Computation

MIRIAD 's CLEAN task -- surprisingly called clean -- implements the Högbom, Clark and SDI algorithms. Given an input dirty image and beam, it produces an output CLEAN component image which has flux density units of Jy/pixel. This CLEAN component image is CLEAN's best guess at what the source really looks like. Invariably its extrapolation at high spatial frequencies is very poor (and probably a reason why AIPS does not make it easy to view it). To reduce the undesirable effects of this extrapolation, and to add in any emission remaining in the residuals, you will need to use the task restor . This gives you what is normally thought of as the `CLEAN' or restored image.

The various input parameters to clean are:

• map, beam: These keywords give the names of the input image and beam datasets -- as generally produced by invert . The image and beam do not need to be the same size, nor do they need to be a power of 2 in size.

• out: This gives the name of the output CLEAN component image.

• mode: MIRIAD clean can perform either Högbom, Clark or SDI CLEANs. The mode parameter can be set to indicate the particular algorithm to use. Possible values are `hogbom', `clark', `steer' or `any' (the default). With `any', clean determines what it believes is the best algorithm for your particular image. In this case, clean can switch between different algorithms, as the nature of the residuals change. This is particularly useful when CLEANing a large image which contains some strong point sources and much lower brightness extended emission. In this case, clean may well switch from a Clark or Högbom algorithm to the SDI algorithm when it finds that the residuals are becoming very smooth.

  Generally you can allow clean to choose the algorithm. Task clean

  will override your choice if you attempt to use the Högbom algorithm on an image that is too big.

• region: As noted above, it is important in clean

  to limit the CLEAN region to an area one quarter that of the beam, and to make it fit snugly around the real emission. The region parameter can be used to describe quite complex CLEANing regions. See Section 6.3 on how to specify this common parameter. Alternatively, you can use task cgcurs to describe the region interactively from a display of an image on a PGPLOT device. Task cgcurs (see Chapter 16.2) produces a text file, cgcurs.region, describing the region selected, which you can then input to clean (see Section 2.5). For example:

       region=@cgcurs.region
  

  The default CLEAN region is the inner quarter of the image.

  For very dirty dirty images, you might initially have only a very poor knowledge of where your emission is. In this case, CLEAN can be somewhat iterative. On your first CLEAN, you make an initial estimate of the region containing emission. Having produced a restored image, you then use this to get a better estimate of the region containing emission. You CLEAN again, and so on.

• gain: A good number for the loop gain is 0.1 (the default). If CLEAN stripes occur, then try lowering the loop gain by a factor of about 10.

• cutoff, niters, options=negstop: You can tell clean when to stop in three ways. You can tell it to quit once the brightest pixel (in an absolute sense) in the residual image reaches some cut-off. This is typically some multiple of the rms noise level (3-- or so). Specify this level with the parameter cutoff. Alternatively, you can tell clean to stop when it encounters the first negative component, by using the options=negstop switch. Otherwise, CLEANing will proceed until niters CLEAN iterations have been performed. When CLEANing a cube, niters is the number of iterations per plane. For small and simple sources, a few hundred iterations are usually sufficient. For complicated and large sources, you can CLEAN forever.

  You can set all three of cutoff, options=negstop and niters. Task clean will stop when any one of these stopping criteria are satisfied.

• phat: This gives Cornwell's Prussian helmet parameter, which is the value of the spike to add to the central lobe of the beam. Using this parameter may give some success at eliminating CLEAN stripes. When using a Prussian helmet, typical values are 0.3 or so. Note that the loop gain should also be reduced when using Prussian helmets.

• model: If you CLEAN an image, and find that you need to CLEAN some more, clean can be restarted from where you left off. Do this by setting the model parameter to the name of the old output CLEAN component image, and setting the out parameter to a new name. Actually the model image need not have been produced by clean -- it can be any image with units of Jy/pixel. Of course, it should be a representation of your source. The model image, however, must align exactly with the region being CLEANed ( i.e. same pixel increments, and same size as the bounding box of the selected region).

• speed: The `speedup factor' is a knob for the Clark algorithm which controls how often major cycles are performed. The more major cycles, the closer a Clark CLEAN is to a Högbom CLEAN. For point sources, this can be about 1. For extended sources, or if you are having trouble with CLEAN stripes, set speed to some small negative number -- typically -1 is good. The default is 0.

• minpatch: For Clark CLEANs, the parameter minpatch is the minimum full width ( not half-width, as it is in AIPS ) of the beam patch used in the minor cycle. The default is 51. However, for ATCA work, where the beam can often have large distant sidelobes, and so this value may be too small. Setting minpatch to larger values will slow the algorithm, but may avoid CLEAN striping problems. The maximum value that clean will accept is 257 -- larger values will be trimmed back to 257.

• clip: For SDI CLEANs, this parameter gives the `clip level'. During an SDI iteration, all pixels greater than clip times the peak flux density are considered to represent true structure, and so are taken as components for that iteration. Typically the clip level is 0.9. The default that clean computes is image dependent, and will be a function of how many pixels there are across the beam. The default value is usually adequate.

• options, as usual, gives several processing options. Possible values are

  negstop

  Stop CLEANing when the first negative component is encountered.

  positive

  Apply a positivity constraint. This constrains the component image to be non-negative. A side-effect of this is that CLEAN will stop iterating if it cannot continue to ensure this. This does not have any effect with SDI iterations. This option does not seem to produce as much improvement as one might hope.

  asym

  Normally clean assumes that the beam has a 180 degree rotational symmetry, which is the norm in radio interferometry. Making such an assumption allows some optimisations. You should instruct clean if this is not the case, by using the asym switch.

  pad

  Double the beam size by padding it with zeros. This will give better stability with Clark and Steer modes if you are daring enough to CLEAN an area which is more than a quarter of the beam area.

Typical inputs are given below:

The total CLEANed flux density ( i.e. the cumulative sum of the CLEAN components) should eventually settle down to a roughly constant number. This indicates that you are just picking up noise, and that there are no sidelobes left to remove. If the total CLEANed flux density starts to decrease again, this usually indicates that you have been a bit heavy handed, and CLEANed too much. You might also look at the result and see if you can see any sidelobes left over.

Having completed clean , you will almost certainly want to ``restore'' your image -- see Section 13.4.