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

Day 3

Warped snapshot imaging for low-frequency dipole arrays (Daniel Mitchell) - direction dependent ionosphere can use facets - with the MWA they can use warped snapshots as the array is small and can look at the ionosphere in the image plane. Can use peeling. See Mitchell  et al. arxiv:0807.1912 - discussion of real time imaging in comparison to the offline ones (e.g. Bhatnagar, et al., arix:0805,0834). CUWARP: CUDA-based wide-field array processor, based on warped-snapshot imaging. Interesting discussion of the future Nvidia cards and how much power the correlator will use (MW). How todo optimal imaging.

Fast W-Projection Gridding on GPUs (John Romein) - use GPUs to get lots of FLOPS - avoid adding to memory in gridding. GPUs are highly parallel, device memory that have "limited" bandwidth - i.e. does not scale with the number of FLOPS. The PCIexpress bus can be a bottleneck but not for gridding? Main goal not to go to memory all the time, idea was to accumulate additions in register. The prototype works on fake data. Performance different betwen CPU and CPU is remarkable, x37 better than a 2 CPU system. CUDA >> OpenCL. Faster than other GPU gridders (van Amesfoort et al.; MWA gridder; Humphreys & Cornwell (SKA memo 132).

Imaging using the GPU (Vamsi Krishna Veligatla) - LOFAR EoR GPU cluster, 2xTesla M1060 GPU, 240 cores @602MHz and 4GB ram. MVDR - minimum variance distortionless response.

Deep observations with LOFAR (Panos Labropoulos) - motivation: epoch of reionization - need to get better sensitivity - look at same field for 100 nights. Have a diagnostic database that contains metadata about the observations. Observe 3C 196. Get dynamic range of 200,000:1 with LOFAR. Direction Dependence, program X and program Y ?? Using Sagecal.  LOFAR data looks like it has been distorted. Direction dependent calibration helps massively to suppress the sidelobes.