The first problem we had was a deviation of the position of sources.
Since the reduction process involves a self-calibration step, the true
positions of sources is not known by definition (the self-calibration
process changes the phase in space, which is equal to a shift
in the Fourier Transform of the
space: the sky position). Only
the relative positions are known. By using a list of very accurately
measured position calibrators, the global shift of the whole field can
be calculated. Extra information is available about sources in
overlapping areas. Since neighboring fields have a strong overlap
(grid points of pointing centres overlap at half halfpower beamwidth)
a lot of information about position differences between pointing
centres can be used to shift the mosaic in a bootstrap method.
Since we now used 4 mosaics which lie next to each other we had a good overview of how well the mosaics were overlapping in terms of accuracy. It turned out that in these fields we had a problem that the overlapping fields from neighboring mosaics were shifted by several arcseconds. Bad positions of one or two position calibrators were ruled out, because this would imply a position difference as function of distance to the calibrator.
At this moment (September 1993) we still haven't traced the cause of the problem yet. A lot of possibilities have been proposed and tested and none of them gave a direct indication. First of all the reduction software was examined.
Somewhere in the code where the different pointings are combined (program MOSCOM in NEWSTAR) an approximation is made for the precession. Errors like these should be small, typically 0.1 arcsec. This is very small compared with the position difference we find. One other option is that the number of position calibrators that are in these mosaics is small compared with earlier low declination fields which didn't show any problem. But if the small number of calibrators is the cause then one would expect that i) large position differences are localized near position calibrators (due to the fact that the position of the calibrator isn't correct, because the centroid at 325 MHz could be different than at 8.4 GHz. This is possible if the source has got a steep spectrum halo which isn't visible at higher frequencies), ii) large position differences are localized far away from position calibrators, because many fields had to be shifted (each with their own small errors) in order to shift that particular field. The differences in positions that we see don't show such a characteristic pattern, so we have doubts about serious problems with this part of the software. A few sources were found to have this property, but the amount of calibrators we have (a few hundred) is so large that this can not be a big effect.
There are two ways how we can test the calibrator quality. We can observe the calibrators ourselves with the WSRT at 21 cm, to get a more realistic idea of the centroid of the source (steep spectrum halos should be picked up at that wavelength), or we can calculate back from the optical identifications what the true positions should have been. This last step requires significant processing time (especially on the Leiden part, since fields have to be processed twice to get the correct answer). It's all right to do this one or twice, but not as a standard way for the whole survey. The first option naturally requires more observation time, although it's not that bad. With about two weeks of WSRT observations we could observe thousands of calibrators, so basically each pointing centre should contain one calibrator. Since each pointing centre with a calibrator in it should be shifted correctly without the building up of errors this should be a good option.
Also the reduction software was checked, to see if it was a bug in the program or a wrong algorithm. Although the program has one part where the positions of sources is calculated in a quick and not very accurate way, this can only account for an error in position of about 0.1''.
As a result of this the maps that are transported to Leiden are not the final frames (6 by 6 degrees maps, centred on the new PSS plates positions), but they are the mosaics (about 10 by 14 degrees). These maps are noisy at the edges, because they are not combined with neighboring mosaics. As a result only fields that are away from the edges can be cut out and processed. The processing software can only handle 1024 by 1024 maps with a smooth noise background (The new software written in IDL solved this problem). Since only a general noise level can be specified and not a local noise level, the procedure would detect noise as true emission if we take a field with a high noise level at the edge. This of course limits the area within the Mini Survey by some amount, but it still gives an acceptable amount of sources, even though the overlap has become smaller.