[time-nuts] Why not TAI? (was: The future of UTC)

[Magnus Danielson]

On 11/08/11 18:20, Tom Van Baak wrote:
>> My Dual Scale Timekeeper will recover TAI from GPS by adding a constant
>> 19 s offset, and it will track and serve out TAI in addition to UTR.
>
> What is usually meant by "TAI" is the single extremely accurate,
> post-processed, paper time-scale managed by BIPM. TAI itself
> is derived from EAL and other inputs. TAI is the basis of UTC.
> No one has copyright on these acronyms and confusion can
> result when TAI is used to mean too many things.
>
> You can "recover TAI" from a GPS timing receiver by adding 19
> seconds in the same crude way I can recover TAI by looking at
> the big clock outside the bank building and adding 7 hours and
> 34 seconds. Or by taking a PC clock and making my web page:
> http://leapsecond.com/java/gpsclock.htm.
> Yes, these look like TAI. But are they really TAI? Or are they all
> just another 6-digit hour:minutes:second clock display that tries
> to be "close to" what TAI would look like if one had access to it?
>
> A question to ask is how many nanoseconds, or milliseconds,
> or even seconds does your TAI clock have to be off before you
> can't rightfully call it TAI anymore? I don't have an answer.
>
> Naming ambiguity is even worse with UTC. You can't have a clock
> at home that is UTC. What you can have at home is a WWVB
> clock that closely follows UTC(NIST) or a GPS display clock that
> closely follows UTC(USNO). But how close is left undefined.
>
> If you put the GPS receiver in holdover mode, when does the
> display stop being UTC?
>
> Most Windows PC's at home are off by seconds. Does that mean
> most of them are running UT1 instead of UTC?
>
> What is missed in many discussions about time scales is intent
> or implied accuracy. If I manually adjust my Pacific Daylight
> Time wrist-watch ahead by 7 hours does it then become a UTC
> watch? If I further adjust it by 0.3 seconds can I now claim it's
> showing UT1? Can I even wear a wrist-watch that displays TAI?
> Is it possible for any clock with analog hands to display UTC?
>
> What we call a time scale is more than just an integer offset. I'm
> working on a paper that explores all these issues.

How close you need to get depends on your application and it's needs.
If you need to be within +/- 1 us, then turning off your outputs when 
you expect to have drifted away 1 us from where the time-scale should be 
is resonable, this is a confidence interval thing.

Whichever time-scale you try to realize, you will have biases, 
deviations and additional noise uncertainty. How much you tolerate and 
the requirement on hold-over depends on your application.

It should not be very surprising.

The concept of hold-over seems to be not as well covered in the 
literature as you would expect.

Cheers,
Magnus