calculate_visibilities¶
Tests for the functions in WODEN/src/calculate_visibilities.cu (there is only one function). calculate_visibilities::calculate_visibilities is the gateway function
to all CUDA functionality in WODEN. It takes in simulations settings and
a sky model, and performs the necessary coorindate and measurement equation calculations, as well as summations over sky model components to generate visibilities.
The tests below are fairly limited in the number of variables tested, as they
are integration tests to make sure all the functionality expected is called by calculate_visibilities::calculate_visibilities, and is able to talk to the
GPU correctly. More rigorous testing of this functionality is included in other
test suites in cmake_testing.
The tests below all use test_calculate_visibilities_common.c to setup a
source_catalogue_t sky model struct with the following sky model setups.
These differing sky models should cover all COMPONENT combinations possible, to
make sure calculate_visibilities is calling the appropriate functions:
Num SOURCEs |
Num POINT in each SOURCE |
Num GAUSS in each SOURCE |
Num SHAPELET in each SOURCE |
Total COMPONENTS |
|---|---|---|---|---|
1 |
1 |
0 |
0 |
1 |
1 |
0 |
1 |
0 |
1 |
1 |
0 |
0 |
1 |
1 |
1 |
1 |
1 |
1 |
3 |
3 |
1 |
0 |
0 |
3 |
3 |
0 |
1 |
0 |
3 |
3 |
0 |
0 |
1 |
3 |
3 |
1 |
1 |
1 |
9 |
3 |
3 |
0 |
0 |
9 |
3 |
0 |
3 |
0 |
9 |
3 |
0 |
0 |
3 |
9 |
3 |
3 |
3 |
3 |
27 |
All sources are given an RA,Dec equal to the phase centre, and a Stokes I flux density of 0.3333333333333333 Jy. Each type of flux component (either POWER_LAW, CURVED_POWER_LAW, or LIST) are set up so they have a perfectly flat spectrum. This way, the output visibilities (in the absence of a primary beam) should be fully real, and equal to the sum of the number of COMPONENTs in the sky model multiplied by 0.3333333333333333. This numerical 1/3 flux is a good test of the precision of the FLOAT and DOUBLE compiled codes.
Note
Given all sources are at phase centre, this means the cross-correlations and the auto-correlations should have exactly the same values. So in all tests described below, calculate_visibilities is run with both woden_settings->do_autos set to 0 and to 1. If the autos are calculated, the number of output visibilities should double (as I only simulate three baselines from three antennas here). The latter half are where the autos are set, and are tested to be equal to the expected values for the crosses.
When the tests have a single COMPONENT per SOURCE, I use a POWER_LAW type flux behaviour. When there are three COMPONENTs, I use one of each of POWER_LAW, CURVED_POWER_LAW, and LIST types.
GAUSSIAN and SHAPELET components with any size cause a reduction of the real part of the visibility due to the extended size on the sky. For these tests, I’ve set their major and minor axes to 1e-10 to make them behave similarly to point sources.
For all tests below, I setup a simulation with three baselines, three frequencies, and two timesteps. For all sky models, frequencies, time steps, and baselines, I check:
The u,v,w are as expected
The real part of the visibilities are equal to number of COMPONENTs times 0.3333333333333333 Jy (modulu the beam repsonse - see below)
To keep testing the u,v,w straight forward, I’ve set the baseline lengths in \(X\) and \(Y\) equal, (i.e. \(X_{\mathrm{diff}} = X_{\mathrm{ant1}} - X_{\mathrm{ant2}} = Y_{\mathrm{diff}}\)), and the length in \(Z\) to zero. With this configuration, the following is true:
\(u = X_{\mathrm{diff}}(\cos(ha_0) + \sin(ha_0))\)
\(v = X_{\mathrm{diff}}\sin(\phi_{\mathrm{lat}})(-\cos(ha_0) + \sin(ha_0))\)
\(w = X_{\mathrm{diff}}\cos(\phi_{\mathrm{lat}})(\cos(ha_0) - \sin(ha_0))\)
where \(ha_0\) is the hour angle of the phase centre, and \(\phi_{\mathrm{lat}}\) the latitude of the array. The allows us to check the u,v,w are changing with time.
For FLOAT compiled code, the absolute tolerance threshold on the u,v,w values is set to 1e-5, and 1e-12 for DOUBLE compiled code.
test_calculate_visibilities_nobeam.c¶
This runs the tests explained above, whilst switching the primary beam off. This really does check that real visibilities are equal to the number of COMPONENTs times 0.3333333333333333 for all 12 sky model configurations, and the imaginary equal to zero.
For FLOAT compiled code, the absolute tolerance threshold on values is set to 1e-6, and 1e-9 for DOUBLE compiled code.
test_calculate_visibilities_gaussbeam.c¶
This runs the same tests as test_calculate_visibilities_nobeam.c, but applies
a Gaussian primary beam model. As all sky model COMPONENTs are set at the same location,
only one beam gain per time step should be applied to visibilities, so for each time
step, we can check whether the visibilities equal this gain times the number of
COMPONENTs. The Gaussian beam is fully real and has no cross-pol values, so only
check that the real XX and YY visibilities have value, and ensure all other
visibility information is zero.
For FLOAT compiled code, the absolute tolerance threshold on values is set to 1e-5, and 1e-8 for DOUBLE compiled code.
test_calculate_visibilities_edabeam.c¶
Runs exactly the same tests as test_calculate_visibilities_gaussbeam.c, but
using the analytic single dipole beam (the EDA2 beam). Obviously tests the
visiblities match the gain values that are expected for the EDA2 beam and not
the Gaussian Beam test.
For FLOAT compiled code, the absolute tolerance threshold on values is set to 1e-5, and 1e-8 for DOUBLE compiled code.
test_calculate_visibilities_mwafeebeam.c¶
Again, runs the same tests as test_calculate_visibilities_gaussbeam.c, but
this time for the coarse resolution MWA FEE primary beam. As this model is
complex and includes mutual coupling, both the real and imaginary values
for all XX, XY, YX, and YY polarisations are tested.
For FLOAT compiled code, the absolute tolerance threshold on values is set to 4e-3, and 1e-7 for DOUBLE compiled code. The expected gains of the MWA FEE beam are taken from the DOUBLE compiled code, and so the large threshold for the FLOAT here is mostly due to the inaccuracy of the FLOAT MWA FEE beam code (see FEE_primary_beam_cuda_cmake for more discussion on this).
test_calculate_visibilities_mwafeebeaminterp.c¶
Same as test_calculate_visibilities_mwafeebeam.c, but
this time for the frequency interpolated MWA FEE primary beam. As this model is
complex and includes mutual coupling, both the real and imaginary values
for all XX, XY, YX, and YY polarisations are tested.
For FLOAT compiled code, the absolute tolerance threshold on values is set to 3e-2, and 1e-7 for DOUBLE compiled code. The expected gains of the MWA FEE beam are taken from the DOUBLE compiled code, and so the large threshold for the FLOAT here is mostly due to the inaccuracy of the FLOAT MWA FEE beam code (see FEE_primary_beam_cuda_cmake for more discussion on this).
test_calculate_visibilities_mwaanalybeam.c¶
Same as test_calculate_visibilities_mwafeebeam.c, but
this time for the analytic primary beam. As this model is real only but contains
leakage terms, all real values are tested to match expectations, and all
imaginary tested to be zero.
For FLOAT compiled code, the absolute tolerance threshold on values is set to 1e-5, and 1e-9 for DOUBLE compiled code.