离散天空算子（DiSkO）合成成像

作者：Tim Molteno tim@elec.ac.nz

它太酷了，以至于没有意义。使用望远镜算子跟踪望远镜零空间和范围空间来获取图像。DiSkO在天空空间中使用场视域的离散表示（作为healpix网格或非结构化网格）。这意味着我们可以成像任意宽的视场（包括整个球体），以及任意形状的视场（圆形更受欢迎）。

DiSkO可以通过正则化进行稀疏性降低并控制天空解的体积。结果是敏感于漫射宽带源的成像算法，并且不需要像CLEAN那样的恢复。即将发表的出版物：）

如何做

disko --display --show-sources

从测量集中加载数据

disko --ms test_data/test.ms --tikhonov --nside 32 --PDF

使用DiSkO进行VLA成像

从NRAO网站下载VLA 5GHZ连续波测量集。AG733_A061209.xp1。

根据CASA教程[https://casaguides.nrao.edu/index.php/VLA_5_GHz_continuum_survey_of_Seyfert_galaxies]校准并分割测量集。

disko --fov 0.01 --ms NGC1194.split.ms --SVG --arcmin 0.0025 --tikhonov

更具挑战性

本教程应生成一个包含大量漫射辐射的文件。[https://casaguides.nrao.edu/index.php/VLA_Continuum_Tutorial_3C391-CASA5.5.0]

wget http://casa.nrao.edu/Data/EVLA/3C391/3c391_ctm_mosaic_10s_spw0.ms.tgz
wget https://github.com/jaredcrossley/CASA-Guides-Script-Extractor/blob/master/extractCASAscript.py

python extractCASAscript.py 'https://casaguides.nrao.edu/index.php?title=VLA_Continuum_Tutorial_3C391-CASA5.5.0'

然后在CASA中

execfile("VLAContinuumTutorial3C391-CASA5.5.0.py")

这将生成一个合适的测量集进行成像。

待办事项

• 添加一个--full-sphere选项，该选项固定天球坐标中的球体，然后正确地指向观测的相位中心。需要指定波束图案（或至少是半球波束）。波束是天空矢量掩码。即，应该在外部降至零。
• 处理标记，以可被测量集中使用的残差输出方式标记。这意味着一些会改变MS的CASA表达式。

更新日志

1.0.0b5 Fix up the inclusion of non tart stuff.
1.0.0b4 Remove tart2ms include.
1.0.0b3 Remove dask-ms dependency.
1.0.0b2 Make tart  dependencies optional so allow direct imaging code.
1.0.0b1 Change Sphere to Fov. I.e. HealpixFoV (Field of View). Doing this because we will use
        square images 
        Add new SquareFoV class for square images (work in progress)
0.9.6b2 Fix the --elevation limit to actually implement this for disko draw
0.9.6b1 Add a minimum elevation to the sphere.el_min_r. This is for setting bounds in imagers.
        Explicitly manage the tart2ms logging
0.9.5b4 Add a timestamp to images (or a title if specified) in SVG mode
        clean up logging so that only happens when --debug is present.
0.9.5b3 Use gmsh rather than optimesh... (WIP)
        Use much faster measurement set reading via casa_read_ms() about 200x faster!
        disko_draw timestamps the image
0.9.5b2 Fix RA direction in generated FITS files (thanks Ben Hugo)
0.9.5b1 Add --min and --max to disko_draw so allow manual setting the range of images
0.9.4b6 Fix bug in drawing PDF.
0.9.4b5 Import Resolution in disko to get array beam width.
        Fix sphere power.
0.9.4b4 Expose parent parsers.
        Refer to min_res() rather than nside for spheres
        Fix bugs in display of mesh spheres
        add disko.fov namespace
        serialize to hdf5 files
        New disko_draw CLI tool
        Conjugate visibilities from files.
0.9.4b3 Move sphere args parser to the sphere object.
0.9.4b2 Add helper method to calculate beam size. 
        Add area(), get_power() method to sphere. Make it work.
        add rms(), copy() methods for rms() and deep copying of spheres
0.9.4b1 Use read_ms from tart2ms (moved there). 
0.9.3b6 Use speed of light from astropy.constants. 
        Add a --version option to print the current version and exit.
0.9.3b5 Fix bug in the Matrix Free Linear Operator which wasn't conjugated.
0.9.3b4 Raise nicer errors when arguments aren't provided
0.9.3b2 Fix indexing error in read_ms when the number of visibilities requested exceeded the number available.
        clean up the meshing 
        rework the command line interface. New resolution specification
        Output residuals to the terminal (still no way to include them back in...). The ms needs to be flagged externally.
        Use Natural weighting when reading from measurement sets
0.9.3b1 Add --h5 option to allow sequential inference from a visiblity file.
0.9.2b1 No longer require arcmin for construction of spheres
0.9.1b1 Remove constraint that nside is a power of two now that healpy has accepted the pull request.
        Add new parameter l1_ratio
        Dont scale the alpha parameter.
        Allow negative solutions for Tikhonov regression.
        Allow full skies using --nside option
        Add a colour bar to the SVG output.
0.9.0b4 Improve measurement set reading. 
        Use the mean RMS value for a single noise estimate on visibilities.
        Use the correct rank value in overdetermined skies.
        Truncate the SVD to keep the condition number of the telescope less than 50.
0.9.0b3 Full Bayesian Inference is workin. Fix bug in meshio (after upgrade beyond 4)
0.9.0b2 Add a multivariate gaussian object. Fix ms_helper. Handle the case where the rank of the telescope operator is not full.
0.9.0b1 Move to a real telescope operator.
0.8.0b5 Allow FISTA to calculate its own largest eigenvalue if negative values are passed in.
0.8.0b4 Clean up code and avoid recalculating harmonics. 
        Added a DirectImagingOperator that performs the discrete Fourier Transform.
0.8.0b3 Add --fista command line option to use the FISTA solver.
0.8.0b2 Add an lsqr option to force the slightly slower lsqr algorithm in place of lsmr.
0.8.0b1 Add a matrix-free operator that actually works. Process UVW in meters.
0.7.0b10 Clean up tests. Rename the DiSkOOperator and get it going.  Fix up timestamp loading, use the correct frequency (based on channel parameter)
0.7.0b9 Fix up timestamp loading
0.7.0b8 Optimize mesh at each stage of refinement.
0.7.0b7 Better refinement.
0.7.0b6 Limit gradient calculation to cells above nyquist limit
0.7.0b5 Improve channel selection
0.7.0b4 Allow selection of the channel number
0.7.0b2 New adaptive meshing on gradient
0.7.0b1 Add adaptive meshing and --adaptive option
0.6.0b9 Report Nyquist resolution
0.6.0b7 MS were being read incorrectly - the UVW are measured in meters, not wavelengths!
0.6.0b6 Correct field pointing from measurement sets.
0.6.0b5 Reduce memory requirements by around 25%.
0.6.0b4 Report the r^2 value.
0.6.0b2  Use dask for very large jobs (use the --dask switch)
0.6.0b1  Get data from Measurement Sets!
0.5.0b5 Allow sources not to be shown.
0.5.0b4 Override plot in HPSubSphere to allow for non-normal pixels.
0.5.0b3 Added elliptical source circle projections in SVG.
0.5.0 Getting imaging logic better. Added L2 regularization, and cross-validation