Naming convention ----------------- First character: K = (Kogan) ring type array C = (Conway) spiral type array Second character: A, C or E, denoting resolution of array Characters 3-5: Member of SIL- M51, CYG, MAR, SDC or MPD Characters 6 and 7: SN (snapshot) or track, e.g. _4 for 4hr track Characters 8-10: signed declination Characters 11 and 12: Wavelength, e.g. _1 or _3 for 1 and 3mm respectively or B7 corresponding to 345GHz frequency
Difficulty has been experienced when trying to generate difference images for the C array (and indeed the B, D and E arrays) due to uncertainty in the registration of the images. The A array has not been problematic as the input model image and the final CLEAN image are the same size and the same pixel scale. However for the other resolutions of array, the images being compared are of different size and pixel scale.
This requires that the convolved model image should be downsampled in some manner e.g. binning, and this by now smaller image subtracted from some equal sized subimage of the CLEAN image. The SUBIM task may be used to perform these operations, but my carefully calculated start coordinates for the subimage gave the results shown in the first difference image below.
This seemed to show some an embossed effect characteristic of differentiation, suggesting some misalignment. I thought then that I should shift the maps about a bit until I got a difference image with the minimum range if pixel values, as this is likely to correspond to the correct registration. This gave the second image below, which also gave the same results as aligning on the reference pixel using the DOALIGN=-1 option in SUBIM (the other images use DOALIGN =-2 which forces a straightforward pixel by pixel subtraction).
Great, I thought, just align on the reference pixel. But when I went back to the A arrays and did a difference image with DOALIGN=-1 just to check, I did not get same values as for DOALIGN=-2 as I had expected. And to confuse things further, the range of pixel values when aligning on the reference pixel was greater than for the straightforward pixel-by-pixel subtraction, suggesting that the latter was indeed correct, at least for the A arrays. Also, the difference image for the C array when aligning on the reference pixel still looked embossed, and finally, Min recalled problems with alignment using the reference pixel when mosaicing when I mentioned this to John Conway and him. It should be noted that both the difference images shown so far differ in registration by a single pixel along the horizontal axis, but that the first would appear to be grossly misaligned to one side, with the second slightly misaligned to the other. This perhaps suggests that the start point for the binning operation may be the main problem. John suggested the back-to-basics approach of recreating the model image at the same pixel scale and size as the resulting CLEAN image for each resolution, using that to generate the visibility data for imaging and deconvolution, and then doing a direct pixel-by-pixel subtraction of the convolved (by a beam the same size as the CLEAN beam) version of this model and the resulting CLEAN image, as for the A array. The result of this is shown in the image below, and is much more plausible. Disregard any comparison with the pixel values for this image as the scaling factor associated with the original model image is of course different.
Interestingly, the CLEAN image and convolved model for this approach is visually indistinguishable from that of the previous approach, suggesting that the approach will give valid results and that my problem does indeed lie in registering the images of different size and scale.
Thus I propose to redo all the imaging and CLEANing for the array resolutions other than A, using resized and rescaled models as described above. I also propose to leave the existing image data for these array resolutions on my website, as the new images will look the same so why waste time uploading them (I will of course check this).