21 cm cosmology with radio interferometers

The Universe expanded from an infinitely dense and hot singularity, the Big Bang, roughly 13.7 billion years ago. The most recent observations have shown that this expansion is even accelerating. The energy that fuels the acceleration is known as ''dark energy''. Dark energy makes up most of the mass of galaxies and affects the Universe on the largest scales. But little is known about its nature and origin. One of the most promising way of studying dark energy is the 21 cm hydrogen line intensity mapping. Hydrogen is the most commen ingredient of our Universe and hydrogen can be found in galaxies and galaxy clusters. Observing hydrogen gas in distant galaxies will map the large-scale structure and the expansion history of the Universe. CHIME (Canadian Hydrogen Intensity Mapping Experiment) is a novel Canadian radio telescope designed to observe hydrogen gas in distant galaxies that carries information about dark energy. Studying the formation of the first stars and galaxies in the infant Universe is my main research interest.


Timeline of the Universe (Credit: NASA/WMAP).


Calibrating and processing radio astronomy data

Radio interferometers are new-generation radio telescopes. Typically, the resolution of a dish telescope is decided by the size of the dish. The larger the dish size, the higher the resolution. Radio interferometers are composed of simple antennas. The signals received by individual antennas are combined by supercomputers. The maximum resolution of an interferometer is decided by the maximum distance between antennas. This way, an interferometer can mimic a large dish telescope and effectively increase its resolution without actually building an expensive large dish telescope. Interferometers are often referred to as "software telescopes" due to the heavy calculations involved in data processing. One of my research topics are improvement of calibration and studying artefacts in the radio astronomy data. I studied the correlations between the excess variance and various artefacts in LOFAR-EoR power spectra (H.Gan et al, 2022a) and the impact of direction dependent calibration methods on LOFAR EoR data (H.Gan et al, 2022b). Currently, I am working on improvement of data processing pipeline for CHIME, mainly focusing on RFI, focal line expansion and foreground removal.


Exploring big data using computer vision and machine learning

Radio telescopes produce a massive amount of data each day. Efficiently finding patterns in the data is crucial for further analysis. I am developing a general purpose data analysis tool for big data sets. Especially, I am interested in exploring connected components in Max-tree data structures, using dimension reduction methods such as self-organising maps (SOMs) and t-distributed stochastic neighbor embedding (tSNE): Self-organising vector attribute maps and pattern spectra. Combining the two concepts enables the automatic exploration of morphological features and increases the interpretability of training output. The application is not only limited to radio astronomy images. My new data analysis tool is tested on three-dimensional PET scans and satelite images (H.Gan et al, submission in prep.).