Raw data

Here's a look at the raw data files. 

C:\tvb\corby> wc log2.txt

log2.txt: 0.865 MB, 886 KB, 906580 chars, 91620 words, 45144 lines

C:\tvb\corby> dir log2.txt
03/29/2008  11:37 AM           906,580 log2.txt

C:\tvb\corby> head -5 log2.txt

0.999,999,978,7  s
0.999,999,978,9  s
0.999,999,978,9  s
0.999,999,978,9  s
0.999,999,978,4  s

C:\tvb\corby> tail -5 log2.txt

0.999,999,511,5  s
0.999,999,511,4  s
0.999,999,511,9  s
0.999,999,511,3  s
0.999,999,511,0  s

Observations: Data file size looks like half a day at 1 Hz, or one day at 2 Hz. Time interval values are all one second less some nanoseconds. This symptom means the UUT 1PPS leads instead of lags the GPS reference 1 PPS. Suggest re-sync'ing the UUT 1PPS to be later by a few microseconds. Alternative solution is to swap TIC channel cables. Meanwhile, we can still use this data if we use tau 2 seconds instead of tau 1 second.

Clean data

Remove stuff related to HP counter format. The colons are statistics lines. The commas are thousands separators. With these gone we have simple time interval numbers with units of 1 second. 

C:\tvb\corby> sed < log2.txt -e /:/d -e "s/,//g" -e "s/ .*s//" > log2.dat

C:\tvb\corby> wc log2.dat

log2.dat: 0.593 MB, 608 KB, 621656 chars, 44404 words, 44404 lines

C:\tvb\corby> head -3 log2.dat

0.9999999787
0.9999999789
0.9999999789

C:\tvb\corby> tail -3 log2.dat

0.9999995119
0.9999995113
0.9999995110

Phase plots

Open file with stable32, using tau 2 seconds, offset -1.0 second; scale 1.0.

Raw phase shows about 500 ns of phase drift over a day, mostly a straight line, some wiggles. This is good so far.

Remove linear phase slope, which is a frequency offset of -5.36×10-12.

The plot below, then, is phase error residuals. Now it's getting interesting. Since the range is on the order of nanoseconds, less than about 10 ns peak-to-peak, we have to be mindful that the reference is a GPSDO. We cannot quickly assume the instability in phase is either the GPSDO or the UUT. It puts an bound on stability, but doesn't give an absolute measurement.

Frequency plots

A simple conversion from phase to frequency. Mostly measurement noise.

Average by 30 points (which represents 60 seconds) gives one point per minute.

Average instead by 150 (which represents 5 minutes).

Stability plots

Do both overlapped and modified Allan deviation. Note calculated frequency drift is -7.2×10-13/day. At this level, consider this value bogus until many days or weeks of data is collected.

Here's the resulting ADEV plot.

Live stability plot

Here's a text-mode stability plot obtained directly from the raw data file. 

C:\tvb\corby> adev2 /p:2 /pix:10,5 /b:10 /plot log2.dat
40000
 -9 +---------+---------+---------+---------+---------+
    |         |         |         |         |         |
    |         |         |         |         |         |
    |  o      |         |         |         |         |
    |         |         |         |         |         |
-10 +-----o---+---------+---------+---------+---------+
    |       oo|         |         |         |         |
    |         oo        |         |         |         |
    |         | ooo     |         |         |         |
    |         |    oo   |         |         |         |
-11 +---------+------oo-+---------+---------+---------+
    |         |        oo         |         |         |
    |         |         |ooo      |         |         |
    |         |         |   ooooooooo       |         |
    |         |         |         |  ooo    |         |
-12 +---------+---------+---------+-----ooo-+---------+
    |         |         |         |        oo         |
    |         |         |         |         |ooo      |
    |         |         |         |         |   ooo   |
    |         |         |         |         |         |
-13 +---------+---------+---------+---------+---------+
                      m                 h            d
    1        10        100      1000       10k      100k

44404 (2 s) samples (= 24.67 hours = 1.0279 days)

Questions/Comments: /tvb