E5 Lock Losses Investigation
(Preliminary report - September 7, 2001)
P. Shawhan (Caltech)
D. Chin, R. Gustafson, K. Riles (Michigan)
M. Ito (Oregon)
W. Butler (Rochester)
(with much appreciated assistance from W. Kells, D. Sigg)
Three main categories of lock losses observed:
-
13 Hz runaway oscillation in Michelson servo control signal
-
Saturation of error signal for common mode servo after tidal drift
-
"Deliberate" software railing of servo control signals after tidal drift
(plus one smoking gun earthquake from south pacific)
13 Hz runaway oscillation:
-
For about the first half of the run we had many lock losses preceded by
a runaway oscillation of about 13 Hz seen most dramatically in the Michelson
servo control signal, but also seen in the recycling cavity control signal.
Figure
1 gives a typical example where the upper plot shows the Michelson
control signal (green curve) going haywire shortly before lock is lost
(indicated by drop in transmitted arm powers (red and blue curves in bottom
plot).
-
13 Hz is a natural vertical bounce mode of all of the large optics, including
the folding mirrors (excluding the beam splitter). The mode is not damped,
but digital filters can be enabled to give a resonant gain at that frequency
to help suppress the motion. Looking at power spectra of osem sensors for
all the large optics suggests that the X arm folding mirror is the optic
most prone to 13 Hz excitation. Figure 2
shows power spectra for the FMX sensors (top plot) about 30 seconds before
a 13-Hz-induced lock loss and for various control and power signals. An
enhancement around 13 Hz is seen here but not in the FMY sensors at the
same time ( figure 3 ). Both the Michelson
and Recycling servo control signals actuated on the folding mirrors during
E5.
-
Empirically it was found that turning on the 13-Hz resonant gain filters
helped at the start of a lock stretch (13 Hz noise on many signals was
reduced), but that later in the lock stretch, after tidal drift, the filters
seem to make things worse. Eventually the filters were routinely left off.
-
Hypothesis: excitation of 13 Hz in Michelson and Recycling servo signals
is a result of optical gain loss in the recycling cavity caused by drifting
alignment in that cavity. The runaway behavior is presumably caused by
a servo instability as the unity gain frequency approaches the 13 HZ region
from above. The optical gain of the Michelson and Recycling servos is expected
to be quite sensitive to relative alignment between the recycling mirror
and the other optics defining the cavity because the cavity is nearly degenerate
(large radii of mirror curvature). Loss of sideband power in the cavity
is indirectly measured by the "nspob" signal which is proportional to the
"2-Omega" RF strength seen from a pickoff in the cavity. That gain is seen
qualitatively to decrease in level and undergo large oscillations shortly
before lock loss. As a test, we tried deliberate misalignment of the recycling
mirror to induce lock loss and indeed produced a loss of the same nature
as described above (figure 4 ). Note
that in this example the 2-Omega signal has dropped dramatically as a result
of the misalignment.
-
The 13-Hz problem was much less severe in the second half of the E5 run
after the recycling mirror alignment biases were tweaked on Saturday night.
-
If the above hypothesis is correct, then this apparent instability should
go away in the future when we have 1) tidal compensation to reduce longitudinal
drifts that couple to orientation, 2) wave front sensing control of the
recycling cavity orientation degrees of freedom, and 3) higher unity gain
frequencies in the Michelson and Recycling cavity servos.
Saturation of common mode error signal:
-
A second class of lock losses was seen throughout the run, but it became
much more apparent when the 13 Hz problem above was ameliorated by alignment
tweaking. The error signal for the common mode servo control of the arms
was found to hit its positive or negative rails after tidal drifting. The
ambient 13 Hz oscillation leaking into that error signal made the railing
happen a bit earlier, but the fundamental problem was the drift itself.
Figures 5, 6,
and 7 show a very typical example at
increasing time magnification, where the REFL_I signal approaches -32K
and then goes haywire. Note that the signal does not quite reach the ADC
limit, suggesting a hardware rail kicking in first. The distinctive pattern
of signals seen in this example was reproduced remarkably well many times
during the run, occurring most frequently during periods of steep tidal
slope.
-
Again, this is a problem that should be helped if not cured by implementing
feed-forward tidal actuation. Also, increasing the gain in the common mode
servo should reduce considerably the absolute range over which the error
signal can vary.
Software railing of servo control signals:
-
Some lock losses were caused by artificial software limits on servo control
signals. In fact, there was some confusion initially in the run because
the control signals read out by the DAQ appeared to be well outside reasonable
values. What was fed back to the masses, however, was constrained to values
half or less of what the hardware could tolerate. As a result, lock was
lost sooner than necessary. Figure 8
shows an example trend plot of the four main longitidunal control signals.
For technical reasons, only the differental mode (DARM) signal shows the
effect of the software limits. During this period the DARM limits were
+/- 8000 or +/- 6000 counts (note that the -6000 limit appears hardwired,
not controlled by a MEDM panel), and it's clear that hitting that limit
caused several lock losses shown here. Once this problem was realized,
the artificial limits were increased to prolong lock stretches.
-
Again, this problem should be helped considerably by feed-forward tidal
actuation.
Information on locked stretches
-
The cumulative live-time fractions for two-arm locking during E5 was about
65%
-
List
of all 1-arm and 2-arm lock stretches at least 1 second long
Conclusion (perhaps optimistic)
-
Although a variety of lock loss classes were observed during E5, nearly
all were results of operation with an incomplete servo control system (missing
full tidal compensation, full wave front sensing control, and full gain
in the common mode servo). Prospects for long lock stretches at Hanford
with a fully commissioned interferometer seem promising.