AIPS Developments in the Nineties

Gustaaf van Moorsel, Athol Kemball

National Radio Astronomy Observatory, Socorro

Eric Greisen

National Radio Astronomy Observatory, Charlottesville

Abstract:

The Astronomical Image Processing System () recently celebrated its year of existence. In those years, its popularity among the radio astronomical community has continued to increase. is now anticipated to be the prime data reduction system for radio interferometric data for at least another 3--5 years. This paper describes some of the more recent developments in the areas of interferometric imaging and support for NRAO's VLBA telescope.

1. Description of

The Astronomical Image Processing System (, Bridle & Greisen 1994) was developed at the NRAO in the late seventies. It was the first major system of its kind to embrace the concept of portability, by rigorously separating architecture dependent and independent code. In the past, this has helped to make the transition from VMS to UNIX with relative ease, and currently greatly simplifies the support of various flavors of UNIX. To date, is fully supported on operating systems like SunOS 4, Sun Solaris, IBM AIX, Linux, DEC OSF/1, HP-UX, and SGI Irix. Releases of take place twice per year, and each release is shipped to around 100 sites, either on tape or via ftp, and with or without executables. We estimate that over 200 sites worldwide actively run . Three non-NRAO sites, with a strong VLBA emphasis, run the ``midnight-job''. This means that apart from the latest released , they also run the test version, which is updated every night, and in which all software development takes place.

is the major package for reducing data from synthesis arrays like the VLA and VLBA. Data from the individual antennas are first correlated, resulting in complex voltage products for each antenna pair sampled at regular intervals. It takes a Fourier transform to convert these uv data to the more familiar representation in the image plane. A major fraction of the software deals with data in the uv domain (data visualization and editing, calibration, Fourier transformation). Another fraction deals with data in the image plane (deconvolution, image arithmetic, visualization), and is more similar to software found in optical image processing.

2. Interferometric Imaging

The new task IMAGR is intended to become the primary image making task in , replacing MX and several less well-known tasks. It offers all of the capabilities of the older tasks, including applying calibration to data from multi-source uv files. It does the Schwab-Cotton uv-plane subtraction form of the Clark Clean, in up to 16 fields, and can apply a variety of wide-field and wide-bandwidth corrections. The really new parts of the program lie in its ability to make 8192x8192 images, to sort the data (if needed), to use the TV display interactively, and to weight the data flexibly. Previous imaging tasks required the user to pre-sort the data, to accept poor forms of uniform weighting, and/or to put up with very inefficient multiple passes through the input data; IMAGR sorts the data if needed to avoid these things. The older tasks offered, at most, a TV display of one of the residual images and the option to terminate the Clean at the end of the current major cycle. IMAGR does its display before each major cycle, allowing the user to interact with the dirty images or the current residual images. The user may zoom and enhance the display and select both circular and rectangular Clean windows for each of the fields. The choice of field to display and window is made from a menu displayed on the TV. The menu offers numerous familiar functions including display of image values and coordinates under the cursor, zoom and pan, image intensity enhancement, pseudo-coloring, and window setting.

IMAGR offers a large number of ``knobs,'' in the form of adverbs, with which you may adjust the data weighting. We do not know what the optimum setting of the knobs might be, but we do know that they can make a significant difference in the signal-to-noise on images, can alter the synthesized beam width and sidelobe pattern, and can produce bad striping in the data when mildly wrong samples get substantially large weights. We suspect that there is no true optimum setting and that the range of good settings depends on the type of source and on the data sampling in the uv plane. IMAGR allows the user to control the size of the ``cells'' in the uv plane used for counting samples in uniform weighting. It offers both circular and rectangular functions to control how a sample is counted as a function of distance from its location in the uv plane. It allows modification of the input data weights by various exponents for counting and/or weighting and performs the usual Gaussian tapering. Finally, it offers a variation of Dan Briggs' ``robust weighting'' scheme to temper the wide divergence of weighting factors attempting to make the weights ``uniform.'' The effect of this `` ROBUSTness'' parameter on synthesized beam patterns is illustrated in the accompanying figure taken from the . IMAGR computes the effect of all of this weighting on the expected noise (compared to that expected from ``natural'' weighting) in the image and reports it in the image headers as a keyword parameter. The values of this parameter found for the beams in the Figure are shown in the accompanying tables.

 
Figure 1: Slices taken through the centers of synthesized beams for various values of the ROBUST parameter. Plot at left for a VLA A- and B-array data set, while the plot at right is for a VLBA data set. Tables give noise increase over natural weighting ( large ROBUST)
Figure 1 (left): PS 41 Kb, Figure 1 (right): PS 41 Kb

3. Object-based Fortran in

The writing of IMAGR was made easier---in part---by the use of the ``OOP'' package which first appeared in the 15OCT92 version of (Cotton 1992). This package is a set of data interface routines based on the concepts of Object-Oriented methodology and provides greatly simplified Fortran access to both the contents of data structures (images, tables and uv data sets) and operations on entire data structures (e.g., image arithmetic, uv data self-calibration). Access to components of data structures (e.g., image pixels) is provided without requiring knowledge of I/O or catalog structures. Performance of these routines is generally comparable to the equivalent in standard , in part because the real computational loads are handled by the same bottom-level routines. The 15JAN96 version of now contains 35 tasks written using this package.

The OOP package does operations on images, including image arithmetic (), FFTs of images, convolutions, deconvolution (both image-plane and uv-subtraction type Cleans including round Clean windows), interpolation (including geometric conversions/corrections), and operations on complex images (allows for ``real'' only images). Table operations include sort and merge contents of a table. uv-data operations include application of calibration and editing on read, time averaging, imaging (DFT or FFT with optional uniform weighting), self-Calibration, sorting, and arithmetic () with Fourier transform of an image model (including the effects of first-order bandwidth smearing and/or the frequency-dependent primary beam shape).

4. VLBI-Specific Applications

Consistent with the aim of supporting general radio-interferometric calibration and imaging, allows the reduction of data observed using the technique of Very Long Baseline Interferometry (VLBI). These arrays differ from connected-element instruments such as the Very Large Array (VLA) in that the antennas are separated by large distances and the signals are recorded for later, rather than real-time, correlation. This places greater demands on calibration both before and during imaging.

The aim of the VLBI software within is to support full astronomical reduction of correlated VLBI data, from calibration in the uv-plane through final imaging and image analysis. At present this support includes data correlated at the MKII, MKIII or Very Long Baseline Array (VLBA) correlators. This covers most astronomical data observed by the VLBA, the European VLBI Network (EVN) or other global networks recording in the standard formats. The primary emphasis is the calibration and imaging of astronomical VLBI data, as opposed to geodetic analysis.

The VLBI capability is fully integrated into and relies on the common infra-structure provided by the design for general interferometric reduction. This is particularly true for data examination, imaging and image analysis. VLBI data conform to the internal data format used for general interferometric data and are accessible to all tasks which act on such data. All VLBI software is coded subject to the same standard and thus meets the portability standards set by the package as a whole.

The tasks specific to VLBI are predominantly concerned with the loading, editing and calibration of VLBI data in the uv-plane. Both spectral-line and continuum data reduction are supported, as well as polarization calibration and imaging. Separate tasks exist to load data from each of the correlators mentioned above, and to make the digital corrections appropriate to the specific instrument. The data can be examined using a range of data listing or graphing tasks and editing can be performed using an interactive graphical editor or in a batch mode. Further software exists to read and apply external a priori amplitude calibration, solve for the instrumental bandpass response, remove residual fringe-rates and delays and perform self-calibration imaging, amongst others. Separate tasks exist specific to spectral-line calibration and polarization calibration. Almost all calibration is implemented in tables separate from the underlying uv data, in keeping with the general calibration formalism.

Several new tasks have been added to recently to support the reduction of data from Space VLBI. The first orbiting VLBI satellite (VSOP) is expected to be launched in 1996 and customized fringe-fitting and model-fitting tasks have been written which address the specific challenges of calibrating such data. This is an active area of work within at present. The advent of the VLBA has also precipitated the development of new VLBI software within , and the present emphasis is on making the reduction of VLBA data as routine and accessible as possible with the aim of increasing data throughput and expanding the size and range of the VLBI user community.

References:

Cotton, W. D. 1992, Object-Oriented Programming in Fortran, Memo 78