Subtraction in the image and visibility domains combined

  

To overcome the problem of the beam changing with frequency, van Gorkom and Ekers (1983) suggested a method which models the continuum emission, Fourier transforms the model correctly for each channel (thus dealing with the frequency-dependent beam problem), and subtracts the Fourier transform from the visibilities for each channel. You then re-image the subtracted visibility data base. This procedure is quite CPU intensive, but, provided you can make a good model of the continuum emission, produces a better result than the simple image subtraction method described above, especially if the line is weak and the continuum is very strong. If the continuum emission is very simple, you may be able to model it by a collection of discrete point sources. More typically, we use CLEAN (see § 16 describing deconvolution of images) to produce this model. It is sometimes necessary to estimate a residual continuum image after this procedure and subtract it as well. This method suffers from the following difficulties.

1. It assumes that the spectral index of the continuum emission is zero.
2. The model may not fully represent all the continuum structure. For example, CLEAN does not perform well on large regions of extended emission. You might also try to produce the model with a maximum entropy technique such as VTESS in AIPS.
3. The computational cost is large (deconvolution plus lots of Fourier transforms).

Let us proceed with the implementation of this technique. First, you must produce a dirty image of the continuum emission with the methods presented in § 15. It may be necessary for you to image all the channels so as to ascertain which channels are line free and suitable for continuum (use ISPEC to plot spectra averaged over some spatial window or POSSM directly on the visibilities).

Next you should deconvolve the continuum with one of the CLEAN programs as described in §16 and produce a model of the continuum emission. The CLEAN model is the CLEAN component list. You must decide whether all of the CLEAN components that CLEAN found are suitable to be included in the model, or whether you should truncate the list at some point (for example, you may have CLEANed into the noise and there is not much point to Fourier transforming and subtracting noise).

The Fourier transformation of the model and subtraction from the visibilities is done with the task UVSUB. You should run this only on a single-source visibility data base. If you don't have one, go and make it with SPLIT, as described in § 12.

inname,inclass Input visibility

inseq,indisk file

nmaps=1 One input model file

channel=0 Subtract from all channels

in2name,in2class Image associated with CLEAN

in2seq,in2disk component list

outname,outclass Output subtracted

outseq,outdisk visibility file

bcomp=1 Start subtracting CLEAN components at this one

ncomp=0 Subtract all components ?

cmethod= ' ' Allow program to choose subtraction method

cmodel='COMP' Model is CLEAN component list

factor=1 Subtract components

opcode=' ' Subtract components

smodel=0 Used for point source models

baddisk Keep scratch off NFS mounted disks

You can now image the subtracted visibility data base with the methods described in § 15.6 to see the line-only signal.