[time-nuts] How close can you trim a Cs?

Chuck Harris cfharris at erols.com
Wed Mar 9 11:31:22 EST 2005

John Ackermann N8UR wrote:
> Chuck Harris wrote:
>> Hi John,
>> The manual for my old Sulzer 2.5A crystal oscillator says to not make
>> frequency adjustments more often than every 72 hours.
>> I never understood exactly why, but that is what they said.
>> I know that with my Austron 2100F/T loran receivers, the displayed
>> frequency offset reading takes about 12 hours before it is strictly
>> correct.  Last night I had just finished surgery on my Sulzer 2.5A
>> and was tracking it with two 2100's, one set for the Master, and one
>> set for a Secondary transmitter.  I started the two receivers tracking
>> about 30 minutes apart.  At the end of 3 hours, one 2100 said the
>> standard was 10E -2.8, and the other 2100 said it was 10E +3.2.
>> By morning, they both agreed that it was 10E +5.8.  If I had started both
>> receivers tracking at the same time, they would have both been in exact
>> agreement.
>> My point is the software that calculates the frequency offset in these
>> receivers (and probably your GPS receiver too) is kind of funny.  Because
>> the tracking loops have very long time constants, you need to give the
>> receivers plenty of time to settle down.
>> -Chuck
> Hi Chuck --
> Thanks for the response!  I'm using the raw 1pps output from a Motorola 
> UT+ receiver as the "stop" input to a time interval counter (with 2ns 
> resolution), and the 1pps from the 5061A as the "start" input.  I then 
> log the time interval at 100 second averaging and plot that.  So, my 
> situation is a bit different than using the Loran receiver (which I 
> agree takes a long time to stabilize).  The real issue is the noise on 
> the GPS signal, which with 100 second averaging is around 10 nanoseconds 
> with occasional spikes of 20ns or so.  But over several hours you can 
> get a pretty good view of what the slope is -- the question is whether 
> it's good enough to measure parts in the low 13s or even better, high 14s.
> John
Hi John,

But the problem is still quite similar.  The short term stability of the
GPS is poor.  A good rubidium should be better, but the long term stability
is very good.  So, that means averaging with a long time constant loop.
Once you get into the long time constant loops, you are back to having to
wait a long period of time to see changes that are small parts in 10E13.

Also, at 2ns, your counter probably isn't high enough resolution to do a good
job with a 1 second gate.  Up your gate to 100 seconds, and it will quiet
things right down, and your ability to see small parts in 10E13 will vastly

The reason I say this, is a counter with a 2ns resolution will have biases
in its timing that are a significant portion of that 2ns.  These
biases will change with temperature/voltage, etc. And will skew your record
even after extensive averaging.  (eg. 100*(f(t) + bias)/100 = avg(f(t)) + bias)


More information about the time-nuts mailing list