The University of Manchester Research
Hi intensity mapping with MeerKAT: Calibration pipeline
for multi-dish autocorrelation observations
DOI:
10.1093/mnras/stab1365
Document Version
Final published version
Link to publication record in Manchester Research Explorer
Citation for published version (APA):
Wang, J., Santos, M. G., Bull, P., Grainge, K., Cunnington, S., Fonseca, J., Irfan, M. O., Li, Y., Pourtsidou, A.,
Soares, P. S., Spinelli, M., Bernardi, G., & Engelbrecht, B. (2021). Hi intensity mapping with MeerKAT: Calibration
pipeline for multi-dish autocorrelation observations. Monthly Notices of the Royal Astronomical Society.
https://doi.org/10.1093/mnras/stab1365
Published in:
Monthly Notices of the Royal Astronomical Society
Citing this paper
Please note that where the full-text provided on Manchester Research Explorer is the Author Accepted Manuscript
or Proof version this may differ from the final Published version. If citing, it is advised that you check and use the
publisher's definitive version.
General rights
Copyright and moral rights for the publications made accessible in the Research Explorer are retained by the
authors and/or other copyright owners and it is a condition of accessing publications that users recognise and
abide by the legal requirements associated with these rights.
Takedown policy
If you believe that this document breaches copyright please refer to the University of Manchester’s Takedown
Procedures [http://man.ac.uk/04Y6Bo] or contact uml.scholarlycommunications@manchester.ac.uk providing
relevant details, so we can investigate your claim.
Download date:10. Aug. 2022
© 2021 The Author(s) Published by Oxford University Press on behalf of Royal Astronomical Society
ORIGINAL UNEDITED MANUSCRIPT
MeerKAT multi-dish autocorrelation calibration 1
Hi intensity mapping with MeerKAT: Calibration pipeline for
multi-dish autocorrelation observations
Jingying Wang
?1
, Mario G. Santos
1,2
, Philip Bull
3,1
, Keith Grainge
4
, Steven Cunnington
3
,
José Fonseca
5,6,3,1
, Melis O. Irfan
1,3
, Yichao Li
1
, Alkistis Pourtsidou
3,1
, Paula S. Soares
3
,
Marta Spinelli
7,8
, Gianni Bernardi
9,10,2
, Brandon Engelbrecht
1
1
Department of Physics and Astronomy, University of the Western Cape, Cape Town 7535, South Africa
2
South African Radio Astronomy Observatory (SARAO), 2 Fir Street, Observatory, Cape Town, 7925, South Africa
3
Astronomy Unit, Queen Mary University of London, Mile End Road, London E1 4NS, United Kingdom
4
Jodrell Bank Centre for Astrophysics, Department of Physics and Astronomy, The University of Manchester, Manchester M13 9PL, UK
5
Dipartimento di Fisica “G. Galilei”, Università degli Studi di Padova, Via Marzolo 8, 35131 Padova, Italy
6
INFN – Istituto Nazionale di Fisica Nucleare, Sezione di Padova, Via Marzolo 8, 35131 Padova, Italy
7
INAF-Osservatorio Astronomico di Trieste, Via G.B. Tiepolo 11, 34143 Trieste, Italy
8
IFPU - Institute for Fundamental Physics of the Universe, Via Beirut 2, 34014 Trieste, Italy
9
INAF - Istituto di Radioastronomia, via Gobetti 101, 40129 Bologna, Italy
10
Department of Physics and Electronics, Rhodes University, PO Box 94, Grahamstown, 6140, South Africa
Accepted XXX. Received YYY; in original form ZZZ
ABSTRACT
While most purpose-built 21cm intensity mapping experiments are close-packed interferometer arrays,
general-purpose dish arrays should also be capable of measuring the cosmological 21cm signal. This can
be achieved most efficiently if the array is used as a collection of scanning autocorrelation dishes rather
than as an interferometer. As a first step towards demonstrating the feasibility of this observing strategy,
we show that we are able to successfully calibrate dual-polarisation autocorrelation data from 64 MeerKAT
dishes in the L-band (856 – 1712 MHz, 4096 channels), with 10.5 hours of data retained from six nights
of observing. We describe our calibration pipeline, which is based on multi-level RFI flagging, periodic
noise diode injection to stabilise gain drifts and an absolute calibration based on a multi-component sky
model. We show that it is sufficiently accurate to recover maps of diffuse celestial emission and point sources
over a 10
◦
× 30
◦
patch of the sky overlapping with the WiggleZ 11hr field. The reconstructed maps have
a good level of consistency between per-dish maps and external datasets, with the estimated thermal noise
limited to 1.4 × the theoretical noise level (∼ 2 mK). The residual maps have rms amplitudes below 0.1
K, corresponding to < 1% of the model temperature. The reconstructed Galactic Hi intensity map shows
excellent agreement with the Effelsberg-Bonn Hi Survey, and the flux of the radio galaxy 4C+03.18 is
recovered to within 3.6%, which demonstrates that the autocorrelation can be successfully calibrated to
give the zero-spacing flux and potentially help in the imaging of MeerKAT interferometric data. Our results
provide a positive indication towards the feasibility of using MeerKAT and the future SKA to measure the
Hi intensity mapping signal and probe cosmology on degree scales and above.
Key words: cosmology: observations – large-scale structure of Universe – radio lines:
galaxies –methods: statistical – data analysis – instrumentation: spectrographs
1 INTRODUCTION
Intensity mapping provides a way of measuring the cosmological
clustering signal from large numbers of galaxies without having to
resolve them individually. Using a spectral line, such as the 21cm
line of neutral hydrogen, it is possible to accurately recover the
spatial distribution of brightness temperature fluctuations of the
line in three dimensions (as a function of angle and redshift), and
therefore – assuming that the neutral hydrogen is a biased tracer
?
astro.jywang@gmail.com
of the dark matter distribution – recover familiar clustering signals
such as the baryon acoustic oscillations (BAO) and redshift space
distortions (RSDs) in the late Universe, or the evolving ionisation
state of the intergalactic medium during Cosmic Dawn and the
Epoch of Reionisation. This method, known as Hi intensity map-
ping (Hi IM hereafter), is a promising technique for efficiently map-
ping the large-scale structure of the Universe and thus delivering
precision constraints on cosmological models (Chang et al. 2008;
Loeb & Wyithe 2008; Mao et al. 2008; Pritchard & Loeb 2008;
Wyithe & Loeb 2008; Wyithe et al. 2008; Peterson et al. 2009;
Downloaded from https://academic.oup.com/mnras/advance-article/doi/10.1093/mnras/stab1365/6276733 by University of Manchester user on 28 May 2021
ORIGINAL UNEDITED MANUSCRIPT
2 Wang et al.
Bagla et al. 2010; Seo et al. 2010; Lidz et al. 2011; Ansari et al.
2012; Battye et al. 2013; Bull et al. 2015; Kovetz et al. 2017).
Hi IM experiments differ from traditional spectroscopic
galaxy surveys in their ability to recover 3D clustering informa-
tion on large scales with a high survey speed over a very wide
redshift range, potentially allowing us to survey extremely large
fractions of the total available comoving volume of the cosmos
along our past lightcone using this method. While the 21cm signal
itself is much fainter than many foreground sources – by a factor
of around 10
4
− 10
5
compared to Galactic synchrotron emission
for example – the advent of large radio telescope arrays, consisting
of many elements with low-noise, high-bandwidth receivers, has
now made it practical to perform large cosmological surveys with
sufficient depth for detection within a reasonable observing time.
The majority of radio telescopes currently targeting the cos-
mological 21cm signal are large interferometric arrays observing at
low frequencies, i.e., at z > 6. These include the Low-Frequency
Array (LOFAR)
1
, the Murchison Widefield Array (MWA)
2
, the
Precision Array for Probing the Epoch of Reionization (PAPER)
3
,
the Hydrogen Epoch of Reionization Array (HERA)
4
and the 21
Centimetre Array (21CMA)
5
. Several arrays targeting intermediate
redshifts are also either operating or under construction, including
the Canadian Hydrogen Intensity Mapping Experiment (CHIME)
6
and Tianlai
7
, and Hydrogen Intensity and Real-Time Analysis ex-
periment (HIRAX)
8
.
While these interferometric experiments have tremendous
sensitivity in principle, the only positive detections of the cos-
mological 21cm signal to date have come from observations by
“single-dish” (or autocorrelation) telescopes. In particular, the
Green Bank Telescope (GBT)
9
and Parkes telescope
10
have pi-
oneered the detection of the Hi signal in cross-correlation with
optical galaxy surveys (Chang et al. 2010; Masui et al. 2013;
Wolz et al. 2017; Anderson et al. 2018; Li et al. 2021; Wolz et al.
2021). While neither Parkes nor GBT were able to definitively de-
tect the auto-power spectrum of the Hi signal (but see Switzer et al.
2013 for upper limits), other single-dish IM projects are planned,
including the purpose-built BINGO
11
instrument, and surveys on
existing or planned telescopes such as the Five-hundred-meter
Aperture Spherical Telescope (FAST)
12
, and the Square Kilometre
Array (SKA)
13
.
In the case of the SKA, an Hi IM survey has been proposed
that would use SKA1-MID in autocorrelation mode. SKA1-MID
is a mid-frequency array of 197 parabolic-dish antennas that will
be built in South Africa, and a large 21cm autocorrelation survey
has been ranked as one of its science priority cases (Santos et al.
2015). While an interferometric intensity mapping survey would
also be possible with SKA1-MID, this is expected to be insuffi-
ciently sensitive to large-scale cosmological features such as the
BAO (Bull et al. 2015).
MeerKAT
14
, as a pathfinder (and, ultimately, a component)
1
http://www.lofar.org/
2
http://www.mwatelescope.org/
3
http://eor.berkeley.edu/
4
http://reionization.org
5
http://21cma.bao.ac.cn
6
https://chime-experiment.ca/en
7
http://tianlai.bao.ac.cn/
8
https://hirax.ukzn.ac.za/
9
https://greenbankobservatory.org/
10
https://www.parkes.atnf.csiro.au/
11
http://www.bingotelescope.org/en/
12
https://fast.bao.ac.cn/
13
https://www.skatelescope.org/
14
https://www.sarao.ac.za/
of SKA1-MID, will play a crucial role in the development of a suc-
cessful multi-dish autocorrelation survey method. In this work, we
address the most critical calibration and systematics issues that are
expected to affect dual-polarisation autocorrelation observations
with MeerKAT. With SKA1 scheduled to begin full operations
in 2028, there is a good window of opportunity for MeerKAT to
robustly demonstrate the single-dish survey method and make a sig-
nificant first detection before the SKA comes online (Santos et al.
2017).
This paper is organized as follows:
Sect. 2 describes the basic setup of our MeerKAT pilot survey.
Sect. 3 gives a detailed overview of our calibration pipeline.
Sect. 4 explains how the calibrated data is combined into multi-
frequency maps.
Sect. 5 studies the characteristics of the calibrated autocorrelation
data, and presents a variety of quality and consistency checks on
the calibrated data.
Sect. 6 presents a variety of comparisons with previous observa-
tions of diffuse Galactic emission and point sources.
Sect. 7 presents our conclusions.
Appendix A presents expressions for weighting the flux from point
sources.
2 DESCRIPTION OF THE SURVEY DATA
In this Section, we review the basic parameters of our survey.
2.1 Target field
To check our observation and calibration strategy, we proposed a
pilot survey through the MeerKAT open time call. The observation
was done in the L-band and targeted a single patch of about 200
deg
2
. The main factors affecting the choice of the sky region were
to avoid the strong galactic emission and to pick an area with good
spectroscopic coverage in the redshifts we are observing (z < 0.5).
This spectroscopic coverage will ultimately allow us to attempt a
detection of the cross-correlation power spectrum between the Hi
IM signal and optical galaxy surveys, which is easier to achieve
than a direct autocorrelation detection of thecosmological Hi signal
because residual foregrounds and other systematics that bias the
autocorrelation drop out in cross-correlation with optical surveys
as they are uncorrelated, (e.g., see Chang et al. 2010; Masui et al.
2013; Wolz et al. 2017; Anderson et al. 2018).
As in Masui et al. (2013), we set our scans to cover the 11hr
field of the WiggleZ Dark Energy Survey (Drinkwater et al. 2010),
which is a large-scale spectroscopic survey of emission-line galax-
ies selected from UV and optical imaging. The footprint of our
observation and the WiggleZ coverage is shown in Figure 1. Note
that this area is also covered by Baryon Oscillation Spectroscopic
Survey (BOSS) (Reid et al. 2016). Our observations took place
between February and July 2019, when the WiggleZ 11h field
appeared with high elevation (> 40
◦
) during the night. The to-
tal usable observation time was about 10.5 hours, which corre-
sponds to around 630 hours in total (before any flagging) when
combined over 60 dishes on average. This can be compared to
Masui et al. (2013), which covered the same region with a total
area of ∼ 41 deg
2
and 190 hrs of total integration time, although
at lower frequencies (700-900 MHz) and higher angular resolution
due to the large (∼100m) diameter of the GBT dish.
2.2 Scanning Strategy
Observations with single dishes typically require a scanning strat-
egy where the dishes are rapidly moved across the sky. The aim is
Downloaded from https://academic.oup.com/mnras/advance-article/doi/10.1093/mnras/stab1365/6276733 by University of Manchester user on 28 May 2021
ORIGINAL UNEDITED MANUSCRIPT
MeerKAT multi-dish autocorrelation calibration 3
145150155160165170175180
R.A.(J2000) [
◦
]
−2
0
2
4
6
8
10
Dec.(J2000) [
◦
]
MeerKAT WiggleZ 11hr WiggleZ 9hr
Figure 1. The footprint of the MeerKAT survey field and the WiggleZ survey fields.
0 1000 2000 3000 4000 5000 6000 7000
60
40
20
0
az (deg)
track-I
track-IIscan
azimuth and elevation against time
0 1000 2000 3000 4000 5000 6000 7000
time (s)
40
45
50
55
el (deg)
track-I
track-IIscan
Figure 2. An example of the azimuth and elevation against time during
a single scan for MeerKAT dish m000 in obs190225. Blue denotes the
scanning of the target field, orange denotes the tracking of the calibrator
source 3C 273.
to cover the relevant angular scales within the stability time scale
of the instrument, when the gains are approximately constant (set
by the so-called 1/f noise knee, e.g., Harper et al. 2018). This re-
quired the development of a new observing mode not previously
available with MeerKAT. The MeerKAT antennas were set to scan
in azimuth at constant elevation (see an example in Figure 2) to
minimise fluctuations of ground spill and airmass.
The telescope scan speed was set to 5 arcmin/s along azimuth,
corresponding to a projected speed on the sky of 5 cos(el) arcmin/s
(where el is the elevation angle). This ensures that the telescope
pointing does not move significantly compared to the width of
the primary beam (∼ 1 deg) during a single time dump. With a
time resolution of 2 sec, this gives a scan speed of no more than
10 arcmin per time sample, which is well within the beam size.
The system-induced 1/ f -type variations for the MeerKAT receiver
are well under the thermal noise fluctuations over ∼ 100 second
time scales after appropriate frequency filtering is performed (see
Li et al. 2020 for details). For a scan speed of 5 arcmin/s, we can
scan about 10 deg in 100 seconds while retaining gain stability.
In order to maintain gain stability over longer time scales, noise
diodes attached to each receiver were fired for 1.8 s once every
20 s during the observation to provide a relative time-ordered data
(TOD) calibration reference. The 1.8 s duration was chosen so that
we can clearly see when the noise diode overlaps two time dumps,
145150155160165170175180
R.A. (J2000)
0
2
4
6
8
Dec. (J2000)
rising
setting
Figure 3. Coverage map of the cross routes for the field rising (blue) and
setting (orange) scans in the same night.
since we are currently unable to synchronise the noise diode fires
with the correlator dumps (see Section 3.6).
The dishes are moved back and forth with a slew of 18 deg
in each direction, corresponding to an observing time of about
200 s per stripe. At fixed elevation, two scans can be performed of
each stripe per night, corresponding to when the field is rising and
setting respectively. The two scans will cross each other as shown
in Figure 3, to achieve good sky coverage in the region of overlap.
The duration of each set of scans is about 1.5 hours, and before
and after each scan we spent about 15 minutes tracking a nearby
celestial point source to use as a bandpass calibrator and absolute
flux calibrator. The technical details of the observations are listed
in Table 1. Not all planned observations were successful due to
equipment issues. After basic data quality checks, we were able
to retrieve seven blocks of data that were suitable for further data
analysis. These blocks are labelled by their observation start Unix
time, as listed in Table 2.
3 DATA REDUCTION PIPELINE
The data reduction pipeline for the time-ordered data includes steps
for flagging of human-made radio frequency interference (RFI),
bandpass and absolute calibration using a known point source, and
calibration of receiver gain fluctuations based on interleaved signal
injection from a noise diode (map making is described in section
Section 4). Figure 4 shows the flowchart of the pipeline, with an
indication of the different stages of calibrated data and a reference
to the sections in the paper where they are described.
Downloaded from https://academic.oup.com/mnras/advance-article/doi/10.1093/mnras/stab1365/6276733 by University of Manchester user on 28 May 2021
ORIGINAL UNEDITED MANUSCRIPT
4 Wang et al.
Table 1. Specifications of the MeerKAT Observations
Antennas All 64 MeerKAT dishes
Observation mode Single-dish
Polarisation Linear (horiz. + vert.) feeds
System temperature ∼16 K
Target field WiggleZ 11hr field (10
◦
× 30
◦
)
Frequency range 856-1712 MHz
Frequency resolution 0.2 MHz
Number of channels 4096
Time resolution 2s
Exposure time 1.5hr x 7 scans
Extent of azimuth scan 18 deg
Scan speed (along azimuth) 5 arcmin/s
Diode injection Pattern mode (1.8s per 20s)
Level 1: Strong
RFI flagged data
Raw signal
Track part
Sec 3.9.1 Overall
bandpass and absolute
flux scale calibration
Scan part
Level 3: Calibrated scans
Sec 3.9.2 Noise
diode-based calibration
of the scanning data
Level 2: Tdiode result
Sec 4.2 From
polarisation to intensity
Level 4: Calibrated
intensity scans
Sec 4.3 Map
pixelisation
Level 5: Pixelised
intensity cube
(per dish and scan)
Level 6: Final
data cube
Sec 4.1 Per-channel outlier
removal for RFI (Round 2)
Sec 4.4 Removal of
low-level RFI features in
the maps (Round 3)
Sec 3.1 Strong RFI
flagging (Round 1)
Figure 4. Flowchart showing each step in the calibration and map-making
pipeline. The numbers of the sections from this paper where each step is
described are also shown.
3.1 Strong RFI flagging (Round 1)
The RFI from modern telecommunications and satellites is an
important contaminant of radio observations (Harper & Dickinson
2018). In Figure 5 weshow the time-averaged raw signal in atypical
observation. Several RFI groups can be seen clearly. MeerKAT
has RFI mitigation systems to prevent RFI before and during the
observation (Jonas & MeerKAT Team 2016). However, as even the
best RFI mitigation methods cannot completely prevent all RFI
(Baan 2010), we must employ methods to reduce the effect of RFI
after observation. One method is to flag those parts of the spectrum
which are dominated by RFI and not use those frequencies/times
in scientific data analysis. In this work we use the RFI package of
Signal Extraction and Emission Kartographer (SEEK; Akeret et al.
2017), which follows the SumThreshold algorithm (Offringa et al.
2010).
We denote the raw input data as
ˆ
T
raw
(t, ν), where the hat indi-
cates data in arbitrary correlator units. We group the data in time
according to the diode status (t
0
nd
, t
1a
nd
, and t
1b
nd
, corresponding to
diode off, diode on/starting, and diode on/stopping; see Section 3.6
for more details), as well as its scan state (whether tracking a cal-
ibrator source or scanning across the survey field). The SEEK
900 1000 1100 1200 1300 1400 1500 1600 1700
Frequency (MHz)
10
1
10
0
10
1
10
2
raw signal
obs190224
obs190225
obs190330
obs190401
obs190423
obs190424
obs190711
00.10.20.30.40.50.6
redshift
Figure 5. The normalised time-averaged raw signal for m010h, all seven
observations. The clusters of bright spike features correspond to RFI-
dominated bands. Unshaded regions are the frequency bands for further
data analysis.
flagging algorithm is then applied to these six groups of raw data
separately, followed by an additional filter that discards any fre-
quency channels or time dumps where > 70% of the data are
flagged.
This strong RFI flagging progress is applied to the data in
two steps: First, on all of the raw data directly, to flag the strongly
RFI-contaminated channels that are occupied by satellite commu-
nications; and second, focusing only on the target frequency bands,
971–1075 MHz (channels 550–1050) and 1305–1504 MHz (chan-
nels 2150-3100). In the second step, the value of
ˆ
T
raw
at each
frequency is normalised by the mean value along the time direc-
tion to account for the shape of the bandpass. At the same time,
the locations of bright point sources are protected by temporary
masks, as otherwise they would show as transient peaks in the time
stream as the dishes scan across the sky, potentially leading to them
being erroneously flagged.
This two-step SEEK flagging is applied to the per-dish HH
and VV raw polarisation signals separately. The union of the flags
from the two polarisations is then adopted as the primary set of
flags. Typically, about 35% of the raw data are flagged in this step.
Note that additional rounds of RFI flagging are performed during
and after the calibration to prevent contamination from weaker
RFI signals that can only be detected after further processing (see
Section 4.4).
In the following calibration process, we focus on the 971-1075
MHz (channels 550-1050) and 1305-1504 MHz (channels 2150-
3100) bands, considering the RFI distribution and the bandpass
flatness. In these frequency ranges, most of the satellite communi-
cations are avoided, with only a few strong RFIs and some weak
RFIs appearing intermittently.
3.2 Calibration model and strategy
Our autocorrelation calibration strategy is based on propagating
an absolute flux and bandpass calibration obtained from tracking
observations of bright point celestial point sources through to scan-
ning observations of the survey field by using periodic noise diode
fires as a relative calibration reference. We use a Bayesian approach
to jointly fit the free parameters in our calibration and sky models,
as described in the next section.
Bright celestial sources (e.g., AGN, supernova remnants) are
commonly taken as bandpass calibrators and absolute flux cali-
brators for single-dish observations. Despite being resolved in in-
Downloaded from https://academic.oup.com/mnras/advance-article/doi/10.1093/mnras/stab1365/6276733 by University of Manchester user on 28 May 2021
