Radio data reduction... How I love thee...

So this was going to be my procrastination from flagging RFI from my radio data; I was going to ramble on about the 'fun' that is radio interferometer data reduction, that is how you go from a bunch of numbers representing correlations between every antenna and every other antenna in your radio interferometer array to something resembling a decent image of your source at radio wavelengths with a resolution far far better than what you can achieve with just one dish. Instead, I actually flagged the data instead... Go figure, but hey, I'll start now.

So (assuming that one's actually got data from a telescope and the telescope was actually pointed where you wanted it to be pointed), how exactly does one go about going from 'bunch of numbers that represent a Fourier transform of the sky' to pretty image?

Well first you need to prepare the data. You start with having a source of known flux, and a source which is a point source so that the phase variation of the radiation from it is just intrinsic and not atmospheric. So before you start your calibration, you load up your data and look for evidence of RFI... Spikes of high intensity radiation at a given time and/or a given frequency channel. And you do this for each and every baseline (i.e. for each antenna correlated with every other antenna... There are 30 GMRT antennas... So there are 435 baselines to go through...) at two different polarisations (so 870 arrays of data to look at and its tedious... The data basically look like a load of white on black barcodes... I can see them in my sleep!), flagging anything that looks suspicious. And then you calibrate -- first the antenna gains for that particular observation. Then you calibrate for the phases, and figure out what your bandpass should look like. Then you apply all this calibration to the data of your source. Then you remove the noise spikes from the source (looking at 870 arrays of data again!).

Then and only then can you start to consider how to do an inverse Fourier transform of your data to create an image... A problem which is further compounded because your sampling of the Fourier plane isn't perfect or complete, so you can't actually do a proper inversion... The fun is endless...


And in case you hadn't realised... Do you really think I love this rigamarole? ;-)