Comparison of GPS clocks in 2003 data

As a side issue in the studies of live- and dead-time the observation that the two GPS clocks present in AMANDA are not identical has been discussed. At the start of these studies the reader used GPS2 rounded to the nearest microsecond which was used in the live- and dead-time studies. This is not generally true since the selection of the GPS latch depends on the detector configuration and the time quality and status information reported by the various types of GPS latches (quoting Steffen Richter). Se the reader manual for details on this and other matters. With the reader option "--clock=g111:u1" we may reveal the individual GPS1 and GPS2 with 0.1 microsecond precision and this was done in this analysis.

In the table below we show the GPS1-GPS2 comparison for each run. The data used are the selected raw data samples brought to Wisconsin. Plots are either 5 histograms, one total and four where the four peaks are zoomed in, or in the odd cases (7170-7173) three histograms in different x-scales, the first in linear- and the other two in log-scale for the y-axis.

Note: The change in day 188 was due to a known and planned change. I quote "run 7170 is the start of the 'old' DAQ for troubleshooting at pole, using the 2002 winter version and CAMAC GPS latches" (Glenn). It was last summer we switched to "new VME GPS latches" (Steffen). Another quote from Steffens respons: "The modules are the identical to the ones used before the switch to the VME latch. The only difference in the setup is the use of Kalle's GPS distribution box. The box also caused trouble with the old VME GPS latch." Although both Steffen and I seem to recall having seen big differences before I cannot spot this now in earlier 2002- or 2003-data checks below. In 2000- and 2001-data we had problems in the beginning of the runs but otherwise it seems ok.

Comments:

• The four peaks may be understood if the two GPSs each have two possible distinct settings, GPS1 at 0.0 µs and 4.8 µs, GPS2 at 0.0 µs and 3.5 µs (for runs 7160-7161). This would cause peaks at exactly -3.5, 0.0, 1.3 and 4.8 µs as observed. There are minor contributions ±0.1 µs from these values which comes from the 10 MHz clock signal. Also for runs 7163-7164 a small shift of 0.1 µs in the above values is seen.
  
  
• While the difference for the first tested runs (7160-7164) were within 5 µ suddenly run 7170 was dramatically worse ranging from -20703 µs to -404 µs!
  
  Similarly run 7171 ranges from -13884 µs to -374 µs and has on top of that two GPS1-rollover bugs causing dt~-1 s.
  
  
• Earlier tests indicated that the second/microsecond rollover bug seen earlier in live- and dead-time studies was affecting both GPSs simutaneously. Obviously this is not the case:
  
  
  • For run 7163 there are 3 outliers for events 801407, 1580208 and 3136121 with dt~1 s caused by the second/microsecond-rollover only being present for GPS2.
    
    
  • For run 7164 event 104028 the second/microsecond-rollover was only present for GPS1 causing a dt=-999995.3 µs ~ -1 s outlier.
    
    
  • Indeed in this investigation we are insensitive to the case where both GPSs have the rollover bug simultaneously. From the vast number of such occurrences seen in the live- and dead-time studies and the few seen here we may conclude that this is still the normal case.
    
    
• Steffen Richter et al at the Pole found problem with latches and from run 7200 things look much better. Run 7200 was perfect but for run 7201 the first event was ok, then came 15 events with dt = -1048 s, the next event had -20 ms whereafter it was perfect. This is known bugs which we had earlier (before we changed to the new VMS GPS latches) where GPS1 will have the times wrong until the second change for the first time while GPS2 gives the first event time wrong (but in run 7201 in perfect agreement with GPS1!). Run 7200 most probably looked perfect since I did not have file 000 with the first events in the run as I had for run 7201. For run 7205 we got the 000 file and here the first event is ok while 75 subsequent events is 68 seconds off.
  
  
• It seems that for runs 7170-7199 GPS1 would have been a better choice than the default GPS2 for the event times. It was GPS2 which turned out to be the faulty one. Steffen Richter plans to introduce a special reader flag to force the use of one of the GPSs but care has to be taken when analyzing those runs.
  
  
• Run 7247 starts with event 0 looking OK but it is followed by 33 events having GPS1-GPS2=-68 s which are seen as underflows in the first plot for this run. This is abnormal for 2003 data but cf the even messier situation for run 7241.
  
  
• In runs 7250 to 7277, besides discrepancies between GPS1 and GPS2 creating two or three peaks in the plots, occasionally there are under- and overflows with even more extreme differences. We see many occurrences of GPS1-GPS2 = -1 s (almost exactly) but also strange things like GPS1 suddenly jumping 10000 seconds for one event (3521941 in run 7276) but being ok again in the next event. In runs 7260 and 7263 long sequences are off by tenths of seconds.
  
  
• Run 7294 has 253 events in three peaks at -18.2, -17.9 and -15.3 but also peaks at -9.6, 0.3 and 8.5 µs.
  
  
• Run 7296 has many peaks at approximately 0.3, 1.7, 4.0, 6.4, 7.1, 8.3, 10.6, 14.7, 17.8 and 19.2 µs.
  
  
• Run 7297 has many peaks at approximately -1 s and -0.3, 0.3, 3.6, 6.9, 7.5, 12.5, 14.1, 15.6, and 16.7 µs.

Earlier 2002 and 2003 data:

A comparison to a few runs (selected files) from 2002 data. Note that the 2003-scripts are not well adjusted for those (yet). I have also made a selection of earler 2003 runs where we see e.g. that the 4-peak structure is inverted in the early season as compared to later on. Somewhere between run 7025 and run 7051 it looks like the two GPSs has been switched.

2000 and 2001 data:

In 2000 and 2001 we had three GPSs and I have reported on different "GPS-problems" seen during the years mainly in the beginning of each run (file 000). However, for a raw data sample (run 3800) from 2001 all three GPS's looked equal on the ±0.1 µs level! For another sample from 2001 (run 3100) GPS3 was exactly 131 seconds off GPS1 and GPS2 (but decimals still correct!). Should perhaps be investigated further.