DRAFT


Time Transfer Issues at Arecibo Observatory - 2000



D. C. Backer (UC Berkeley), B. Genter, M. Davis, A. Vazquez, A. Venkataraman, M. Nolan, K. Xilouri (UVa), D. Nice (Princeton), Trudi Peppler (NIST) ...
2000 March 15



Outline


  • Introduction
  • Schematic
  • Unix Time
  • Maser Status
  • AO Clock Comparator System (CCS)
  • Princeton Time Monitoring System
  • NIST Monitoring
  • Miscellaneous Notes, ToDo's
  • Propagation in Earth's Atmosphere

    Introduction

    Precision pulsar timing, multi-observatory single pulse measurements, radar astronomy and other programs demand accurate transfer of Atomic Time from standards laboratories to the observatory via GPS. The greatest demand on this time transfer is for precision pulsar timing owing to the current precision which can reach down to 25-50 ns on several objects in 30-60 minutes of observing. Stability of time transfer over a decade of time is particularly demanding owing to changing equipment and cables. This memo provides an assessment of where we stand at the Arecibo Observatory, and makes suggestions for the future maintenance of this level of precision.

    Schematic

    The atomic standards, GPS receivers, time transfer systems and time/frequency distribution at the Observatory are summarized in the official schematic . Need larger version for clarity xxx? See Bill Genter for a hard copy or print one from /share/timing/timing.eps.

    There are three atomic standards: Maser, Rubidium 1 & Rubidium 2. There are two GPS receivers: NIST (National Institute of Standards and Technology) which is also known as GPS1 & TAC (Totally Accurate Clock=Thomas A. Clark) which is also known as GPS2. There are two clocks which create 1 PPS signals from the Maser 5 MHz: TRAK & EECO; EECO is about to be retired and replaced by True Time 1 PPS generator. There are three time interval measurement systems: Clock Comparator System/CCS (E. Castro) which uses an HP 5334A counter, NIST unit which is part of their GPS rig, & Princeton Unit (J. Taylor) which use a home brew counter. There is one WWV receiver which is a PC radio system. Operation of these many units was exhaustively documented by Donna Kubik in her too brief tenure at the observatory; linking to this pdf file will allow you to copy it/read via acrobat. A hard copy will be kept in the control room?

    Unix Time

    (from Arun) Mike Nolan is the prime architect of the current NTP-based synchronisation scheme for the Sun network. NTP uses a tree hierarchy to distribute time. "Peers" in this hierarchy can synchronize with each other - each one dynamically picks a synchronization source out of several available (various sanity checks are built into the algorithm). The stratum-1 server (currently "cuca" an Ultra-30) receives IRIG code from the TRAK clock driven from the Maser and broadcasts The Time to our local network. A 1PPS signal from the Equipment Room buffer is input to a few other machines (SPARC IPC, LX, 20) which also independently broadcast The Time. Other machines can pick one of the above to synchronize their clocks. A few external servers are also available. Synchronization accuracy is typically within 1ms. Mike has also been thinking about using GPS input to further stabilize the system.

    Useful utilities on the unix system would be "lst" and "mjd" which, respectively, give current LST and MJD. In addition, it is often useful to compute lst and mjd for chosen date so such routines with -? option would be very useful system wide utilities. I have installed the Berkeley versions of "mjd" and "rolex" (fancier version by Tony Wong) in /share/abpp/bin directory and am working on the version of lst (which we exported from NRAO Green Bank).

    Maser Status

    The current state of the maser is that it is drifting seriously following a failure of the thermal servo on maser cavity on about 2000 January 20. The drift is 100 ns/day - see below. Time steps will need to be inserted to maintain proximity of the house 1 PPS to UTC. These steps need to avoid critical observations and observers need to be informed of status quo.

    AO Clock Comparator System (CCS)

    The CCS samples time intervals of the following pairs of 1 PPS signals: MASER-MASER, MASER-GPS2, MASER-EECO, MASER-TRAK, MASER-RUB1, MASER-RUB2, and MASER-TTIME. The CCS sits on each pair of signals for 1 minute every 10 minutes and takes 30 time interval samples spaced by 2 s. The average data are stored in daily files every 10 m. The file nomenclature is {mm-dd-yy}.txt. Following a recent request Bill Genter copies the CCS PC files ( e.g., file for 2000 March 7). to /share/timing on the unix system. A sample program to read these files, which annoyingly separate columns with tabs, is available.

    Princeton Time Monitoring System

    The Princeton TAC monitoring halted with a Y2K problem in the PC bios. During the pulsar Y2K Workshop visit David Nice reset the PC clock back 20 y to allow continued operation of its critical function of recording the time interval between the TAC/GPS2 and the TRAK clock 1 PPS derived from the maser 5 MHz and distributed to pulsar backends and elsewhere as the house 1 PPS. There is some confusion here about the active cumulative file of MASER-GPS time differences in ~joe/tac: aoclk.1.bak is complete to end of 1999 when Y2K bug hit; aoclk.1 has recent results after David restarted the PC monitoring system.

    The Princeton system samples the MASER-TAC/GPS2 time interval every second and stores an average over 10 min along with a histogram of different readings about the mean in {mjd}.clk files; ~joe/tac/*.clk; e.g. for MJD 51610, 2000 March 07). A daily value and current frequency are determined in a chron job, probably by ~joe/autofit/tacday.f; e.g., first lines of aoclk.1). I have compared recent values of these two time interval measurements of the *same* GPS signals. One might think that the agreement would be perfect, in the ns's range. But remember that the two systems don't sample at the same instant, nor do they cover the same range of the 10 min cycle, and the GPS signals have considerable jitter owing to averaging over satellites as well as intrinsic snr. I don't then really know what to expect. Longer term real features should track and the maser drift certainly is seen equally be both. A plot of AO CCS vs Princeton measurement of TAC (GPS 2) 1 PPS vs that of maser during 7-12 March 2000 shows a short example of the equivalence of the two systems (although CCS has a couple noise spikes). This plot has the two measures and their difference via a spline interpolation.

    NIST Monitoring

    The NIST time interval measurement is done in "common view" mode. The same satellites are viewed at the same time from Arecibo and from NIST. This removes the DoD installed dither in the otherwise more accurate GPS time codes. (The TAC system suppresses this by looking simultaneously at *all* satellites above the horizon.) "Soon" we are told that the dither (SA, selective availability) will be turned off and this technique will be moot, at least as far as SA is concerned. Time files each day are generated. Recent versions have been placed in /share/timing: {mjd}.dat. These are of limited use. What is important is the results of the common view analysis which are provided to the observatory in hard copy form by Trudi Peppler (303-497-3276; tpeppler@boulder.nist.gov). Annual costs for this service is $5000 according to M. Davis. Trudi has provided the recent results (Table 1 entries from monthly reports as cumulative files of:
    daily time error MASER-GPS
    daily time error with Kalman filter MASER-GPS
    daily freq error with Kalman filter MASER-GPS
    See also
    README file

    Kiriaki has entered the daily data from monthly NIST reports in files that we are attempting to recover. Trudi also may be able to extend her contribution backwards in time. I need to check that files received compare accurately to monthly reports (even though I am sure they will).

    I have started to compare the NIST results with other time interval tables. The first effort is with the Princeton table in aoclk.1.bak which covers through to end of 1999. The plot shows a frequency step around MJD 51330 and the large jump in frequency discussed above around MJD 51560 (2000 Jan 20). What I don't understand are the two jumps in aoclk.1.bak around MJD 51340 and 51370. These may be 5-MHz cycle jumps in the TRAK clock which won't be in the NIST reports which sample a 1 PPS direct from the MASER (see master diagram above). I need to recover the CCS data to see the independent sampling of the TRAK xxx.

    Miscellaneous Notes, ToDo's

    I note for tutorial purposes that the definition of a time interval is accurately stated in English as "the algebraic difference of two clocks read at the same instant". And I therefore need to update Donna's quote on her Salvador Dali illustrated cover page of her manual, "A man with a watch knows what time it is. A man with two watches is never sure." The update is that while "time epoch" may not be known (or have only arbitrary meaning), time interval is a matter of measurement which is to say good engineering.

    The job is not done with this discussion of the clock and time transfer systems. We would like to measure the time interval between a well defined fiducial point in the measurement system and the time interval measurement system in a pulsar backend where PULSAR-TRAK(MASER) is determined. This can be done by bringing a cable of known (measured) propagation length directly to the backend. Further specificity can be obtained by defining rise times of actual 1 PPS pulses and the point on the rising slope to which time interval is referred -- both in the clock and time transfer system and in the pulsar backend.

    I leave to another forum the discussion of time transfer from the pulsar backend to a fiducial point in the telescope receiving optics. Current understanding is that this is ~7 microseconds which Phil et al. measured by looking at delay of pulsed calibration signal. Eddy Castro plans to loop a fiber up and back as soon as splicer arrives so that a monitoring of the round trip on this path can be done to see diurnal and cable twist changes.

    Propagation in Earth's Atmosphere

    The zenith delay through the dry part of troposphere is ~7 ns. This term will increase approximately as sec(ZA); reference: Sovers, Fanselow & Jacobs (1998 RMP 70 1393) as quoted by Walker (1999 ASP CS 180 461). To keep known effects under 10 ns it would be appropriate for full sky tracking telescopes to include a correction for nominal dry troposphere. The zenith propagation delay through the ionosphere is ~few to ~75 ns at 1 GHz for low Earth longitude; reference as above and also GPS data for 1998 July 16 at 01:00 from GPS TEC0 movie. This delay also scales roughly with secant(ZA), has strong diurnal and solar cycle terms and can be corrected with seemingly sufficient accuracy via GPS data bases. If a dual frequency GPS receiver system is installed at Arecibo for stabilization of absolute linear polarization angle studies, as recommended during Pulsar Y2K workshop, then this data will be available locally. The ionospheric scientists (Sixto Gonzalez, Mike Kelly) are also interested in this dual frequency data.