Re: Timing Definitions

On Feb 18, 2007, at 10:32 PM, David L. Mills wrote:

> There is no such thing as accuracy in the product, only
a  
> statistical estimation of the resolution, precision and
other  
> statistics. There is no specific accuracy statement
possible in  
> this thing. It's like saying, how pretty is your
girlfriend.

(One is tempted to point out that the way to answer the
latter  
question is to ask one's wife.)

Isn't the same issue as traceability?  Perhaps convolved
with the non- 
parametric nature of the clock discipline?  The whole point
of  
timekeeping is to tie a remote clock to a master time
standard.  This  
is distinct, say, from identifying physical limits on the
accuracy of  
absolute intervals returned by an egg timer.

In general, unless a trusted measurement of whatever
quantity is  
available, an estimate of accuracy is impossible.  The
mental picture  
usually used to convey the difference between accuracy and
precision  
is target practice.  Tight clustering of bullet holes is
precision.   
Having the average pattern spread coincide with the center
of the  
target is accuracy.

The two things missing from that last statement are first,
an  
indication of the repeatability and reliability of the
shooting/ 
measurement.  What did the same marksman do last time (to
provide  
confidence for next time)?

And secondly, the point in contention with this thread is
how do you  
know where the center of the target lies?  With target
shooting the  
answer is that the target is much larger than the bullet in
both  
absolute size and solid angle.  With timekeeping, we demand
that our  
fiducial systems of time and our portable/remote estimates
of same  
look and behave like the same things.  Our bullets and our
targets  
are the same size.

There would be (relatively) little difficulty in using a
nanosecond  
standard timekeeper to measure the accuracy of a
tenth-second laptop  
clock - at least, little difficulty if you remove the
vagaries of the  
intervening network in favor of something other than NTP...

Rob Seaman
National Optical Astronomy Observatory



(http://www.globalsecurity.org/military/library/p
olicy/navy/misc/ 
image434.gif)
_______________________________________________
ntpwg mailing list
ntpwglists.ntp.isc.org
http
s://lists.ntp.isc.org/mailman/listinfo/ntpwg

On Mon, Feb 19, 2007 at 07:35:40AM -0700, Rob Seaman wrote:
> On Feb 18, 2007, at 10:32 PM, David L. Mills wrote:
> 
> >There is no such thing as accuracy in the product,
only a  
> >statistical estimation of the resolution, precision
and other  
> >statistics. There is no specific accuracy statement
possible in  
> >this thing. It's like saying, how pretty is your
girlfriend.
> 
> Isn't the same issue as traceability?  Perhaps
convolved with the non- 
> parametric nature of the clock discipline?  The whole
point of  
> timekeeping is to tie a remote clock to a master time
standard.  This  
> is distinct, say, from identifying physical limits on
the accuracy of  
> absolute intervals returned by an egg timer.
> 
> In general, unless a trusted measurement of whatever
quantity is  
> available, an estimate of accuracy is impossible.  The
mental picture  
> usually used to convey the difference between accuracy
and precision  
> is target practice.  Tight clustering of bullet holes
is precision.   
> Having the average pattern spread coincide with the
center of the  
> target is accuracy.

The maximum error in a measurement is the sum of the error
caused by
accuracy and precision.  What we do is calculate the maximum
error (and
estimate the error).  From this number you can't split it up
in which
part is caused by the accuracy and which by the precision
error.

If we could somehow calculate the accuracy error, and the
resolution of
our clock is smaller than that, we could reduce it to
something in the
order of our resolution.

Anyway, depending on the application, you're either
interrested in the
maximum total error or in the precision.  I can't think of
something
were you're interrested in the accuracy.


Kurt

_______________________________________________
ntpwg mailing list
ntpwglists.ntp.isc.org
http
s://lists.ntp.isc.org/mailman/listinfo/ntpwg