Correlation-based Beam Calibration of 21cm Intensity Mapping (2024)

Jiacheng DingSchool of Physics and Astronomy, Sun Yat-Sen University,
No.2 Daxue Road, Zhuhai 519082, China
Xin WangSchool of Physics and Astronomy, Sun Yat-Sen University,
No.2 Daxue Road, Zhuhai 519082, China
CSST Science Center for the Guangdong-Hong Kong-Macau Greater Bay Area, SYSU, China
Ue-Li PenInstitute of Astronomy and Astrophysics, Academia Sinica, Astronomy-Mathematics Building,
No.1, Sec.4, Roosevelt Road, Taipei 10617, Taiwan
Canadian Institute for Theoretical Astrophysics, 60 St. George Street, Toronto, ON M5S 3H8, CanadaCanadian Institute for Advanced Research, 180 Dundas St West, Toronto, ON M5G 1Z8, CanadaDunlap Institute for Astronomy and Astrophysics, University of Toronto,
50 St George Street, Toronto, ON M5S 3H4, Canada
Perimeter Institute of Theoretical Physics, 31 Caroline Street North, Waterloo, ON N2L 2Y5, Canada
Xiao-Dong LiSchool of Physics and Astronomy, Sun Yat-Sen University,
No.2 Daxue Road, Zhuhai 519082, China
CSST Science Center for the Guangdong-Hong Kong-Macau Greater Bay Area, SYSU, China

Abstract

Foreground removal presents a significant obstacle in both current and forthcoming intensity mapping surveys. While numerous techniques have been developed that show promise in simulated datasets, their efficacy often diminishes when applied to real-world data. A primary issue is the frequency-dependent variations in the instrumental response. In this paper, we propose a novel approach utilizing the internal cross-correlation among different frequencies to calibrate the beam’s frequency fluctuations. Using a simulated dataset that incorporates frequency-dependent random fluctuations into the beam model, we illustrate that our method can achieve considerable improvements over traditional techniques. Our results represent a step forward in enhancing the precision and reliability of foreground removal in intensity mapping surveys.

cosmology, 21cm intensity mapping

software: CAMB(Lewis etal., 2000),NumPy(Van DerWalt etal., 2011; Harris etal., 2020),SciPy(Virtanen etal., 2020),Astropy(AstropyCollaboration etal., 2013, 2018, 2022),HEALPix/healpy(Górski etal., 2005; Zonca etal., 2019)mpi4py(Dalcin & Fang, 2021),Matplotlib(Hunter, 2007),OpenMPI(Gabriel etal., 2004)

1 Introduction

The large-scale structure of the Universe holds invaluable information, providing insights into various aspects such as galaxy evolution, fundamental physics including dark matter, dark energy, alternative theories of gravity, and the early Universe, etc. For over four decades, galaxy surveys have played an important role in enhancing our understanding of the Universe, including the 2dF Galaxy Redshift Survey111http://www.2dfgrs.net/ Colless etal. (2001), the 6dF Galaxy Survey222http://www.6dfgs.net/Jones etal. (2004), the WiggleZ Dark Energy Survey333https://wigglez.swin.edu.au/ Drinkwater etal. (2010), and SDSS-III BOSS444https://www.sdss3.org/Anderson etal. (2014), DES555https://www.darkenergysurvey.org/Collaboration: etal. (2016), DESI666https://www.desi.lbl.gov/Collaboration etal. (2023), and LSST777https://www.lsst.org/Ivezić etal. (2019) in the future. In recent years, 21cm intensity mapping has emerged as a alternative technique to probe the large-scale structure (Scott & Rees, 1990). This method, focusing on the 21cm line emission from neutral hydrogen (HI), differs from conventional galaxy surveys that observe individual galaxies. Instead, 21cm intensity mapping captures the collective emission from multiple galaxies, mapping the intensity fluctuation without resolving individual galaxies. This efficient approach enables us to rapidly map matter distribution over large volumes of Universe. Significant achievements have been made in this area over the past decade. These include the successful cross-correlation of 21cm maps with galaxy samples(Masui etal., 2013; Wolz etal., 2022; Cunnington etal., 2023) and the measurement of the auto-power spectrum using interferometer(Abdurashidova etal., 2022; Paul etal., 2023). Consequently, numerous ongoing surveys and forthcoming next-generation radio telescopes (SKA888https://www.skao.int/en/, HERA999https://reionization.org/(DeBoer etal., 2017)) are poised to significantly enhance the detection capabilities, and will eventually lead to cosmological constraints using this technique.

Despite these advancements, significant challenges remain, particularly with foreground contamination. Being several orders of magnitude brighter than the cosmological signal, these foregrounds substantially hinder progress in auto-correlation measurements, especially when compared to cross-correlation with galaxy surveys – a milestone achieved over a decade ago. Recent breakthrough in detecting the auto-power spectrum using the MeerKAT interferometry mode and foreground avoidance techniques(Paul etal., 2023) marks significant progress. However, the long baselines of the MeerKAT telescope are confined to observing only smaller scales, which limits their ability in deriving meaningful cosmological constraints.

Many different techniques have been devised to overcome this challenge, including methods like spectral fitting, Principal Component Analysis (PCA) (Masui etal., 2013; Switzer etal., 2013, 2015; Bigot-Sazy etal., 2015), Independent Component Analysis (ICA) (Chapman etal., 2012; Wolz etal., 2014), and the Karhunen-Loève transform (KL) (Bégin etal., 2022) and other machine learning based methods (Mangena etal., 2020; Makinen etal., 2021; Wadekar etal., 2021; Villanueva-Domingo &Villaescusa-Navarro, 2021) etc. However, while many of these methods demonstrate reasonable effectiveness in extracting signals from idealized, simulated datasets, real-world complexities often hinder their practical application.

One complexity arises from the chromaticity of the instrument. In interferometric observations, the chromatic response causes a mixing of low-wavelength foreground modes with small-scale HI signal within a wedge-like region, further restricting the Fourier space where foreground avoidance techniques can be applied. While this effect is less a issue in telescopes operating in single-dish mode, chromaticity can still be a significant factor. This is largely due to the inherent imperfections in any instrument we construct, which introduce a multiplicative, frequency-dependent noise into the response beam. The convolution of this noisy beam with the observed sky can significantly degrade the effectiveness of most, if not all, existing foreground removal methods. This includes techniques ranging from blind PCA/ICA and parametric fitting to others that necessitate knowledge of the instrumental response, such as the Karhunen-Loève m-modes and various machine learning-based methods.

At the first glance, the most straightforward solution to mitigate beam variation appears to be simply improving the calibration or measurement accuracy of the beam. Indeed, this approach has been widely pursued, with significant efforts dedicated to improving beam understanding through methods such as drone measurement (Chang etal., 2015) and holographic measurements (Hunter etal., 2011; Iheanetu etal., 2019). In this paper, however, we introduce a novel approach to address this issue. Our core concept is that it is the relative beam differences across frequencies that primarily influence the success of foreground removal. An unknown, frequency-independent deviation in the beam only introduces an additional window function to the detected signal, with minimal impact on the efficiency of the foreground removal process. Therefore, we propose a Correlation-Based Beam Calibration method (CBC), designed to internally calibrate the frequency variations among different frequency channels.

The structure of this paper is as follows: Section II provides a concise overview of our method, outlining the fundamental concept. Section III delves into the detailed methodology. Section IV describes the process of generating the simulated data, followed by the presentation of our results in Section V. The paper concludes with a discussion in Section VI. In this paper, we assume a fiducial cosmology with parameters of (Ωm=0.3subscriptΩm0.3\Omega_{\rm m}\!=\!0.3roman_Ω start_POSTSUBSCRIPT roman_m end_POSTSUBSCRIPT = 0.3, Ωb=0.049subscriptΩb0.049\Omega_{\rm b}\!=\!0.049roman_Ω start_POSTSUBSCRIPT roman_b end_POSTSUBSCRIPT = 0.049, ΩΛ=0.7subscriptΩΛ0.7\Omega_{\Lambda}\!=\!0.7roman_Ω start_POSTSUBSCRIPT roman_Λ end_POSTSUBSCRIPT = 0.7, w=1.0𝑤1.0w\!=\!-1.0italic_w = - 1.0, σ8=0.8subscript𝜎80.8\sigma_{8}\!=\!0.8italic_σ start_POSTSUBSCRIPT 8 end_POSTSUBSCRIPT = 0.8, h=0.670.67h\!=\!0.67italic_h = 0.67, ns=0.96subscript𝑛𝑠0.96n_{s}=0.96italic_n start_POSTSUBSCRIPT italic_s end_POSTSUBSCRIPT = 0.96)

2 Instrumental Chromaticity and Foreground Removal

To demonstrate the fundamental principle of our method, let us begin by presuming that the observed data have undergone processing up to the map-making stage, yielding an intensity map. This map, represented as the temperature Tν(𝐧^)subscript𝑇𝜈^𝐧T_{\nu}(\mathbf{\hat{n}})italic_T start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ( over^ start_ARG bold_n end_ARG ), is a function of the line-of-sight (LoS) direction 𝐧^^𝐧\mathbf{\hat{n}}over^ start_ARG bold_n end_ARG and the frequency ν𝜈\nuitalic_ν, which is composed of foregrounds and cosmological signal

Tν(𝐧^)=TνHI(𝐧^)+TνFG(𝐧^).subscript𝑇𝜈^𝐧subscriptsuperscript𝑇HI𝜈^𝐧subscriptsuperscript𝑇FG𝜈^𝐧\displaystyle T_{\nu}(\mathbf{\hat{n}})=T^{\rm HI}_{\nu}(\mathbf{\hat{n}})+T^{%\rm FG}_{\nu}(\mathbf{\hat{n}}).italic_T start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ( over^ start_ARG bold_n end_ARG ) = italic_T start_POSTSUPERSCRIPT roman_HI end_POSTSUPERSCRIPT start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ( over^ start_ARG bold_n end_ARG ) + italic_T start_POSTSUPERSCRIPT roman_FG end_POSTSUPERSCRIPT start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ( over^ start_ARG bold_n end_ARG ) .(1)

To proceed, we make the basic assumption that the observed sky map is a convolution of the true sky Tν(𝐧^)subscript𝑇𝜈^𝐧T_{\nu}(\mathbf{\hat{n}})italic_T start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ( over^ start_ARG bold_n end_ARG ) with a frequency-dependent power beam Bν(𝐧^)subscript𝐵𝜈^𝐧B_{\nu}(\mathbf{\hat{n}})italic_B start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ( over^ start_ARG bold_n end_ARG )

T~ν(𝐧^)=𝑑𝐧^Bν(𝐧^𝐧^)Tν(𝐧^)+nν(𝐧^),subscript~𝑇𝜈^𝐧differential-dsuperscript^𝐧subscript𝐵𝜈^𝐧superscript^𝐧subscript𝑇𝜈superscript^𝐧subscript𝑛𝜈^𝐧\displaystyle\widetilde{T}_{\nu}(\mathbf{\hat{n}})=\int d\mathbf{\hat{n}}^{%\prime}B_{\nu}(\mathbf{\hat{n}}-\mathbf{\hat{n}}^{\prime})T_{\nu}(\mathbf{\hat%{n}}^{\prime})+n_{\nu}(\mathbf{\hat{n}}),over~ start_ARG italic_T end_ARG start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ( over^ start_ARG bold_n end_ARG ) = ∫ italic_d over^ start_ARG bold_n end_ARG start_POSTSUPERSCRIPT ′ end_POSTSUPERSCRIPT italic_B start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ( over^ start_ARG bold_n end_ARG - over^ start_ARG bold_n end_ARG start_POSTSUPERSCRIPT ′ end_POSTSUPERSCRIPT ) italic_T start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ( over^ start_ARG bold_n end_ARG start_POSTSUPERSCRIPT ′ end_POSTSUPERSCRIPT ) + italic_n start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ( over^ start_ARG bold_n end_ARG ) ,(2)

where T~ν(𝐧^)subscript~𝑇𝜈^𝐧\widetilde{T}_{\nu}(\mathbf{\hat{n}})over~ start_ARG italic_T end_ARG start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ( over^ start_ARG bold_n end_ARG ) represents the observed sky temperature, with the tilde indicating the quantity as post-instrumentally ‘observed’.

As will be seen, this assumption is essential here, as our methodology relies on filtering within the Fourier space. For many radio telescopes, this assumption holds reasonably true, although the observational strategy or subsequent data analysis procedures might introduce some position-dependent response functions. A case in point is the 19-beam L-band receivers of the Five-hundred-meter Aperture Spherical radio Telescope (FAST). In its drift scan mode (Zhang etal., 2019; Hu etal., 2021; Li etal., 2023), disparate sky regions may be observed by different subsets of these 19 beams. Consequently, without proper compensation during the map-making process, any disparities among the receivers might lead to a violation of our assumption.

Correlation-based Beam Calibration of 21cm Intensity Mapping (1)

In the Fourier space, Equation (2) is expressed as

T~ν(𝐮)=Bν(𝐮)Tν(𝐮)+nν(𝐮),subscript~𝑇𝜈𝐮subscript𝐵𝜈𝐮subscript𝑇𝜈𝐮subscript𝑛𝜈𝐮\displaystyle\widetilde{T}_{\nu}(\mathbf{u})=B_{\nu}(\mathbf{u})\,T_{\nu}(%\mathbf{u})+n_{\nu}(\mathbf{u}),over~ start_ARG italic_T end_ARG start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ( bold_u ) = italic_B start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ( bold_u ) italic_T start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ( bold_u ) + italic_n start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ( bold_u ) ,(3)

where 𝐮=ν𝐛/c𝐮𝜈𝐛𝑐\mathbf{u}=\nu\mathbf{b}/cbold_u = italic_ν bold_b / italic_c is the Fourier conjugate of sky coordinates 𝐧^^𝐧\mathbf{\hat{n}}over^ start_ARG bold_n end_ARG. Here c𝑐citalic_c is the speed of light and 𝐛𝐛\mathbf{b}bold_b is the spatial separation between elements of the instrument. In an ideal radio telescope observation, the beam function scales linearly with frequency. In Fourier space, this implies that the dependency on ν𝜈\nuitalic_ν is expressed through the relation

𝐮=ν𝐛/c.𝐮𝜈𝐛𝑐\displaystyle\mathbf{u}=\nu\mathbf{b}/c.bold_u = italic_ν bold_b / italic_c .(4)

In reality, no instrument is flawless, and these imperfections could significantly influence the performance of foreground removal techniques. Many existing methods for foreground subtraction rely on having at least some knowledge of the beam characteristics. An imperfect or noisy telescope beam can compromise the accuracy of these techniques. Here, the ‘imperfect beam’ refers to any deviation from the ideal antenna response pattern, which can stem from a range of issues including manufacturing inaccuracies, mechanical structural deformations of the antenna, errors in feed positioning, and frequency-dependent responses of electronic components, etc. Environmental conditions such as temperature fluctuations and wind can induce additional deformations. While these factors indeed influence the beam, their temporal variability also challenges our assumption of a consistent convolution process. As such, they fall outside the scope of this paper.

To gain a clearer understanding of how frequency-dependent variations in the beam response can cause distortion and potential intermixing of foreground and cosmological signals, it is instructive to examine specific cleaning techniques. For our analysis, existing foreground removal methods can be divided into two categories. The first category includes methods that explicitly require knowledge of the beam function. This group encompasses forward-modeling methods (Bernardi etal., 2011; Sullivan etal., 2012), the m-mode formalism which relies on the input beam function to construct a signal-to-contaminant-ratio ordered Karhunen–Loève basis (Shaw etal., 2014), and various semi-blind methods that assume prior knowledge of both the foreground and the instrument (Zuo etal., 2023). Additionally, this category includes machine learning-based approaches (Gillet etal., 2019; LaPlante & Ntampaka, 2019; Makinen etal., 2021; Ni etal., 2022; Gheller & Vazza, 2022; Shi etal., 2023). For these methods, a precise understanding of the instrumental response is essential. The second category includes methods that do not explicitly require an instrumental response. However, they can still be significantly affected by a noisy beam. For instance, some techniques that rely on the spectrum’s smoothness, such as parametric and certain non-parametric fitting methods, are particularly vulnerable. A noisy beam disrupts the underlying assumption of smoothness, rendering these techniques less effective or inapplicable.

Another type of method within this category is the blind methods, notably PCA and ICA, as detailed in various studies (Masui etal., 2013; Alonso etal., 2015; Bigot-Sazy etal., 2015; Stone, 2002; Alonso etal., 2015; Tharwat, 2021). These techniques are generally more adept at handling instrumental systematics than parametric fitting approaches and are widely employed in real data analysis, making them the primary focus of our paper. In these methods, the data cube 𝐃=𝐅+𝐒+𝐍𝐃𝐅𝐒𝐍\mathbf{D}=\mathbf{F}+\mathbf{S}+\mathbf{N}bold_D = bold_F + bold_S + bold_N is interpreted as a combination of the foregrounds 𝐅𝐅\mathbf{F}bold_F, the 21 cm signal 𝐒𝐒\mathbf{S}bold_S, and instrumental noise 𝐍𝐍\mathbf{N}bold_N. Here, 𝐃𝐃\mathbf{D}bold_D is expressed as a two-dimensional matrix in the notation n×psuperscript𝑛𝑝\mathbb{R}^{n\times p}blackboard_R start_POSTSUPERSCRIPT italic_n × italic_p end_POSTSUPERSCRIPT, where n𝑛nitalic_n is the number of frequency channels and p𝑝pitalic_p the number of pixels. The goal is to derive a cleaned map 𝐓clsuperscript𝐓cl\mathbf{T}^{\rm cl}bold_T start_POSTSUPERSCRIPT roman_cl end_POSTSUPERSCRIPT by excluding the foreground elements from the observed data. This is achieved as follows:

𝐓cl=𝐃(𝐔𝐈m𝐔T)𝐃,superscript𝐓cl𝐃subscript𝐔𝐈msuperscript𝐔T𝐃\displaystyle\mathbf{T}^{\rm cl}=\mathbf{D}-\left(\mathbf{U}\mathbf{I}_{\rm m}%\mathbf{U}^{\rm T}\right)\mathbf{D},bold_T start_POSTSUPERSCRIPT roman_cl end_POSTSUPERSCRIPT = bold_D - ( bold_UI start_POSTSUBSCRIPT roman_m end_POSTSUBSCRIPT bold_U start_POSTSUPERSCRIPT roman_T end_POSTSUPERSCRIPT ) bold_D ,(5)

where 𝐔𝐔\mathbf{U}bold_U represents the components matrix, which could be either the most dominant modes (PCA) or the most non-Gaussian modes (ICA). The first mm\rm mroman_m-components are selected with matrix 𝐈msubscript𝐈m\mathbf{I}_{\rm m}bold_I start_POSTSUBSCRIPT roman_m end_POSTSUBSCRIPT, signifying approximate foregrounds modeling as 𝐅(𝐔𝐈m𝐔T)𝐃𝐅subscript𝐔𝐈msuperscript𝐔T𝐃\mathbf{F}\approx\left(\mathbf{U}\mathbf{I}_{\rm m}\mathbf{U}^{\rm T}\right)%\mathbf{D}bold_F ≈ ( bold_UI start_POSTSUBSCRIPT roman_m end_POSTSUBSCRIPT bold_U start_POSTSUPERSCRIPT roman_T end_POSTSUPERSCRIPT ) bold_D. However, chromatic fluctuations in the instrument may modify the decomposed modes, leading to a confusion between the 21cm signal and beam fluctuations, thereby reducing the efficacy of these blind methods. To illustrate this, consider a simplified noisy beam model. In this model, three-dimensional random noise δBν(𝐮)𝛿subscript𝐵𝜈𝐮\delta B_{\nu}(\mathbf{u})italic_δ italic_B start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ( bold_u ) is added in the frequency-𝐮𝐮\mathbf{u}bold_u space to a Gaussian beam BνG(𝐮)subscriptsuperscript𝐵𝐺𝜈𝐮B^{G}_{\nu}(\mathbf{u})italic_B start_POSTSUPERSCRIPT italic_G end_POSTSUPERSCRIPT start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ( bold_u ):

Bν(𝐮)subscript𝐵𝜈𝐮\displaystyle B_{\nu}(\mathbf{u})italic_B start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ( bold_u )=\displaystyle==BνG(𝐮)+δBν(𝐮)=BνG(𝐮)(1+ϵν(𝐮))subscriptsuperscript𝐵𝐺𝜈𝐮𝛿subscript𝐵𝜈𝐮subscriptsuperscript𝐵𝐺𝜈𝐮1subscriptitalic-ϵ𝜈𝐮\displaystyle B^{G}_{\nu}(\mathbf{u})+\delta B_{\nu}(\mathbf{u})=B^{G}_{\nu}(%\mathbf{u})(1+\epsilon_{\nu}(\mathbf{u}))italic_B start_POSTSUPERSCRIPT italic_G end_POSTSUPERSCRIPT start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ( bold_u ) + italic_δ italic_B start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ( bold_u ) = italic_B start_POSTSUPERSCRIPT italic_G end_POSTSUPERSCRIPT start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ( bold_u ) ( 1 + italic_ϵ start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ( bold_u ) )(6)

where ϵν(𝐮)subscriptitalic-ϵ𝜈𝐮\epsilon_{\nu}(\mathbf{u})italic_ϵ start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ( bold_u ) represents a uniformly distributed random number. The specifics of this model are further elaborated in section 4.2.

In Figure (1), we illustrate the impact of a noisy beam on PCA foreground removal. The upper panel illustrates the eigenvalues resulting from the SVD decomposition, where a rapid decrease in eigenvalues indicates that the spectral-smooth foreground can be effectively captured by the initial few modes. As shown, with a frequency-independent beam (marked by red squares), the first four eigen-modes sufficiently describe the foregrounds over six orders of magnitude. This sets the baseline result for the most ideal scenario. Similarly, a frequency rescaling of the Gaussian beam (see Eq. 4) does not significantly alter the results (shown as red diamons). More crucially, a frequency-independent noisy beam, where δBν(𝐮)𝛿subscript𝐵𝜈𝐮\delta B_{\nu}(\mathbf{u})italic_δ italic_B start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ( bold_u ) varies only in the uv𝑢𝑣uvitalic_u italic_v plane but not in the frequency direction (indicated by blue crosses), aligns closely with our ideal beam model. This observation is further supported in the lower panel, which presents the cross-correlation ratio between the true cosmological signal and the ‘cleaned’ map after the subtraction of three SVD modes.

Conversely, with a increasing level of the frequency-dependent beam noise ϵνsubscriptitalic-ϵ𝜈\epsilon_{\nu}italic_ϵ start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT, the eigenvalues descend more gradually, indicating a more complex data structure. As a result, more modes must be removed to reduce the map amplitude, leading to a more significant loss of the signal. At the same time, the cross-correlation ratio between the cleaned map (with three subtracted SVD modes) and the HI map starts to decline, a consequence of increased residual power. Notably, even a beam noise level of 0.1%percent0.10.1\%0.1 % can substantially reduce the cross-correlation ratio, dropping it from near 1111 to about 0.60.60.60.6. This highlights the substantial impact of beam noise on effective foreground removal.

3 Correlation-based Beam Calibration

The above discussion clearly demonstrates that a frequency-independent fluctuations in the beam pattern have a minimal impact on foreground cleaning. This observation highlights the importance of understanding the relative beam variation across frequencies, rather than focusing solely on accurately measuring its absolute pattern. We would like to remind readers that this concept is not entirely new. The traditional SVD method (denoted as ‘TM’) has previously sought to convolve the observed dataset with a common beam function, albeit under the assumption that the beam is well described by a Gaussian function (Masui etal., 2013)

fνG(𝐮)subscriptsuperscript𝑓G𝜈𝐮\displaystyle f^{\rm G}_{\nu}(\mathbf{u})italic_f start_POSTSUPERSCRIPT roman_G end_POSTSUPERSCRIPT start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ( bold_u )=\displaystyle==exp[(γσref2σν2)u22]𝛾subscriptsuperscript𝜎2refsubscriptsuperscript𝜎2𝜈superscript𝑢22\displaystyle\;\exp\left[-\frac{(\gamma\sigma^{2}_{\rm ref}-\sigma^{2}_{\nu})u%^{2}}{2}\right]roman_exp [ - divide start_ARG ( italic_γ italic_σ start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT start_POSTSUBSCRIPT roman_ref end_POSTSUBSCRIPT - italic_σ start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ) italic_u start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT end_ARG start_ARG 2 end_ARG ](7)
=\displaystyle==BνrefGBνGexp[(γ1)σref2u22].subscriptsuperscript𝐵𝐺subscript𝜈refsubscriptsuperscript𝐵𝐺𝜈𝛾1superscriptsubscript𝜎ref2superscript𝑢22\displaystyle\frac{B^{G}_{\nu_{\rm ref}}}{B^{G}_{\nu}}\exp\left[-\frac{(\gamma%-1)\sigma_{\rm ref}^{2}u^{2}}{2}\right].divide start_ARG italic_B start_POSTSUPERSCRIPT italic_G end_POSTSUPERSCRIPT start_POSTSUBSCRIPT italic_ν start_POSTSUBSCRIPT roman_ref end_POSTSUBSCRIPT end_POSTSUBSCRIPT end_ARG start_ARG italic_B start_POSTSUPERSCRIPT italic_G end_POSTSUPERSCRIPT start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT end_ARG roman_exp [ - divide start_ARG ( italic_γ - 1 ) italic_σ start_POSTSUBSCRIPT roman_ref end_POSTSUBSCRIPT start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT italic_u start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT end_ARG start_ARG 2 end_ARG ] .

Here σ(ν)𝜎𝜈\sigma(\nu)italic_σ ( italic_ν ) represents the size of the Gaussian beam at frequency ν𝜈\nuitalic_ν, with σrefsubscript𝜎ref\sigma_{\rm ref}italic_σ start_POSTSUBSCRIPT roman_ref end_POSTSUBSCRIPT being the beam width at the reference frequency which is also the maximum value of σ(ν)𝜎𝜈\sigma(\nu)italic_σ ( italic_ν ). The scaling factor, γ𝛾\gammaitalic_γ, is selected from a range of values, such as 1.21.21.21.2 in MeerKAT study (Cunnington etal., 2023) or 1.41.41.41.4 in GBT measurement (Condon & Ransom, 2016), to further smooth out any noise arising from this process. It becomes evident that applying the filter fGsuperscript𝑓𝐺f^{G}italic_f start_POSTSUPERSCRIPT italic_G end_POSTSUPERSCRIPT to the observed map results in a map convolved with the same Gaussian beam at all frequencies

Tν(𝐮)subscript𝑇𝜈𝐮\displaystyle{T}_{\nu}(\mathbf{u})italic_T start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ( bold_u )=\displaystyle==fνG(𝐮)T~ν(𝐮)=fνG(𝐮)BνG(𝐮)Tν(𝐮)subscriptsuperscript𝑓𝐺𝜈𝐮subscript~𝑇𝜈𝐮subscriptsuperscript𝑓𝐺𝜈𝐮subscriptsuperscript𝐵𝐺𝜈𝐮subscript𝑇𝜈𝐮\displaystyle f^{G}_{\nu}(\mathbf{u})\widetilde{T}_{\nu}(\mathbf{u})=f^{G}_{%\nu}(\mathbf{u})\,B^{G}_{\nu}(\mathbf{u})\,T_{\nu}(\mathbf{u})italic_f start_POSTSUPERSCRIPT italic_G end_POSTSUPERSCRIPT start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ( bold_u ) over~ start_ARG italic_T end_ARG start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ( bold_u ) = italic_f start_POSTSUPERSCRIPT italic_G end_POSTSUPERSCRIPT start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ( bold_u ) italic_B start_POSTSUPERSCRIPT italic_G end_POSTSUPERSCRIPT start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ( bold_u ) italic_T start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ( bold_u )(8)
=\displaystyle==BνrefGexp[(γ1)σref2u22]Tν(𝐮).subscriptsuperscript𝐵𝐺subscript𝜈ref𝛾1superscriptsubscript𝜎ref2superscript𝑢22subscript𝑇𝜈𝐮\displaystyle B^{G}_{\nu_{\rm ref}}\exp\left[-\frac{(\gamma-1)\sigma_{\rm ref}%^{2}u^{2}}{2}\right]T_{\nu}(\mathbf{u}).italic_B start_POSTSUPERSCRIPT italic_G end_POSTSUPERSCRIPT start_POSTSUBSCRIPT italic_ν start_POSTSUBSCRIPT roman_ref end_POSTSUBSCRIPT end_POSTSUBSCRIPT roman_exp [ - divide start_ARG ( italic_γ - 1 ) italic_σ start_POSTSUBSCRIPT roman_ref end_POSTSUBSCRIPT start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT italic_u start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT end_ARG start_ARG 2 end_ARG ] italic_T start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ( bold_u ) .

Subsequently, in this section, we will expand on this concept and introduce our correlation-based approach, designed to effectively calibrate for the frequency-dependent variations in the beam pattern.

3.1 Frequency-dependent Beam Fluctuation

Generally, a realistic beam function Bν(𝐮)subscript𝐵𝜈𝐮B_{\nu}(\mathbf{u})italic_B start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ( bold_u ) typically has a complex functional form, showing variations in both 𝐮𝐮\mathbf{u}bold_u and ν𝜈\nuitalic_ν. As demonstrated in the previous section, the efficacy of foreground removal depends on the beam’s relative response. This prompts the decomposition of the beam into two distinct components:

Bν(𝐮)=B¯(𝐮)Bνrel(𝐮),subscript𝐵𝜈𝐮¯𝐵𝐮subscriptsuperscript𝐵rel𝜈𝐮\displaystyle B_{\nu}(\mathbf{u})=\bar{B}(\mathbf{u})B^{\rm rel}_{\nu}(\mathbf%{u}),italic_B start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ( bold_u ) = over¯ start_ARG italic_B end_ARG ( bold_u ) italic_B start_POSTSUPERSCRIPT roman_rel end_POSTSUPERSCRIPT start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ( bold_u ) ,(9)

where B¯¯𝐵\bar{B}over¯ start_ARG italic_B end_ARG denotes the frequency-independent part, aligned with the beam at the reference frequency, i.e., B¯=Bνref¯𝐵subscript𝐵subscript𝜈ref\bar{B}=B_{\nu_{\rm ref}}over¯ start_ARG italic_B end_ARG = italic_B start_POSTSUBSCRIPT italic_ν start_POSTSUBSCRIPT roman_ref end_POSTSUBSCRIPT end_POSTSUBSCRIPT. On the other hand, the frequency-dependent relative beam Brelsuperscript𝐵relB^{\rm rel}italic_B start_POSTSUPERSCRIPT roman_rel end_POSTSUPERSCRIPT is defined as Bνrel=Bν/B¯subscriptsuperscript𝐵rel𝜈subscript𝐵𝜈¯𝐵B^{\rm rel}_{\nu}=B_{\nu}/\bar{B}italic_B start_POSTSUPERSCRIPT roman_rel end_POSTSUPERSCRIPT start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT = italic_B start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT / over¯ start_ARG italic_B end_ARG. In the following, the primary objective of our method is to determine the relative beam Bνrelsubscriptsuperscript𝐵rel𝜈B^{\rm rel}_{\nu}italic_B start_POSTSUPERSCRIPT roman_rel end_POSTSUPERSCRIPT start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT with existing data.

To achieve this, we aim to develop a filtering function designed to uniformize the beam response across all frequencies

fνrefν(𝐮)=Bνref(𝐮)Bν(𝐮)=1Bνrel(𝐮)subscript𝑓subscript𝜈ref𝜈𝐮subscript𝐵subscript𝜈ref𝐮subscript𝐵𝜈𝐮1subscriptsuperscript𝐵rel𝜈𝐮\displaystyle f_{\nu_{\rm ref}\nu}(\mathbf{u})=\frac{B_{\nu_{\rm ref}}(\mathbf%{u})}{B_{\nu}(\mathbf{u})}=\frac{1}{B^{\rm rel}_{\nu}(\mathbf{u})}italic_f start_POSTSUBSCRIPT italic_ν start_POSTSUBSCRIPT roman_ref end_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ( bold_u ) = divide start_ARG italic_B start_POSTSUBSCRIPT italic_ν start_POSTSUBSCRIPT roman_ref end_POSTSUBSCRIPT end_POSTSUBSCRIPT ( bold_u ) end_ARG start_ARG italic_B start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ( bold_u ) end_ARG = divide start_ARG 1 end_ARG start_ARG italic_B start_POSTSUPERSCRIPT roman_rel end_POSTSUPERSCRIPT start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ( bold_u ) end_ARG(10)

Consequently, applying this filter fνrefν(𝐮)subscript𝑓subscript𝜈ref𝜈𝐮f_{\nu_{\rm ref}\nu}(\mathbf{u})italic_f start_POSTSUBSCRIPT italic_ν start_POSTSUBSCRIPT roman_ref end_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ( bold_u ) to the observed map T~ν(𝐮)subscript~𝑇𝜈𝐮\widetilde{T}_{\nu}(\mathbf{u})over~ start_ARG italic_T end_ARG start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ( bold_u )

fνrefν(𝐮)T~ν(𝐮)subscript𝑓subscript𝜈ref𝜈𝐮subscript~𝑇𝜈𝐮\displaystyle f_{\nu_{\rm ref}\nu}(\mathbf{u})\,\widetilde{T}_{\nu}(\mathbf{u})italic_f start_POSTSUBSCRIPT italic_ν start_POSTSUBSCRIPT roman_ref end_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ( bold_u ) over~ start_ARG italic_T end_ARG start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ( bold_u )=\displaystyle==fνrefν(𝐮)Bν(𝐮)Tν(𝐮)subscript𝑓subscript𝜈ref𝜈𝐮subscript𝐵𝜈𝐮subscript𝑇𝜈𝐮\displaystyle f_{\nu_{\rm ref}\nu}(\mathbf{u})\,B_{\nu}(\mathbf{u})\,T_{\nu}(%\mathbf{u})italic_f start_POSTSUBSCRIPT italic_ν start_POSTSUBSCRIPT roman_ref end_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ( bold_u ) italic_B start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ( bold_u ) italic_T start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ( bold_u )
=\displaystyle==B¯(𝐮)Tν(𝐮).¯𝐵𝐮subscript𝑇𝜈𝐮\displaystyle\bar{B}(\mathbf{u})\,T_{\nu}(\mathbf{u}).over¯ start_ARG italic_B end_ARG ( bold_u ) italic_T start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ( bold_u ) .

results in a map B¯(𝐮)Tν(𝐮)¯𝐵𝐮subscript𝑇𝜈𝐮\bar{B}(\mathbf{u})\,T_{\nu}(\mathbf{u})over¯ start_ARG italic_B end_ARG ( bold_u ) italic_T start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ( bold_u ) that is convolved with the same reference beam Bνrefsubscript𝐵subscript𝜈refB_{\nu_{\rm ref}}italic_B start_POSTSUBSCRIPT italic_ν start_POSTSUBSCRIPT roman_ref end_POSTSUBSCRIPT end_POSTSUBSCRIPT at every frequencies.

Correlation-based Beam Calibration of 21cm Intensity Mapping (2)

3.2 Correlation-based Beam Calibration

Therefore, the core idea of our algorithm revolves around obtaining the relative beam Bνrelsubscriptsuperscript𝐵rel𝜈B^{\rm rel}_{\nu}italic_B start_POSTSUPERSCRIPT roman_rel end_POSTSUPERSCRIPT start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT through internal calibration.First, let us re-express the temperature map at a given frequency ν𝜈\nuitalic_ν and 𝐮𝐮\mathbf{u}bold_u cell as the product of the frequency-dependent relative beam Bνrel(𝐮)subscriptsuperscript𝐵rel𝜈𝐮B^{\rm rel}_{\nu}(\mathbf{u})italic_B start_POSTSUPERSCRIPT roman_rel end_POSTSUPERSCRIPT start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ( bold_u ) and the remaining component

T~ν(𝐮)=Bνrel(𝐮)B¯(𝐮)Tν=Bνrel(𝐮)T¯ν(𝐮).subscript~𝑇𝜈𝐮subscriptsuperscript𝐵rel𝜈𝐮¯𝐵𝐮subscript𝑇𝜈subscriptsuperscript𝐵rel𝜈𝐮subscript¯𝑇𝜈𝐮\displaystyle\widetilde{T}_{\nu}(\mathbf{u})=B^{\rm rel}_{\nu}(\mathbf{u})\bar%{B}(\mathbf{u})T_{\nu}=B^{\rm rel}_{\nu}(\mathbf{u})\bar{T}_{\nu}(\mathbf{u}).over~ start_ARG italic_T end_ARG start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ( bold_u ) = italic_B start_POSTSUPERSCRIPT roman_rel end_POSTSUPERSCRIPT start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ( bold_u ) over¯ start_ARG italic_B end_ARG ( bold_u ) italic_T start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT = italic_B start_POSTSUPERSCRIPT roman_rel end_POSTSUPERSCRIPT start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ( bold_u ) over¯ start_ARG italic_T end_ARG start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ( bold_u ) .(12)

Here we have ignored the noise term nν(𝐮)subscript𝑛𝜈𝐮n_{\nu}(\mathbf{u})italic_n start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ( bold_u ) and absorbed the frequency-independent beam in the temperature map and define

T¯ν=B¯(𝐮)Tν.subscript¯𝑇𝜈¯𝐵𝐮subscript𝑇𝜈\displaystyle\bar{T}_{\nu}=\bar{B}(\mathbf{u})T_{\nu}.over¯ start_ARG italic_T end_ARG start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT = over¯ start_ARG italic_B end_ARG ( bold_u ) italic_T start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT .(13)

To proceed, we aim to construct a filter denoted as Fνrefν(𝐮)subscript𝐹subscript𝜈ref𝜈𝐮F_{\nu_{\rm ref}\nu}(\mathbf{u})italic_F start_POSTSUBSCRIPT italic_ν start_POSTSUBSCRIPT roman_ref end_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ( bold_u ). Applying this filter to observed temperature T~ν(𝐮)subscript~𝑇𝜈𝐮\widetilde{T}_{\nu}(\mathbf{u})over~ start_ARG italic_T end_ARG start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ( bold_u ) at each frequency channel effectively inverts the relative beam, yielding an estimator for T¯νsubscript¯𝑇𝜈\bar{T}_{\nu}over¯ start_ARG italic_T end_ARG start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT

T¯νWF(𝐮)=Fνrefν(𝐮)T~ν(𝐮).subscriptsuperscript¯𝑇WF𝜈𝐮subscript𝐹subscript𝜈ref𝜈𝐮subscript~𝑇𝜈𝐮\displaystyle\bar{T}^{\rm WF}_{\nu}(\mathbf{u})=F_{\nu_{\rm ref}\nu}(\mathbf{u%})\,\widetilde{T}_{\nu}(\mathbf{u}).over¯ start_ARG italic_T end_ARG start_POSTSUPERSCRIPT roman_WF end_POSTSUPERSCRIPT start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ( bold_u ) = italic_F start_POSTSUBSCRIPT italic_ν start_POSTSUBSCRIPT roman_ref end_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ( bold_u ) over~ start_ARG italic_T end_ARG start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ( bold_u ) .(14)

Consequently, our goal is to design the filter Fνrefν(𝐮)subscript𝐹subscript𝜈ref𝜈𝐮F_{\nu_{\rm ref}\nu}(\mathbf{u})italic_F start_POSTSUBSCRIPT italic_ν start_POSTSUBSCRIPT roman_ref end_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ( bold_u ) so that it functions similarly to the relative filter fνrefν(𝐮)subscript𝑓subscript𝜈ref𝜈𝐮f_{\nu_{\rm ref}\nu}(\mathbf{u})italic_f start_POSTSUBSCRIPT italic_ν start_POSTSUBSCRIPT roman_ref end_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ( bold_u ) as described in Eq.(10).This could be achieved via the minimal variance estimator, derived by examining the variance

r2(𝐮)superscript𝑟2𝐮\displaystyle r^{2}(\mathbf{u})italic_r start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT ( bold_u )=\displaystyle==(T¯νWFT~νref)(T¯νWFT~νref)(𝐮)delimited-⟨⟩subscriptsuperscript¯𝑇WF𝜈subscript~𝑇subscript𝜈refsubscriptsuperscript¯𝑇WF𝜈subscript~𝑇subscript𝜈ref𝐮\displaystyle\left\langle\left(\bar{T}^{\rm WF}_{\nu}-\widetilde{T}_{\nu_{\rmref%}}\right)\left(\bar{T}^{\rm WF}_{\nu}-\widetilde{T}_{\nu_{\rm ref}}\right)%\right\rangle(\mathbf{u})~{}~{}~{}⟨ ( over¯ start_ARG italic_T end_ARG start_POSTSUPERSCRIPT roman_WF end_POSTSUPERSCRIPT start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT - over~ start_ARG italic_T end_ARG start_POSTSUBSCRIPT italic_ν start_POSTSUBSCRIPT roman_ref end_POSTSUBSCRIPT end_POSTSUBSCRIPT ) ( over¯ start_ARG italic_T end_ARG start_POSTSUPERSCRIPT roman_WF end_POSTSUPERSCRIPT start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT - over~ start_ARG italic_T end_ARG start_POSTSUBSCRIPT italic_ν start_POSTSUBSCRIPT roman_ref end_POSTSUBSCRIPT end_POSTSUBSCRIPT ) ⟩ ( bold_u )(15)

where delimited-⟨⟩\langle\cdots\rangle⟨ ⋯ ⟩ presents the averaging within specific grid cell in 𝐮𝐮\mathbf{u}bold_u. Minimizing r2(𝐮)superscript𝑟2𝐮r^{2}(\mathbf{u})italic_r start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT ( bold_u ) leads to the well-known Wiener filter (WF) 101010hence this explains the meaning of the superscript WFWF{\rm WF}roman_WF in Eq. (14)

Fνrefν(𝐮)subscript𝐹subscript𝜈ref𝜈𝐮\displaystyle F_{\nu_{\rm ref}\nu}(\mathbf{u})italic_F start_POSTSUBSCRIPT italic_ν start_POSTSUBSCRIPT roman_ref end_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ( bold_u )=\displaystyle==T~νT~νrefT~νT~ν(𝐮)=CννrefT~CννT~(𝐮)delimited-⟨⟩subscript~𝑇𝜈subscript~𝑇subscript𝜈refdelimited-⟨⟩subscript~𝑇𝜈subscript~𝑇𝜈𝐮subscriptsuperscript𝐶~𝑇𝜈subscript𝜈refsubscriptsuperscript𝐶~𝑇𝜈𝜈𝐮\displaystyle\frac{\left\langle\widetilde{T}_{\nu}\,\widetilde{T}_{\nu_{\rm ref%}}\right\rangle}{\left\langle\widetilde{T}_{\nu}\,\widetilde{T}_{\nu}\right%\rangle}(\mathbf{u})=\frac{C^{\widetilde{T}}_{\nu\nu_{\rm ref}}}{{C^{%\widetilde{T}}_{\nu\nu}}}(\mathbf{u})divide start_ARG ⟨ over~ start_ARG italic_T end_ARG start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT over~ start_ARG italic_T end_ARG start_POSTSUBSCRIPT italic_ν start_POSTSUBSCRIPT roman_ref end_POSTSUBSCRIPT end_POSTSUBSCRIPT ⟩ end_ARG start_ARG ⟨ over~ start_ARG italic_T end_ARG start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT over~ start_ARG italic_T end_ARG start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ⟩ end_ARG ( bold_u ) = divide start_ARG italic_C start_POSTSUPERSCRIPT over~ start_ARG italic_T end_ARG end_POSTSUPERSCRIPT start_POSTSUBSCRIPT italic_ν italic_ν start_POSTSUBSCRIPT roman_ref end_POSTSUBSCRIPT end_POSTSUBSCRIPT end_ARG start_ARG italic_C start_POSTSUPERSCRIPT over~ start_ARG italic_T end_ARG end_POSTSUPERSCRIPT start_POSTSUBSCRIPT italic_ν italic_ν end_POSTSUBSCRIPT end_ARG ( bold_u )(16)

Here, we have introduced the correlation function Cν1ν2T~subscriptsuperscript𝐶~𝑇subscript𝜈1subscript𝜈2C^{\widetilde{T}}_{\nu_{1}\nu_{2}}italic_C start_POSTSUPERSCRIPT over~ start_ARG italic_T end_ARG end_POSTSUPERSCRIPT start_POSTSUBSCRIPT italic_ν start_POSTSUBSCRIPT 1 end_POSTSUBSCRIPT italic_ν start_POSTSUBSCRIPT 2 end_POSTSUBSCRIPT end_POSTSUBSCRIPT, defined as

Cν1ν2T~=T~ν1T~ν2.subscriptsuperscript𝐶~𝑇subscript𝜈1subscript𝜈2delimited-⟨⟩subscript~𝑇subscript𝜈1subscript~𝑇subscript𝜈2\displaystyle C^{\widetilde{T}}_{\nu_{1}\nu_{2}}=\left\langle\widetilde{T}_{%\nu_{1}}\,\widetilde{T}_{\nu_{2}}\right\rangle.italic_C start_POSTSUPERSCRIPT over~ start_ARG italic_T end_ARG end_POSTSUPERSCRIPT start_POSTSUBSCRIPT italic_ν start_POSTSUBSCRIPT 1 end_POSTSUBSCRIPT italic_ν start_POSTSUBSCRIPT 2 end_POSTSUBSCRIPT end_POSTSUBSCRIPT = ⟨ over~ start_ARG italic_T end_ARG start_POSTSUBSCRIPT italic_ν start_POSTSUBSCRIPT 1 end_POSTSUBSCRIPT end_POSTSUBSCRIPT over~ start_ARG italic_T end_ARG start_POSTSUBSCRIPT italic_ν start_POSTSUBSCRIPT 2 end_POSTSUBSCRIPT end_POSTSUBSCRIPT ⟩ .(17)

Substituting Eq.(3) of the observed temperature map T~~𝑇\widetilde{T}over~ start_ARG italic_T end_ARG and temporarily neglecting the noise matrix Nν1ν2=nν1nν2subscript𝑁subscript𝜈1subscript𝜈2delimited-⟨⟩subscript𝑛subscript𝜈1subscript𝑛subscript𝜈2N_{\nu_{1}\nu_{2}}=\langle n_{\nu_{1}}n_{\nu_{2}}\rangleitalic_N start_POSTSUBSCRIPT italic_ν start_POSTSUBSCRIPT 1 end_POSTSUBSCRIPT italic_ν start_POSTSUBSCRIPT 2 end_POSTSUBSCRIPT end_POSTSUBSCRIPT = ⟨ italic_n start_POSTSUBSCRIPT italic_ν start_POSTSUBSCRIPT 1 end_POSTSUBSCRIPT end_POSTSUBSCRIPT italic_n start_POSTSUBSCRIPT italic_ν start_POSTSUBSCRIPT 2 end_POSTSUBSCRIPT end_POSTSUBSCRIPT ⟩, we obtain

Cν1ν2T~(𝐮)subscriptsuperscript𝐶~𝑇subscript𝜈1subscript𝜈2𝐮\displaystyle C^{\widetilde{T}}_{\nu_{1}\nu_{2}}(\mathbf{u})italic_C start_POSTSUPERSCRIPT over~ start_ARG italic_T end_ARG end_POSTSUPERSCRIPT start_POSTSUBSCRIPT italic_ν start_POSTSUBSCRIPT 1 end_POSTSUBSCRIPT italic_ν start_POSTSUBSCRIPT 2 end_POSTSUBSCRIPT end_POSTSUBSCRIPT ( bold_u )=\displaystyle==Bν1Tν1Bν2Tν2(𝐮)delimited-⟨⟩subscript𝐵subscript𝜈1subscript𝑇subscript𝜈1subscript𝐵subscript𝜈2subscript𝑇subscript𝜈2𝐮\displaystyle\left\langle B_{\nu_{1}}\,T_{\nu_{1}}\,B_{\nu_{2}}\,T_{\nu_{2}}%\right\rangle(\mathbf{u})⟨ italic_B start_POSTSUBSCRIPT italic_ν start_POSTSUBSCRIPT 1 end_POSTSUBSCRIPT end_POSTSUBSCRIPT italic_T start_POSTSUBSCRIPT italic_ν start_POSTSUBSCRIPT 1 end_POSTSUBSCRIPT end_POSTSUBSCRIPT italic_B start_POSTSUBSCRIPT italic_ν start_POSTSUBSCRIPT 2 end_POSTSUBSCRIPT end_POSTSUBSCRIPT italic_T start_POSTSUBSCRIPT italic_ν start_POSTSUBSCRIPT 2 end_POSTSUBSCRIPT end_POSTSUBSCRIPT ⟩ ( bold_u )(18)
=\displaystyle==Bν1relBν2relB¯2Tν1Tν2(𝐮).delimited-⟨⟩subscriptsuperscript𝐵relsubscript𝜈1subscriptsuperscript𝐵relsubscript𝜈2superscript¯𝐵2subscript𝑇subscript𝜈1subscript𝑇subscript𝜈2𝐮\displaystyle\left\langle B^{\rm rel}_{\nu_{1}}\,B^{\rm rel}_{\nu_{2}}\,\bar{B%}^{2}\,T_{\nu_{1}}\,T_{\nu_{2}}\right\rangle(\mathbf{u}).⟨ italic_B start_POSTSUPERSCRIPT roman_rel end_POSTSUPERSCRIPT start_POSTSUBSCRIPT italic_ν start_POSTSUBSCRIPT 1 end_POSTSUBSCRIPT end_POSTSUBSCRIPT italic_B start_POSTSUPERSCRIPT roman_rel end_POSTSUPERSCRIPT start_POSTSUBSCRIPT italic_ν start_POSTSUBSCRIPT 2 end_POSTSUBSCRIPT end_POSTSUBSCRIPT over¯ start_ARG italic_B end_ARG start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT italic_T start_POSTSUBSCRIPT italic_ν start_POSTSUBSCRIPT 1 end_POSTSUBSCRIPT end_POSTSUBSCRIPT italic_T start_POSTSUBSCRIPT italic_ν start_POSTSUBSCRIPT 2 end_POSTSUBSCRIPT end_POSTSUBSCRIPT ⟩ ( bold_u ) .

It is apparent that

Fνrefν(𝐮)subscript𝐹subscript𝜈ref𝜈𝐮\displaystyle F_{\nu_{\rm ref}\nu}(\mathbf{u})italic_F start_POSTSUBSCRIPT italic_ν start_POSTSUBSCRIPT roman_ref end_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ( bold_u )=\displaystyle==1Bνrel(𝐮)FνrefνT(𝐮),1delimited-⟨⟩subscriptsuperscript𝐵rel𝜈𝐮subscriptsuperscript𝐹𝑇subscript𝜈ref𝜈𝐮\displaystyle\frac{1}{\left\langle B^{\rm rel}_{\nu}(\mathbf{u})\right\rangle}%\,F^{T}_{\nu_{\rm ref}\nu}(\mathbf{u}),divide start_ARG 1 end_ARG start_ARG ⟨ italic_B start_POSTSUPERSCRIPT roman_rel end_POSTSUPERSCRIPT start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ( bold_u ) ⟩ end_ARG italic_F start_POSTSUPERSCRIPT italic_T end_POSTSUPERSCRIPT start_POSTSUBSCRIPT italic_ν start_POSTSUBSCRIPT roman_ref end_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ( bold_u ) ,(19)

where the additional intrinsic temperature factor FνrefνT(𝐮)subscriptsuperscript𝐹𝑇subscript𝜈ref𝜈𝐮F^{T}_{\nu_{\rm ref}\nu}(\mathbf{u})italic_F start_POSTSUPERSCRIPT italic_T end_POSTSUPERSCRIPT start_POSTSUBSCRIPT italic_ν start_POSTSUBSCRIPT roman_ref end_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ( bold_u ) is defined as

FνrefνT(𝐮)subscriptsuperscript𝐹𝑇subscript𝜈ref𝜈𝐮\displaystyle F^{T}_{\nu_{\rm ref}\nu}(\mathbf{u})italic_F start_POSTSUPERSCRIPT italic_T end_POSTSUPERSCRIPT start_POSTSUBSCRIPT italic_ν start_POSTSUBSCRIPT roman_ref end_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ( bold_u )\displaystyle\equivCννrefT(𝐮)CννT(𝐮).subscriptsuperscript𝐶𝑇𝜈subscript𝜈ref𝐮subscriptsuperscript𝐶𝑇𝜈𝜈𝐮\displaystyle\frac{C^{T}_{\nu\nu_{\rm ref}}(\mathbf{u})}{C^{T}_{\nu\nu}(%\mathbf{u})}.divide start_ARG italic_C start_POSTSUPERSCRIPT italic_T end_POSTSUPERSCRIPT start_POSTSUBSCRIPT italic_ν italic_ν start_POSTSUBSCRIPT roman_ref end_POSTSUBSCRIPT end_POSTSUBSCRIPT ( bold_u ) end_ARG start_ARG italic_C start_POSTSUPERSCRIPT italic_T end_POSTSUPERSCRIPT start_POSTSUBSCRIPT italic_ν italic_ν end_POSTSUBSCRIPT ( bold_u ) end_ARG .(20)

It is apparent that if we ignore the frequency variation of the temperature map, causing the factor FνrefνT(𝐮)subscriptsuperscript𝐹𝑇subscript𝜈ref𝜈𝐮F^{T}_{\nu_{\rm ref}\nu}(\mathbf{u})italic_F start_POSTSUPERSCRIPT italic_T end_POSTSUPERSCRIPT start_POSTSUBSCRIPT italic_ν start_POSTSUBSCRIPT roman_ref end_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ( bold_u ) to become unity, the constructed filter simply represents the average value of the relative beam across each given 𝐮𝐮\mathbf{u}bold_u cell. Therefore, Fνrefν(𝐮)subscript𝐹subscript𝜈ref𝜈𝐮F_{\nu_{\rm ref}\nu}(\mathbf{u})italic_F start_POSTSUBSCRIPT italic_ν start_POSTSUBSCRIPT roman_ref end_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ( bold_u ) is indeed the filter that meets our requirements.However, in reality, the image of the foreground varies with frequency, thus introducing an additional undetermined function on top of the relative beam. Various strategies can be employed to mitigate its influence, including using a foreground model as prior information. As we will discuss in detail in Section 5, the impact of this term is not significant.

Furthermore, observational data contains instrumental noise (Equation 3), making the filter described in Equation (19) usually sub-optimal for analyzing real data.Various methods can be adopted to effectively isolate the beam variations in such situation. We will explore these approaches in more detail and discuss their applications to real data analysis in a separate paper.In this work, however, it is worth mentioning that a straightforward solution involves considering the full correlation matrix CνiνjT~(𝐮)subscriptsuperscript𝐶~𝑇subscript𝜈𝑖subscript𝜈𝑗𝐮C^{\widetilde{T}}_{\nu_{i}\nu_{j}}(\mathbf{u})italic_C start_POSTSUPERSCRIPT over~ start_ARG italic_T end_ARG end_POSTSUPERSCRIPT start_POSTSUBSCRIPT italic_ν start_POSTSUBSCRIPT italic_i end_POSTSUBSCRIPT italic_ν start_POSTSUBSCRIPT italic_j end_POSTSUBSCRIPT end_POSTSUBSCRIPT ( bold_u ) across all frequencies

CνiνjT~(𝐮)subscriptsuperscript𝐶~𝑇subscript𝜈𝑖subscript𝜈𝑗𝐮\displaystyle C^{\widetilde{T}}_{\nu_{i}\nu_{j}}(\mathbf{u})italic_C start_POSTSUPERSCRIPT over~ start_ARG italic_T end_ARG end_POSTSUPERSCRIPT start_POSTSUBSCRIPT italic_ν start_POSTSUBSCRIPT italic_i end_POSTSUBSCRIPT italic_ν start_POSTSUBSCRIPT italic_j end_POSTSUBSCRIPT end_POSTSUBSCRIPT ( bold_u )=\displaystyle==T~νi(𝐮)T~νj(𝐮).delimited-⟨⟩subscript~𝑇subscript𝜈𝑖𝐮subscript~𝑇subscript𝜈𝑗𝐮\displaystyle\left\langle\widetilde{T}_{\nu_{i}}(\mathbf{u})\widetilde{T}_{\nu%_{j}}(\mathbf{u})\right\rangle.⟨ over~ start_ARG italic_T end_ARG start_POSTSUBSCRIPT italic_ν start_POSTSUBSCRIPT italic_i end_POSTSUBSCRIPT end_POSTSUBSCRIPT ( bold_u ) over~ start_ARG italic_T end_ARG start_POSTSUBSCRIPT italic_ν start_POSTSUBSCRIPT italic_j end_POSTSUBSCRIPT end_POSTSUBSCRIPT ( bold_u ) ⟩ .(21)

Here the observed temperature map T~νi(𝐮)subscript~𝑇subscript𝜈𝑖𝐮\widetilde{T}_{\nu_{i}}(\mathbf{u})over~ start_ARG italic_T end_ARG start_POSTSUBSCRIPT italic_ν start_POSTSUBSCRIPT italic_i end_POSTSUBSCRIPT end_POSTSUBSCRIPT ( bold_u ) is a combination of sky Bνrel(𝐮)T¯ν(𝐮)subscriptsuperscript𝐵rel𝜈𝐮subscript¯𝑇𝜈𝐮B^{\rm rel}_{\nu}(\mathbf{u})\bar{T}_{\nu}(\mathbf{u})italic_B start_POSTSUPERSCRIPT roman_rel end_POSTSUPERSCRIPT start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ( bold_u ) over¯ start_ARG italic_T end_ARG start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ( bold_u ) and noise nν(𝐮)subscript𝑛𝜈𝐮n_{\nu}(\mathbf{u})italic_n start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ( bold_u ). Assuming that the amplitude of the sky is much larger than instrumental noise, we can then perform the SVD of matrix 𝑪T~(𝐮)superscript𝑪~𝑇𝐮\bm{C}^{\widetilde{T}}(\mathbf{u})bold_italic_C start_POSTSUPERSCRIPT over~ start_ARG italic_T end_ARG end_POSTSUPERSCRIPT ( bold_u )

𝑪T~(𝐮)=𝑼𝚺𝑽,superscript𝑪~𝑇𝐮𝑼𝚺superscript𝑽\displaystyle\bm{C}^{\widetilde{T}}(\mathbf{u})=\bm{U}\bm{\Sigma}\bm{V}^{%\dagger},bold_italic_C start_POSTSUPERSCRIPT over~ start_ARG italic_T end_ARG end_POSTSUPERSCRIPT ( bold_u ) = bold_italic_U bold_Σ bold_italic_V start_POSTSUPERSCRIPT † end_POSTSUPERSCRIPT ,(22)

where the diagonal matrix 𝚺=diag(λi)𝚺diagsubscript𝜆𝑖{\bm{\Sigma}}={\rm diag}(\lambda_{i})bold_Σ = roman_diag ( italic_λ start_POSTSUBSCRIPT italic_i end_POSTSUBSCRIPT ) store all eigen-values in decending order, and 𝑼𝑼\bm{U}bold_italic_U (and 𝑽𝑽\bm{V}bold_italic_V) lists the corresponding eigen-vectors.As long as the sky, primarily comprised of the foreground, dominates the matrix CνiνjT~(𝐮)subscriptsuperscript𝐶~𝑇subscript𝜈𝑖subscript𝜈𝑗𝐮C^{\widetilde{T}}_{\nu_{i}\nu_{j}}(\mathbf{u})italic_C start_POSTSUPERSCRIPT over~ start_ARG italic_T end_ARG end_POSTSUPERSCRIPT start_POSTSUBSCRIPT italic_ν start_POSTSUBSCRIPT italic_i end_POSTSUBSCRIPT italic_ν start_POSTSUBSCRIPT italic_j end_POSTSUBSCRIPT end_POSTSUBSCRIPT ( bold_u ), one could approximate the observed temperature with the value of the first eigen-mode.Therefore, in a similar approach to constructing the filter in Eq.(19), we can define the filter using only the first eigenvector Uν(1)(𝐮)subscriptsuperscript𝑈1𝜈𝐮U^{(1)}_{\nu}(\mathbf{u})italic_U start_POSTSUPERSCRIPT ( 1 ) end_POSTSUPERSCRIPT start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ( bold_u ), which can be expressed as:

Fνrefν(𝐮)subscript𝐹subscript𝜈ref𝜈𝐮\displaystyle F_{\nu_{\rm ref}\nu}(\mathbf{u})italic_F start_POSTSUBSCRIPT italic_ν start_POSTSUBSCRIPT roman_ref end_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ( bold_u )=\displaystyle==Uνref(1)(𝐮)Uν(1)(𝐮).superscriptsubscript𝑈subscript𝜈ref1𝐮subscriptsuperscript𝑈1𝜈𝐮\displaystyle\frac{U_{\nu_{\rm ref}}^{(1)}(\mathbf{u})}{U^{(1)}_{\nu}(\mathbf{%u})}.divide start_ARG italic_U start_POSTSUBSCRIPT italic_ν start_POSTSUBSCRIPT roman_ref end_POSTSUBSCRIPT end_POSTSUBSCRIPT start_POSTSUPERSCRIPT ( 1 ) end_POSTSUPERSCRIPT ( bold_u ) end_ARG start_ARG italic_U start_POSTSUPERSCRIPT ( 1 ) end_POSTSUPERSCRIPT start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ( bold_u ) end_ARG .(23)

The only underlying assumption here is the dominance of the sky temperature relative to the instrumental noise, which is achievable with sufficient integration time.Consequently, Eq. (23), representing the ratio between the first eigen-modes at two different frequencies, characterizes the desired filter with reduced noise influence.Therefore, this approach has the advantage of requiring minimal information about the noise covariance matrix.

4 Simulation Data

In this section, we will provide a detailed overview of the technical specifics of our simulated data set, covering both the sky and beam models.

Correlation-based Beam Calibration of 21cm Intensity Mapping (3)

4.1 Sky Model

Correlation-based Beam Calibration of 21cm Intensity Mapping (4)

For simplicity, the sky model for our simulation is generated using the publicly available 𝙲𝚁𝙸𝙼𝙴𝙲𝚁𝙸𝙼𝙴\mathtt{CRIME}typewriter_CRIME code Alonso etal. (2014) 111111http://intensitymapping.physics.ox.ac.uk/CRIME.html,which includes five distinct components:(a) HI signal,(b) Galactic synchrotron emission,(c) Galactic free-free emission,(d) Extra-galactic free-free emission,(e) Extra-galactic point sources emission.As demonstrated in Alonso etal. (2014), the HI signal here is derived from the lognormal model (Coles & Jones, 1991) of matter fluctuation δ(𝐱)𝛿𝐱\delta(\mathbf{x})italic_δ ( bold_x ), which is then transformed into a map of 21cm brightness temperature fluctuations

THI(𝐧^,z)=(0.19055K)hΩb(1+z)2xHI(z)Ωm(1+z)3+ΩΛ(1+δHI),superscript𝑇HI^𝐧𝑧0.19055KsubscriptΩ𝑏superscript1𝑧2subscript𝑥HI𝑧subscriptΩ𝑚superscript1𝑧3subscriptΩΛ1subscript𝛿HIT^{\rm HI}(\mathbf{\hat{n}},z)=(0.19055\,\mathrm{K})\frac{h\,\Omega_{b}(1+z)^{%2}x_{\rm HI}(z)}{\sqrt{\Omega_{m}(1+z)^{3}+\Omega_{\Lambda}}}(1+\delta_{\rm HI%}),italic_T start_POSTSUPERSCRIPT roman_HI end_POSTSUPERSCRIPT ( over^ start_ARG bold_n end_ARG , italic_z ) = ( 0.19055 roman_K ) divide start_ARG italic_h roman_Ω start_POSTSUBSCRIPT italic_b end_POSTSUBSCRIPT ( 1 + italic_z ) start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT italic_x start_POSTSUBSCRIPT roman_HI end_POSTSUBSCRIPT ( italic_z ) end_ARG start_ARG square-root start_ARG roman_Ω start_POSTSUBSCRIPT italic_m end_POSTSUBSCRIPT ( 1 + italic_z ) start_POSTSUPERSCRIPT 3 end_POSTSUPERSCRIPT + roman_Ω start_POSTSUBSCRIPT roman_Λ end_POSTSUBSCRIPT end_ARG end_ARG ( 1 + italic_δ start_POSTSUBSCRIPT roman_HI end_POSTSUBSCRIPT ) ,(24)

where δHIsubscript𝛿HI\delta_{\rm HI}italic_δ start_POSTSUBSCRIPT roman_HI end_POSTSUBSCRIPT represents the HI overdensity, smoothed over the volume element defined by δν𝛿𝜈\delta\nuitalic_δ italic_ν and δΩ𝛿Ω\delta\Omegaitalic_δ roman_Ω. Here the HI fraction xHI(z)=0.008(1+z)subscript𝑥HI𝑧0.0081𝑧x_{\rm HI}(z)=0.008(1+z)italic_x start_POSTSUBSCRIPT roman_HI end_POSTSUBSCRIPT ( italic_z ) = 0.008 ( 1 + italic_z ) is assumed to be fixed. Although a more complex modeling approach might be beneficial, it is not critical for the purposes of our study. In this paper, we consider the frequency range between 700900MHz700900MHz700\!-\!900\,\rm MHz700 - 900 roman_MHz corresponding to the redshift range 0.58<z<1.030.58𝑧1.030.58\!<\!z\!<\!1.030.58 < italic_z < 1.03 in our choice of cosmology.

In terms of foreground modeling, 𝙲𝚁𝙸𝙼𝙴𝙲𝚁𝙸𝙼𝙴\mathtt{CRIME}typewriter_CRIME models the large-scale Galactic synchrotron radiation (b) by extrapolating the 408MHz408MHz408\,\rm MHz408 roman_MHz Haslam map(Haslam etal., 1982). This extrapolation utilizes a direction-dependent spectral index derived from the Planck Sky model(Delabrouille etal., 2013), formulated as:

Tsync(ν,𝐧^)=THaslam(𝐧^)(νHν)β(𝐧^),superscript𝑇sync𝜈^𝐧subscript𝑇Haslam^𝐧superscriptsubscript𝜈𝐻𝜈𝛽^𝐧T^{\rm sync}(\nu,\mathbf{\hat{n}})=T_{\rm Haslam}(\mathbf{\hat{n}})\left(\frac%{\nu_{H}}{\nu}\right)^{\beta(\mathbf{\hat{n}})},italic_T start_POSTSUPERSCRIPT roman_sync end_POSTSUPERSCRIPT ( italic_ν , over^ start_ARG bold_n end_ARG ) = italic_T start_POSTSUBSCRIPT roman_Haslam end_POSTSUBSCRIPT ( over^ start_ARG bold_n end_ARG ) ( divide start_ARG italic_ν start_POSTSUBSCRIPT italic_H end_POSTSUBSCRIPT end_ARG start_ARG italic_ν end_ARG ) start_POSTSUPERSCRIPT italic_β ( over^ start_ARG bold_n end_ARG ) end_POSTSUPERSCRIPT ,(25)

where β(𝐧^)𝛽^𝐧\beta(\mathbf{\hat{n}})italic_β ( over^ start_ARG bold_n end_ARG ) is the synchrotron spectral index, varying with the line-of-sight (LoS) direction 𝐧^^𝐧\mathbf{\hat{n}}over^ start_ARG bold_n end_ARG. To capture structures on scales smaller than Haslam resolution, 𝙲𝚁𝙸𝙼𝙴𝙲𝚁𝙸𝙼𝙴\mathtt{CRIME}typewriter_CRIME incorporates a constrained Gaussian realization with parameters from (Santos etal., 2005; Alonso etal., 2014). Similarly, for the Galactic free-free (c) and Extra-galactic point sources (e) emissions, 𝙲𝚁𝙸𝙼𝙴𝙲𝚁𝙸𝙼𝙴\mathtt{CRIME}typewriter_CRIME also uses Gaussian realizations based on specific angular power spectra (Alonso etal., 2014).

From the full-sky simulated maps, we then cutout a mock data cube measuring 40×40superscript40superscript4040^{\circ}\times 40^{\circ}40 start_POSTSUPERSCRIPT ∘ end_POSTSUPERSCRIPT × 40 start_POSTSUPERSCRIPT ∘ end_POSTSUPERSCRIPT in angular direction and spanning a frequency range of 700900MHz700900MHz700-900{\rm MHz}700 - 900 roman_M roman_H roman_z. Due to computational constraints, the angular resolution of the maps is set at 3.44arcmin3.44arcmin3.44~{}\rm arcmin3.44 roman_arcmin, corresponding to 𝚗𝚜𝚒𝚍𝚎=1024𝚗𝚜𝚒𝚍𝚎1024\mathtt{nside}=1024typewriter_nside = 1024 in the 𝚑𝚎𝚊𝚕𝚙𝚒𝚡𝚑𝚎𝚊𝚕𝚙𝚒𝚡\mathtt{healpix}typewriter_healpix format.

As illustrated in Figure (2), we have chosen a relatively clean region with right ascension 130RA170superscript130RAsuperscript170130^{\circ}\leq{\rm RA}\leq 170^{\circ}130 start_POSTSUPERSCRIPT ∘ end_POSTSUPERSCRIPT ≤ roman_RA ≤ 170 start_POSTSUPERSCRIPT ∘ end_POSTSUPERSCRIPT and declination 10DEC50superscript10DECsuperscript5010^{\circ}\leq{\rm DEC}\leq 50^{\circ}10 start_POSTSUPERSCRIPT ∘ end_POSTSUPERSCRIPT ≤ roman_DEC ≤ 50 start_POSTSUPERSCRIPT ∘ end_POSTSUPERSCRIPT. In the frequency domain, we have divided the data into 128 frequency channels spanning the range from 700MHz700MHz700\,{\rm MHz}700 roman_MHz to 900MHz900MHz900\,{\rm MHz}900 roman_MHz. Subsequently, we proceed to grid this dataset into a three-dimensional cube with dimensions (Nν,NRA,NDEC)=(128,400,400)subscript𝑁𝜈subscript𝑁RAsubscript𝑁DEC128400400(N_{\nu},N_{\rm RA},N_{\rm DEC})=(128,400,400)( italic_N start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT , italic_N start_POSTSUBSCRIPT roman_RA end_POSTSUBSCRIPT , italic_N start_POSTSUBSCRIPT roman_DEC end_POSTSUBSCRIPT ) = ( 128 , 400 , 400 ). This raw data cube will then be convolved with beam functions, as detailed in the following subsection, Sec.4.2.

Correlation-based Beam Calibration of 21cm Intensity Mapping (5)

Correlation-based Beam Calibration of 21cm Intensity Mapping (6)

Furthermore, to evaluate the impact of the intrinsic frequency variations in the sky (as described in Eq. 20), we consider a simplified foreground model in which the spatial patterns at a reference frequency νrefsubscript𝜈ref\nu_{\rm ref}italic_ν start_POSTSUBSCRIPT roman_ref end_POSTSUBSCRIPT are held fixed, while the overall amplitudes are adjusted according to the respective frequency. In the following, this model is labeled to as “𝙵𝙶:𝚏𝚛𝚎𝚚𝚒𝚗𝚍𝚎𝚙:𝙵𝙶𝚏𝚛𝚎𝚚𝚒𝚗𝚍𝚎𝚙\mathtt{FG:freq-indep}typewriter_FG : typewriter_freq - typewriter_indep”, whereas the original foreground model is denoted as “𝙵𝙶:𝚏𝚛𝚎𝚚𝚍𝚎𝚙𝚎𝚗𝚍.:𝙵𝙶𝚏𝚛𝚎𝚚𝚍𝚎𝚙𝚎𝚗𝚍\mathtt{FG:freq-depend.}typewriter_FG : typewriter_freq - typewriter_depend .”. In Figure (3), we display the map of temperature ratio between two frequencies: T~ν(𝐧^)subscript~𝑇𝜈^𝐧\widetilde{T}_{\nu}(\mathbf{\hat{n}})over~ start_ARG italic_T end_ARG start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ( over^ start_ARG bold_n end_ARG ) at 800.2MHz800.2MHz800.2{\rm MHz}800.2 roman_MHz and the reference frequency νref=900MHzsubscript𝜈ref900MHz\nu_{\rm ref}=900{\rm MHz}italic_ν start_POSTSUBSCRIPT roman_ref end_POSTSUBSCRIPT = 900 roman_M roman_H roman_z. The left panel displays the original 𝚏𝚛𝚎𝚚𝚍𝚎𝚙𝚎𝚗𝚍.𝚏𝚛𝚎𝚚𝚍𝚎𝚙𝚎𝚗𝚍\mathtt{freq-depend.}typewriter_freq - typewriter_depend . sky map, while the right panel showcases the simplified 𝚏𝚛𝚎𝚚𝚒𝚗𝚍𝚎𝚙.𝚏𝚛𝚎𝚚𝚒𝚗𝚍𝚎𝚙\mathtt{freq-indep.}typewriter_freq - typewriter_indep . model. In the simplified model, we observe some noise around the value of 1.41.41.41.4, which is attributed to the independently added HI signal.

4.2 Beam Model

In this subsection, we introduce the noisy beam model utilized in this study.For simplicity, we consider a single dish telescope with a Gaussian model as the unperturbed beam

BG(θ)=exp(θ22σ2)superscript𝐵𝐺𝜃superscript𝜃22superscript𝜎2\displaystyle B^{G}(\theta)=\exp\left(-\frac{\theta^{2}}{2\sigma^{2}}\right)italic_B start_POSTSUPERSCRIPT italic_G end_POSTSUPERSCRIPT ( italic_θ ) = roman_exp ( - divide start_ARG italic_θ start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT end_ARG start_ARG 2 italic_σ start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT end_ARG )(26)

with σ=θFWHM/(22ln2)𝜎subscript𝜃FWHM222\sigma=\theta_{\rm\tiny FWHM}/(2\sqrt{2\ln 2})italic_σ = italic_θ start_POSTSUBSCRIPT roman_FWHM end_POSTSUBSCRIPT / ( 2 square-root start_ARG 2 roman_ln 2 end_ARG ). The full-width at half-maximum (FWHM) θFWHMsubscript𝜃FWHM\theta_{\rm\tiny FWHM}italic_θ start_POSTSUBSCRIPT roman_FWHM end_POSTSUBSCRIPT is determined by

θFWHM=cνD,subscript𝜃FWHM𝑐𝜈𝐷\theta_{\rm\tiny FWHM}=\frac{c}{\nu D},italic_θ start_POSTSUBSCRIPT roman_FWHM end_POSTSUBSCRIPT = divide start_ARG italic_c end_ARG start_ARG italic_ν italic_D end_ARG ,(27)

where D𝐷Ditalic_D is the diameter of the telescope dish. In many instances, this model offers a sufficiently accurate approximation. For example, as demonstrated in Liao etal. (2016), it provides a robust first-order approximation for both the intensity and polarization of the Green Bank Telescope.

Correlation-based Beam Calibration of 21cm Intensity Mapping (7)

As demonstrated in section 3, our method is implemented in Fourier space. In this domain, the Gaussian beam is represented as

BνG(𝐮)subscriptsuperscript𝐵𝐺𝜈𝐮\displaystyle B^{G}_{\nu}(\mathbf{u})italic_B start_POSTSUPERSCRIPT italic_G end_POSTSUPERSCRIPT start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ( bold_u )=\displaystyle==exp(σ(ν)2u22).𝜎superscript𝜈2superscript𝑢22\displaystyle\exp\left(-\frac{\sigma(\nu)^{2}u^{2}}{2}\right).roman_exp ( - divide start_ARG italic_σ ( italic_ν ) start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT italic_u start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT end_ARG start_ARG 2 end_ARG ) .(28)

In the following, we focus on a single dish telescope resembling the Green Bank Telescope (GBT) with a 100100100100-meter diameter, corresponding to the angular resolution θFWHMsubscript𝜃FWHM\theta_{\rm\tiny FWHM}italic_θ start_POSTSUBSCRIPT roman_FWHM end_POSTSUBSCRIPT that varies from 14141414 to 18arcmin18arcmin18~{}{\rm arcmin}18 roman_arcmin across the frequency range ν=700900MHz𝜈700900MHz\nu=700\!-\!900~{}{\rm MHz}italic_ν = 700 - 900 roman_MHz.

In this paper we consider two simple models of beam fluctuation. The first model (Model-I) treats the beam fluctuation as a three-dimensional random field in the 𝐮ν𝐮𝜈\mathbf{u}-\nubold_u - italic_ν space,

Bν(𝐮)subscript𝐵𝜈𝐮\displaystyle B_{\nu}(\mathbf{u})italic_B start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ( bold_u )=\displaystyle==BνG(𝐮)+δBν(𝐮)subscriptsuperscript𝐵𝐺𝜈𝐮𝛿subscript𝐵𝜈𝐮\displaystyle B^{G}_{\nu}(\mathbf{u})+\delta B_{\nu}(\mathbf{u})italic_B start_POSTSUPERSCRIPT italic_G end_POSTSUPERSCRIPT start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ( bold_u ) + italic_δ italic_B start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ( bold_u )
=\displaystyle==BνG(𝐮)(1+ϵν(𝐮)),subscriptsuperscript𝐵𝐺𝜈𝐮1subscriptitalic-ϵ𝜈𝐮\displaystyle B^{G}_{\nu}(\mathbf{u})(1+\epsilon_{\nu}(\mathbf{u})),italic_B start_POSTSUPERSCRIPT italic_G end_POSTSUPERSCRIPT start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ( bold_u ) ( 1 + italic_ϵ start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ( bold_u ) ) ,

where δBν(𝐮)𝛿subscript𝐵𝜈𝐮\delta B_{\nu}(\mathbf{u})italic_δ italic_B start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ( bold_u ) is the beam noise, and ϵν(𝐮)subscriptitalic-ϵ𝜈𝐮\epsilon_{\nu}(\mathbf{u})italic_ϵ start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ( bold_u ) is a zero mean random field in 𝐮ν𝐮𝜈\mathbf{u}-\nubold_u - italic_ν space, defined as

ϵν(𝐮)=εrand(𝐮,ν).subscriptitalic-ϵ𝜈𝐮𝜀rand𝐮𝜈\displaystyle\epsilon_{\nu}(\mathbf{u})=\varepsilon~{}{\rm rand}(\mathbf{u},%\nu).italic_ϵ start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ( bold_u ) = italic_ε roman_rand ( bold_u , italic_ν ) .(30)

Here, ε𝜀\varepsilonitalic_ε is the amplitude coefficient of the noise, and randrand{\rm rand}roman_rand is a uniformly distributed random variable within the range of 11-1- 1 to 1111. Given the entirely random nature of the noises across frequencies, this model represents a pessimistic scenario regarding potential beam noise.

In our second model of beam noise (Model-II), we consider a more realistic scenario where the noise originates within the telescope’s spatial configuration. This noise manifests as a two-dimensional random field

ϵd(𝐝)=εrand(𝐝)superscriptitalic-ϵ𝑑𝐝𝜀rand𝐝\displaystyle\epsilon^{d}(\mathbf{d})=\varepsilon~{}{\rm rand}(\mathbf{d})italic_ϵ start_POSTSUPERSCRIPT italic_d end_POSTSUPERSCRIPT ( bold_d ) = italic_ε roman_rand ( bold_d )(31)

where 𝐝𝐝\mathbf{d}bold_d represents the distance within the telescope’s frame. This noise field is subsequently rescaled and projected onto the 𝐮𝐮\mathbf{u}bold_u space at various frequencies using the relation 𝐮=𝐝/λ𝐮𝐝𝜆\mathbf{u}={\mathbf{d}}/\lambdabold_u = bold_d / italic_λ

ϵνd(𝐮)=ϵd(𝐝/λ)subscriptsuperscriptitalic-ϵ𝑑𝜈𝐮superscriptitalic-ϵ𝑑𝐝𝜆\displaystyle\epsilon^{d}_{\nu}(\mathbf{u})=\epsilon^{d}(\mathbf{d}/\lambda)italic_ϵ start_POSTSUPERSCRIPT italic_d end_POSTSUPERSCRIPT start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ( bold_u ) = italic_ϵ start_POSTSUPERSCRIPT italic_d end_POSTSUPERSCRIPT ( bold_d / italic_λ )(32)
Correlation-based Beam Calibration of 21cm Intensity Mapping (8)

In Figure (4), present our noisy beam model for a 100100100100 meter single-dish telescope, showcasing in both Fourier space (panels a-d) and configuration space (panel e). Panels (a) and (b) illustrate the noisy beam Bν(𝐮)subscript𝐵𝜈𝐮B_{\nu}(\mathbf{u})italic_B start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ( bold_u ) at 800.2,MHz800.2MHz800.2,{\rm MHz}800.2 , roman_MHz and its noise component δBν(𝐮)𝛿subscript𝐵𝜈𝐮\delta B_{\nu}(\mathbf{u})italic_δ italic_B start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ( bold_u ), respectively. The resolution of δBν(𝐮)𝛿subscript𝐵𝜈𝐮\delta B_{\nu}(\mathbf{u})italic_δ italic_B start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ( bold_u ) in Fourier space is about 1.3deg11.3superscriptdegree11.3~{}{\deg}^{-1}1.3 roman_deg start_POSTSUPERSCRIPT - 1 end_POSTSUPERSCRIPT, as shown in panel (b), with positive values indicated in red and negative values in blue. Model-I is characterized by completely random fluctuations in the frequency direction. This is exemplified in panel (c), which depicts a slice of the one-dimensional beam as a function of frequency. Panel (d) displays the one-dimensional beam at a specific frequency (800.2MHz800.2MHz800.2{\rm MHz}800.2 roman_MHz) under various noise levels ε𝜀\varepsilonitalic_ε, while keeping the underlying random field constant. Here, ε=2%𝜀percent2\varepsilon=2\%italic_ε = 2 % suggests that the beam fluctuates within a range of 0.020.02-0.02- 0.02 to 0.020.020.020.02. Finally, panel (e) illustrates the noisy side-lobes in configuration space, resulting from the beam noise δB𝛿𝐵\delta Bitalic_δ italic_B. The normalization chosen here ensures B(θ=0)=1𝐵𝜃01B(\theta=0)=1italic_B ( italic_θ = 0 ) = 1 at the center. We notice that these side-lobes in configuration space, emerged from simple Fourier space random noise, do resemble features in actual telescope beams, highlighting the practicality of our beam noise model.

Correlation-based Beam Calibration of 21cm Intensity Mapping (9)

In Figure (5), we showcase the one-dimensional beam fluctuation in Fourier space, with different colors representing three distinct frequency channels around 800MHz800MHz800\,{\rm MHz}800 roman_MHz. The upper panel illustrates our ‘pessimistic’ Model-I with ε=0.4%𝜀percent0.4\varepsilon=0.4\%italic_ε = 0.4 %, which is the three-dimensional random noise model, described by Eq. (30). In this model, the beam fluctuations are completely independent across frequencies, indicating a potentially larger impact on foreground removal. The lower panel shows a similar plot for Model-II with ε=1.6%𝜀percent1.6\varepsilon=1.6\%italic_ε = 1.6 %, where the noise resides in the telescope’s spatial domain. In this model, the frequency-dependence of the noise arises from the chromatic relationship described in Eq. (32). From the plot, it is clear that the noise profiles appear similar at lower k𝑘kitalic_k values for nearby frequencies, with deviations only becoming noticeable at higher k𝑘kitalic_k values.Our choice of ε𝜀\varepsilonitalic_ε here is consistent with real-world beam fluctuations. For instance, Liao etal. (2016) measured the beam pattern of the Green Bank Telescope (GBT) using astrophysical sources such as quasars and pulsars, characterized by two-dimensional Gauss-Hermite coefficients.Using their beam model, we estimate that the root-mean-square (RMS) of the beam fluctuation along the frequency direction could reach as high as 0.8%percent0.80.8\%0.8 %.

4.3 Mock Observation

After convolving our simulated datacube with the noisy beam model, we derive the mock observation data T~ν(𝐮)subscript~𝑇𝜈𝐮\widetilde{T}_{\nu}(\mathbf{u})over~ start_ARG italic_T end_ARG start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ( bold_u ) in Fourier space

T~ν(𝐮)subscript~𝑇𝜈𝐮\displaystyle\widetilde{T}_{\nu}(\mathbf{u})over~ start_ARG italic_T end_ARG start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ( bold_u )=Bν(𝐮)Tν(𝐮)absentsubscript𝐵𝜈𝐮subscript𝑇𝜈𝐮\displaystyle=B_{\nu}(\mathbf{u})\,T_{\nu}(\mathbf{u})= italic_B start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ( bold_u ) italic_T start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ( bold_u )(33)

as well as the corresponding map in real space T~ν(𝐧^)subscript~𝑇𝜈^𝐧\widetilde{T}_{\nu}(\mathbf{\hat{n}})over~ start_ARG italic_T end_ARG start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ( over^ start_ARG bold_n end_ARG ).

Figure (6) offers a comparative view of the mock observation data against the raw temperature map. The upper panels show the raw sky map (upper-left), denoted as Tνsubscript𝑇𝜈T_{\nu}italic_T start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT, alongside its convolution with the noisy telescope beam T~νsubscript~𝑇𝜈\widetilde{T}_{\nu}over~ start_ARG italic_T end_ARG start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT (upper-right), under beam Model-I (Eq.4.2), at a frequency of 800.2MHz800.2MHz800.2{\rm MHz}800.2 roman_MHz. In the middle-left panel, we illustrate the difference between these two plots, effectively highlighting the temperature variations caused by the beam noise of the telescope. Similarly, the middle-right panel displays the HI variation caused by the beam noise. Finally, the bottom part of the figure demonstrates the temperature fluctuation, δT~𝛿~𝑇\delta\widetilde{T}italic_δ over~ start_ARG italic_T end_ARG, along the frequency axis. It is evident from this display that such frequency variations will significantly challenge the effectiveness of foreground subtraction methods.

Correlation-based Beam Calibration of 21cm Intensity Mapping (10)

5 Results

Before proceeding to foreground subtraction and statistical analysis, our CBC method constructs a frequency-dependent filter Fνrefν(𝐮)subscript𝐹subscript𝜈ref𝜈𝐮F_{\nu_{\rm ref}\nu}(\mathbf{u})italic_F start_POSTSUBSCRIPT italic_ν start_POSTSUBSCRIPT roman_ref end_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ( bold_u ). As detailed in (3), this filter is generated through the cross-correlation of maps in the 𝐮𝐮\mathbf{u}bold_u space, comparing each frequency ν𝜈\nuitalic_ν with a reference frequency νrefsubscript𝜈ref\nu_{\rm ref}italic_ν start_POSTSUBSCRIPT roman_ref end_POSTSUBSCRIPT. This section aims to demonstrate the effectiveness of this approach. We will first discuss the calibration of the beam, followed by showcasing the results achieved through foreground removal.

5.1 Beam Calibration

In an ideal scenario, after applying the filter Fνrefνsubscript𝐹subscript𝜈ref𝜈F_{\nu_{\rm ref}\nu}italic_F start_POSTSUBSCRIPT italic_ν start_POSTSUBSCRIPT roman_ref end_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT, observational maps at different frequencies should all be convolved with the same beam, specifically the one at the reference frequency Bνrefsubscript𝐵subscript𝜈refB_{\nu_{\rm ref}}italic_B start_POSTSUBSCRIPT italic_ν start_POSTSUBSCRIPT roman_ref end_POSTSUBSCRIPT end_POSTSUBSCRIPT. However, in practice, intrinsic temperature variations across frequencies complicate this process, as they introduce an additional temperature factor (Eq. 19). To dissect the separate impacts of these two factors, we also consider a simplified ‘𝚏𝚛𝚎𝚚𝚒𝚗𝚍𝚎𝚙.𝚏𝚛𝚎𝚚𝚒𝚗𝚍𝚎𝚙\mathtt{freq-indep.}typewriter_freq - typewriter_indep .’ foreground model. In this model, the temperature factor FνrefνTsubscriptsuperscript𝐹𝑇subscript𝜈ref𝜈F^{T}_{\nu_{\rm ref}\nu}italic_F start_POSTSUPERSCRIPT italic_T end_POSTSUPERSCRIPT start_POSTSUBSCRIPT italic_ν start_POSTSUBSCRIPT roman_ref end_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT is held constant in the 𝐮𝐮\mathbf{u}bold_u space. Meanwhile, the ν𝜈\nuitalic_ν-dependent overall temperature amplitude of each channel is adjusted to aligns with that of the actual foreground model.

In the upper panel of Figure (7), we illustrate a one-dimensional slice of the filter Fνrefνsubscript𝐹subscript𝜈ref𝜈F_{\nu_{\rm ref}\nu}italic_F start_POSTSUBSCRIPT italic_ν start_POSTSUBSCRIPT roman_ref end_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT along with its individual components. Here, the sky temperature ratio FνrefνTsubscriptsuperscript𝐹𝑇subscript𝜈ref𝜈F^{T}_{\nu_{\rm ref}\nu}italic_F start_POSTSUPERSCRIPT italic_T end_POSTSUPERSCRIPT start_POSTSUBSCRIPT italic_ν start_POSTSUBSCRIPT roman_ref end_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT remains constant (blue dash-dotted line) at a value of approximately 1.51.51.51.5. Consequently, the normalized constructed filter Fνrefνnsubscriptsuperscript𝐹𝑛subscript𝜈ref𝜈F^{n}_{\nu_{\rm ref}\nu}italic_F start_POSTSUPERSCRIPT italic_n end_POSTSUPERSCRIPT start_POSTSUBSCRIPT italic_ν start_POSTSUBSCRIPT roman_ref end_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT (red solid line) closely resembles the relative beam 1/Bνrel1subscriptsuperscript𝐵rel𝜈1/B^{\rm rel}_{\nu}1 / italic_B start_POSTSUPERSCRIPT roman_rel end_POSTSUPERSCRIPT start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT (dotted line) that we introduced in the beam model. This convincingly validates the efficacy of our method.

In the lower panel of Figure (7), we present the filter for the more realistic ‘𝚏𝚛𝚎𝚚𝚍𝚎𝚙𝚎𝚗𝚍.𝚏𝚛𝚎𝚚𝚍𝚎𝚙𝚎𝚗𝚍\mathtt{freq-depend.}typewriter_freq - typewriter_depend .’ foreground. In contrast to the ‘𝚏𝚛𝚎𝚚𝚒𝚗𝚍𝚎𝚙.𝚏𝚛𝚎𝚚𝚒𝚗𝚍𝚎𝚙\mathtt{freq-indep.}typewriter_freq - typewriter_indep .’ model, the intrinsic variation in sky temperature (blue dash-dotted line) has a substantial influence on the estimation of the relative beam.As depicted, the fluctuations in the temperature ratio are considerably larger than the beam noise we introduced (black dashed line). As a result, the constructed filter (red solid line) bears a closer resemblance to the temperature ratio than the relative beam.However, this extra factor actually does not invalidate our method. In fact, as we will demonstrate in the next subsection, the foreground-cleaned map following the aforementioned procedure still exhibits considerable improvements compared to traditional methods. This is primarily because the variation in sky temperature, as shown in Figure (8), changes gradually along the frequency axis. In contrast, while the beam noise we introduced might appear small, it is either completely random (beam Model-I) or somewhat random (Model-II) along frequency. This more pronounced frequency dependence has a much more impactful influence on the result of foreground removal.

Correlation-based Beam Calibration of 21cm Intensity Mapping (11)
Correlation-based Beam Calibration of 21cm Intensity Mapping (12)
Correlation-based Beam Calibration of 21cm Intensity Mapping (13)
Correlation-based Beam Calibration of 21cm Intensity Mapping (14)

5.2 Foreground Removal and Statistical Measurements

After estimating and applying the CBC filter to our mock data, we now proceed with foreground removal and statistical analysis. Specifically, we adopt the blind Principal Component Analysis (PCA) cleaning technique following the methods detailed in Masui etal. (2013) and Switzer etal. (2013).We begin by representing the temperature map as a matrix 𝑻𝑻\bm{T}bold_italic_T, with dimensions Nν×Nθsubscript𝑁𝜈subscript𝑁𝜃N_{\nu}\times N_{\theta}italic_N start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT × italic_N start_POSTSUBSCRIPT italic_θ end_POSTSUBSCRIPT. The PCA modes are then obtained through the decomposition of 𝑪=𝑻𝑻T/Nθ=𝑼𝚲𝑼T𝑪𝑻superscript𝑻Tsubscript𝑁𝜃𝑼𝚲superscript𝑼T\bm{C}=\bm{TT}^{\rm T}/N_{\theta}=\bm{U\Lambda U}^{\rm T}bold_italic_C = bold_italic_T bold_italic_T start_POSTSUPERSCRIPT roman_T end_POSTSUPERSCRIPT / italic_N start_POSTSUBSCRIPT italic_θ end_POSTSUBSCRIPT = bold_italic_U bold_Λ bold_italic_U start_POSTSUPERSCRIPT roman_T end_POSTSUPERSCRIPT, where 𝚲𝚲\bm{\Lambda}bold_Λ is a diagonal matrix arranged in descending order. The ‘cleaned’ map, obtained by subtracting the first mm\rm mroman_m PCA modes, is calculated using Eq. (5).We continue by calculating the auto-power spectrum of the cleaned map

Pcl(k)=TclTcl(k).superscript𝑃cl𝑘delimited-⟨⟩superscript𝑇clsuperscript𝑇cl𝑘\displaystyle P^{\rm cl}(k)=\left\langle T^{\rm cl}T^{\rm cl}\right\rangle(k).italic_P start_POSTSUPERSCRIPT roman_cl end_POSTSUPERSCRIPT ( italic_k ) = ⟨ italic_T start_POSTSUPERSCRIPT roman_cl end_POSTSUPERSCRIPT italic_T start_POSTSUPERSCRIPT roman_cl end_POSTSUPERSCRIPT ⟩ ( italic_k ) .(34)

Additionally, we also compute the cross-power spectrum with the injected true HI signal. Specifically, our interest lies in the cross-correlation ratio between the two fields, defined as

f(k)=Pcl×HI(k)THITHITclTcl(k)𝑓𝑘superscript𝑃clHI𝑘delimited-⟨⟩superscript𝑇HIsuperscript𝑇HIdelimited-⟨⟩superscript𝑇clsuperscript𝑇cl𝑘\displaystyle f(k)=\frac{P^{\rm cl\times HI}(k)}{\sqrt{\langle T^{\rm HI}T^{%\rm HI}\rangle\langle T^{\rm cl}T^{\rm cl}\rangle}(k)}italic_f ( italic_k ) = divide start_ARG italic_P start_POSTSUPERSCRIPT roman_cl × roman_HI end_POSTSUPERSCRIPT ( italic_k ) end_ARG start_ARG square-root start_ARG ⟨ italic_T start_POSTSUPERSCRIPT roman_HI end_POSTSUPERSCRIPT italic_T start_POSTSUPERSCRIPT roman_HI end_POSTSUPERSCRIPT ⟩ ⟨ italic_T start_POSTSUPERSCRIPT roman_cl end_POSTSUPERSCRIPT italic_T start_POSTSUPERSCRIPT roman_cl end_POSTSUPERSCRIPT ⟩ end_ARG ( italic_k ) end_ARG(35)

Here THIsuperscript𝑇HIT^{\rm HI}italic_T start_POSTSUPERSCRIPT roman_HI end_POSTSUPERSCRIPT represents the injected HI signal and Pcl×HI(k)superscript𝑃clHI𝑘P^{\rm cl\times HI}(k)italic_P start_POSTSUPERSCRIPT roman_cl × roman_HI end_POSTSUPERSCRIPT ( italic_k ) is the cross-power spectrum between the signal and the cleaned map.

As detailed in Section 4.1, our mock observational data encompass the region spanning from 700MHz700MHz700{\rm MHz}700 roman_M roman_H roman_z to 900MHz900MHz900{\rm MHz}900 roman_M roman_H roman_z in frequency, 130.0superscript130.0130.0^{\circ}130.0 start_POSTSUPERSCRIPT ∘ end_POSTSUPERSCRIPT to 170.0superscript170.0170.0^{\circ}170.0 start_POSTSUPERSCRIPT ∘ end_POSTSUPERSCRIPT in right ascension (RA), and 10.0superscript10.010.0^{\circ}10.0 start_POSTSUPERSCRIPT ∘ end_POSTSUPERSCRIPT to 50.0superscript50.050.0^{\circ}50.0 start_POSTSUPERSCRIPT ∘ end_POSTSUPERSCRIPT in declination (DEC). With our chosen cosmological model, this region translate into a comoving box with dimensions of (Lν,LRA,LDEC)=(860.8,1343.7,1343.7)Mpc/hsubscript𝐿𝜈subscript𝐿RAsubscript𝐿DEC860.81343.71343.7Mpch\left(L_{\nu},L_{\rm RA},L_{\rm DEC}\right)=\left(860.8,1343.7,1343.7\right)~{%}{\rm Mpc/h}( italic_L start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT , italic_L start_POSTSUBSCRIPT roman_RA end_POSTSUBSCRIPT , italic_L start_POSTSUBSCRIPT roman_DEC end_POSTSUBSCRIPT ) = ( 860.8 , 1343.7 , 1343.7 ) roman_Mpc / roman_h.To assess the effectiveness of our method, we compare the outcomes achieved with the CBC approach to those obtained using a traditional method.The latter involves convolving data with a common Gaussian beam. This traditional approach, as described in Masui etal. (2013), includes the convolution with an additional kernel, equivalent to Eq. (7). In this paper, we apply this filter in real space.Following the discussion in Section 4.3, we will conduct a comprehensive comparison that incorporates two foreground models: the simplified ‘𝚏𝚛𝚎𝚚𝚒𝚗𝚍𝚎𝚙.𝚏𝚛𝚎𝚚𝚒𝚗𝚍𝚎𝚙\mathtt{freq-indep.}typewriter_freq - typewriter_indep .’ model and the realistic ‘𝚏𝚛𝚎𝚚𝚍𝚎𝚙𝚎𝚗𝚍.𝚏𝚛𝚎𝚚𝚍𝚎𝚙𝚎𝚗𝚍\mathtt{freq-depend.}typewriter_freq - typewriter_depend .’ foregrounds, two beam noise models (Model I and II), and two methods of beam calibration: the traditional method (TM) and our own correlation-based calibration method (CBC).

Correlation-based Beam Calibration of 21cm Intensity Mapping (15)
Correlation-based Beam Calibration of 21cm Intensity Mapping (16)

To establish a baseline result, Figure (9) presents both the auto power spectrum of the cleaned map (upper panels) and the cross-correlation ratio with true cosmic signal (lower panels), assuming that the relative beam fνrefν(𝐮)=Bνref/Bνsubscript𝑓subscript𝜈ref𝜈𝐮subscript𝐵subscript𝜈refsubscript𝐵𝜈f_{\nu_{\rm ref}\nu}(\mathbf{u})=B_{\nu_{\rm ref}}/B_{\nu}italic_f start_POSTSUBSCRIPT italic_ν start_POSTSUBSCRIPT roman_ref end_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT ( bold_u ) = italic_B start_POSTSUBSCRIPT italic_ν start_POSTSUBSCRIPT roman_ref end_POSTSUBSCRIPT end_POSTSUBSCRIPT / italic_B start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT is precisely known. As expected, both metrics agree with the expectation of an ‘ideal’ foreground removal, with only a few modes’ subtraction resulting in a low auto-power spectrum and a cross-correlation ratio nearing unity. This observation remains consistent across different foreground models and beam calibration methods. However, the results from the simplified ‘𝚏𝚛𝚎𝚚𝚒𝚗𝚍𝚎𝚙𝚏𝚛𝚎𝚚𝚒𝚗𝚍𝚎𝚙\mathtt{freq\!-\!indep}typewriter_freq - typewriter_indep’ (left panels) model demonstrate better performance. Particularly, subtracting as few as 2222 modes (blue-dashed line) is sufficient to remove this simple foreground contamination. Furthermore, as shown in the cross-correlation, the more realistic foreground model tends to result in an over-subtraction of the signal at the large-scale (k0.1h/Mpcless-than-or-similar-to𝑘0.1hMpck\lesssim 0.1\mathrm{h/Mpc}italic_k ≲ 0.1 roman_h / roman_Mpc). Notice that we did not differentiate between two beam models in this context, as they yield very similar results when the ‘true knowledge’ of beam fluctuations is known.

We then turn to assessing the efficacy of our CBC method against the traditional approach of the Gaussian common beam regularization. Figure (10) illustrates the result for the three-dimensional random beam model, i.e. Model-I, with noise amplitude ε=0.4%𝜀percent0.4\varepsilon=0.4\%italic_ε = 0.4 %.As shown, our CBC method consistently outperforms the traditional approach, regardless of the number of subtracted modes (Nsubsubscript𝑁subN_{\rm sub}italic_N start_POSTSUBSCRIPT roman_sub end_POSTSUBSCRIPT), or the choice of foreground models. For each chosen value of Nsubsubscript𝑁subN_{\rm sub}italic_N start_POSTSUBSCRIPT roman_sub end_POSTSUBSCRIPT, the CBC method significantly reduces the amplitude of the auto-power spectrum by a factor of several and enhances the cross-correlation ratio f𝑓fitalic_f by approximately 50%percent5050\%50 %. These outcomes suggest that, with such a noisy beam, the CBC method excels in removing foregrounds more effectively and in preserving the HI signals more efficiently.

For both methods, a sudden shape change appears in the auto-power spectrum around k0.1h/Mpcsimilar-to𝑘0.1hMpck\sim 0.1{\rm h/Mpc}italic_k ∼ 0.1 roman_h / roman_Mpc, resulting in a bump in the cross-correlation ratio f𝑓fitalic_f. This signature is the residual foreground caused by the mixing of beam noise with the intrinsic structure in the foreground. Such foreground structure can be partially observed in Figure 7. In the traditional method, we observe that the cross-correlation ratio f𝑓fitalic_f undergoes slightly more degradation than in the CBC method, especially at larger scales (k0.5h/Mpcless-than-or-similar-to𝑘0.5hMpck\lesssim 0.5\mathrm{h/Mpc}italic_k ≲ 0.5 roman_h / roman_Mpc). Additionally, the performance impact when using the realistic ‘𝚏𝚛𝚎𝚚𝚍𝚎𝚙𝚏𝚛𝚎𝚚𝚍𝚎𝚙\mathtt{freq\!-\!dep}typewriter_freq - typewriter_dep’ foreground model is surprisingly insignificant compared to the simpler ‘𝚏𝚛𝚎𝚚𝚒𝚗𝚍𝚙𝚏𝚛𝚎𝚚𝚒𝚗𝚍𝚙\mathtt{freq\!-\!indp}typewriter_freq - typewriter_indp’ model. This can be partly attributed to the slow frequency variation of the temperature ratio FνrefνT=Tν/Tνrefsubscriptsuperscript𝐹𝑇subscript𝜈ref𝜈subscript𝑇𝜈subscript𝑇subscript𝜈refF^{T}_{\nu_{\rm ref}\nu}=T_{\nu}/T_{\nu_{\rm ref}}italic_F start_POSTSUPERSCRIPT italic_T end_POSTSUPERSCRIPT start_POSTSUBSCRIPT italic_ν start_POSTSUBSCRIPT roman_ref end_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT = italic_T start_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT / italic_T start_POSTSUBSCRIPT italic_ν start_POSTSUBSCRIPT roman_ref end_POSTSUBSCRIPT end_POSTSUBSCRIPT. Despite the significant spatial variation of FνrefνTsubscriptsuperscript𝐹𝑇subscript𝜈ref𝜈F^{T}_{\nu_{\rm ref}\nu}italic_F start_POSTSUPERSCRIPT italic_T end_POSTSUPERSCRIPT start_POSTSUBSCRIPT italic_ν start_POSTSUBSCRIPT roman_ref end_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT relative to the beam noise, as illustrated in Figure (7), its frequency dependence is actually quite smooth. This is evident in Figure (8), where the right panel shows a different but slow frequency change across various spatial u𝑢uitalic_u values.

To examine how the performance of CBC method varies with the noise levels of the beam, we also analyze the results for two different amplitude, ε𝜀\varepsilonitalic_ε, in Model-I. Figure 11 and 12 illustrate the outcomes for ε=0.2%𝜀percent0.2\varepsilon=0.2\%italic_ε = 0.2 % and ε=0.6%𝜀percent0.6\varepsilon=0.6\%italic_ε = 0.6 %, respectively. For smaller beam noise (ε=0.2%𝜀percent0.2\varepsilon=0.2\%italic_ε = 0.2 %), we are able to subtract fewer modes to achieve a similar level of the cross-power spectrum, and the auto-power spectrum also approaches more closely to the injected signal for both the traditional and CBC methods. Although the CBC method continues to outperform the traditional approach, the correlation ratio for the CBC method stabilizes around f0.8similar-to𝑓0.8f\sim 0.8italic_f ∼ 0.8 at large k𝑘kitalic_k, compared to f0.6similar-to𝑓0.6f\sim 0.6italic_f ∼ 0.6 for the traditional method. This indicates a reduced improvement factor of about one-third.In contrast, with larger noise (ε=0.6%𝜀percent0.6\varepsilon=0.6\%italic_ε = 0.6 %), greater residual foregrounds lead to a higher auto-power spectrum and a lower cross-correlation ratio for both methods. Regardless, the CBC method still improves the cross-correlation ratio by approximately 50%percent5050\%50 %, which is a similar level of improvement compared to our fiducial model with ε=0.4%𝜀percent0.4\varepsilon=0.4\%italic_ε = 0.4 %.

Figure 13 displays the results for beam Model-II with a noise level of ε=1.6%𝜀percent1.6\varepsilon=1.6\%italic_ε = 1.6 %. The higher noise amplitude compared to Model-I is due to the significantly reduced frequency dependence of the noise in Model-II, as evidenced in Figure 5.A notable distinction from Figure 10 is the increased over-subtraction at larger scales (k0.15h/Mpcless-than-or-similar-to𝑘0.15hMpck\lesssim 0.15\mathrm{h/Mpc}italic_k ≲ 0.15 roman_h / roman_Mpc), observed in both the traditional approach and the CBC method, regardless of the foreground model considered. This difference arises because, in this beam model, the two-dimensional noise is rescaled along the frequency direction with ϵμd(𝐮)=ϵd(𝐝ν/c)subscriptsuperscriptitalic-ϵ𝑑𝜇𝐮superscriptitalic-ϵ𝑑𝐝𝜈𝑐\epsilon^{d}_{\mu}(\mathbf{u})=\epsilon^{d}(\mathbf{d}\nu/c)italic_ϵ start_POSTSUPERSCRIPT italic_d end_POSTSUPERSCRIPT start_POSTSUBSCRIPT italic_μ end_POSTSUBSCRIPT ( bold_u ) = italic_ϵ start_POSTSUPERSCRIPT italic_d end_POSTSUPERSCRIPT ( bold_d italic_ν / italic_c ), making the frequency dependence of the noise also scale-dependent. This behavior is apparent in Figure 5, where at lower 𝐮𝐮\mathbf{u}bold_u values, corresponding to lower k𝑘kitalic_k and shorter distance 𝐝𝐝\mathbf{d}bold_d, the dependency on ν𝜈\nuitalic_ν is less pronounced than at higher k𝑘kitalic_k values. Such a condition leads to a unique interplay between the scales and the frequency of fluctuations. As a result, the SVD modes, derived from the frequency singular modes, do not perform uniformly across all scales. The maps produced are smoother in ν𝜈\nuitalic_ν at larger angular scales, leading to over-removal in the SVD-cleaned maps. Despite this distinctive characteristic, our CBC method continues to outperform the traditional approach. As one can see, the level of improvement observed in Figure 13 is similar to what is observed in Figure 10.

We also examine the performance of beam Model-II at various noise levels. The results are illustrated in Figures 14 and 15 for ε=0.8%𝜀percent0.8\varepsilon=0.8\%italic_ε = 0.8 % and ε=2.4%𝜀percent2.4\varepsilon=2.4\%italic_ε = 2.4 %, respectively. The overall feature of over-subtracted large scale modes persists in both situations. Similar to Model-I, the performance difference between CBC and the traditional method is much smaller for the low-noise beam model, with the improvement in the cross-correlation ratio ranging from ten percent to a few percent. Conversely, for a noise level of ε=2.4%𝜀percent2.4\varepsilon=2.4\%italic_ε = 2.4 %, the improvement reaches approximately one-third.

6 Discussion and Conclusions

Foreground removal plays a pivotal role in the analysis of 21cm intensity mapping data. Most existing techniques require at least a basic understanding of the foreground, the instrument, or both. Even in blind methods, such as Principal Component Analysis (PCA), the specifics of the instrument significantly influence the efficiency of the results. Therefore, it is not surprising that these methods often show diminished performance in real-world applications, where actual instrumental beams are frequently noisy. Considerable efforts have been made towards accurate beam modeling and measurement. In this paper, we introduce a novel method that utilizing internal cross-correlation between different frequencies to calibrate the beam’s variation with frequency.

To simulate the real-world beam fluctuations, we consider two specific models of beam noise. The first is a random noise in the three-dimensional 𝐮ν𝐮𝜈\mathbf{u}\!-\!\nubold_u - italic_ν space, while the second is a two-dimensional random noise in the instrumental space, which is then projected along the frequency axis with varying 𝐮𝐮\mathbf{u}bold_u coordinates. Interestingly, the performances of these two models are somewhat distinct and complementary. Beam Model-II, in particular, experiences an over-subtraction at larger scales. However, when compared with traditional methods that overlook beam variation, our approach shows a considerable improvement. It achieves an approximate 50%percent5050\%50 % increase in the cross-correlation ratio and a significant reduction in the auto-power spectrum by several factors.

The filter constructed using our method incorporates two main factors: the relative change in the beam between two frequencies and the intrinsic temperature variation. To assess the impact of the later, we consider a simplified foreground model where the spatial pattern of the foreground remains constant. Interestingly, even substantial spatial fluctuations in the foreground temperature factor FνrefνTsubscriptsuperscript𝐹𝑇subscript𝜈ref𝜈F^{T}_{\nu_{\rm ref}\nu}italic_F start_POSTSUPERSCRIPT italic_T end_POSTSUPERSCRIPT start_POSTSUBSCRIPT italic_ν start_POSTSUBSCRIPT roman_ref end_POSTSUBSCRIPT italic_ν end_POSTSUBSCRIPT have a minimal effect on the performance. This resilience can be attributed to the smooth frequency variation of the foreground. Therefore, these findings further validate the effectiveness of our method.

Acknowledgments

X.W. is supported by the National SKA Program of China (Grants Nos. 2022SKA0110200 and 2022SKA0110202). U-L.P. receives support from Ontario Research Fund—research Excellence Program (ORF-RE), Natural Sciences and Engineering Research Council of Canada (NSERC) [funding reference number RGPIN-2019-067, CRD 523638-201, 555585-20], Canadian Institute for Advanced Research (CIFAR), Canadian Foundation for Innovation (CFI), the National Science Foundation of China (Grants No. 11929301), Simons Foundation, Thoth Technology Inc, and Alexander von Humboldt Foundation.

References

  • Abdurashidova etal. (2022)Abdurashidova, Z., Aguirre, J.E., Alexander, P., etal. 2022, TheAstrophysical Journal, 925, 221
  • Alonso etal. (2015)Alonso, D., Bull, P., Ferreira, P.G., & Santos, M.G. 2015, Monthly Noticesof the Royal Astronomical Society, 447, 400
  • Alonso etal. (2014)Alonso, D., Ferreira, P.G., & Santos, M.G. 2014, Monthly Notices of theRoyal Astronomical Society, 444, 3183
  • Anderson etal. (2014)Anderson, L., Aubourg, E., Bailey, S., etal. 2014, Monthly Notices of theRoyal Astronomical Society, 441, 24
  • AstropyCollaboration etal. (2018)AstropyCollaboration, Price-Whelan, A.M., Sipőcz, B., Günther, H.,etal. 2018, The Astronomical Journal, 156, 123
  • AstropyCollaboration etal. (2022)AstropyCollaboration, Price-Whelan, A.M., Lim, P.L., Earl, N., etal.2022, The Astrophysical Journal, 935, 167
  • AstropyCollaboration etal. (2013)AstropyCollaboration, Robitaille, T.P., Tollerud, E.J., Greenfield, P.,etal. 2013, Astronomy & Astrophysics, 558, A33
  • Bégin etal. (2022)Bégin, J.-M., Liu, A., & Gorce, A. 2022, Phys. Rev. D, 105, 083503,doi:10.1103/PhysRevD.105.083503
  • Bernardi etal. (2011)Bernardi, G., Mitchell, D.A., Ord, S.M., etal. 2011, Monthly Noticesof the Royal Astronomical Society, 413, 411,doi:10.1111/j.1365-2966.2010.18145.x
  • Bigot-Sazy etal. (2015)Bigot-Sazy, M.-A., Dickinson, C., Battye, R.A., etal. 2015, Monthly Noticesof the Royal Astronomical Society, 454, 3240
  • Chang etal. (2015)Chang, C., Monstein, C., Refregier, A., etal. 2015, Publications of theAstronomical Society of the Pacific, 127, 1131
  • Chapman etal. (2012)Chapman, E., Abdalla, F.B., Harker, G., etal. 2012, Monthly Notices of theRoyal Astronomical Society, 423, 2518
  • Coles & Jones (1991)Coles, P., & Jones, B. 1991, Monthly Notices of the Royal AstronomicalSociety, 248, 1, doi:10.1093/mnras/248.1.1
  • Collaboration etal. (2023)Collaboration, D., Adame, A., Aguilar, J., etal. 2023, arXiv preprintarXiv:2306.06308
  • Collaboration: etal. (2016)Collaboration:, D. E.S., Abbott, T., Abdalla, F., etal. 2016, MonthlyNotices of the Royal Astronomical Society, 460, 1270
  • Colless etal. (2001)Colless, M., Dalton, G., Maddox, S., etal. 2001, Monthly Notices of theRoyal Astronomical Society, 328, 1039
  • Condon & Ransom (2016)Condon, J.J., & Ransom, S.M. 2016, Essential radio astronomy, Vol.2(Princeton University Press)
  • Cunnington etal. (2023)Cunnington, S., Li, Y., Santos, M.G., etal. 2023, Monthly Notices of theRoyal Astronomical Society, 518, 6262
  • Dalcin & Fang (2021)Dalcin, L., & Fang, Y.-L.L. 2021, Computing in Science & Engineering, 23, 47
  • DeBoer etal. (2017)DeBoer, D.R., Parsons, A.R., Aguirre, J.E., etal. 2017, Publications ofthe Astronomical Society of the Pacific, 129, 045001
  • Delabrouille etal. (2013)Delabrouille, J., Betoule, M., Melin, J.-B., etal. 2013, Astronomy &Astrophysics, 553, A96
  • Drinkwater etal. (2010)Drinkwater, M.J., Jurek, R.J., Blake, C., etal. 2010, Monthly Notices ofthe Royal Astronomical Society, 401, 1429
  • Gabriel etal. (2004)Gabriel, E., fa*gg, G.E., Bosilca, G., etal. 2004, in Recent Advances inParallel Virtual Machine and Message Passing Interface: 11th European PVM/MPIUsers’ Group Meeting Budapest, Hungary, September 19-22, 2004. Proceedings11, Springer, 97–104
  • Gheller & Vazza (2022)Gheller, C., & Vazza, F. 2022, Monthly Notices of the Royal AstronomicalSociety, 509, 990
  • Gillet etal. (2019)Gillet, N., Mesinger, A., Greig, B., Liu, A., & Ucci, G. 2019, Monthly Noticesof the Royal Astronomical Society, 484, 282
  • Górski etal. (2005)Górski, K.M., Hivon, E., Banday, A.J., etal. 2005, ApJ, 622,759, doi:10.1086/427976
  • Harris etal. (2020)Harris, C.R., Millman, K.J., Van DerWalt, S.J., etal. 2020, Nature, 585,357
  • Haslam etal. (1982)Haslam, C., Salter, C., Stoffel, H., & Wilson, W. 1982, Astronomy andAstrophysics Supplement Series, 47, 1
  • Hu etal. (2021)Hu, W., Li, Y., Wang, Y., etal. 2021, MNRAS, 508, 2897,doi:10.1093/mnras/stab2728
  • Hunter (2007)Hunter, J.D. 2007, Computing in science & engineering, 9, 90
  • Hunter etal. (2011)Hunter, T.R., Schwab, F.R., White, S.D., etal. 2011, Publications of theAstronomical Society of the Pacific, 123, 1087
  • Iheanetu etal. (2019)Iheanetu, K., Girard, J., Smirnov, O., etal. 2019, Monthly Notices of theRoyal Astronomical Society, 485, 4107
  • Ivezić etal. (2019)Ivezić, Ž., Kahn, S.M., Tyson, J.A., etal. 2019, TheAstrophysical Journal, 873, 111
  • Jones etal. (2004)Jones, D.H., Saunders, W., Colless, M., etal. 2004, Monthly Notices of theRoyal Astronomical Society, 355, 747
  • LaPlante & Ntampaka (2019)LaPlante, P., & Ntampaka, M. 2019, The Astrophysical Journal, 880, 110
  • Lewis etal. (2000)Lewis, A., Challinor, A., & Lasenby, A. 2000, The Astrophysical Journal, 538,473
  • Li etal. (2023)Li, Y., Wang, Y., Deng, F., etal. 2023, arXiv preprint arXiv:2305.06405
  • Liao etal. (2016)Liao, Y.-W., Chang, T.-C., Kuo, C.-Y., etal. 2016, ApJ, 833, 289,doi:10.3847/1538-4357/833/2/289
  • Liao etal. (2016)Liao, Y.-W., Chang, T.-C., Kuo, C.-Y., etal. 2016, The AstrophysicalJournal, 833, 289, doi:10.3847/1538-4357/833/2/289
  • Makinen etal. (2021)Makinen, T.L., Lancaster, L., Villaescusa-Navarro, F., etal. 2021, Journalof Cosmology and Astroparticle Physics, 2021, 081
  • Mangena etal. (2020)Mangena, T., Hassan, S., & Santos, M.G. 2020, Monthly Notices of the RoyalAstronomical Society, 494, 600
  • Masui etal. (2013)Masui, K., Switzer, E., Banavar, N., etal. 2013, The Astrophysical JournalLetters, 763, L20
  • Ni etal. (2022)Ni, S., Li, Y., Gao, L.-Y., & Zhang, X. 2022, The Astrophysical Journal, 934,83
  • Paul etal. (2023)Paul, S., Santos, M.G., Chen, Z., & Wolz, L. 2023, arXiv preprintarXiv:2301.11943
  • Santos etal. (2005)Santos, M.G., Cooray, A., & Knox, L. 2005, The Astrophysical Journal, 625,575
  • Scott & Rees (1990)Scott, D., & Rees, M.J. 1990, Monthly Notices of the Royal AstronomicalSociety, vol. 247, p. 510, 247, 510
  • Shaw etal. (2014)Shaw, J.R., Sigurdson, K., Pen, U.-L., Stebbins, A., & Sitwell, M.2014, ApJ, 781, 57, doi:10.1088/0004-637X/781/2/57
  • Shi etal. (2023)Shi, F., Chang, H., Zhang, L., etal. 2023, arXiv preprint arXiv:2310.06518
  • Stone (2002)Stone, J.V. 2002, Trends in cognitive sciences, 6, 59
  • Sullivan etal. (2012)Sullivan, I.S., Morales, M.F., Hazelton, B.J., etal. 2012, ApJ,759, 17, doi:10.1088/0004-637X/759/1/17
  • Switzer etal. (2013)Switzer, E., Masui, K., Bandura, K., etal. 2013, Monthly Notices of theRoyal Astronomical Society: Letters, 434, L46
  • Switzer etal. (2015)Switzer, E.R., Chang, T.-C., Masui, K.W., Pen, U.-L., & Voytek, T.C. 2015,The Astrophysical Journal, 815, 51
  • Tharwat (2021)Tharwat, A. 2021, Applied Computing and Informatics, 17, 222
  • Van DerWalt etal. (2011)Van DerWalt, S., Colbert, S.C., & Varoquaux, G. 2011, Computing in science& engineering, 13, 22
  • Villanueva-Domingo &Villaescusa-Navarro (2021)Villanueva-Domingo, P., & Villaescusa-Navarro, F. 2021, The AstrophysicalJournal, 907, 44
  • Virtanen etal. (2020)Virtanen, P., Gommers, R., Oliphant, T.E., etal. 2020, Nature methods, 17,261
  • Wadekar etal. (2021)Wadekar, D., Villaescusa-Navarro, F., Ho, S., & Perreault-Levasseur, L. 2021,The Astrophysical Journal, 916, 42
  • Wolz etal. (2014)Wolz, L., Abdalla, F., Blake, C., etal. 2014, Monthly Notices of the RoyalAstronomical Society, 441, 3271
  • Wolz etal. (2022)Wolz, L., Pourtsidou, A., Masui, K.W., etal. 2022, Monthly Notices of theRoyal Astronomical Society, 510, 3495
  • Zhang etal. (2019)Zhang, K., Wu, J., Li, D., etal. 2019, Science China Physics, Mechanics &Astronomy, 62, 1
  • Zonca etal. (2019)Zonca, A., Singer, L., Lenz, D., etal. 2019, Journal of Open SourceSoftware, 4, 1298, doi:10.21105/joss.01298
  • Zuo etal. (2023)Zuo, S., Chen, X., & Mao, Y. 2023, ApJ, 945, 38,doi:10.3847/1538-4357/acb822
Correlation-based Beam Calibration of 21cm Intensity Mapping (2024)
Top Articles
Ati System Disorder Hypertension
Mybiglots Net Associates
Craigslist San Francisco Bay
Http://N14.Ultipro.com
Soap2Day Autoplay
Truist Park Section 135
Heska Ulite
Seafood Bucket Cajun Style Seafood Restaurant in South Salt Lake - Restaurant menu and reviews
Nier Automata Chapter Select Unlock
Chicken Coop Havelock Nc
Colts seventh rotation of thin secondary raises concerns on roster evaluation
Sams Early Hours
Magicseaweed Capitola
2024 U-Haul ® Truck Rental Review
Walmart End Table Lamps
iLuv Aud Click: Tragbarer Wi-Fi-Lautsprecher für Amazons Alexa - Portable Echo Alternative
boohoo group plc Stock (BOO) - Quote London S.E.- MarketScreener
Jinx Chapter 24: Release Date, Spoilers & Where To Read - OtakuKart
Accident On 215
Aes Salt Lake City Showdown
Mega Personal St Louis
Inbanithi Age
How To Tighten Lug Nuts Properly (Torque Specs) | TireGrades
Scripchat Gratis
Bleacher Report Philadelphia Flyers
Vht Shortener
Pokémon Unbound Starters
Things to do in Pearl City: Honolulu, HI Travel Guide by 10Best
Mastering Serpentine Belt Replacement: A Step-by-Step Guide | The Motor Guy
Gridwords Factoring 1 Answers Pdf
Aid Office On 59Th Ashland
Craigslist Gigs Norfolk
2024 Coachella Predictions
Beaver Saddle Ark
The Pretty Kitty Tanglewood
Old Peterbilt For Sale Craigslist
Moxfield Deck Builder
Federal Student Aid
House Of Budz Michigan
SF bay area cars & trucks "chevrolet 50" - craigslist
Ludvigsen Mortuary Fremont Nebraska
Barber Gym Quantico Hours
Craigslist Tulsa Ok Farm And Garden
Questions answered? Ducks say so in rivalry rout
Sams Gas Price Sanford Fl
Blue Beetle Showtimes Near Regal Evergreen Parkway & Rpx
About Us
Phmc.myloancare.com
Legs Gifs
Is TinyZone TV Safe?
Renfield Showtimes Near Regal The Loop & Rpx
Generator für Fantasie-Ortsnamen: Finden Sie den perfekten Namen
Latest Posts
Article information

Author: Tuan Roob DDS

Last Updated:

Views: 5843

Rating: 4.1 / 5 (62 voted)

Reviews: 85% of readers found this page helpful

Author information

Name: Tuan Roob DDS

Birthday: 1999-11-20

Address: Suite 592 642 Pfannerstill Island, South Keila, LA 74970-3076

Phone: +9617721773649

Job: Marketing Producer

Hobby: Skydiving, Flag Football, Knitting, Running, Lego building, Hunting, Juggling

Introduction: My name is Tuan Roob DDS, I am a friendly, good, energetic, faithful, fantastic, gentle, enchanting person who loves writing and wants to share my knowledge and understanding with you.