[time-nuts] NIST UT1 NTP server results
Tom Van Baak
tvb at LeapSecond.com
Sat Jul 23 19:02:32 EDT 2016
> Interesting plots. A couple of points.
> 1. These look like the data points are taken at 0h and without intermediary
> measurements as the data points are connected by straight line segments.
> If we don’t know what the intermediary data points are, the plots, to my mind,
> should be presented « with steps ».
Here's the two IERS sites I frequent. Maybe you can locate data with greater precision or shorter tau:
I like the plots a lot. These are phase measurements of a continuous oscillator (earth). It's ok to connect time/phase points. You can argue about the rate/frequency data: For this, Stable32 uses steps; TimeLab uses cubic spine. I would suggest you work with the raw data (which includes error bars as well). It's the same data I used for:
Are you asking for fraction of a day resolution at tens of microseconds accuracy? I don't think VLBI gives that. This is not unlike any of us trying to measure a quartz oscillator -- there's a trade-off between resolution and accuracy; between short-term noise and long-term precision. We all face this, whether using nixie counter or a TimePod. Every oscillator (including earth) and every measurement system (including VLBI) has an ADEV curve.
I believe this is what Judah was talking about when he mentioned interpolation; that is, the sub-day precision of UTx measurements. So see what you can find, look into different interpolation schemes, see what practical effect it would have on the average computer running NTP. There might be a best case; there might be a simplest case. It's probably worth looking into.
> Steps aren’t good enough for reasons I previously outlined.
Take a step back and consider time & frequency measurement 101. The earth is an oscillator. You cannot measure it continuously or with perfect resolution. I know you don't like steps. Yes, oscillators are analog. But when you make periodic, digital measurements you get steps, in both time and value. Even a GPSDO has steps (the DAC).
> I suppose that the answer to that depends on the objective of having a readily
> available accurate UT1 timescale realization. It has an intrinsic value unrelated
> to leap seconds so we should have it.
Earth is a very noisy, wandering, drifting, incredibly-expensive-to-measure, low-precision (though high-Q) clock. I think it's cool to try to discipline the quartz oscillator in a computer against earth. But I'm still wondering if you or someone else could outline the set of advantages this provides.
More information about the time-nuts