EveryBeam Testing and Development

Click these links to go to notebooks of integration tests

• Defining and applying parallactic rotations
  • The beam Jones matrix
  • Parallactic rotations
  • Putting things into practice
• EveryBeam MWA integration tests
  • What does the beam look like on the sky?
  • EveryBeam MWA
  • Stokes recovery
  • Now test an RM recovery
  • Test antenna locations
  • Test source positions
  • Try off zenith pointings
  • Profiling the code
  • Interpolated beam
• EveryBeam LOFAR LBA integration tests
  • Image the beam on the sky
  • Off cardinal dipoles
  • Stokes recovery with off cardinal dipoles
  • RM recovery
  • Test unique primary beam per station
  • Check positions of components via image
  • LoBES vs Hamaker Beams
• EveryBeam LOFAR HBA integration tests
  • Image the beam on the sky
  • Stokes recovery with off cardinal dipoles
  • RM recovery
  • Test unique primary beam per station
  • Check positions of components via image
• EveryBeam OSKAR “MWA” integration tests
  • What does the beam look like on the sky?
  • Stokes IQUV recovery
  • Now test an RM recovery
  • Test antenna locations
  • Let’s see if we can make flagged dipole beams
• EveryBeam OSKAR SKA integration tests
  • Image the beam on the sky
  • Stokes recovery
  • RM recovery
  • Test unique primary beam per station
  • Let’s make an image

Installing EveryBeam

The build instructions for EveryBeam live on the everybeam insallation page. However, you need to clone my branch as currently it’s the only branch with an MWA Python wrapper:

$ git clone -b mwa_python_wrapper --recursive -j4 https://git.astron.nl/RD/EveryBeam.git

When I ran cmake, I did:

cmake .. -DCMAKE_INSTALL_PREFIX=/home/jline/software/installed/ \
  -DBUILD_WITH_PYTHON=ON

so I would know where the installation went. I also did all of this inside a conda environment called woden_dev, but I think cmake still found the system Python. Once installed, I had to do:

export PYTHONPATH=$PYTHONPATH:"/home/jline/software/installed/lib/python3.12/site-packages"
ln -s /home/jline/software/installed/share/everybeam /home/jline/software/anaconda3/envs/woden_dev/share/everybeam

which let me conda environment see everything it needed to.

When I was running notebooks, which don’t load stuff from system, only the conda environment, I had to do:

conda install -c conda-forge libstdcxx-ng
conda install hdf5

for certain things to work.

Work still to be done on EveryBeam in WODEN

• I’ve managed to read the antenna locations from a measurement set, needed to calculate u,v,w coords. But I haven’t worked out where the central latitude/longitude are stored in the measurement set. At the moment you have to put that in via command line arguments. Something to figure out.

• You can only run one WODEN frequency band at a time. This is because WODEN is designed to calculate the beam values on GPU, iteratively band by band. It’s not the end of the world but a slight design flaw.

• Only certain frequencies exists in certain models. This means we don’t need to call EveryBeam for every single frequency. So create some kind of cache or map for the existing beam models to save compute.

• Things are threaded over EveryBeam now, but we still need more optimisation. CPU-only version?

• Unit tests. Mostly everything is an integration test at the moment. Best practice is to break down the testing into units.

Adding new beam models

You need to update the following Python functions to add a new EveryBeam model (in general, look for anywhere EB_OSKAR is already used, and make sure whatever is done with EB_OSKAR is also done with your new beam model):

• Update the help for --primary_beam in the parser defined in WODEN/wodenpy/wodenpy_setup/run_setup.get_parser to show a new recognised beam model. A.k.a add a “my_new_beam” as an accaptable argument to --primary_beam.

• Add this “my_new_beam” to the eb_args list in WODEN/wodenpy/wodenpy_setup/run_setup.check_args.

• Add a new beamtype to the BeamTypes and BeamGroups enums in WODEN/wodenpy/use_libwoden/beam_settings.py. You must also update the equivalent e_beamtype in WODEN/include/woden_struct_defs.h. E.g. MWA already has a value of EB_MWA, so try adding in a new EB_YOURBEAM. If the beam model you add in has off-cardinal dipoles, you should add in to the BeamGroups.off_cardinal_beam_values list.

• Make sure your beam model is added to woden_settings in the WODEN/wodenpy/use_libwoden/woden_settings.create_woden_settings function.

• Add a new function to load in the beam model in WODEN/wodenpy/primary_beam/use_everybeam.py (similar to load_OSKAR_telescope or load_LOFAR_telescope).

• In WODEN/wodenpy/skymodel/read_fits_skymodel.read_fits_skymodel_chunks, ensure the new beam model initiated correctly.

You need to update the following GPU functions:

• Ensure your beam model is added to beam_settings in WODEN/src/primary_beam.c

• At the start of calculate_visibilities in WODEN/src/calculate_visibilities.cpp, ensure use_twobeams, num_beams are set correctly for new beam.

• Wherever copy_CPU_beam_gains_to_GPU is called in WODEN/src/calculate_visibilities.cpp, ensure that happens for your new beam model.

• Add your model to WODEN/src/source_components.cpp.free_beam_gains function to ensure the leakage terms are freed correctly.