JIVE Wiki

The DS5-EMBRACE [1] prototype production of an SKA concept for the aperture array station is essentially a square layout of Vivaldi elements of single-polarization. The element beam is designed to include cross-talk among adjacent units in its response. Data from actual EM-simulations at ASTRON were provided upon request, with detailed plots on the nature of the station-level beam, and MATLAB [2] codes for generation of the same for the 2D case. Equipped with these details, a model for the element beam was derived in the form of a power-law expression. Initial modeling was performed extensively in the IDL [3] environment, and subsequently a robust implementation was made in MEqTrees. Figures 1(a,b) depict the results from this simulations exercise – a comparison of the EMBRACE-tile simulated to the station level in IDL, and reproduction of the results in a MeqTrees environment is remarkable to the finest detail.

(a)	(b)
Fig. 1 MeqTrees visualizations of SKA station beam, pointed to zenith, incorporating details of the EMBRACE element pattern

Interactions with the DS2-T2 group of SKADS at MPIfR, Bonn [4], to get a detailed acquaintance with the array configuration studies being pursued there, and the means of fine-tuning the activities to better suit portability of their software package into the larger MEqTrees framework, were conducted.

Fig. 2 MeqBrowser run post-integration of the MPIfR module depicted as a screen-shot. The MeqBrowser predominates the view, with the G-Jones plots on the right, the run-time menu window in the mid-foreground, the simulations script behind it, and the trees / nodes created as listed on the left frame. The two options to validate the array invoke a C++ script external to the browser, results of which are seen appearing on the lower-left console window

A greater hands-on knowledge had been gained out of this, and considerable feedback advanced towards refining the scheme of recommendation [5] provided from such a study in application to generic array configurations as warranted by the science problem being pursued. The configuration studies for the SKA station layout that are being pursued at Bonn were then integrated to the larger simulations pipeline deployed at JIVE (Figure 2).

Simulations of the radio sky in the mid-frequency range currently make extensive use of the MEqTrees software. One particular stumbling block in the simulations exercise has been the computational taxing involving processors and memory utilization. A simple inclusion of the station beam (E-Jones) apart from the visibility kernel matrix (K_p) had led to a stretch on computational power currently available for the majority of users.

In an 2008-Q1 effort to benchmark this performance on a conventional hardware platform that runs MEqTrees simulations (Table 1), the processor clock-rate and memory usage were assessed for progressively upscaled quantities of observing stations, frequency channels over the available bandwidth, and the radio sources simulated. Inclusion of a subset of corrupt influences on the radio signals in the image & uv plane was also made a part of the exercise.

Fig. 3 Single frequency channel AA simulation results – point-source counts vs. simulator run-time, for a 30-station compact array configuration; dotted lines are extrapolations in the computationally limiting regime of MEqTrees simulator

The method involved progressive addition of station, source and frequency channel numbers in the simulations, which then found the processor and memory wanting when further enhancements to combined larger configurations of the variables was attempted.

This particular task had been crucial in that, results from this led to a considerable understanding of the scale to which complexity could be attempted regarding simulations.

A definite requirement for parallelization of the simulation algorithms in a manner that would agree with similar yet large-scale exercises to be pursued in course of time later has been advanced, while the alternative would be to downsize the exercise to execute flawlessly on current infrastructure. One particular result from this exercise is depicted in Figure 3. This quantitative benchmarking study on MeqTrees' performance has subsequently led to a revamping of the application with several fixes, automated tests for system-wide validation of scripts, accent on multi-threaded implementations of simulation code, and optimizations like message-passing interface (MPI) for over-sized simulation runs.

We wish to thank the MEqTrees developers team of Jan Noordam, Oleg Smirnov, et al. at ASTRON, Dwingeloo, The Netherlands.

[1] http://webmail.jb.man.ac.uk/skadswiki/DesignStudy5
[2] http://www.mathworks.com
[3] http://www.ittvis.com
[4] http://www.mpifr.de
[5] Lal, D.V., Lobanov, A.P., Array configuration studies for the SKA - Implementation of figures of merit based on SDR, SKADS Memorandum S20 (http://www.skads-eu.org/p/memos.php) & http://arxiv.org/abs/0810.0943v1