The SETHI database consists of about 16 million HI spectra, recorded by the SERENDIP spectrometer at Arecibo. A typical spectrum looks like this:
If one wants to make maps or datacubes with these spectra, we need to combine them all in a clever and scientifically valid way. Spectra of better quality should be weighted higher than lower quality spectra, for example. Quantities like system temperature are important for determining the weighting factors to apply to each spectrum. Removal of interference spikes is a necessary evil too.
Since SETHI records spectra while the feed is moving with respect to the rest frame of the sky (the feed is not TRACKING), the beam of the telescope changes position during the 5.033 second integration. If the beam was tracking, it might look essentially like a gaussian:
Since the beam moves, it really looks more like this:
When the telescope moves really fast, sometimes it'll look like this:
In all these cases, it's important to know the relationship between this beam and the pixels you're going to assign to your map. If the pixels are near the beam center, then they should be weighted quite high. The figure below shows the contours of a typical beam overlaid on some sky pixels in its vicinity. The pixels look kind of funny because they are HEALPix pixels with an angular size of about 1.7 arcminutes, while the resolution of the image is 12 arcseconds per (smaller) pixel. So there are close to 70 little pixels inside each 'qpix' pixel. The light green pixel near the top of the image shows a typical qpix pixel. I made this image using the 'pseudocolour' setting in KARMA , a great visualisation tool.
After applying the right weights to the spectra, we average those that occupy the same area on the sky, and try to remove any baseline ripple that might arise from the reflections of radio waves bouncing around the radio telescope. In the figure below, the white shows the original average spectrum for 5 spectra corresponding to qpix number 830000, with celestial coordinates RA=2.6484375 hours, DEC=24.480787 degrees. Overlaid in green is the baseline-corrected spectrum, which is found by first subtracting out the HI signal, then doing a polynomial fit to what's left. The polynomial is subtracted from the spectrum, smoothing it out. The blue overlay corresponds to the Leiden-Dwingeloo spectrum for this point on the sky. Not a bad match, considering LDS's resolution is about 10 times worse.
Update: June 22, 2007: Today the 144 "raw" SETHI datacubes were completed. I have made some mosaics of these in tiles approximately 22 degrees per side, so each tile contains 9 of the SETHI cubes (in a 3x3 grid). Below is a link to an image of each of these mosaics, showing one channel of the cube. The velocity is not the same from one mosaic to another, so there is no guarantee that structure can be traced from one mosaic to the next.
Region ABC01 Region ABC04 Region ABC07 Region ABC10
Region ABC13 Region ABC16 Region ABC19 Region ABC22
Region ABC25 Region ABC28 Region ABC31 Region ABC34
Region ABC37 Region ABC40 Region ABC43 Region ABC46
to Kevin's Research page