Logged on 10/07/2009 12:54:12 PM
NOTE: FITS files and Measurement Sets have been removed so that the file was small enough to email
Based on Filipe's presentation at the Nancay SKADS workshop. No responsiblity is taken for inaccurate representation of the material!
Using supersim.py, a modified simulation script (attached).
The UV Brick options are under Sky Model > Siamese.OMS.fitsimage_sky. Set the FITS image (s3-brick-all.fits) and padding factor (2).
The initial measurement set VLAAA4h.MS (.tar.gz of this is attached in case you don't already have it).
Intitial dirty image is dominated by a single point source and a fairly poor image. Compare the dirty fits image with input sky model attached below.
Logged on 10/07/2009 12:59:34 PM
Follow the same procedure with a different MS, more like the VLA. The new measurement set is attached below.
Still see strong effects of the PSF on the image, but better than before, so we can see more of the other sources, not just the dominating bright point source.
|
||||||||||||||||
Logged on 10/07/2009 01:02:18 PM
Now make a clean image of the same simulation. By eye we see a much closer match to the original image.
|
Logged on 10/07/2009 02:16:53 PM
Set the padding to 1 instead of 2 and see what difference it makes to the simulated images. All other settings and the measurement set are the same.
Recompile script supersim.py with different padding (1 instead of 2). Simulate MS, Make dirty image and Make clean image.
|
||||||||||||||||
|
Logged on 10/07/2009 02:29:52 PM
Load both of the cleaned images into kvis, from the simulations with two different padding factors. By eye they look quite similar, so we need to make a difference image to highlight the changes.
Select 'Make Data' to do a subtraction to see the difference (ie. a + next to one image and a - next to the other). Export the resulting image to FITS (attached below). We can see the main differences between the images are around the edges of the input field, as expected.
I will attempt to explain this with my limited understanding, feel free to correct me!
Without the padding factor, the edges of the image are influenced by the gridding convolution kernal (in uv space) and gridding correction function division (in image space). With the padding factors, the errors from this tapering only really effects the part that was padded with zeros and thrown away.
|
Logged on 10/07/2009 02:33:08 PM
Now use a different sky model, testeor.fits and a LOFAR measurement set (both attached below). I think the EOR image is courtesy of Mario.
Go back to a padding factor of two, since we've seen this is more accurate. Make a dirty image (set a much large field size, 600 arcminutes).
We can see some extended structure from the initial input map in the dirty image.
|
||||||||||||||||
|
||||||||||||||||
Logged on 10/07/2009 03:06:34 PM
Do again with an emerlin measurement set (attached). Simulate MS and make dirty image.
Don't get anything displayed in kvis. Looking at the terminal output we see that the image had zeros everywhere (min 0, max 0).
This is because we have put in an input sky with extended EOR structure, and tried to sample it with the merlin antennas, which only have long baselines. The baselines are therefore only sampling high spacial frequencies, fine structure, while the input sky only contains extended structure. Since the UV tracks were way outside the UV brick from the input sky, no signal was found. So something large and extended in the sky is only seen by the short baselines.
Another way to think of this is in the image plane. With very long baselines, for this input, we end up with a point spread function that is much smaller than a single pixel. I think for once it's actually easier to think about this one in UV space.
In reality a small amount of signal would be detected but because of the way we calculate it, if te UV tracks do not pass through the brick, there is no sampling at all, rather than a very-small-but-not-zero-amount.
|
||||||||||||||||