Accelerated implementations of the RIME for DDE calibration and source modelling
- Authors: Van Staden, Joshua
- Date: 2021
- Subjects: Radio astronomy , Radio inferometers , Radio inferometers -- Calibration , Radio astronomy -- Data processing , Radio inferometers -- Data processing , Radio inferometers -- Calibration -- Data processing
- Language: English
- Type: text , Thesis , Masters , MSc
- Identifier: http://hdl.handle.net/10962/172422 , vital:42199
- Description: Second- and third-generation calibration methods filter out subtle effects in interferometer data, and therefore yield significantly higher dynamic ranges. The basis of these calibration techniques relies on building a model of the sky and corrupting it with models of the effects acting on the sources. The sensitivities of modern instruments call for more elaborate models to capture the level of detail that is required to achieve accurate calibration. This thesis implements two types of models to be used in for second- and third-generation calibration. The first model implemented is shapelets, which can be used to model radio source morphologies directly in uv space. The second model implemented is Zernike polynomials, which can be used to represent the primary beam of the antenna. We implement these models in the CODEX-AFRICANUS package and provide a set of unit tests for each model. Additionally, we compare our implementations against other methods of representing these objects and instrumental effects, namely NIFTY-GRIDDER against shapelets and a FITS-interpolation method against the Zernike polynomials. We find that to achieve sufficient accuracy, our implementation of the shapelet model has a higher runtime to that of the NIFTY-GRIDDER. However, the NIFTY-GRIDDER cannot simulate a component-based sky model while the shapelet model can. Additionally, the shapelet model is fully parametric, which allows for integration into a parameterised solver. We find that, while having a smaller memory footprint, our Zernike model has a greater computational complexity than that of the FITS-interpolated method. However, we find that the Zernike implementation has floating-point accuracy in its modelling, while the FITS-interpolated model loses some accuracy through the discretisation of the beam.
- Full Text:
- Date Issued: 2021
Design patterns and software techniques for large-scale, open and reproducible data reduction
- Authors: Molenaar, Gijs Jan
- Date: 2021
- Subjects: Radio astronomy -- Data processing , Radio astronomy -- Data processing -- Software , Radio astronomy -- South Africa , ASTRODECONV2019 dataset , Radio telescopes -- South Africa , KERN (omputer software)
- Language: English
- Type: text , Thesis , Doctoral , PhD
- Identifier: http://hdl.handle.net/10962/172169 , vital:42172 , 10.21504/10962/172169
- Description: The preparation for the construction of the Square Kilometre Array, and the introduction of its operational precursors, such as LOFAR and MeerKAT, mark the beginning of an exciting era for astronomy. Impressive new data containing valuable science just waiting for discovery is already being generated, and these devices will produce far more data than has ever been collected before. However, with every new data instrument, the data rates grow to unprecedented quantities of data, requiring novel new data-processing tools. In addition, creating science grade data from the raw data still requires significant expert knowledge for processing this data. The software used is often developed by a scientist who lacks proper training in software development skills, resulting in the software not progressing beyond a prototype stage in quality. In the first chapter, we explore various organisational and technical approaches to address these issues by providing a historical overview of the development of radioastronomy pipelines since the inception of the field in the 1940s. In that, the steps required to create a radio image are investigated. We used the lessons-learned to identify patterns in the challenges experienced, and the solutions created to address these over the years. The second chapter describes the mathematical foundations that are essential for radio imaging. In the third chapter, we discuss the production of the KERN Linux distribution, which is a set of software packages containing most radio astronomy software currently in use. Considerable effort was put into making sure that the contained software installs appropriately, all items next to one other on the same system. Where required and possible, bugs and portability fixes were solved and reported with the upstream maintainers. The KERN project also has a website, and issue tracker, where users can report bugs and maintainers can coordinate the packaging effort and new releases. The software packages can be used inside Docker and Singularity containers, enabling the installation of these packages on a wide variety of platforms. In the fourth and fifth chapters, we discuss methods and frameworks for combining the available data reduction tools into recomposable pipelines and introduce the Kliko specification and software. This framework was created to enable end-user astronomers to chain and containerise operations of software in KERN packages. Next, we discuss the Common Workflow Language (CommonWL), a similar but more advanced and mature pipeline framework invented by bio-informatics scientists. CommonWL is supported by a wide range of tools already; among other schedulers, visualisers and editors. Consequently, when a pipeline is made with CommonWL, it can be deployed and manipulated with a wide range of tools. In the final chapter, we attempt something unconventional, applying a generative adversarial network based on deep learning techniques to perform the task of sky brightness reconstruction. Since deep learning methods often require a large number of training samples, we constructed a CommonWL simulation pipeline for creating dirty images and corresponding sky models. This simulated dataset has been made publicly available as the ASTRODECONV2019 dataset. It is shown that this method is useful to perform the restoration and matches the performance of a single clean cycle. In addition, we incorporated domain knowledge by adding the point spread function to the network and by utilising a custom loss function during training. Although it was not possible to improve the cleaning performance of commonly used existing tools, the computational time performance of the approach looks very promising. We suggest that a smaller scope should be the starting point for further studies and optimising of the training of the neural network could produce the desired results.
- Full Text:
- Date Issued: 2021
Machine learning methods for calibrating radio interferometric data
- Authors: Zitha, Simphiwe Nhlanhla
- Date: 2019
- Subjects: Calibration , Radio astronomy -- Data processing , Radio astronomy -- South Africa , Karoo Array Telescope (South Africa) , Radio telescopes -- South Africa , Common Astronomy Software Application (Computer software)
- Language: English
- Type: text , Thesis , Masters , MSc
- Identifier: http://hdl.handle.net/10962/97096 , vital:31398
- Description: The applications of machine learning have created an opportunity to deal with complex problems currently encountered in radio astronomy data processing. Calibration is one of the most important data processing steps required to produce high dynamic range images. This process involves the determination of calibration parameters, both instrumental and astronomical, to correct the collected data. Typically, astronomers use a package such as Common Astronomy Software Applications (CASA) to compute the gain solutions based on regular observations of a known calibrator source. In this work we present applications of machine learning to first generation calibration (1GC), using the KAT-7 telescope environmental and pointing sensor data recorded during observations. Applying machine learning to 1GC, as opposed to calculating the gain solutions in CASA, has shown evidence of reducing computation, as well as accurately predict the 1GC gain solutions representing the behaviour of the antenna during an observation. These methods are computationally less expensive, however they have not fully learned to generalise in predicting accurate 1GC solutions by looking at environmental and pointing sensors. We call this multi-output regression model ZCal, which is based on random forest, decision trees, extremely randomized trees and K-nearest neighbor algorithms. The prediction error obtained during the testing of our model on testing data is ≈ 0.01 < rmse < 0.09 for gain amplitude per antenna, and 0.2 rad < rmse <0.5 rad for gain phase. This shows that the instrumental parameters used to train our model more strongly correlate with gain amplitude effects than phase.
- Full Text:
- Date Issued: 2019