2. Observations and data reduction
The observations were conducted at the WSRT during 29 sessions between 9 March and 14 May 1989, and used a special correlator configuration which provided 28 auto-correlations (14 telescopes 2 polarizations) in addition to a number of cross-correlations between the different telescopes. The cross-correlation data were not used in the data analysis. The observations were done in a frequency switched mode, with the magnitude of the frequency throw equal to the total bandwidth of 0.156 MHz ( km s-1). Each spectrum had 256 uniformly tapered channels. The ON spectrum was centered on km s-1. Either before or after each source observation (which typically lasted a few hours) a standard calibrator was observed in the same frequency switched mode for about an hour.
The data reduction was done in WASP (Chengalur 1996), a package suited to the automated reduction of large quantities of spectral data. For each observation, the data for each polarization of each telescope were reduced separately. Every 60 second ON spectrum was baselined using the corresponding OFF spectrum. Next the following flagging and baselining operations were done iteratively, until the the rms of the remaining unflagged data converged (which it typically did within 3 iterations). The rms over all channels and all time was computed, all points which exceeded 6 sigma were flagged. A linear baseline was then fit and subtracted from each spectrum (excluding these flagged points). Spectra whose rms (over channel numbers) exceeded 2.5 times the global rms were flagged, and finally channels whose rms (over time) exceeded 3 times the global rms were flagged. For all but the first iteration, the previously fit baseline was first added back to each spectrum. These criteria were selected after trial runs on a few sample observations and were tailored for efficient removal of narrow-band EMI and of spectra in which the EMI was strong enough to cause ringing effects. The data from telescopes 4, 5, 6 and 7 (which are the closest to the control building) showed a large amount of EMI, and are almost entirely flagged out. Finally, all the unflagged data were averaged together to produce one average spectrum per polarization per telescope per observation.
Since low-level intermittent EMI sometimes survived this flagging process, this average spectrum was inspected visually and telescopes with broad multi-channel EMI or with low-level ringing were flagged. All the remaining data were averaged together, separately for each polarization of each telescope. These processing steps were carried out separately for the source and the calibrator observations. A order polynomial baseline was then fit to the calibrator spectrum and then this same baseline was removed from the source observations (after allowing for a small scaling difference between the source and calibrator observation). The spectra from two telescopes (both polarizations of RTD and one polarization of RT3) showed substantial baseline structure even after this correction and were dropped from further processing. The spectra from the remaining telescopes were then averaged together and smoothed with a Gaussian smoothing function of 11.2 km s-1 FWHM. A few channels which still showed unresolved peaks, presumably due to EMI, were blanked before smoothing. Since the smoothing length was channels, this had very little effect on the spectrum.
The final spectrum (excluding about 10% of the band on either edge
for which the noise level is higher and the calibrator baseline is
more poorly determined) is shown in Figure 1. The peak-to-peak
variation is about 4 mK, approximately a factor of two smaller
than that of the spectrum in Blitz & Heiles (1987). It is
interesting to note that similar to Blitz & Heiles, there is a
feature of low significance ( peak brightness)
at the velocity km s-1.
We have overlaid a Gaussian profile with peak brightness 2.4 mK
and the same velocity centroid,
km s-1 and FWHM of
18.7 km s-1 as the HI emission feature in this
direction. Excluding this possible feature, the noise level is
0.85 mK over 11.2 km s-1, again about a
factor of 2 lower than the 1.6 mK quoted in Blitz & Heiles
1987 (who also appear to have excluded the putative feature before
computing the noise level). Including this feature in the calculation
of the noise gives an rms of 1.06 mK, while the expected thermal
noise is 0.8 mK.
© European Southern Observatory (ESO) 1997
Online publication: July 8, 1998