100 Mhz reference unlocks
Observatory devices use 5, 10, 20, and 100Mhz
reference signals to lock themselves to the hydrogen maser. The maser provides
5 Mhz (or is it 1..). This is converted to 10,20 and 100 Mhz. The
100 Mhz reference is used to drive the correlators (interim and wapp) as
well as all of the equipment upstairs in the dome (transmitter and synthesizer
for the first lo).
On 25nov02 peter backus noticed that the seti test
tone was drifting. Since the drift rate did not match that of the earth's
rotation (and it was monday) he decided it was probably a problem with
the LO system rather than ET. On 28nov02 we found that the AO 100 MHz reference
was drifting. We temporarily replaced it with a synthesizer locked to the
The 100 MHz reference is generated using a PLL driven
from the 5 Mhz. Ganesh pointed out that most VCO's drift with time and
all that was needed was to tune the oscillator. The loop bandwidth
was about 60 Hz. We adjusted the VCO putting the 5 Mhz at the center
of the loop bandwidth. After this, the system locked up ok.
The question is when did the 100 Mhz unlock? A time
line of the events are:
20nov02: sband radar ranging to titan with echo --> it's locked
22nov02: vlba holography run. Need to see if they get fringes.
23nov02: sband radar run with narrow echo but 50 Hhz from the ephemeris
25nov02: peter finds the system unlocked
28nov02: 17:00 replace pll with synthesizer.
02dec02: take data to characterize drift, then retune vco.
The 23nov02 echo probably
means that it was locked on the 23nov02.
I took some data on 02dec02 before we adjusted the
vco. I let the system warm up for about 1 hour before running. A test tone
was transmitted from the birdie maker (in the bowl) at 2380.005 Mhz. This
was mixed to base band and sampled at 100 Khz bandwidth for 60 seconds.
show the wandering of the tone at sband:
When peter backus looked at the signal on 25nov02 he
saw a 22 second period between the sweeping. On 27nov02 we saw a 15 second
period between the sweeps. On 02dec02 after an hour warmup, the sweep period
was 2.8 seconds. So the unlocked oscillator was definitely changing with
time (probably do to temperature).
Fig 1 top. Transforming the entire 60 second period at 1 shot. The average
offset is about -126.5 hz with a motion of about 10 Hz in 60 seconds.
Fig 1 bottom. I broke the data set up into 60 1 second records and transformed
each second separately. I then fit a gaussian to find the center frequency
of each second. The offsets from the expected frequency are plotted. There
is a cycle that lasts for about 3 seconds.
Fig 2 top. This is the voltage time series for 12 seconds. In software
i mixed the average frequency of the tone to base band. You can see that
the signal is being chirped (swept in frequency) during this 3 second
Fig 2 middle. To resolve the frequency chirp, i broke the data set up into
.2 second records, transformed each record, fit to the peaks, and the plotted
the time versus peak frequency offset from the true frequency.
Fig 3 bottom. To look at the fast chirp region (red line in top plot) I
broke the data into 5 millisecond records, transformed each record, and
the fit for the peak. The plot covers .2 seconds of time and shows the
frequency offset (in Khz) from the expected frequency. The VCO is
sweeping across 15 Khz in about 60 milliseconds. The middle plot is reversed
in frequency because of under sampling in the time domain.