A Cyberinfrastructure platform to meet the needs of data intensive radio astronomy on route to the SKA

SKA Calim 2011 - Day 1

Square Kilometre Calibration Meeting 2011 at the University of Manchester, UK.

Day 1

LOFAR Status (Ronald Nijboer) - LBA (10/30 - 80 MHz), HBA (120-240 MHz). Superterp ~ 350m, Core +18 stations ~ 3km, NL Remote ~ 80km, EU remote ~ 1000km. Currently:  2. 0cores, 7 remote, 6 EU. 102 compute nodes (2x 12 cores (2.1GHz)) with 20 TB per node. Capacity is currently 20 PB and 20 TFlops, though the space is filled in 2 weeks of observing. They also have a multi treed distributed archive (currently 700TB of data, but rapid (looks exponential) growth of data so far this year). Example of image: 3C196, 10x10deg - dynamic range of 250 000: 1. So far the ionosphere appears to not produce a huge effect on the tens of km baselines. Using AWprojection can recover IQUV to better than 1% - this is now in CASA. Interesting cosmic ray detection results (TerVeen et al.) Comissinion survey - Million Source Shallow Survey (MSSS) - used for calibration and pipelines (15mJy per field for LBA and 5mJy for HBA).

ASKAP Status (Tim Cornwell) - Full observing 2013 with 36 anteenas. 3 axis antenna - extra roation very useful, can be used up to 10GHz. 8800 core HP cluster called EPIC ~88 on top 500. 100 TFLOPS. There is a calibration database that is passed to the main pipeline.

MeerKAT Status (Ludwig Schwardt) - KAT 7 test arrays and paper site nearby. Intel sponsored, i7core 2600 3x300GB SSD. Takes in 2x400MHz bandwidth voltage streams. 10 TB archive. Apache OODT for data archive toolset. IBM infosphere streams work on RFI flagging.

MWA (Daniel Mitchell) - "cheap antennas". arcminute resolution, baselines less than a couple of km to use 2D ionospheric models. Built near ASKAP.  2 FPGA systems. arxiv:0903.1828 for their analysis pipeline. Wide-field corrections, c.f. Cornwell et al. arxiv:0807.4161. Good fit for warped snapshots. Use GPU correlator solution in their real time system. 32 IBM serves (x2x6 Xeon X5650, 2.66GHz, 2x NVIDIA Tesla Fermi GPU) - embarrassingly parallel - paralelised over frequency with MPI. Aim to get 128 tile array in the next year. Infrastructure tender in progress.

Beam Quality and Stability of PAF systems (Stefan Wijnholds) - ideal polarimetric beamformer should provide maximum sensitivity and preserver polarimetric properties of observed signal. Polarimetric behaviour over FoV is important. PAF calibration, Wijnholds et al., IEEE TAP, in prep to discuss issues. Generic model of a phased array, Ivashina, Maaskant & Woestenburg, IEEE AWPL 2008. Optimal polarimetric calibration (Warnick et al., IEEE TAP, accepted) - assume voltage response to pure u and v signal is known. use 2 by 2 beamformer matrix. Want to minimise noise and preserver polarization. Example given - APERTIF (300MHz BW - 1000-1650MHz, 8deg**2 FOV) - beam spec 1% error at HPBW rel. to main beam due to WSRT mechanical pointing error. Off axis beams have more severe error propagation.

EVLA and algorithms (Sanjay Bhatnagar) - Seamless frequency coverage from 1 to 50 GHz. Two observed modes: Open Shared Risk Observing - available to all through proposal project; Resident Shared Risk Observing. End 2012 for completion of telescope.  Wide band RFI - situation rapidly improves at you increase in frequency from L to S band - its rather strong and narrow band RFI. Early science results: RSRO project AB1345 - ApJL EVLA Special Issue, Bhatnagar, Rau, Green and Rupen in press - rms noise for 4 hours - 10microJy/beam). NRAO Algorithms dominated by: wide-band wide-field imaging for EVLA, heterogeneous array - ALMA and WIDAR correlator capabilities.  Highest EVLA image so far, dynamic range 900,000:1. EVLA survey speed is better than ASKAP and MeerKAT... maybe. All are very similar. Data volume is large, inevitable consequence of improved sensitivity. Multiscale imaging now in CASA. Algorithms work in progess. Strong overlap with EVLA and MeerKAT. Need direction dependent solvers for high dynamic range imaging. One DD-solver is a pointing error solver, doing this seperately for the L and R polarizations. JAWS: mathematicla frameworks using AW degridding. Can not ignore the off diagonal contributions of the Muller matrix - work by C. Tasse.

Main beam representation in non-regular array (Christophe Craeye) - SKA AA-lo (70-450MHz), place the elements in a non-regular pattern getting the smallest number of antennas but largest collecting area. Discussion of apertures vs arrays.

Radio interferometer calibratability and limits to polarimetry (Tobia Carozzi) - Does expensive computational power always give you the image you want to get? There is a relationship between visibilities and brightness'. Calibration could just be seen as a process of inverting gains. See: Carozzie, Woan IEEE TAP special issue "Future radio telescopes" June 2011)

Calibration pipelines ASKAP (Max Voronkov) - What needs to be calibrated? Visibilit-plane effects, image-plane effects (hopefully not), operations-specific calibration (not something we can do in real time). Calibration data - keep in a database and use the latest calibration database from reduction pipeline and apply. Assume calibration is a least square fit. We can do the calibration analytically - but painful to calculate the derviatives. Use AutoDiff and SpareDiff. 5 minute observation took 666 seconds - therefore brute force does not work. We need to keep up. Using analytical approach takes 23 seconds and only 11 seconds if leakages are not dealt with.