Our algorithm yields very satisfactory results for long PSPC sequences
(exposure time 
ksec), since in these cases the
observations have high probability to be contaminated by particle
and/or solar background. Figure 5 (click here) shows the results of the
screening procedure on RP 200020, a sequence pointed at the Hyades open
cluster with 40412 s exposure time. The sources have been detected
using the Wavelet transform algorithm developed by Damiani et al.
(1997a). In this run, our algorithm has rejected 
s (8%
of the total exposure time). In the figure, we show a scatter plot of
(SNR of the sources in the screened observation) versus 
(SNR of the sources in the unscreened observation). The detection
threshold (
) is marked by the dashed lines
, while the solid line divides the plot in the
region where 
(upper left) and the region where
(lower right). We have displayed only the sources with
4<SNR<6, to stress the behavior of the screening algorithm on
faint sources. In the plot, we have 15 sources with , of
which 8 were not detected in the unscreened data but were detected in the
screened data when running the detection algorithm at the 4.35 
threshold. These sources are in the upper left shaded area. On the
other hand, we have 9 sources with , of which none moved
below the detection threshold after the screening (lower right shaded
area). We therefore have 8 new detections in the screened data
set. Since there are 6 sources detected with 4.35<SNR<6 in the
unscreened sequence, we have obtained a gain of more than 100% in the
number of faint detected sources by running our screening
algorithm. In addition, we have an average SNR gain of 
% for
the matched sources with 4.35<SNR<6 (2.5% considering all sources
with SNR>4.35). This is in agreement with the results of previous
tests on simulated data sets which showed the advantages of tailoring
the algorithm for the case of a faint template source. We note that at
the chosen threshold, the algorithm is expected to detect one spurious
source per field (Damiani et al. 1997b), and therefore
essentially all the new sources in the screened sequence should be real. To
provide more confidence on this result we have tried to find counterparts
for the 8 new sources using the SIMBAD database and the MPE ROSAT PSPC
Catalogue, finding counterparts in 5 cases and none in the remaining.
Figure 5: 
scatter plot of RP 200020. Symbols represent off
axis angles of displayed sources in arcmin (legend in the plot).
Arrows mark upper limits for sources detected only, for instance, in
the screened observation and not detected in the unscreened one or
vice versa. Dashed lines correspond to the typical detection threshold
used in the Wavelet transform detection algorithm of Damiani
et al. (1997a,b), while the solid line marks 
. Sources moved above detection threshold after the screening process lie
in the upper-left dashed area with a dotted line pattern (8 new sources),
while sources moved below the detection threshold would be in lower left
dashed area (none in this case). New sources are not concentrated at large
off-axis angles, where the PSPC PSF is not circular simmetric, but they are
at a wide range of off-axis angles
We have verified that for short PSPC exposures (
ksec), our
algorithm tends to reject less than 
% of the total exposure
time, in agreement with the expectations. In these cases, the average
SNR gain is very small, and the average number of gained sources is
negligible.