aips++ EVLA TST2.0 Walter Brisken May 2005 Running test on computer imager-c using AIPS++ Version 1.9 Build 1056 Working on flank fields removed data: 1. Fill data -- used function filldata() in shk116.wfb.g 2. Make "small" test dirty image to see whats there, about 3x3 degrees. Both IFs. Used function smalldirtyimage() Output looks fine. Image : smalldirty.im After experimentation, I came up with parameters that gave a nice image: 7500x6912 image at 0.36" resolution --> 0.75 x 0.69 degree image robust 0.4 wprojplanes=256 algorithm='wfhogbom' (mfclark fails with memory error abouve ~6000^2) gain=0.2 niter=120000) Imaging on imager-c took about 50000 seconds. A very large fraction of time was used in the residual calculation and I saw no sign that a gain of 0.2 was too large for this well sampled data. A large speedup with a gain of 0.3 or even 0.4 may be possible without loss in image quality. Imaging both IFs simultaneously (with otherwise identical settings) took about 60000 seconds. I got trapped twice by carelessly misspelling function parameter names. Imager would run until either completion or crash before telling me in a not-so-clear manner: NORMAL: Finished imager::clean 4392.64 real 3657 user 140.04 system NORMAL: Successfully closed empty server: imager : Method setoptions fails! Parameter ftmachine cannot convert between numbers and strings File: servers.g, Line 1009 Stack: .(), imager.g line 244 .(), im.g line 40 This particular error resulted from not quoting 'wproject' in the command: imgr.setoptions(ftmachine='wproject',wprojplanes=256) A test for such values should be added. (I realize that the new architecture should address this automatically) Except for this issue, I am _very_ pleased with the functionaality of imager in this wide-field uuse case. Image noise statistics: I chose a blank part of the image containing 208170 pixels and computed RMS noise on the images with the following results: IF 1 3.81 muJy IF 2 4.19 muJy Averaged image 2.84 muJy IF 12 cleaned together 2.92 muJy Note that the expected RMS of the averaged image was 2.83 indicating no detection of systematic noise (ie due to common sidelobes or large scale emission). Spectral index variations across the image or relative flux calibration errors appear to have caused the imaging of both IFs together to have a higher noise background. The noise differed in detail but didn't show substantially different structures except around a couple bright sources. Brightest source at J2000 10:46:27.3 +59:04:57.1 IF1 7.303 mJy IF2 7.313 mJy Averaged 7.308 mJy IF 12 cleaned together 7.278 mJy Images are stored in /home/imager-c/wbrisken/tst2.0: IF1 huger.1.im IF2 huger.2.im Averagd image huger.im IF 12 cleaned together huger.12.im In order to do a proper job imaging a full A-array dataset with sources visible well into second and third sidelobes images than than ~7200^2 need to be made. Thus bigger memory machines are needed to properly image such fields without the tedious and less effective use of flanking fields. Even for a data set such as the one imaged here processor speed is not really the biggest issue, but rather the memory size. Requirements Evaluation: EVLA numbered requirements (based on tst2.0 subset) ESec. ENo. Evaluation 5.1 1 Only tested in part. I think this is the wrong place to speak of supported data formats, rather this requirement should be "ability to image any combination of (e)VLA data with or without supplemental single dish data. The data formats should be addressed in a different requirement. 2.1 Tested. Made dirty images of various weightings. Worked. 2.2 Residual images were properly made during imaging. 2.3 Single scale clean using Clark and Hogbom clean algortithms were tested. I didn't test Cotton-Schwab clean. Cleaining converged and produced good images in both test cases. 2.4 MEM was not tested. This particular dataset is not a good test case for MEM. 3.1 Selection by source -- not tested here, but previously. It is functional. 3.2 Various cell sizes were used. Functional 3.3 Various image sizes (500x500 to 7500x6912) were tested. Functional 3.4 Images were made for IF1, IF2 separately. Also an image was made using both IFs simultaneously. Selection appears to do the right thing. 3.5 Spectral channels were not selected here. I have tested this selection criterium in the past and it is functional. 3.6 Various robust imaging was tested ranging from -1 to 1. Functional. Uniform and Natural were tested on other data sets. 3.7 Clean gain of 0.1 and 0.2 was tested. 0.2 was used exclusively when cleaning large images as the speed-up made the cleaning faster. Functional. 3.8 Only the iteration threshhold was tested. It is functional. 4.1 Masking of any sort was not used in my testing. 4.2 " 5.1 Same as 4.1? 5.2 Same as 5.2? 6.1 Tested. I subtracted IF2 image from IF1 image to see differences. 6.2 Not tested. 6.3 Not tested. 7.1 Not tested. 8 Tested only Stokes I in this test. 5.2 1 Wide field imaging using w-projection was tested. Functional. Impressive! 2.1 Faceted imaging was not tested 2.2 " 2.3 " 3.1 Only Stokes I imaging (no squit correction needed) was tested. 5.4 1.1 The dynamic range was about 3000:1 . This was sensitivity limited though as the brightest source was ~7 mJy. Much higher dynamic range is clearly possible. 1.2 Not tested -- almost certainly limited by VLA correlator. Need to await WIDAR to test this 2.1 Single field imaging of primary beam out past first sidelobe was tested for stokes I. Stokes Q/U not available. Works. Image fidelity is limited by bandwidth smearing. 2.2 " Suggestions: Regarding 5.1 req 3.8 -- it would be good to include number of major cycles as a stopping criterium as that is the slowest part of cleaning. Entering a new major cycle for 10 additional minor cycles is rediculous when cleaning 120000 iterations!