much
larger (bad relay contacts) as you repeatedly
Most of my 1 PPS sources are deliberately set
(or allowed to be) many microseconds advanced
or retarded, to insure the phase never drifts across
zero during a measurement run. It's not that it's
a big problem, it just adds some pain to the data
analysis. Start/stop channel choice is arbitrary
and I usually pick them so that I get all positive
numbers rather than negative or mixed sign.
At the level of your test I don't think cable types
matter but at least secure them with tie wraps to
prevent phase changes due to vibration, wiggling,
and the occasional lab visitor. BNC connectors
are not as good as TNC, SMA, or N so I think
some mechanical tie down is advised if you want
clean data for a year.
----- Original Message -----
From: "John Ackermann N8UR" <jra at febo.com>
To: <time-nuts at febo.com>
Sent: Saturday, June 18, 2005 17:37
Subject: [time-nuts] Data collection cycle -- multiple sources to
> I want to monitor all my atomic devices over the long term against GPS.
> Rather than devoting one counter to each source and gathering a lot more
> data than is useful, I'm going to use a GPIB-controlled coax switch to
> route the 1pps from each source into the counter START input. That way
> I can alternately read each source for X minutes against GPS.
> My question is whether there is an optimum "X" for the time to dwell on
> each source. I've been thinking that 15 minutes might make sense. That
> means I'd start with device A, average it against GPS for 15 minutes of
> 1pps samples, record that to the log file, then switch the input to
> device B, dwell there for 15 minutes and record, then to device C for 15
> minutes and record, then repeat the cycle.
> With three devices, that would give me averages of 900 readings and a
> tau of 45 minutes. Seems reasonable, but I suspect a lot of labs do
> something similar and I wonder if there's tribal knowledge of the
> optimum cycle.
> Any thoughts?
More information about the time-nuts