NRAO ALBiUS Report, April 8 2011
Report assembled by Walter Brisken
Based on input from Kumar Golap, Sanjay Bhatnagar and Jeff Kern

STATUS OF DELIVERABLES

6.2.1 “New implementation of Global Fringe Fitting algorithm”

NRAO has retained only a facilitory role in the fringe fitting development after handing this deliverable over to JIVE. NRAO hosted the visit of Stephen Bourke and Ian Stewart in Socorro in February 2011. A full report on this work will be provided by JIVE. NRAO will be happy to host a second visit of fringe fitting developers this year if deemed useful.

6.2.3 “Software for mosaic imaging including primary beam correction”

The delivery date for this milestone was Month 25 of RadioNet FP7, or February 2011. This deadline was not met, but considerable work has been made toward the goal. The key components are contained within the A-projection algorithm as described below. Code supporting A-projection will be included in CASA version 3.2, which is to be released late April or early May 2011. This code will not be directly accessable by users at this point but will be for CASA version 3.3 to be released sometime around October 2011. It is anticipated that both the spirit and the word of the goals will be fully met at this time.
Most of the work done in this cycle in the area of algorithms to correct for various Primary Beam (PB) effects (effects of wide-band sensitivity pattern, time-dependent gain variations due to rotation of the PB and pointing errors, and PB effects for heterogeneous arrays like ALMA) fall in the following categories:
1. Implementation of the software framework for the A-Projection approach to allow handling of heterogeneous arrays
2. Implementing the components of this framework for ALMA to use simulated antenna aperture illumination patterns.
3. Begin prototyping work on multi-threaded gridding utilizing multi-core CPUs resources to improve run-time performance.
Prior to re-factoring of the software framework, the implementation of the A-Projection algorithm supported only EVLA primary beams. Code required to include PBs for other telescopes, in particular ALMA, was re-factored to allow inclusion of other PBs such that only a single telescope-specific specialization of a software component needs to be written. This component for the EVLA was implement and the software framework was regressively tested against existing data from the EVLA. The ALMA-specific component has also been written by Dirk at ESO and is pending software testing. Some more work on this front for the extra bookkeeping for heterogeneous arrays might still be required to fully integrate this into the framework. Further work is planned when Dirk visits us here.
The technique for correcting antenna pointing errors utilizes the A-Projection algorithm and has a relatively high computing foot print. Work on corrections for the antenna pointing errors will follow after the work on integration of ALMA PBs and the work for deploying the A-Projection algorithm on the HPC platform.
In support of efficient use, effort towards parallelization has continued on 2 fronts: multi-process (cluster computing) and threaded (multi-core on a single machine).
Multi-process imaging for continuum and spectral cube was achieved last year. This year work on this has been refinements and bug fixing. The work this year was focussed on dealing with data-parallel optimization for problems that require minimal inter-process communication. Imaging with multi-process parallelization will work on single or multiple datasets. This infrastructure has lead to an easy path for implementing parallelization of different modules of a data reduction pipeline (e.g., flagging, calibration application or continuum subtraction) by distributing the work on different datasets to different processes.
Multi-threaded parallelization had already been applied to the time-consuming gridding process of continuum imaging. We have been experimenting with multi-threaded approach for Multiscale clean and a first version is available using OpenMP. Work is being done on non-blocking multi-threaded I/O access and multi-threaded gridding of data in imaging. A-Projection, and in general techniques for wide-band imaging, require more computing compared to standard techniques for narrow band imaging. To solve the problem of higher computing requirements, the current efforts for parallelization of CASA uses the approach of multiple processes per node on a multi-node cluster. The memory footprint for this scales linearly with the number of processes and is not indefinitely scalable, motivating parallelization across many computers within a cluster.
On the mosaicing front, we passed for the first time real interferometric on-the-fly mosac data from ALMA through CASA. The code was shown to work.

Global Fringe Fitting (GFF):

Substantive work commenced on this package in January 2011. Although in view of the short remaining time available to complete this package, the formal objectives had been scaled back to just the production of a report, it is nevertheless felt that it may still be possible to make some inroads into the initial desirables of introducing some GFF capability into CASA on the one hand and exploring some new and extended algorithms on the other.
Near the start of the work period, Stephen Bourke and Ian Stewart visited the VLAOC at Socorro. The aims of this visit were information transfer to Stephen and Ian on the one hand and discussion and agreeing on a GFF interface with CASA on the other. Both these objectives were successfully met.
Since this Socorro visit, Stephen has been working mostly on the interface with CASA. He has written a simple framework fringe-fitting solver along the lines agreed at Socorro. This has lately been tested with real data and been found to perform satisfactorily. Ian has been investigating a new algorithm, which performs a fast linear fit to the imaginary part of visibilities which have been coarsely phase-flattened, and has extended the familiar Fourier-transform algorithm to return uncertainties in the returned values of delay and rate.