Draft version December 1, 2021
Typeset using L
A
T
E
X
twocolumn
style in AASTeX63
COMAP Early Science: III. CO Data Processing
Marie K. Foss
,
1
H
̊
avard T. Ihle
,
1
Jowita Borowska,
1
Kieran A. Cleary
,
2
Hans Kristian Eriksen
,
1
Stuart E. Harper
,
3
Junhan Kim
,
2
James W. Lamb
,
4
Jonas G. S. Lunde,
1
Liju Philip
,
5
Maren Rasmussen,
1
Nils-Ole Stutzer
,
1
Bade D. Uzgil
,
6
Duncan J. Watts
,
1
Ingunn K. Wehus
,
1
David P. Woody,
4
J. Richard Bond
,
7
Patrick C. Breysse
,
8
Morgan Catha,
4
Sarah E. Church,
9
Dongwoo T. Chung
,
7, 10
Clive Dickinson
,
3
Delaney A. Dunne
,
2
Todd Gaier,
5
Joshua Ott Gundersen,
11
Andrew I. Harris
,
12
Richard Hobbs,
4
Charles R. Lawrence,
5
Norman Murray,
7
Anthony C. S. Readhead
,
2
Hamsa Padmanabhan
,
13
Timothy J. Pearson
,
2
Thomas J. Rennie
,
3
(COMAP Collaboration)
1
Institute of Theoretical Astrophysics, University of Oslo, P.O. Box 1029 Blindern, N-0315 Oslo, Norway
2
California Institute of Technology, Pasadena, CA 91125, USA
3
Jodrell Bank Centre for Astrophysics, Alan Turing Building, Department of Physics and Astronomy, School of Natural Sciences, The
University of Manchester, Oxford Road, Manchester, M13 9PL, U.K.
4
Owens Valley Radio Observatory, California Institute of Technology, Big Pine, CA 93513, USA
5
Jet Propulsion Laboratory, California Institute of Technology, 4800 Oak Grove Drive, Pasadena, CA 91109, USA
6
California Institute of Technology, 1200 E. California Blvd., Pasadena, CA 91125, USA
7
Canadian Institute for Theoretical Astrophysics, University of Toronto, 60 St. George Street, Toronto, ON M5S 3H8, Canada
8
Center for Cosmology and Particle Physics, Department of Physics, New York University, 726 Broadway, New York, NY, 10003, USA
9
Kavli Institute for Particle Astrophysics and Cosmology & Physics Department, Stanford University, Stanford, CA 94305, USA
10
Dunlap Institute for Astronomy and Astrophysics, University of Toronto, 50 St. George Street, Toronto, ON M5S 3H4, Canada
11
Department of Physics, University of Miami, 1320 Campo Sano Avenue, Coral Gables, FL 33146, USA
12
Department of Astronomy, University of Maryland, College Park, MD 20742, USA
13
Departement de Physique Th ́eorique, Universite de Gen`eve, 24 Quai Ernest-Ansermet, CH-1211 Gen`eve 4, Switzerland
ABSTRACT
We describe the first season COMAP analysis pipeline that converts raw detector readouts to cali-
brated sky maps. This pipeline implements four main steps: gain calibration, filtering, data selection,
and map-making. Absolute gain calibration relies on a combination of instrumental and astrophys-
ical sources, while relative gain calibration exploits real-time total-power variations. High efficiency
filtering is achieved through spectroscopic common-mode rejection within and across receivers, result-
ing in nearly uncorrelated white noise within single-frequency channels. Consequently, near-optimal
but biased maps are produced by binning the filtered time stream into pixelized maps; the corre-
sponding signal bias transfer function is estimated through simulations. Data selection is performed
automatically through a series of goodness-of-fit statistics, including
χ
2
and multi-scale correlation
tests. Applying this pipeline to the first-season COMAP data, we produce a dataset with very low lev-
els of correlated noise. We find that one of our two scanning strategies (the Lissajous type) is sensitive
to residual instrumental systematics. As a result, we no longer use this type of scan and exclude data
taken this way from our Season 1 power spectrum estimates. We perform a careful analysis of our data
processing and observing efficiencies and take account of planned improvements to estimate our future
performance. Power spectrum results derived from the first-season COMAP maps are presented and
discussed in companion papers.
1.
INTRODUCTION
Corresponding author: Marie K. Foss
m.k.foss@astro.uio.no
Understanding the evolution of galaxies and the in-
tergalactic medium (IGM) over the largest spatial and
temporal scales is one of the principal goals of cosmol-
ogy. Galaxy surveys address this challenge by resolving
and detecting individual galaxies, a technique that nec-
arXiv:2111.05929v3 [astro-ph.IM] 30 Nov 2021
2
Foss et al.
essarily favors brighter galaxies and smaller cosmic vol-
umes. Spectral line intensity mapping (LIM) (Madau
et al. 1997; Battye et al. 2004; Peterson et al. 2006;
Loeb & Wyithe 2008) is a complementary technique (see
Kovetz et al. 2017 or Kovetz et al. 2019 for a review)
that holds the potential to characterize the global prop-
erties of galaxies and their evolution by surveying the
aggregate emission from all galaxies over large volumes.
This technique uses redshifted line emission (e.g., 21-
cm, Ly
α
, CO, or C
ii
) as a tracer for the underlying den-
sity field. Large volumes along a given line-of-sight may
be surveyed simultaneously with a single spectrometer
at relatively low spatial resolution, and by scanning this
spectrometer across the sky a full 3D density map may
be derived. Despite multiple different modeling efforts
(Righi et al. 2008; Visbal & Loeb 2010; Lidz et al. 2011;
Pullen et al. 2013; Breysse et al. 2014; Li et al. 2016;
Padmanabhan 2018; Moradinezhad Dizgah & Keating
2019; Sun et al. 2019; Yang et al. 2021; Moradinezhad
Dizgah et al. 2021; Chung et al. 2021a) and significant
progress on the observational front (Keating et al. 2016;
Riechers et al. 2018; Keating et al. 2020; Keenan et al.
2021), the overall level of the CO signal, especially in
the clustering regime, is still unknown.
The CO Mapping Array Project (COMAP; Cleary
et al. 2021) is an intensity mapping experiment that
aims to use emission from carbon monoxide (CO) to
trace the aggregate properties of galaxies over cosmic
time, back to the Epoch of Reionization. A Pathfinder
experiment, consisting of a 19-feed 26–34 GHz receiver,
has been fielded on a 10.4 m single-dish telescope at the
Owens Valley Radio Observatory (OVRO). In this fre-
quency range, the receiver is sensitive to CO(1–0) at
z
= 2
.
4–3.4, with a fainter contribution from CO(2–1)
at
z
= 6–8. The main goal of the Pathfinder is to detect
the CO(1–0) signal and use it to constrain the properties
of galaxies at the Epoch of Galaxy Assembly. A future
phase will add a second receiver at 12–20 GHz in order to
detect CO(1–0) from around
z
= 5–9, cross-correlating
with the CO(2–1) signal from the 26–34 GHz receiver
and constraining the properties of galaxies towards the
end of the Epoch of Reionization.
The receiver’s detector chain is based on cryogeni-
cally cooled HEMT low-noise amplifiers (LNA) which
contribute to a typical system temperature of about
44 K across the full frequency range. The predicted
signal from high-redshift CO emission is expected to
be no more than a few microkelvin per COMAP spa-
tial/spectral resolution element (or “voxel”). Thus, the
raw instrumental noise must be reduced by many orders
of magnitude before a statistically significant detection
may be achieved. In practice, this is done by repeatedly
observing the same part of the sky using multiple de-
tectors, and thereby gradually increasing the sensitivity
per voxel. For this to succeed, however, it is necessary
to suppress systematic contributions from atmospheric
temperature variations, sidelobe contamination, ground
pickup, standing waves, Galactic foregrounds, etc. by a
corresponding amount.
The first season COMAP science observations started
in June 2019 and lasted until August 2020. This pa-
per describes the first season COMAP data analysis
pipeline, which aims to produce clean maps from raw
time-ordered COMAP observations. This includes cali-
bration, data selection, filtering, and map-making. The
rest of this paper is organized as follows: First, in or-
der to establish useful notation and conventions, we give
a brief introduction to the COMAP instrument in Sec-
tion 2, while referring the interested reader to Lamb
et al. (2021) for full details. Next, we provide a high-
level overview of the analysis pipeline in Section 3.1,
before specifying each step in Sections 3.3–3.6. Data se-
lection and efficiency is discussed in Sections 4 and 5.
The results are presented in Section 6, and we summa-
rize and conclude in Section 7.
2.
INSTRUMENT AND DATA MODEL
Before describing the COMAP analysis pipeline, we
provide a brief overview of the instrument itself, and
define an explicit data model. A more detailed descrip-
tion of the instrument can be found in a separate paper
(Lamb et al. 2021).
2.1.
Instrument overview
The COMAP Phase I instrument observes in the K
a
band, at 26–34 GHz and is located at the Owens Valley
Radio Observatory (OVRO) in California, USA. It is
mounted on a 10.4 m telescope that was originally built
for the Millimeter Array at OVRO, then used as a part
of the Combined Array for Research in Millimeter-wave
Astronomy (CARMA) experiment, and has now been
repurposed for COMAP. The telescope’s primary and
secondary reflectors have diameters of 10.4 m and 1.1 m,
respectively, and the beam FWHM is about 4.5 arcmin
at 30 GHz.
The receiver comprises 19 independent detector
chains, called “feeds”. The signal chain of each feed
consists of individual feed horns, polarizers, low noise
amplifiers, two stages of downconversion, frequency sep-
aration and digitization. For the observations described
in this paper, 15 feeds have a two-stage polarizer, two
feeds have a single-stage polarizer, and two feeds have
no polarizer. The digitization happens in two CASPER
“ROACH-2” FPGA-based spectrometers for each sig-
nal chain, giving us four 2 GHz-wide sidebands (SB),
COMAP Early Science: CO Data Processing
3
0
5
10
15
20
Local Sidereal Time [h]
0
10
20
30
40
50
60
70
80
90
Elevation [deg]
Field 1
Field 2
Field 3
Tau A
Cas A
Cyg A
Figure 1.
Elevation of CO (pink/purple) and calibration
(orange) fields as a function of Local Sidereal Time.
each of which has 1024 frequency channels, resulting in
a native frequency resolution of approximately 2 MHz.
The two sidebands of each band (A and B) are labelled
“lower” (LSB) or “upper” (USB). For more details on
the instrument see Lamb et al. (2021).
To support frequent and accurate gain estimation,
COMAP employs an ambient temperature load that is
directly attached to the environmental shroud housing.
This “calibration vane” is automatically moved in front
of the feed horn array at the beginning and end of each
observation (each lasting for about one hour; see Sec-
tion 2.3), fully filling the field of view of each pixel. The
temperature of the calibration vane is monitored with
sensors, allowing the system temperature to be calcu-
lated and applied to calibrate the gain (see Section 3.4
for more details).
2.2.
Field Selection
COMAP observes several parts of the sky. Table 1
lists all CO science fields and calibrators
1
. In Figure 1
we plot the elevation of the CO and calibration fields as
a function of Local Sidereal Time, indicating when the
fields are available for observation. Figures 2 and 3 show
the position of the three CO fields observed by COMAP.
These were selected to maximize the observing efficiency,
avoid bright 30 GHz point sources (
&
1 Jy), and overlap
with the coverage of Hobby-Eberly Telescope Dark En-
ergy eXperiment (HETDEX; Hill et al. 2008; Gebhardt
1
Since COMAP began observing, the boundaries of the HETDEX
Spring field coverage changed, with the result that one COMAP
field no longer overlaps with the main HETDEX survey although
we hope to also fill in this field with additional HETDEX obser-
vations.
et al. 2021; Hill et al. 2021), a galaxy survey target-
ing Ly-alpha emission from galaxies in the same red-
shift. Although COMAP’s observing strategy has been
designed to permit the direct detection of CO fluctua-
tions from galaxies at
z
= 2
.
4
−
3
.
4, cross-correlation
with a galaxy survey such as HETDEX can increase the
detection significance by at least a factor of two (Chung
et al. 2019; Silva et al. 2021) as well as provide valida-
tion for the origin of detected signal in galaxies at the
target redshift.
In addition to the main science fields, we are also con-
ducting a survey of the Galactic plane covering longi-
tudes 20
◦
< l <
220
◦
, details of which can be found in
Rennie et al. (2021).
To facilitate calibration with astrophysical sources, we
observe a handful of radio sources, including Jupiter, the
supernova remnants Taurus A (TauA) and Cassiopeia A
(CasA), and the radio galaxy Cygnus A (CygA), all of
which are somewhat extended compared to the beam
except for Jupiter.
2.3.
Observation Strategy
Telescope scans of the science fields follow a harmonic
motion described by
az =
A
sin(
at
+
φ
); el =
B
sin(
bt
)
,
(1)
where
A,B
are amplitude parameters that define the
size of the field, the ratio
a/b
determines the shape of
the curve, and
φ
is a phase parameter. Two different
scan types were used: “constant elevation scans (CES)”
(
b
= 0) and “Lissajous” (varying parameters), alternat-
ing between each on a daily basis. At the start of a
scan, the telescope is positioned at the leading edge of
the field. The telescope then executes the scan while the
field drifts through the pattern. This typically takes 3–
10 minutes, after which the telescope is repointed to the
leading edge of the field again in preparation for the next
scan. An example of the scanning path for about one
hour of continuous observations with a Lissajous scan
and a CES is shown in Figure 4. Testing the relative
performance of the CES and Lissajous scanning strate-
gies in terms of final data quality is an important goal
of the first-season COMAP survey.
2.4.
Data model
As described by Lamb et al. (2021), the COMAP de-
tector readout for a single frequency channel may be
modelled as
P
out
=
k
B
G
∆
νT
sys
,
(2)
where
k
B
is the Boltzmann constant,
G
is the gain, ∆
ν
is the bandwidth, and
T
sys
is the system temperature of
4
Foss et al.
Table 1.
COMAP fields and calibrators
Field Name RA (J2000) Dec (J2000) Notes
Field 1
01:41:44.4
+00:00:00.0
CO science field - lies within the HETDEX Fall field
Field 2
11:20:00.0
+52:30:00.0
CO science field - lies within the HETDEX Spring field
Field 3
15:04:00.0
+55:00:00.0
CO science field
TauA
05:34:31.9
+22:00:52.2
Pointing calibrator - supernova remnant (Crab Nebula)
CasA
23:23:24.0
+58:48:54.0
Pointing calibrator - supernova remnant
CygA
19:59:28.4
+40:44:02.1
Pointing calibrator - radio galaxy
Jupiter
Pointing calibrator
Field 1
Field 2
Field 3
0
1200
K
CMB
Figure 2.
The three CO fields observed by the telescope overplotted as contours with radii of
∼
1
◦
, centered at the field centers
(in Galactic coordinates) (lon
,
lat) = (149
.
0
◦
,
−
60
.
3
◦
)
,
(150
.
64
◦
,
59
.
53
◦
) and (91
.
35
◦
,
53
.
22
◦
) for Fields 1, 2 and 3 respectively,
on top of the
Planck
LFI 30 GHz full-mission map (downloaded from the
Planck Legacy Archive
Planck Collaboration et al.
2020).
the instrument. The system temperature may be further
modeled as
2
T
sys
=
T
receiver
+
T
atmosphere
+
T
ground
+
T
CMB
+
T
foregrounds
+
T
CO
,
(3)
where
T
reciever
is the effective noise temperature of the
receiver,
T
atmosphere
is the noise contribution from the
2
In this section we are writing all the contributions to
T
sys
in
terms of their effective noise contribution, rather than any physi-
cal temperatures. See Section 3.4 for a definition of
T
sys
in terms
of physical quantities.
atmosphere,
T
ground
is ground pickup from far sidelobes,
T
CMB
is the contribution from the CMB,
T
foregrounds
are
continuum foregrounds (typically from the galaxy), and
T
CO
is the line emission signal from extragalactic CO,
which is the main scientific target of the COMAP in-
strument.
To understand the challenges involved in measuring
the cosmological CO signal, it is instructive to consider
the order of magnitude and stability of each term in
Equation (3). The largest single contribution is that
of the receiver temperature, which is usually about 10–
COMAP Early Science: CO Data Processing
5
30°
20°
10°
10°
5°
0°
-5°
Right Ascension (J2000)
Declination (J2000)
HETDEX
Fall Field
Field 1
240°
210°
180°
50°
40°
Right Ascension (J2000)
Declination (J2000)
Field 2
Field 3
HETDEX
Spring Field
0
125
250
375
500
K
CMB
Figure 3.
The three CO fields observed. The contours, illustrating the rough coverage of each field, have radii of
∼
1
◦
. In the
left and right panels respectively we have drawn in the approximate coverage of the HETDEX Fall and Spring fields presented
by Gebhardt et al. (2021). The map in the background is the same
Planck
LFI 30 GHz full-mission map (downloaded from the
PLA, Planck Collaboration et al. 2020) as seen in Figure 2.
50
51
52
53
54
Elevation [degrees]
180
185
190
195
200
205
Azimuth [degrees]
50
51
52
53
54
Elevation [degrees]
0
10
20
30
40
50
60
Time [minutes]
Figure 4.
Movement of the telescope boresight in azimuth and elevation for an observation employing Lissajous scans (top)
and an observation employing CES (bottom). Both observations consist of 15 individual scans of Field 1.
30 K. For the COMAP receiver, with HEMT LNA tech-
nology, this is very stable.
The second-largest contribution is from the atmo-
sphere, which typically adds 15–25 K. This term varies
significantly on all time scales longer than a few sec-
onds, and depends on external conditions including ele-
vation, humidity, cloud coverage, ambient temperature
and wind speed. It is also strongly correlated between
detectors and frequencies, since all feeds observe through
essentially the same atmospheric column at any given
time; fortunately, the phase structures of the atmo-
spheric fluctuations are uncorrelated on long time scales.
Next, ground pickup typically accounts for 5–6 K, and
this term can be particularly problematic because it de-
pends sensitively on the instrument pointing: If a side-
lobe happens to straddle a strong signal gradient, such
as the horizon or the Sun, several mK variations may
be measured on very short timescales and with a time-
dependency that appears nearly sky synchronous.
The fourth term represents the CMB temperature
of 2.7 K, which is both isotropic and stationary, while
the fifth term represents astrophysical foregrounds, ex-
pected to contribute at most 1 mK; for instance syn-
chrotron, free-free, and dust emission from the Galaxy.
Although these are sky synchronous, and in principle
could confuse potential CO measurements, they also
have very smooth frequency spectra (Keating et al.
2015), and are therefore relatively easy to distinguish
from the cosmological CO signal, which varies rapidly
with frequency. An important potential exception is line
emission from other molecules redshifted to our band
from galaxies at other epochs. The hydrogen cyanide
(HCN) line is expected to be one of the brightest such
lines. Emission from HCN in galaxies towards our CO
6
Foss et al.
fields at redshift
z
= 1
.
6–2.4 will appear in our frequency
range. However, this contribution is expected to be an
order of magnitude lower than that from CO (Chung
et al. 2017).
Finally, the cosmological CO line emission signal is
expected to account for
O
(1
μ
K). Whether it is possible
to detect such a weak signal depends directly on the sta-
bility and sensitivity of the instrument. In this respect,
the fundamental quantity of interest is the overall noise
level of the experiment, which is dominated by random
thermal noise.
The magnitude of these random thermal fluctuations
is proportional to
T
sys
, with a standard deviation that
is given by the so-called radiometer equation,
σ
N
=
T
sys
√
∆
ν τ
,
(4)
where
τ
is the integration time. Thus, since both the
system temperature and the bandwidth are essentially
fixed experimental parameters, the only way of reduc-
ing the total uncertainty is by increasing the integration
time. As a concrete and relevant example, we note that
an integration time of 45 hours is required to achieve a
standard deviation of 20
μ
K with a system temperature
of 45 K and a bandwidth of 31.25 MHz.
In addition to the thermal and uncorrelated noise de-
scribed by the radiometer equation, there are three main
sources of correlated noise, namely gain fluctuations in
the low-noise amplifiers, atmospheric temperature fluc-
tuations, and time-dependent standing waves. All of
these are expected to have a roughly 1
/f
-type spectrum,
although with different particular properties
3
. The fact
that these sources of correlated noise are also strongly
correlated between frequencies is very useful in order to
filter out this noise in the analysis.
Equation (2) describes the detector output at any
given time. To connect this to the actual measurements
recorded by the detector, we adopt the following data
model,
d
i
ν
(
t
) =
〈
d
i
ν
〉
(1 +
δ
i
G
(
t
))
[
1 +
P
i
cel
(∆
s
cont
+ ∆
s
ν
CO
)
+
P
i
tel
∆
s
ground
+
n
corr
(
t
) +
n
νi
w
(
t
)
]
.
(5)
Here
d
i
ν
(
t
) denotes the raw data recorded at time
t
for frequency channel
ν
in feed
i
;
〈
d
i
ν
〉
represents the
corresponding time average and basically corresponds
to
〈
T
iν
sys
(
t
)
〉〈
G
i
ν
(
t
)
〉
;
δ
i
G
(
t
) denotes feed dependent gain
fluctuations;
P
i
cel
and
P
i
tel
are pointing matrices in ce-
lestial and telescope coordinate systems, respectively;
3
There are several different sources of standing waves, some of the
main ones give rise to 1
/f
-like spectra, but others do not.
∆
s
cont
denotes the celestial continuum source fluctua-
tions, mainly from the CMB and Galactic foregrounds;
∆
s
ν
CO
is the CO line emission fluctuation; ∆
s
ground
is the
ground signal fluctuation picked up by the far sidelobes;
and
n
corr
(
t
) are the correlated temperature fluctuations,
mostly consisting of atmosphere fluctuations and stand-
ing waves. Factors with no feed or frequency index are
assumed to be similar (or at least strongly correlated)
at different frequencies and feeds, while factors with a
ν
label indicate parts of the model that are assumed to
have non-smooth frequency dependence. The main pur-
pose of the COMAP analysis pipeline is to characterize
∆
s
ν
CO
given
d
i
ν
(
t
).
2.5.
Data overview
Before presenting the analysis pipeline, we provide
a preview of the raw time-ordered data (TOD) gener-
ated by the COMAP instrument, with the goal of build-
ing intuition that will be useful for understanding the
purpose of each component of the analysis pipeline de-
scribed in this paper. Figures 5 and 6 show examples
of such raw time-ordered data (TOD) from the instru-
ment using the CES (left column) and Lissajous (right
column) scanning strategies. Perhaps the most obvious
features in these plots are step-wise changes in power as
the telescope changes elevation during repointings be-
tween scans; see Section 2.3. The Lissajous scans ad-
ditionally show oscillations in power as the telescope
changes elevation during the scan, since the telescope
looks through a thicker slab of atmosphere at lower ele-
vations, and this increases the atmospheric contribution
to the system temperature.
The top panels in Figure 6 show an individual fre-
quency channel for a single scan (i.e., stationary obser-
vation period), while the bottom panel shows the cor-
responding power spectral density (PSD). For the CES
case, the PSD is relatively featureless, with an overall
shape that looks consistent with a typical 1
/f
noise spec-
trum. For the Lissajous case, an additional strong peak
is seen around 0.007 Hz, which matches the scanning
period of 14 sec, and this corresponds to the periodic
atmospheric variations seen in the panels above.
Figure 7 shows the time averaged data for all fre-
quency channels of a single feed for one scan. The spec-
tral shape is mostly determined by the average gain as a
function of frequency, due to the combined effect of the
various components of the receiver chain. This average
gain is a purely instrumental effect, not associated with
the true sky signal, and therefore simply corresponds
to a normalization factor that should be calibrated out
before higher-level analysis. However, some of the spec-
tral shape is also determined by the fact that the sys-
COMAP Early Science: CO Data Processing
7
0.138
0.140
0.142
Power [du]
Constant elevation scan
feed 2, A:USB
0.136
0.138
0.140
0.142
Power [du]
Lissajous scan
0
2
4
6
8
10
12
14
Time [m]
0.1000
0.1025
0.1050
0.1075
0.1100
Power [du]
= 28.488 GHz
= 28.490 GHz
= 28.492 GHz
= 28.494 GHz
0
2
4
6
8
10
12
14
Time [min]
0.100
0.105
Power [du]
Figure 5.
Raw data from the COMAP instrument (in arbitrary digital units of power) Here we see data averaged over a single
2 GHz-wide sideband (top) and examples of data from four individual frequency channels in that sideband (bottom). These
data were taken using two different scan patterns: CES (left) and Lissajous (right).
0
50
100
150
200
Time [s]
0.100
0.102
0.104
Power [MW Hz
1
]
Constant elevation scan
Feed 2, A:USB, = 28.488 GHz
0
50
100
150
200
Time [s]
0.098
0.100
0.102
Power [MW Hz
1
]
Lissajous scan
10
2
10
1
10
0
10
1
Frequency [Hz]
10
10
10
8
10
6
PSD [MW
2
Hz
3
]
Feed 2, A:USB, = 28.488 GHz
10
2
10
1
10
0
10
1
Frequency [Hz]
10
10
10
8
10
6
PSD [MW
2
Hz
3
]
Figure 6.
Raw data from an individual frequency channel of the COMAP instrument. Power is shown as a function of time
(top), and the corresponding power spectral density (PSD) is also shown (bottom). We show data from a CES scan (left) and
a Lissajous scan (right).
tem temperature also changes with frequency, and in
some cases exhibits large spikes within specific frequency
ranges (see Lamb et al. (2021) for more details). Sepa-
rating the gain variation as a function of frequency from
the system temperature as a function of frequency is
a main goal of the calibration procedures described in
Section 3.4.
In Figure 8 we plot the correlation,
C
ij
=
〈
ˆ
d
i
ˆ
d
j
〉
√
〈
ˆ
d
i
ˆ
d
i
〉〈
ˆ
d
j
ˆ
d
j
〉
,
(6)
between the power,
ˆ
d
i
recorded by any two feeds,
i
and
j
,
after averaging over all frequencies within each sideband
for each radiometer. Here we first note that the data
8
Foss et al.
26
27
28
29
30
31
32
33
34
Frequency [GHz]
0.0
0.2
0.4
0.6
0.8
1.0
Power [du]
A:LSB
A:USB
B:LSB
B:USB
Figure 7.
Time-averaged raw data from each frequency channel on a single feed of the COMAP instrument. The colors
represent the four 2 GHz-wide sidebands. Note that a few of the frequency channels at at the edges and middle of sidebands
tend to be unstable and are masked out in the analysis.
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
Feed
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
Feed
0.0
0.2
0.4
0.6
0.8
1.0
Correlation
Figure 8.
Correlation between the sideband-averaged data
from the 19 feeds of the COMAP instrument for a single
constant elevation scan. For this observation, as for much
of the observing campaign, the LNAs for feeds 4 and 7 were
turned off because those feeds, as a test, did not have a
polarizer and so had large standing waves due to reflections
between the receiver and the secondary reflector.
from different sidebands of the same feed are strongly
correlated. This is because both main sources of cor-
related noise in the COMAP data, namely gain fluc-
tuations and atmospheric fluctuations, are common for
sidebands within a given feed. In contrast, sidebands for
different feeds mostly share the atmospheric fluctuations
(and also some standing waves), but have independent
gain fluctuations, and this results in lower overall cor-
relations, but still typically in the 10–40% range. Ac-
counting for and mitigating such correlations will clearly
be essential in order to extract robust science from these
observations.
The quality of the COMAP data depends strongly on
the observing conditions, as illustrated in Figure 9. The
top panel shows an observation made under normal con-
ditions, while the middle panel shows an observation
made during poor weather, with thick cloud coverage.
The bottom panel shows a data segment with strong
“spikes”, a feature of some data taken in summer, pos-
sibly associated with insects flying in front of the focal
plane. Automatic identification and removal of prob-
lematic data is clearly an important and necessary com-
ponent of the pipeline.
Finally, Figure 10 shows the calibration vane observa-
tions that are made at the beginning and end of each
observation period. Since the ambient temperature is
about one order of magnitude higher than
T
sys
, the mea-
sured power is also correspondingly about one order of
magnitude higher, and this bright and known signal al-
lows for a precise estimate of
T
sys
. Note that these data
segments are removed prior to data analysis, as they
would otherwise compromise any filtering that may be
applied to the data.
3.
COMAP ANALYSIS PIPELINE
3.1.
Pipeline Overview
We are now ready to present the COMAP analysis
pipeline, which is designed to process the raw data dis-
cussed in Section 2.5 into calibrated and cleaned CO
maps. The main steps of this pipeline are schematically
illustrated in Figure 11.
The processing starts with “Level 1” files, which con-
tain raw data as recorded by the instrument, together
with pointing information and house-keeping data. Each
COMAP Early Science: CO Data Processing
9
0
5
10
15
Time [min]
0.24
0.26
0.28
Power [du]
Feed
01
02
03
04
05
06
07
08
09
10
11
12
13
14
15
16
17
18
19
0
10
20
30
40
50
Time [min]
0.5
1.0
1.5
Power [du]
0
10
20
30
40
50
Time [min]
0.2
0.3
0.4
Power [du]
Figure 9.
Feed averaged COMAP TOD recorded under
various observing conditions. The top panel shows data ob-
served under normal conditions, and is dominated by instru-
mental noise. The middle panel shows data observed under
poor weather conditions with a thick cloud coverage, result-
ing in large coherent power fluctuations observed by all feeds.
This third panel shows data with strong spikes, which may
for instance happen during rare periods with high insect ac-
tivity.
35
40
45
50
Time [s]
2
4
Power [du]
3610
3615
3620
3625
Raw TOD
P
hot
Figure 10.
The calibration vane is inserted in front of the
receiver at the beginning and end of one observation of a CO
science field. The time between calibration vane insertions is
typically about an hour, a period set by the preferred data
file size for the CO field observations.
of these files typically contain about one hour of obser-
vation time, including calibration vane observations at
the beginning and end. We denote each (rougly) one
hour of data as one observation, and assign it an indi-
vidual observation ID (abbreviated obsID). Each obser-
vation consists of several scans, where one scan is the
period between two re-pointings of the telescope, during
which the telescope performs the same motions around
a fixed point in azimuth and elevation while the tar-
get field drifts through. The instrumental properties
are consequently assumed to be stationary within each
scan. The module denoted
scan
detect
in Figure 11
indicates a dedicated code that partitions each obser-
vation into individual scans, based on pointing informa-
tion, and records information of each scan in a database.
The main processing takes place in the
l2gen
mod-
ule, which generates calibrated and cleaned TOD and
stores them in so-called “Level 2” files. This is achieved
through the application of a series of filters (see Sec-
tion 3.3) and a time-varying gain normalization (see Sec-
tion 3.4). This stage also evaluates basic goodness-of-
fit statistics and defines a frequency channel mask that
excludes missing or broken data for the current scan,
before reducing the spectral resolution of the data to
a spectral resolution suitable for map-making. In our
main analysis, we reduce the resolution from
∼
2 MHz
to
∼
31 MHz, resulting in the computational speed-up of
subsequent steps and a memory saving for storing final
maps by a factor of 16.
Next, the
accept
mod
module reads in the statis-
tics (including goodness-of-fit) and basic frequency mask
produced by
l2gen
and produces a list of accepted ob-
servations as defined by user-specified thresholds for
each statistic (see Section 4). Examples of relevant
statistics used for this purpose are
χ
2
per observation,
correlated noise knee-frequency (
f
knee
), and Solar elon-
gation. The output from this process is called an
accept
list
, which determines what data to use for mapmaking.
Converting time-ordered data into pixel-ordered data
is done by a map-maker called
tod2comap
(see Sec-
tion 3.6).
As shown in the following sections, the
adopted filters result in very nearly uncorrelated white
noise, and the current implementation of
tod2comap
ac-
cordingly adopts simple binning into voxels. Finally,
from these maps we can estimate the CO power spec-
trum using the module
comap2ps
(see Ihle et al. 2021
for details).
3.2.
Data Segmentation
As described above, we define a
scan
to be the observ-
ing period between re-pointings of the telescope. The
purpose of the
scan
detect
code is to identify all scans