Visibility Comparison Plots
Last update: June 19, 2017
This document describes the visibility comparison plots that are available in the RHESSI software as of June 7, 2017.
The visibility comparison plots show the observed visibility amplitudes and errors overlaid by the visibilities computed from an image. These plots are similar to the plot that the VIS FWDFIT image reconstruction algorithm produces, but can be generated for any image reconstruction algorithm, both visibility-based and not. The observed visibilities have been filtered (edited for outliers, conjugate combined, and normalized) if those options are selected and regularized (photon or electron) if selected. The visibilities from the image are computed using the vis_map2vis IDL routine.
The plot above shows the format of the new plot for an image generated by the VIS_FWDFIT algorithm. The amplitudes of the visibilities for all detectors are displayed. The red outline identifies the detectors that were used to make the image (1-6 in this case). For visibility-based image algorithms, the white Xs for the detectors used in the image are the visibilities used to make the image. For non-visibility-based algorithms, the white Xs show the visibilities computed using the current settings of the detector, edit, combine, normalize, and vis type options in the image object. The normalization for the detectors used in the image is based only on those detectors. For the detectors not used in the image, the normalization is based on all detectors.
The blue vertical bars show the 1-sigma error on the visibility amplitudes. The red diamonds show the visibilities computed from the image using vis_map2vis. The green triangles show the vector difference between the observed amplitudes and the image amplitudes ( sqrt(xdiff^2 + ydiff^2) ).
A chi-square comparison is computed and displayed for all detectors, for each detectors separately, and for the detectors used in the image. The chi-square is computed as:
chi-square = total( ( (vismapx-visx)^2 + (vismapy-visy)^2 ) / sigamp^2 ) / (nvis-1)
where vismapx, vismapy are the real, imaginary parts of the visibilities from the image, visx,visy are the real, imaginary parts of the observed visibilities, sigamp is the sigma on the amplitude of the observed visibilities, and nvis is the number of visibilities.
In the RHESSI GUI, there is a button selection to send the
visibility comparison plot to the screen or a PS file WHILE the images are
being generated. (Unfortunately the plot for each image goes to a
separate PS file for each image in a cube, not one multi-page file.) There
is also a button under the Display pull-down menu to display the plot on
the screen for the most recent image made.
From the command line, you can call the hsi_image_vis_plot routine for any
image in the image object. There are a number of keywords arguments that
you can call hsi_image_vis_plot with - please see the header of the
routine for details. For example:
viscomp_show_plot | If set, generate the visibility comparison
plot while making the image(s) |
viscomp_window | Window ID to draw viscomp plot in (set to -1
to open a new window) |
viscomp_ps_plot | If set, output plot to PS file
(auto-generated name) |
viscomp_ps_dir | Directory to write PS file in (default is
current directory) |
By default, the y axis maximum is based on the amplitudes of the observed visibilities in the coarsest detectors to avoid scaling to bad values from the image. Use the yrange keyword argument to set it yourself.
If we make a CLEAN image for 25-Sep-2011 03:30:07.467-03:30:47.261 12-25 keV using the default detectors (4,5,6,7,8), (note: we also set clean_beam_width_factor=2 and clean_regress_combine=1) we get the following image and visibility comparison plot.
We can see that the detector selection should be modified since the visibilities for Detectors 1-3 show structure and Detectors 7-9 don't contribute much.
If we remake the image with Detectors 1-6, we get the following:
The image now shows the fine structure, but we can see in the visibility comparison plot that for Detectors 1 and 2, the image peaks are not as narrow and high as the observed peaks.
If we make the same image with the loop option in VIS_FWDFIT, we get a very narrow loop (~2 arcsec FWHM) and the visibility plot shows that the image visibilities track the structure in the observed visibilities for the finer detectors better.