Challenge Abstracts
Visualizing and Reacting to Data in Real Time
Jayce Dowell, University of New Mexico
Abstract: Radio interferometers produce data at rates of 100s of Gbps to Tbps and need real-time processing to reduce the data into something more manageable. However, even this “more manageable” data volume can still require hours of additional processing into order to create images that can be viewed and used for science. More recently, progress has been made towards trying to bring real-time processing to these latter stages of data processing. This increase in the ability to process at high data rates will also allow for quick identification of problems with the instrument, on-the-fly flagging of interference, or novel searches for short duration transients in the data. In this talk I will present the Bifrost computing framework. The goal of Bifrost is to bring this type of high throughput computing users with a low barrier of entry. I will also provide a few examples of how Bifrost is used to increase the capabilities of the Long Wavelength Array.
Spatio-Spectral Modeling of Massive Spectral Data Cubes
Eric Koch, Center for Astrophysics | Harvard & Smithsonian
Abstract: The spectral data cube is a common data product between medical imaging and radio astronomy, where each spatial position contains an associated spectrum and/or polarization information. These spectral data cubes are rich data products that raise an interesting challenge for source identification and modeling due to their mixture of axis types. In radio astronomy, typical analyses of spectral data cubes use unsupervised source identification through "blob"' spatio-spectral connectivity or the modeling of individual spectra without considering spatial information. With current and next-generation radio telescopes generating increasingly complex and detailed spectra, the next step for radio astronomy is spatio-spectral modeling, whereby fitting physical spectral models also incorporate the spatial connectivity from source identification. This challenge is a clear opportunity for radio astronomy to learn from the medical imaging community, where this combination in modeling larger-than-memory spectral data cubes has been more thoroughly explored (e.g., Kelm et al. 2009). The goal of this challenge is to explore (i) applications of current spatio-spectral modeling in medical imaging to radio astronomy data; (ii) the interface between implementations of larger-than-memory data handling in both fields to facilitate complex modeling; and (iii) where standard regimes in radio astronomy data cubes (e.g., low S/N recovery; poor model selection constraints) may have applications for medical imaging.
References and relevant literature:
- Kelm BM, Menze BH, Nix O, Zechmann C, Hamprecht FA. Estimating Kinetic Parameter Maps
- From Dynamic Contrast-Enhanced MRI Using Spatial Prior Knowledge. IEEE T. Med. Imaging 2009; 28: 1534–1547. doi:10.1109/TMI.2009.2019957.
- Marchal, A., Miville-Deschenes, M.A., Orieux, F., et al. 2019, Astronomy & Astophysics, 626, A101. doi:10.1051/0004-6361/201935335
Real Time Dynamic Imaging in Radio Interferometry and MRI
Hendrik Müller, Max-Planck-Institut für Radioastronomie
Abstract: Radio interferometric observations typically stretch over longer periods of time in order to fill up the uv-plane (Fourier domain) with the help of Earth rotation. This makes it difficult to target rapidly evolving sources whose emission and structure are varying on intraday time scales. In such situations, the observation needs to be split into a number of snapshot frames with durations not exceeding a fraction of the variability time scales. This inevitably increases the noise and sparsity of the data. To address this problem, we need to design novel imaging algorithms for dynamic imaging that take the intrinsic source variability into account and overcome the extreme sparsity of short snapshots. One example of such algorithms is the imaging scheme developed by the EHT collaboration for their groundbreaking results on SgrA* which is variable on time scales down to about 20 minutes. Similar problems arise in real-time MRI as well, with frontline applications to such various fields such as cardiovascular MRI, turbulent flows or speech production. Deriving from this commonality, the project proposed here will explore areas of synergy between real-time MRI and dynamic VLBI imaging. In particular, we would like to:
- study how regularization in temporal domain can be described and implemented in various imaging approaches (e.g., RML, Bayesian, CLEAN) and applications with special emphasis on neural network aided reconstructions and compressed sensing,
- explore the synergy between parallel imaging in MRI and hybrid imaging in radio interferometry (imaging + self-calibration) with special focus on how temporal variation can be separated from varying instrumental response,
- point out system requirements and data processing strategies for short time data processing and data visualization.
Shape - Orientation - Size Conservation and Image Plane Self-Calibration
Christopher Carilli (NRAO), Nithyanandan Thyagarajan (CSIRO), Bojan Nikolic (MRAO, University of Cambridge)
Abstract: Closure phase is the phase of a closed-loop product of spatial coherences formed by a 3-element interferometer array. Its invariance to phase corruption attributable to individual array elements acquired during the propagation and the measurement processes, subsequent calibration, and errors therein, makes it a valuable tool in interferometry applications that require high-accuracy phase calibration. However, its understanding has remained mainly mathematical and limited to the aperture plane (Fourier dual of the image plane = 'visibilities'). Here, we present a geometrical, image-domain view of closure phase. Using the principal triangle in a 3-element interference image formed by a triad of interferometer elements, we show that the properties of closure phase, particularly its invariance to multiplicative element-based corruption factors (even of a large magnitude) and to translation, relate directly to the conserved properties of the triangle, namely, its shape, orientation, and size, which is referred herein as the ``shape-orientation-size (SOS) conservation principle''. In the absence of a need for element-based amplitude calibration of the interferometer array (as is typical in optical interferometry), the principal triangle in any 3-element interference image formed from phase-uncalibrated spatial coherences remains a true and fully coherent image of the source's morphology, except for an unknown overall spatial translation. Based on the triangle SOS conservation principle, we present two geometric methods to measure the closure phase directly from a 3-element interference image, without resorting to the aperture-plane: (i) the closure phase is directly measurable from any one of the triangle's heights, and (ii) the squared closure phase is proportional to the product of the areas enclosed by the triad of array elements and the principal triangle in the aperture and image planes, respectively. We validate the geometric understanding of closure phase in the image plane using observations with the Jansky Very Large Array, and the Event Horizon Telescope. These results verify the SOS conservation principle across a wide range of interferometric conditions.
Building from SOS conservation, we develop a new process of image plane self-calibration for interferometric imaging data. The SOS principle implies that closed triad images represent true images of the source brightness, modulo an unknown translation. These unknown translations are derived via cross correlation of the observed triad images with a priori model images of source brightness. After correcting for these shifts and summing, a coherent image of the source brightness is generated, recovering source structure. The process is iterative, using improved source models based on previous iterations. The method is implemented completely in the image domain, without resort to aperture plane visibilities. We demonstrate the technique using simulations in the astronomical context, including a realistic interferometric configuration and simple sources models appropriate for e.g. binary AGN or star studies, or dwarf stars hosting giant planets. The technique is generalizable to non-astronomical interferometric imaging applications across the electromagnetic spectrum, such as laboratory laser interferometry. We show that the process converges, although the convergence is slower than for aperture plane self-calibration for large-N arrays. The current process is most relevant for arrays with a small number of elements, particularly those in which measurements are made in the image-plane, and for which accurate phase calibration is required. We discuss future improvements to the process and potential application.