Epoch of Reionization


Members of the EoR Science working group came together in Pisa in March to discuss what is needed to extract reionization science from SKA1-low. Meeting in the ornate surroundings of the Scuola Normale Superiore (SNS), about 30 scientists from around the world assembled for this three day meeting. The discussion was dominated by establishing the key techniques needed for an end-to-end pipeline to analyse SKA1-low observations, with the priority being to identify existing algorithms and techniques that could be connected to provide a simple mock SKA pipeline.

Fig1 - Conference_photo_IMG_5410Figure 1: Meeting of EoR scientists at Scuola Normale Superiore in Pisa. March 2017.

Concrete progress to this goal was made, a consequence of the significant advances in EoR data analysis over the last few years, and a goal of a first mock data challenge was set, aiming to mimic SKA observations of the 21cm signal, without foregrounds, and the recovery of the input parameters. Another topic of discussion was the SKA cost control exercise with members being tasked to work on the likely science implications ahead of SKAO science meetings planned for late May and thereafter.


Alongside the SKA precursors MWA and HERA, LOFAR is one of the low-frequency radio telescopes attempting to detect the redshifted 21cm signal from the epoch of reionization. In March, LOFAR posted their first upper limits on the Epoch of Reionization 21-cm power spectrum using the LOFAR High-Band Antenna (HBA), which are comparable to the best currently achieved. These first results make use of 13 hours of data from observations centred on the North Celestial Pole (NCP). HBA frequencies of 121.8 – 159.3 MHz probe redshifts z=7.9 – 10.6, which correspond to the late stages of reionization close to the values favoured by the Planck optical depth measurement. Although the 2-sigma upper limit of (80 mK)2 is still considerably larger than the expected EoR signal, as expected for a single night of data, they already approach the limits of more extreme models without heating.

Analysis of the LOFAR data is a complex process requiring careful calibration of the instrument and removal of approximately 20,000 sources above a few mJy included in the sky model, as well as remaining diffuse emission. Calibration is one of the key challenges for these EoR experiments and the LOFAR method has considerably evolved and improved over several years. The calibration process begins with a direction independent step, which solves for gains and also absorbs much of the structure of the bandpass response of the stations. To reach the dynamic range required for EoR, this must then be followed by a direction dependent step, which is needed to handle the wide field of view (~10deg), variation of the ionosphere, and other effects, like beam errors, not adequately captured by the first step.

To account for challenges in the calibration, LOFAR baselines are split into “long baselines”, used for calibration, and “short baselines” (50-250 wavelengths) used for the EoR analysis. This approach helps addressing problems of suppressing diffuse structures in the signal, associated with using a sky model based only on point sources and devoid of diffuse components, but at the cost of introducing some excess noise into the short baselines.

This process results in 3 deg x 3 deg boxes at 0.5 arcmin resolution split into three frequency bands between 121.8 – 159.3 MHz. Along with the desired EoR signal, this cube still contains significant residuals from unremoved point sources and errors in the calibration (see Figure 2). To further remove this contamination, a blind signal separation algorithm – Generalised Morphological Component Analysis (GMCA) – is applied. This attempts to separate different components of the signal by working in a wavelet basis and seeking to maximize the statistical independence of a small number of components. This has been shown to work well in simulations at removing galactic radio foregrounds and can also identify unremoved point sources and other errors.

Fig2 - panel-IV-sub

Figure 2: Stokes I and Stokes V images after sky-model subtraction for the baseline ranges 30-800l (top panels) and 50–250l (bottom panels). Sub-bands with frequencies between 121 and 134 MHz went into these images. Note the reduction in the displayed field-of-view from 20° x 20° to 10° x 10°. Intensity units are in mJy/PSF and the scale range is set by plus and minus three times the standard deviation over the full field in all images. Note the noise-like structure in the two Stokes V images. i.e. a lack of any features. The Stokes I images, on the other hand, clearly show the LOFAR-HBA primary beam attenuation effects on the remaining diffuse emission. The level of this emission is limited by the classical confusion noise within the primary beam. The 3° x 3° box delineates the area where LOFAR measures the power spectra.

After this cleaning, the data cube resembles noise and can be weighted appropriately to estimate the 21-cm power spectrum. The Stokes-V power spectrum provides a useful estimate of the thermal noise level and can be compared to the Stokes I power spectrum (see Figure 3), which should contain the EoR signal. At this stage, the Stokes-I power spectrum is typically about 2-3 times that of Stokes V. Unfortunately, this is not an indication of an EoR detection, but more likely some other source of excess noise. The exact source is still unclear, but contributions may come from sidelobe noise due to an incomplete sky model, the effect of ionospheric noise, or the effect of only using the long baselines for calibration.

Fig3 - pspec1d_L90490_3redshiftsFigure 3: The spherically averaged Stokes I and V power spectra after GMCA for L90490; The mean redshifts are indicated in the panels.

The smallest value (2-sigma upper limit) in the Stokes-I power spectrum is (80 mK)2 at k=0.053 h cMpc-1 in the redshift range z=9.6-10.6. This is still 2-3 orders of magnitude larger than the expected EoR signal of ~1 mK2, but represents a competitive upper limit. PAPER (within 1150 hours) achieved a limit of (22 mK)2 at k=0.15-0.5 h cMpc-1, at z=8.4, while MWA (within 32 hours) achieved (164 mK)2 at k=0.27h cMpc-1, at z=7.1. All of these numbers are comparable and although still large, already start to make contact with some theoretical models in which the intergalactic medium (IGM) has cooled without heating, maximizing the 21cm signal.

The LOFAR analysis was conducted on only 13 hours and represents a first pass through the data. Some 2500 hours of data have been collected with LOFAR on both the NCP and 3C196 fields. Improvements in the analysis should allow more of this data to be reduced leading to much improved limits in the near future. The increasing maturity of the EoR analysis pipeline for LOFAR and the SKA precursors is now starting to address some of the crucial questions for how SKA EoR science will be done and what are the key challenges still to be overcome.

Upper limits on the 21-cm Epoch of Reionization power spectrum from one night with LOFAR

A.H. Patil, S. Yatawatta, L.V.E. Koopmans, A.G. de Bruyn, M. A. Brentjens, S. Zaroubi, K. M. B. Asad, M. Hatef, V. Jelic, M. Mevius, A. R. Offringa, V.N. Pandey, H. Vedantham, F. B. Abdalla, W. N. Brouw, E. Chapman, B. Ciardi, B. K. Gehlot, A. Ghosh, G. Harker, I. T. Iliev, K. Kakiichi, S. Majumdar, M. B. Silva, G. Mellema, J. Schaye, D. Vrbanec, S. J. Wijnholds; The Astrophysical Journal, 838:65 (16pp), 2017 March 20

Report provided by the EoR SWG