COAST Wiki

This is all based on information reported in June 2010.

Working group representatives:

WG2 (source-finding): George Hobbs, possibly Zaven Arzoumanian
WG4 (hardware/tied-array beams): Aaron Berndsen, Willem van Straten
WG5 (data formats and access): Willem van Straten, David Champion, Aaron Berndsen

Objectives to May 2010 and progress:

Establish formal interactions with EMU, VAST and CRAFT teams to work toward understand their point-source selection criteria (Feb 2010): George is interacting with EMU via a student.

Develop population/detection simulation software for point- source and UV-search modes. To that purpose a population synthesis code is being developed (largely based on Faucher-Giguere & Kaspi 2006). The code simulates the observable properties and sky distribution of Galactic pulsars, including the recycled, millisecond pulsars. The software constrains the best observing parameters, measured in number of pulsars found vs computing and time requirements, given the detection limits and observational biases inherent to the telescope design and survey configuration.

Liaise with POSSUM to identify common needs for point-source calibration (start in January 2010) (May 2010): Not really accomplished.

Work with ATNF to identify main issues related to calibration of polarised point sources, both on- and off-axis, with an FPA (May 2010): George et al have done some observing with the FPA on the 12m testbed antenna; not sure about calibration work, although PDFB2 was used.

Test (FPA) polarisation observations on Parkes testbed antenna (May 2010): same comment as previous

Report on GPU-based baseband processing system at Parkes (May 2010):

At Swinburne, we have extended our primary digital signal processing software (dspsr.sourceforge.net), to use the Compute Unified Device Architecture (CUDA) developed by NVidia. In collaboration with the Center for Astronomy Signal Processing and Electronics Research (CASPER) at Berkeley, we have completed the development of a new GPU-based baseband recording and processing system, the CASPER Swinburne Parkes Recorder (CASPSR). This instrument performs real-time phase-coherent dispersion removal using four server-class workstations, each equipped with two NVidia Tesla C1060 GPUs. One of the first-light observations is shown in Figure NN.

Figure NN: caspsr_bw.eps sent in email --- start caption --- Five-minute integration of PSR B1937+21 observed with CASPSR. This high-$DM$ ($\sim 71$ pc cm$^{−3}$) small-$P$ ($\sim 1.56$ ms) pulsar is a standard benchmark of phase-coherent dispersion removal fidelity. The top panel plots the mean pulse profile integrated over the entire 400~MHz band and the bottom panel shows the effects of interstellar scintillation as a function of frequency with 500~kHz resolution. The upper 62.5~MHz of the band (not shown) are filtered to eliminate a strong source of radio frequency interference. --- end caption ---

GPU-based pulsar instrumentation affords significant savings in capital cost as well as energy and cooling requirements, as demonstrated by the benchmarks shown in Figure NN+1. Whereas over 10 of the current best Intel-based machines are required, only four NVidia Fermi GPUs can process a 512~MHz band in real time. [WvS: I’m not certain that this represents a reduction in Watts]

Figure NN+1: cpu_bench.eps and fermi_bench.eps --- start caption --- Processing time to real time ratio as a function of dispersion measure $DM$. The centre frequency $\nu$ varies from 375~MHz (bottom panel) to 3~GHz (top panel). In the bottom panel, the bandwidth $\delta\nu$ varies as $\delta\nu=8$\,MHz (triangle), $\delta\nu=16$\,MHz (cross), $\delta\nu=32$\,MHz (square), and $\delta\nu=64$\,MHz (circle). Each symbol corresponds to the same relative bandwidth $\delta\nu/\nu$ in each panel. Processing time is the total time required to perform all steps in the signal processing path typically used for pulsar timing observations. The benchmarks on the left were performed on a workstation with dual \intel\ \xeon\ E5520 (Nehalem-EP) processors, each with 4 cores running at 2.26\,GHz, 8\,MB of \intel\ Smart Cache, and 24\,GB of RAM. The benchmarks on the right were performed on a similar machine equipped with an NVIDIA Fermi architecture Tesla C2050 general purpose graphics processing unit (GPGPU). --- end caption ---

Identify potential (still preliminary) fast-UV-dump disk/memory storage hardware solutions (May 2010): delaying this as there’s no point buying anything now. Aaron Berndsen’s task.

Identify potential funding sources for on-site coherent dedispersion hardware (May 2010). Identify potential funding sources for GPU-based processors of filterbanked data (May 2010): Michael Kramer to discuss with Ingrid in July. Need to have broad discussion, likely via telecon, say in mid-July

Report on millisecond pulsar polarimetry experiment at GMRT (May 2010): no progress to report

Identify existing (suitable, unsearched) point-source catalogs, and propose searches based on preliminary source-identification techniques (May 2010): George and student are looking at ATLAS for this purpose, and Zaven was going to see how he should contribute.

Match existing COAST teams to ASKAP working groups (Feb 2010): Essentially done

Set up COAST wiki and email exploder (Feb 2010): More or less done!

Establish communication channels with ATNF staff in all areas that require close communication (simulations, polarisation/calibration, hardware, archiving) (Feb 2010): This is basically the Working Groups.

Set up COAST public webpage (Feb 2010): Erm, not yet.

Start to define publication/follow-up policies and priorities (May 2010): not yet started

Internal progress review (May 2010): In progress

Define potential student projects (Feb 2010): not so well accomplished, although effectively some projects have started!

Other things that people have reported

UBC: Aaron Berndsen has been working on simulating ASKAP pulsar signals. The simulations currently generate UV rpfits data for tracked sources, and may generate searchable time series in psrfits format. Current efforts focus on the math for a possible search in the UV plane itself (skipping the mapping stage), though he is also testing map-searching plans

ATNF: (George reports) The PULSE@Parkes outreach project has always been designed as a test for outreach on ASKAP. We have a science paper published and have just submitted an analysis of the educational value of the project to the International Journal of Science Education. That paper is full of references to ASKAP etc. Also, ASKAP will be finding pulsars and we have many groups who have developed outreach activities based on pulsar searching (e.g. UTB, pulsar collaboratory ….) I think that Marta should simply coordinate activities and find out from each project what works and what doesn’t. Maybe we should also invite Rob Hollow to join COAST (and maybe lead the outreach activities)

ICRAR: (Richard Dodson) The CRAFT Incoherent GPU transient search engine is working and has been tried on Parkes High Time Res (64usec) data with 1024 channels. On an old card (2/3 years old) we achieve half real-time. Further development is required in more effective peak-detection and the managing of multiple kernels (for the more modern `double-headed’ cards).

Swinburne: (via George) Swinburne has been working with John Reynolds and Keith Bannister to set up an IBOB-based filterbank system for regular monitoring observations on the 12-m ASKAP prototype dish at Parkes. [WvS not sure about this: Somebody somewhere knows whether they cause RFI issues, what’s good and what isn’t. What are problems with these systems on remote sites.] Benchmarking tests exist that could be run for ASKAP.