Performance tests for a lot of 100 Toyocom brand TCO-999
The IceCube collaboration office, at U of Wisconsin, Madison, procured
a lot of 100 Toyocom oscillators for evaluation, and for use in the
first pre-production, Engineering models of the DOM main board.
The frequency sample data files consisted of from about 2000 to 50,000
frequency measurements. Measurements were taken every 5 seconds,
uninterrupted for the duration of the test, with no dead-time
between measurements. The time-base for the measurements, as
usual, was a Stanford Research Systems model FS700 Loran Frequency
Standard with the Option 01 high stability oscillator (specifications).
The Alan Variance of the standard is about 5 times better than
the population being measured.
The Alan Variance analysis program digested blocks of 100 frequency
samples. The analysis program reported the best block of 100
measurements, the next best block of 100 measurements, and the worst
block of 100 measurements.
The above histogram plots the best, and next best result for a
block of 100 contiguous measurements in a particular data
file. If the two histograms plotted here
substantially overlap, then the best measurement results are consistent
with typical performance. If the two histograms are widely
separated from each other, then the 'best' measurement results would
represent a tail on the distribution of measurements.
Since the histograms overlap fairly well, one assumes the assessments
the oscillator performance are fairly reliable. Outliers, those
samples in the 'More' bin, are still
outliers.... Outliers are 6% of this relatively small
If 6% of the oscillator production truly does fall far from the
typical population, then 300 of the DOMs in IceCube could deliver
marginal or unsatisfactory time-stamping performance. That number
seems to be unacceptable. Some scheme for discovering the poor
performers will have to be found. More to come... Stay
The results file
the results summary used for this histogram. The unfiltered
results file, all 46 Megabytes of it, will be provided to anyone
requesting it. Likewise, the raw data files are available to
anyone interested. Just send me an email
The six worst... the outliers... the marginal... the parts that are
not members of the population... for whatever reason...
||Best Result ?f/f
||2nd Best Result ?f/f
As with the parent population, the best result for each of the six
is not significantly better than the next best result, so one would
assume the results are pretty representative of typical
As usual, comments, beefs, suggestions are solicited by the .