cdeconvolver-mpi

This software takes either the gridded visibilities, gridded weights and the PCF grid or the image equivalents of those (dirty image, natural weighted PSF, and PCF image) and runs a minor cycle solver to produce a deconvolved image. The idea is to use this to do multi epoch integrations of large spectral datasets by storing cubes with Nyquist grids in compressed format, adding them together and then deconvolving them. This avoids having to store the large spectral MeasurementSets at the cost of only being able to use minor cycle solvers.

To generate the uv-grid or raw (unnormalised) image input cdeconvolver-mpi expects you need to run imager with the following parameters present:

Cimager.ncycles = 0
Cimager.restore = false
Cimager.nyquistgridding = true # not required, but reduces size of output
Cimager.write.grids = true
Cimager.write.grids.fft = false # for uv grid, or true for raw images
Cimager.write.psfimage = false
Cimager.write.modelimage = false
Cimager.preconditioner.Names = [Wiener] # Wiener preconditioner is required
Cimager.preconditioner.Wiener.robust = 0.0 # specify required value

The imager task should be run with only a single input MeasurementSet at a time, as the grids will only contain data from the first MS specified. The use of the Wiener preconditioner is required at present. Specify the parameters in the imager run to generate the PCF grid cdeconvolver-mpi expects to find.

In addition the SpheroidalFunction parameter alpha (e.g. Cimager.gridder.Wproject.alpha) should be left at the default value of 1.0 as that is what cdeconvolver-mpi assumes when it does the convolution correction.

cdeconvolver-mpi can deal with a few different input types (matching the output grid types in imager): Input can be a complex valued casa image with uv-coordinate system, a complex FITS image split in real and imaginary parts containing uv-grids but with the image coordinates of the corresponding (FFT) image attached or a standard FITS or casa image. In all cases specify the base name only: e.g., “visgrid.epoch1” will look for casa image “visgrid.epoch1” if imagetype=casa and FITS image “visgrid.epoch1.fits”, “visgrid.epoch1.real.fits” and “visgrid.epoch1.imag.fits” if imagetype=fits.

In the case of a casa uv-grid, the user needs to supply the image direction, cellsize and shape used during imaging in the parset. The case of a FITS uv-grid with uv-coordinates is not currently implemented because FITS readers don’t support this very well.

The cdeconvolver-mpi task is, as the name implies, mpi-enabled and can be run on cubes with a number of ranks that divides evenly into the number of channels. E.g., for a 6 channel cube, you can use 1, 2, 3 or 6 ranks (including rank 0).

Parset parameters understood by cdeconvolver-mpi are given in the following table. All parameters must have Cdeconvolver prefix, e.g., Cdeconvolver.solver.Clean.algorithm).

Parameter	Type	Default	Description
imagetype	string	fits	the type (fits or casa) of images
visgrid	vector<string>	none	Input visibility grids or dirty images. Names must start with visgrid, the psf and pcf grids will be opened by replacing this with psfgrid and pcfgrid. All input grids will be added together before preconditioning and deconvolution.
oversampling	float	1.0	Oversampling applied to images before they are passed to the deconvolver. Essential to set this to 1.5 or larger if Nyquist gridding was used in imager. To use the value imager determined and create images in the same resolution as specified in imager search the imager log for “extraoversampling” and use that value here.
solver	string	none	Name of the solver. Currently supported option is “Clean”. Further parameters are given by solver.solver_name.something. See Solver Documentation for details.
solver.Clean.algorithm	string	BasisfunctionMFS	Algorithm option for a solver: Hogbom, Basisfunction, BasisfunctionMFS.
threshold.minorcycle	vector<string>	no threshold	Use to set the clean limit (see Solvers)
preconditioner.Names	vector<string>	none	Preconditioners to use: GaussianTaper and Wiener are available. See solvers preconditioner documentation
Images.Names	vector<string>	None	Base name of output images - only the first entry is used for this task. Image names must start with the word image
Images.direction	vector<string>	None	When using imagetype=casa and a UV grid as input specify the image direction exactly like in the original imaging run that created the UV grids.
Images.cellsize	vector<string>	None	When using imagetype=casa and a UV grid as input specify the size of pixels in the spatial axes in the original imaging run , e.g. [6.0arcsec, 6.0arcsec]. If Nyquist gridding was used in imager, specify the cellsize of the Nyquist grid (search for “Changing cellsize” in the log)
Images.shape	vector<int>	None	When using imagetype=casa and a UV grid as input specify the shape from the original imaging run. If Nyquist gridding was used in imager, specify the shape of the Nyquist grid (search for “Changing shape” in the log)
restore.beam	vector<string>	fit	Either a single word ‘fit’ or a quantity string describing the shape of the clean beam (to convolve the model image with). If quantity is given it must have exactly 3 elements, e.g. [30arcsec, 10arcsec, 40deg]. Otherwise an exception is thrown. This parameter is only used if write.restoredimage is set to True. If restore.beam=fit, the code will fit a 2D gaussian to the PSF image and use the results of this fit. In this case, each channel will have an independently-fitted beam.
restore.beamReference	string	mid	The channel to use as the reference for the beam - this channel’s beam is written to the cube header. Values can be an integer indicating the channel number (0-based), or one of ‘mid’, ‘first’, or ‘last’.
restore.beam.cutoff	double	0.5	Cutoff for the support search prior to beam fitting, as a fraction of the PSF peak. This parameter is only used if restore.beam=fit. The code does fitting on a limited support (to speed things up and to avoid sidelobes influencing the fit). The extent of this support is controlled by this parameter representing the level of the PSF which should be included into support. This value should be above the first sidelobe level for meaningful results.
imageaccess.write	string	serial	For FITS images “parallel” can be used to write the output in parallel which is often faster
write.residualimage	bool	false	If true write the residual image
write.psfimage	bool	false	If true write the preconditioned PSF
write.modelimage	bool	false	If true write the clean model
write.restoredimage	bool	true	If true write the restored image At least one image must be written out
calcstats	bool	false	Write per plane image statistics to image
outputStats	string	“”	Write per plane image statistics to file with given name as well, or not if empty
imageHistory	vector<string>	none	Specify lines to add to the image history
header.<keyword>	vector<string>	[“”,””,”STRING”]	Header keyword to write to the image header. Replace <keyword> with up to 8 character name of the keyword. Second string is the keyword description or comment (default no comment). Third string is the optional type (INT,DOUBLE, or STRING) with STRING the default.

Examples

Example 1:

Example cdeconvolver-mpi parset to first add the 2 input grids in FITS format, use preconditioning with a Wiener filter (robustness=0.5) and then deconvolve the oversampled image using the BasisFunctionMFS cleaner. By default we’re only writing out the resulting restored image cube. Beam shape and statistics information will be embedded in the output file (as records for casa and as binary tables for FITS).

Cdeconvolver.visgrid                          = [visgrid.epoch1,visgrid.epoch2]
Cdeconvolver.Images.Names                     = [image.two-epochs]
Cdeconvolver.oversampling                     = 2.0
Cdeconvolver.solver                           = Clean
Cdeconvolver.solver.Clean.algorithm           = BasisFunctionMFS
Cdeconvolver.solver.Clean.scales              = [0, 3, 10]
Cdeconvolver.solver.Clean.psfwidth            = 256
Cdeconvolver.solver.Clean.gain                = 0.1
Cdeconvolver.solver.Clean.niter               = 1000

Cdeconvolver.preconditioner.Names             = [Wiener]
Cdeconvolver.preconditioner.Wiener.robustness = 0.5