Wednesday, May 16, 2007

Help with continuum subtraction?

I have 23GHz spectral line data with a weak continuum source, and need help making the continuum go away so I can measure the strength of the line emission... for a couple subtle reasons, UVLSF won't do the trick in this case. Any ideas for other ways to remove continuum? More details below.

The source itself is a weak point source. The problem with UVLSF is that spectral line structure is found across much of the spectral window, making it really hard to find line-free channels to average together to use as a model for the continuum. UVSUB uses clean component models, and honestly, I'm not sure how to use it or indeed if this is really what I want.

An additional twist-- I have a higher S/N image of the continuum source taken in an image of the same field at a slightly different central freq (~1.5 GHz offset). I have a feeling I can use this continuum emission as a better S/N image to subtract from the original image, but I'm not sure how to do this...

Ideas? Thanks...


amanda said...

Some random thoughts....

To use UVSUB or the higher S/N image you need to assume that the continuum emission is flatish in this region (i.e., if s~nu^{-alpha}, alpha ~0). If I were observing at lower frequencies (~5GHz), I would be inclined to use UVSUB clean component model rather than the higher S/N image since I would expect the continuum source intensity to change. For your frequencies, it might be a toss up. You could model the source using the higher S/N image and put this into UVSUB as the smodel. I'd try both and see which one you believe.

Laura said...

Ok, I know nothing about this stuff, but looking at the AIPS cookbook, it sounds like this is the one and only time that you would want to do continuum subtraction in the image plane. They discuss a couple of techniques in 8.3

KED said...

this is likely free-free emission, so the emission spectrum is pretty flat up here.

After some thought, I agree that clean components are the way to go. The continuum intensity should remain the same, but here's why CC are better-- at a slightly different frequency with data from a different night and completely different UV coverage, the beam shapes will be very different. Clean components, on the other hand, just say where in the image plane the flux peaks are located, so they can be convolved with the dirty beam and subtracted out.

I think there are also issues with the way the 2 data sets overlap in UV space if you try and use an image as a model, but this isn't totally clear enough to me to explain; I'll keep working on it :)

From the help file:

The task UVSUB can be used to subtract (or add) or divide the Fourier transform of a specified model file(s) from/to/into a visibility data set. The input model may consist of the CLEAN components associated with one of more input files; one or more input images; or, a specified point model. In addition, there are two model methods. The DFT method (CLEAN components and point model only) takes a direct Fourier transform to determine the model visibility values.
The gridded-FFT method does an FFT on either an image consisting of the CLEAN components or another specified image and interpolates the model visibility values. There are at least three principal circumstances for which this is required.
(The division option can be used to solve for "correlator offsets." See VLA Scientific Memo. No. 152.)

The continuum can be subtracted from spectral line channels either in map-plane or in u-v data. If the source size S, the synthesized beam width s, the center frequency F, and the frequency range f satisfy the following constraint

(S/s) * (f/F) < 0.1

then little scaling occurs from channel to channel and the dirty continuum map can be subtracted from the channel maps. The continuum map is generally produced from a sum of the channel
maps which have no significant line radiation or absorption. Cleaning and self-calibration of the differential channel maps may be useful if there is sufficient signal to noise. For the purposes of self-calibration, the sum of the u-v data of the
channels with significant line emission or the original continuum u-v data set can be used to increase the dynamic range. The self-calibration solution can then be applied to each channel.
If the above constraint is not satisfied, then it is more accurate to subtract the continuum model directly from the u-v data for each channel. The continuum model is obtained by summing the channel maps which contain little significant line emission and then cleaning the map to obtain the CC file. UVSUB is then used to subtract this continuum model from each of the continuum u-v data sets. As above, cleaning and self-calibration may be needed for the residual channel data is the signal to noise warrents it.
Because of the differential scaling between channels which may be summed to generate a continuum map, several continuum maps, each composed of relatively closely-spaced frequency channels, may have to be made and the cleaned separately. Each CC file will then have to be subtracted from the channels using UVSUB. Factor will then have to be adjusted so that the sum of the FACTOR's used is 1.0 with each FACTOR akin to a weight of that particular continuum map (i.e. the number of channels).