The XO Project: Searching for Transiting Extra-solar Planet Candidates
ABSTRACT The XO project's first objective is to find hot Jupiters transiting bright stars, i.e. V < 12, by precision differential photometry. Two XO cameras have been operating since September 2003 on the 10,000-foot Haleakala summit on Maui. Each XO camera consists of a 200-mm f/1.8 lens coupled to a 1024x1024 pixel, thinned CCD operated by drift scanning. In its first year of routine operation, XO has observed 6.6% of the sky, within six 7 deg-wide strips scanned from 0 deg to +63 deg of declination and centered at RA=0, 4, 8, 12, 16, and 20 hours. Autonomously operating, XO records 1 billion pixels per clear night, calibrates them photometrically and astrometrically, performs aperture photometry, archives the pixel data and transmits the photometric data to STScI for further analysis. From the first year of operation, the resulting database consists of photometry of 100,000 stars at more than 1000 epochs per star with differential photometric precision better than 1% per epoch. Analysis of the light curves of those stars produces transiting-planet candidates requiring detailed follow up, described elsewhere, culminating in spectroscopy to measure radial-velocity variation in order to differentiate genuine planets from the more numerous impostors, primarily eclipsing binary and multiple stars. Comment: 29 pages, 12 figures, accepted by PASP for Aug 2005 issue
arXiv:astro-ph/0505560v1 27 May 2005
To appear August 2005 issue of PASP
The XO Project: Searching for Transiting Extra-solar Planet
P. R. McCullough1,2J. E. Stys1, J. A. Valenti1, S. W. Fleming3, K. A. Janes4and
J. N. Heasley5
pmcc,jstys,email@example.com; firstname.lastname@example.org; email@example.com;
The XO project’s first objective is to find hot Jupiters transiting bright stars,
i.e. V < 12, by precision differential photometry. Two XO cameras have been
operating since September 2003 on the 10,000-foot Haleakala summit on Maui.
Each XO camera consists of a 200-mm f/1.8 lens coupled to a 1024x1024 pixel,
thinned CCD operated by drift scanning. In its first year of routine operation, XO
has observed 6.6% of the sky, within six 7◦-wide strips scanned from 0◦to +63◦
of declination and centered at RA=0, 4, 8, 12, 16, and 20 hours. Autonomously
operating, XO records 1 billion pixels per clear night, calibrates them photo-
metrically and astrometrically, performs aperture photometry, archives the pixel
data and transmits the photometric data to STScI for further analysis. From the
first year of operation, the resulting database consists of photometry of ∼100,000
stars at more than 1000 epochs per star with differential photometric precision
better than 1% per epoch. Analysis of the light curves of those stars produces
transiting-planet candidates requiring detailed follow up, described elsewhere,
culminating in spectroscopy to measure radial-velocity variation in order to dif-
ferentiate genuine planets from the more numerous impostors, primarily eclipsing
binary and multiple stars.
1Space Telescope Science Institute, 3700 San Martin Dr., Baltimore MD 21218
2University of Illinois, Urbana, IL 61801
3Vassar College, Dept. of Physics and Astronomy, 124 Raymond Ave., Poughkeepsie, NY 12604-0745
4Boston University, Astronomy Dept., 725 Commonwealth Ave., Boston, MA 02215
5University of Hawaii, Inst. for Astronomy, 2680 Woodlawn Dr., Honolulu, HI 96822-1839
– 2 –
Subject headings: instrumentation: miscellaneous – telescopes – techniques: pho-
tometric – stars: planetary systems, variables
Borucki & Summers (1984) proposed detection of planets with the transit technique.
At the time, their proposal was thought to be impractical because astronomers expected
planetary systems to be like our own. In particular the Jovian-sized planets that could
create a readily-detectable photometric transit would occupy orbits many AU in radius with
periods of many years. Since the discovery of the “hot Jupiter” orbiting 51 Peg with a
4.23 day orbit, many attempts have been made to find hot Jupiters that transit stars bright
enough (mV ? 12) for detailed studies to be performed with existing telescopes such as HST,
Keck and the VLT (e.g. Vulcan, Stare, etc). The first such success was Tres-1 (Alonso et al.
2004). The OGLE collaboration’s successes were with fainter stars (Udalski et al. 2003).
The radius of solar-composition objects is expected to change by less than a factor
of two (e.g. Burrows et al. 2001) over the entire range in mass from the bottom of the
main sequence to less than a Jupiter mass. So transit observations by themselves provide no
information about the mass of the planet (or brown dwarf, or red dwarf). On the other hand,
radial velocity studies yield only minimum masses. However, the combination of the Doppler
orbit and the observation of transits yields the gross physical characteristics of the planet:
mass, radius, density, and “surface” gravity. A transiting planet is interesting in at least
two other ways: 1) absorption of starlight is a much larger signal than reflected starlight,
so for example absorption spectroscopy has already permitted detection of an exoplanet’s
atmosphere (Charbonneau et al. 2002), and 2) the rapid and precisely predictable on/off
nature of the transit permits excellent calibration, which among other things, allows one
to search for natural satellites and Saturn-like rings orbiting the transiting planet and to
attempt to measure the planet’s albedo by observing the reflected light of the planet being
blocked by the star (Brown et al. 2001).
This paper describes the XO project’s design, implementation, and verification. Not an
acronym but a name, XO is pronounced as it is in “exoplanet.” We describe the XO design
requirements in Sec. 2, the hardware in Section 3, the observing strategy in Section 4, and
the software in Section 5. Section 6 shows that our system is finding transiting hot Jupiter
(THJ) candidates. In future papers we describe additional observations that test whether a
candidate is definitely an impostor such as an eclipsing binary star, or potentially one of the
THJs that we seek (McCullough et al. 2005).
– 3 –
Pepper et al. (2003) formalize the optimization of systems designed to find THJs and
have selected a 2-inch diameter telescope with a 4kx4k sensor as best for a 2π steradian
survey of stars with V? 10. Independently we designed the XO system with a similar but
simpler analysis for a limiting magnitude of V? 12, requiring a larger diameter aperture, d
≈ 0.1 m, and in order to accommodate drift scanning, a smaller instantaneous field of view,
The XO system was designed to find THJs around stars bright enough to permit signifi-
cant follow up as described in the introduction, i.e. (mV ? 12). The number of THJ-systems
on the sky is estimated below from the number density of stars as a function of their bright-
ness, the frequency of hot Jupiters around those stars, and the geometric probability of the
orbit being inclined such that a transit can occur as seen from Earth. The photometric
precision required is set by the fraction of the star’s area obscured by the THJ, ∼ 0.01. The
cadence of the observations is set by the need to have multiple observations made during the
duration of the transit, ∼ 0.1 day. The duration of an observing sequence is set by the need
to observe multiple transits to define a tentative ephemeris. Given that useful observations
are obtained only at moderate zenith angles during clear nights, the observing sequence must
be much longer than the orbital period in order to witness at least 3 transits (Brown 2003).
Of stars brighter than mV= 12 at the north galactic pole (NGP), there are 2.9 main
sequence stars per 2◦with MV = 4.5 to 5.5, and 1.3 with MV = 5.5 to 6.5 (Bahcall &
Soneira 1981). Solar-type and later stars with MV > 5 and brighter than our limiting
magnitude of mV= 12 must be closer than D = 250 pc, whereas for such stars the scale
height H = 400 pc. For a volume density of stars with exponential scale height H, the ratio
of stars per square degree within a distance D at Galactic latitude b = 0◦to those at b = 90◦
is D/(H × (1 − exp(−D/H)), which equals 1.34 for D/H = 250 pc / 400 pc = 0.625, and
approximately equals 1 + 0.58*D/H for D/H ≤ 1. The density enhancement in the galactic
plane is countered by the disadvantages of crowding and confusion. For XO’s drift-scanned
observations of a variety of Galactic latitudes, we empirically find the maximum surface
density of stars with photometry sufficient for our purposes occurs at b ≈
6). Although stars become more concentrated toward the Galactic plane with earlier type,
early-type stars are larger in physical size and thus are expected to have transits of lower
amplitude that are more difficult to detect. For simplicity, we use the stellar density at
the NGP in order to conservatively estimate that there are ∼160,000 solar type stars with
mV< 12 over the entire celestial sphere. The probability that a given solar type star has a
THJ is 0.00075, because the fraction of stars with Jovian planets is 0.05 (Marcy & Butler
2000), the fraction of those planets with periods less than 7 days is 0.15 (Brown 2003), and
– 4 –
the fraction of those that could exhibit transits seen from Earth is 0.10 (Borucki & Summers
1984). Thus, we expect 120 THJs orbiting solar type stars brighter than 12th magnitude,
or 30 THJs orbiting stars brighter than 11th magnitude. The corresponding predictions are
7.5, 2, and 0.5 stars for mV< 10, 9, and 8, respectively.
There are at least three interesting points implied by the calculations above:
1. the 8th magnitude HD 209458 probably is the brightest star exhibiting hot-Jupiter
2. there is likely one hot Jupiter transiting an 11th magnitude star for every 1400 2◦of
3. at any given time approximately one bright (mV< 11) star is being transited.
For simplicity, we have assumed the probabilities are independent and can be multiplied
together. A thorough analysis of joint probability functions is beyond the scope of this paper,
and indeed some of the joint probabilities are unknown (Brown 2003). Pepper et al. (2003)
estimate there are ∼5 THJs orbiting stars brighter than V=10 mag for 4.5 < MV< 5.5, and
scaling from their Figure 1 we estimate for V<10 an additional ∼6 THJs and ∼3 THJ for
3.5 < MV< 4.5 and 5.5 < MV< 6.5, respectively, if the frequency of THJs is independent
of MV. We have based our requirements upon the statistics from radial velocity surveys,
which are secure for solar type stars. However, if the estimate of Pepper et al. (2003) is
appropriate, then we could hope for of order 200 THJs orbiting approximately solar sized
(or smaller) stars brighter than 12th magnitude.
Succinctly, the XO system is required to image hundreds of square degrees of sky many
times per hour for months, and from those images the software must enable photometry of
stars 9 ? mV ? 12 with a precision of ∼ 10 millimag per measurement. Those require-
ments have been met by the XO Mark I system described below. Given the demonstrated
performance of the Mark I system, we anticipate replicating it to speed the discovery of
The XO Mark I hardware is described in this section. Figure 1 illustrates the cameras
and equatorial mount; Figure 2 gives a block diagram of the system; and Table 1 summarizes
– 5 –
Many systems have been built to discover THJs orbiting bright stars (Horne 2003).
Some of the unique aspects of the XO system are that it uses a broad spectral bandpass (0.4
µm to 0.7 µm) to collect more photons per second, drift scanning to simplify calibration,
aperture photometry for simplicity and reliability, and two identical cameras pointed in the
same direction to collect more photons and to provide redundancy so that observing will not
be entirely interrupted due to failure of a single camera or its control computer.
Sometimes a component is unresponsive and must be reset by cycling its AC power. We
accomplish this either by human intervention either in person or remotely using a network
power strip that permits remote control of 8 AC power outlets independently (Figure 2).
The network power strip is one of the essential components in the XO system and it has
operated 100% reliably.
3.1.Weather Sensing and Roof Control
Gaustad et al. (2001) enlisted a human operator of an adjacent facility to authorize or
to override, via email, the nightly opening and closing of the dome protecting their robotic
telescope. The XO system observes without human assistance. Weather hazardous to the
equipment is sensed in near real time as described below, and weather that is not hazardous
but is unsuitable for observing is sensed after the fact by the science data analysis software.
We use three sources of weather data to determine whether to open our roof. The first
is a CCD camera to detect stars. The second source is a set of weather stations operated by
other tenants on the mountain. The third is a forecasting service operated by yet another
tenant. If all three services are operating, then the data from every one must meet specific
criteria for our system to open its roof. If one of the second or third services is inoperable,
we reconfigure our system to ignore it. These three sources of weather data are described in
A small camera mounted in a weather-proof enclosure is aimed near the north celestial
pole. At regular intervals, the PC takes a short exposure of a ∼ 3-degree field of view. If the
sky is clear, many stars are visible in the image and our software can determine an astrometric
solution in the same manner as it does for our science images (Section 5.2). However, with
time the images became increasingly defocused, due to wind shaking loose the lens’ helical
focus mechanism. To compensated for this issue remotely, we adopted a different and more
robust algorithm that works well with even very defocused images. The algorithm operates
on two images taken twenty minutes apart. It rotates one image about the pole in order to
register it to the other image, and if the two-dimensional cross-correlation function peaks
– 6 –
Fig. 1.— The XO Mark I observatory: two 200mm f/1.8 lenses and 1K×1K CCDs attached
to a German equatorial mount, deployed under a roll-off roof.
within ±3 pixels of the center and is greater than an empirically-determined threshold (0.17),
then we assume the sky is clear. The latter algorithm is simple and effective, so we have
not returned to the original algorithm of matching patterns of stars, even after re-adjusting
the focus. We note that pointing the camera near the pole has some advantages: 1) sunlight
– 7 –
Fig. 2.— The XO Mark I system block diagram.
cannot reach the lens and thereby damage a shutter, 2) rain (or snow) slides off the inclined
window, and 3) clouds tend to come from up from below on Haleakala, so we think aiming
our weather camera at a large zenith angle gives early warning of fog rising up the mountain.
If the data from any of the weather stations triggers any of the following conditions,
then we close the roof, or keep it closed: 1) data are non-existent or stale, i.e. older than 30
minutes, 2) humidity > 75% or dew point < 4◦C from ambient, 3) temperature < −5◦C,
4) wind velocity > 20 m s−1, 5) non-zero rain accumulation in the past 10 minutes, 6)
sunlight detected on a solar cell. Humidity and/or dew point accounts for nearly all closures.
Because the humidity on Haleakala tends to be quite bimodal, either very low or nearly 100%,
and because high humidity should correlate with poor photometric precision, we have not
attempted to optimize the thresholds for humidity and dew point.
Another tenant on Haleakala operates weather sensors and a neural network prediction of
the probability of inclement weather in the near future (a “forecast”) and now (a “nowcast”).
XO closes its roof if the “forecast” or “nowcast” indicates a 95% or greater probability of
– 8 –
Fig. 3.— A representative section of a flat-fielded XO image shows the Pleiades star cluster
(left side). This 1024-column section is as wide as the XO images, 7.2◦but is only 6% of the
9000-row height of the scans. The vertical and horizontal trails from very saturated stars do
not prevent good photometry of most of the stars in each image.
Drift scanning is efficient if the number of rows readout is much larger than the number
of columns, i.e. for long, rectangular fields of view. Its advantage over staring-mode imaging
is that the flat-field and the dark correction are homogenized by shifting the charge through
all the rows. Thus, the drift-scanned vectors are more uniform and smoother than their
staring-mode counter parts, which are 2-D arrays. The drift-scanned PSF is slightly wider
in both directions compared to what it would be with “standard” staring-mode observations,
but this is beneficial in the following sense. Drift scanning makes intra-pixel gain variations
irrelevant; even intra-column variations are homogenized due to the fact that stars do not
track perfectly parallel to columns. The drift-scanned PSF is approximately a Gaussian with
FWHM = 1.8 pixels. During the first year of operation, the correlation coefficient of the
observed PSF with a FWHM=1.8 pixel Gaussian is ∼ 0.80 with no measurable dependence
on time of night, day of year, hour angle, or temperature from 3◦C to 16◦C. The correlation
– 9 –
Fig. 4.— The periods of time spent observing during the first year of operation are plotted.
The dates of full moon are indicated with plus signs.
coefficient is expected to be variable with position on the CCD, because the scan rate cannot
be optimum for all columns of the CCD. Indeed if the scanning of the CCD and mount are
purposefully not synchronized, for example in order to mitigate saturation of very bright
stars by elongating them, the PSF elongates in the center of the CCD and becomes shaped
like a “)” or a “(” on either side of center. With proper synchronization, the elongation is
nearly imperceptible and nearly uniform across the field as intended (see Figure 6).
By rotating the mount at 478′′s−1about the declination axis while tracking at the
sidereal rate in right ascension, the XO system scans repeatedly from 0◦to 63◦of declination.
The repeatability of a star’s position in rows, Y, is directly related to the repeatability of
the timing of the CCD readout and its synchronization with the equatorial mount. Typically
the repeatability is ±1 second, corresponding ±19 rows, peak to valley. Occasionally larger
– 10 –
Fig. 5.— This figure shows a representative section of a calibrated array of differential stellar
magnitudes, with each column corresponding to a single star, and each row to a single epoch.
The mean magnitude of each star has been subtracted from each column. Bad data have
not been flagged. Ideally the entire image would be white noise, less at the left (brighter
stars) and more at the right (fainter stars). In practice data are missing (white) and trends
are evident (see text). Visual inspection of diagrams like this reveals trends and assists in
improving the calibration. Large amplitude variable stars are also apparent; an example
near the middle of the figure is shown inside the ellipse.
excursions occur, e.g. due to missing the daily reset of one of the computers’ clocks. The
clocks on the computers reading the CCDs drift reliably by 4 seconds per day.
The repeatability of a star’s position in columns, X, is directly related to the repeatabil-
ity of the mount’s positioning in RA. The repeatability measured over an observing season
is ±4 columns (±100′′), peak to valley, dominated by a slowly-varying function of hour angle
presumably due to residual misalignments or imperfections in the mount.
– 11 –
Fig. 6.— Images of stars are nearly uniform across the field of view. Subimages centered at
CCD columns 100, 300, 500, 700, and 900 (left to right) are displayed for camera 0 (top) and
camera 1 (bottom). The two cameras observe nearly the same field of view simultaneously;
in their native format, they are misaligned by 13 columns, i.e. 1.3% of the sensor’s width,
but we have shifted one by 13.0 columns for this plot.
We constructed a “flat field” vector from a robust averaging of measurements of “sky”
along columns of a few images selected to be relatively free of stars, saturated stars, Galac-
tic cirrus, and gradients or curvature. The “flat field” for each camera is approximately
parabolic, with the peak within 6 columns of the center of the CCD, and it is almost entirely
due to optical vignetting, because the detectors are intrinsically uniform and are made even
more so by the drift scanning technique. At the edges, a noticeable departure upward occurs
that we attribute to excess scattered light from the sky near the edges of the CCD. We
replaced the edges of the measured “flat field” vector with linear extrapolations from the
interior. Other than a bad column in each of the CCDs, we did not measure any repeatable,
small-scale structure in the observed flat field vector, so we smoothed it with a 31-column
boxcar. The flat field’s peak, near the center column of the CCD, is greater than its minimum
at one of the edges, by 34% and 35% respectively for the two cameras. The absolute value
of the slope of the flat field is everywhere < 0.9 millimag column−1. The slope combined
with the repeatability in X-position implies repeatability of the instrument of ±3.6 millimag,
peak to valley, prior to calibration.
– 12 –
3.3. Mount Control
The equatorial mount selected is a Paramount ME manufactured by Software Bisque,
Inc.1We control the Paramount using a custom visual basic script interacting with The
Sky by the same manufacturer. The script commands the mount to scan every ten minutes
according to a pre-determined nightly schedule. For simplicity, the script and the mount
operate regardless of the roof’s state.
The Paramount has required no physical maintenance, and we have experienced only
a few minor problems with it. The cameras are re-oriented by 180◦on the sky whenever
the German equatorial mount crosses the meridian. This requires scanning northward east
of the meridian and southward west of the meridian; it also requires calibrating the data
separately, because the positions of a given star on the CCD are not the same on opposite
sides of the meridian.
3.4.Charge Coupled Devices
The sensors selected are SITe SI-003AB 10242-pixel back-illuminated CCDs with 24
micron pixels, in the model AP8p camera manufactured by Apogee, Inc. The CCD is cooled
thermoelectrically, with waste heat dissipated by fans on a heat sink. Because the CCD
is illuminated by a broad optical bandpass with a f/1.8 lens, dark current can be nearly
negligible with moderate cooling, so for simplicity and reliability, we operate the CCDs in
MPP mode at −30 C year round. A disadvantage of the MPP mode for these sensors is
that in addition to “bleeding” vertically, bright stars leave trails horizontally. In order to
eliminate the horizontal trails, MPP mode can be suppressed during readout. However that
option is not suitable for drift scanning wherein the CCD is read continuously. Operating
the CCD without MMP mode increases the dark current by a factor of 20, which we decided
would be less desirable than having trails associated with a few very bright stars per image.
Each CCD has one unique bad column, but for surveys such as this, sensors with a single
bad column can be more cost effective than ones with no bad columns. We flag data for
which a star’s photometric aperture contains a bad column.
1We experimented with and rejected a less expensive mount due to its “runaway” behavior described by
Lopez-Morales & Clemens (2004).
– 13 –
4.Observing Strategy and Experience
Our observing strategy is based upon a compromise between observing a region of sky
for many weeks and observing at moderate zenith angles. As stated before, we scan in
declination from 0◦to +63◦, and each night a program selects the right ascensions α of the
scans from a table. Each star is observed by each of the two cameras every 10 minutes. On a
given night, we either concentrate on a particular RA, α0, or split the night at midnight and
observe one RA, α−, before midnight and another RA, α+, after midnight. In the former
case, we observe the primary target whenever its hour angle is within 4 hours of the meridian,
and after/before it is within that range we observe RA=α0± 4 hours. With this strategy,
a season of observing a given target lasts ∼4 months, as follows: it is observed only a few
times in the morning for the first ∼0.5 month; from midnight to morning for the next month;
whenever it is within 4 hours of the meridian for a month; from evening until midnight for a
month; and finally only a few times in the evening for the last ∼0.5 month of its season. In
a calendar year, this strategy targets six RAs, each separated by 4 hours from its neighbors.
The visibility function as a function of period of the transits for a representative star is
given in Figure 7. A second season of observation of the same stars improves the visibility of
transits, especially for longer orbital periods, but at the cost of not observing “new” stars.
With a second observing season on the same stars, the longer time baseline permits more
precise period determination, which is important for predictions of future transits and for
accurately estimating the orbital phase of radial velocities.
XO’s 25′′pixel−1scale is approximately twice that of other similar projects (Figure 3).
This is primarily because our system was designed to scan large regions of sky, typically far
from the Galactic plane, where the stellar density per unit solid angle is small. Also, the
greater surface brightness in the Galactic plane elevates the photometric noise. As stated
earlier, XO’s targets are approximately isotropically distributed on the sky. Therefore, at low
Galactic latitudes we expect that the number of solar type stars to be enhanced somewhat
but the number of false positives to be enhanced greatly, either astrophysically (Brown 2003)
or instrumentally, i.e. that the crowding will make the photometry more error prone.
In the first year, the XO system’s operational readiness was as follows: 1) it missed 36
contiguous nights (10%) due to a roof motor failure caused by warping of the roof’s rails at
the onset of winter, 2) it did not observe 22 additional nights (6%) due to various shorter-
term equipment or software problems throughout the year, 3) it did not observe at all on 62
nights (17%) due to weather, and 4) of the remaining 245 nights (67%), data was gathered
for some fraction of the night, with that fraction being distributed on [0,1] approximately
Figure 4 shows many patterns to the actual observations in the first year. The hour-
– 14 –
Fig. 7.— The visibility of transits of a given period are indicated for observations of at least
1 transit (solid), at least 2 transits (dotted), and at least 3 transits (dashed). In the plot,
the fraction of transits seen indicates the fraction that could be seen with sufficient detection
sensitivity, given the actual times of observation of RA=0hby XO in 2003. Characteristically,
the visibility is excellent at periods of 3 days or less and tapers off rapidly for longer periods.
glass shape is due to the annual pattern of times of darkness for Maui. The sky is more
often cloudy in summer than winter. The roof was inoperable for all of January 2004. The
telescope stops observing for 10-30 minutes when transitioning from one target RA to another
(which often occurs near midnight) and at meridian crossing (ragged slanted lines). Moon
light saturates the detector when the telescope scans near the moon, so some data is lost
around full moon; for example, on nights 94-96 moonlight saturated the southern part of
each scan. The ragged data after midnight on nights 50-55 was caused by a software error:
we had been discarding any image with fewer than 5000 stars detected, but for RA = 12
– 15 –
hours, this threshold was too high, so we adjusted it to be 1500 stars. Our linux operating
system performed housekeeping by default at 4 AM, and until we forced this to occur during
daytime, images at 4 AM were discarded because they were trailed slightly (nights -112 to
-89). During approximately that same period, we had a simplistic schedule of observing only
RA = 0 hours, and we kept the shutter-closed images taken during twilight.
5.1.Readout of CCD
We accomplish the CCD readout in drift-scan mode using Tcl scripts and linux drivers
for the Apogee cameras. We modified the Random Factory Version 0.5 GUI interface to
permit command-line control, because the latter is better suited to autonomous operation.
We used Xvfb to create a virtual frame buffer to work around the code’s default operation
within X windows.
Each CCD is attached to a laptop computer’s parallel port for control and data transfer.
The laptop operates the CCD on a predetermined schedule, initiated by linux’s cron every 10
minutes. During twilight the camera’s shutters are not opened but the CCD is readout just
as it will be during normal (shutter-open) operation. Whenever the sun’s altitude < −15◦,
the control computer checks the weather (Section 3.1) and if it is clear, it opens the roof and
the camera’s shutter and reads the CCD at a rate of 18.8 rows per second. The first 1024
rows are discarded because they are improperly exposed, and the next 9000 rows are stored
as a FITS file. These files are automatically analyzed the following morning as described in
the next Section.
Rarely, one of the linux laptops “freezes” and must be revived by human intervention.
On the summit of Haleakala, the laptops are at their specified limit for ambient atmospheric
pressure (altitude < 3048 m), and for the first year they were routinely operated at too
high an ambient temperature, especially for a few hours each morning when they reduced
data. The frequency of “freezes” has been reduced to once every few months by increasing
ventilation to reduce the temperature of the laptops’ ambient environment by ∼5 C to within
specification (T < 31.3◦C).
Initially we operated the CCD’s only at night, warming them in the morning and cooling
them at night. This was to reduce the risk of a power outage causing an uncontrolled return
to ambient temperature that might crack the CCD or its thermoelectric cooler. However,
occasionally a CCD would not cool after turning it on at nightfall, so we chose to leave them
on and cooled all the time. XO has experienced at least one power outage that outlasted