[ntp:questions] Timing GPS recommendations

unruh unruh at invalid.ca
Tue Aug 21 16:02:58 UTC 2012

On 2012-08-21, Chris Albertson <albertson.chris at gmail.com> wrote:
> Actually can CAN set a remote clock very accurately.  This is how NTP works
> and what it idoes best.    NTP uses the Internet to synchronise clocks to
> MUCH higher precision than the delay in Internet messages.
> Try this:  You try to adjust the rate of an old mechanical clock by
> adjusting a screw that says "faster/slower".  But your "reference clock" is
> across the street and you have to walk back and forth.   This can actually

Not if sometimes your courier stops for coffee, sometmes waits for
traffic, etc. Ie, the noise of the time for walking across the street is
variable, and highly non-gaussian with very fat tails. 
> work very well if you can wait a few months.   First you walk across the
> street and back and measure the time it takes.  Do this many times and keep
> the mean and sigma.  Now you can set your clock to what you read on the
> reference clock but add in 1/2 the mean round trip time.   Now wait 100
> days.   Go and read the reference clock and come back add 1/2 the mean and
> note the error.  Let's say your clock gained one minute.   Now you know the
> rate of your clock relative to the standard to better then one part in
> 100,000.  We adjust the screww to remove some fraction of the observed

Unfortunately it does not get "very small". If you do it long enough and
if the rate of your clock does not vary, you can  stabilize the rate
this way, but not the time itself. And if the clock hardware rate
varies, then by the time you waited 100 days, the rate has changed, and
that measurement is useless, and waiting longer then makes the rate
worse rather than better. (minimum allan variance and all that). 

> error.  Then we wait aother 100 days and do it again.  Eventually the error
> in the rate gets small.    NTP uses this same method but repeats it
> thousands of times.
> The accuracy with which you can set your remote clock by walking down the
> street depends on measuring the mean time to wak down the street.  But we
> are using a computer so measure many tests

That does not help, if the "walking down the street" time varies. You
can never get below some accuracy ( and in this example it would be many
seconds, certainly not ms.)

> On Mon, Aug 20, 2012 at 12:17 PM, unruh <unruh at invalid.ca> wrote:
>>  Imagine that you have a wristwatch
>> that is accurate to the second but your send out messangers by foot to
>> deliver the time to your friend across town. Do you really think that
>> buying a new wrist watch is the way to improve the time your friend
>> gets?
> Chris Albertson
> Redondo Beach, California

More information about the questions mailing list