2 The search

  

Figure 1: The test for the presence of a burst at time 0.0 involves two intervals of 17.408 s in which the background is derived, one ending at time -20.48 s, the other beginning at time 230.4 s, see text

We have tried to automate the search procedure as much as possible, for the sake of both speed and consistency. This will allow simulations to be carried out to investigate various scientific problems. Our search has some similarities to that carried out by the on-board BATSE trigger: using the counts in channels 2+3 covering the energy range 50-300 keV; averaging background counts over 17.408 s; and requiring that two modules see a minimum excess over background. We differ, however, in the evaluation of the background as explained below.

  

Figure 2: Geographic positions of the Compton satellite during 7536 triggers. The geographic areas in the denser parts of the plot were excluded from the search, eliminating 3051 triggers

In the search conducted by the on-board BATSE trigger on the 1024 msec time scale, the background is derived over a given stretch of 17.408 s and the trigger test is carried out for the 1024 msec bin following that stretch. The same background stretch is used for the next sixteen 1024 msec bins, so that there is a separation between the end of the background stretch and the test bin of 0 - 16.384 s.

We use a fixed separation between the background stretch and the test bin, and also introduce a second background stretch allowing a linear interpolation of the background for the duration of the GRB, cf. Fig. 1. For a small or zero separation between background stretch and test bin, slowly rising GRBs may escape detection (Higdon & Lingenfelter 1996). If we increase the separation, more slowly rising GRBs can be detected, but then we find that the number of false triggers caused by higher-order variations in the background increases substantially. In the current search, we used a separation of 20.48 s, and placed the beginning of the second background stretch 230.4 s after the test bin, as shown in Fig. 1. Following a burst, we disabled the trigger mechanism for 230.4 s.

  

Figure 3: Equatorial coordinates of 4485 triggers. The effects of CygX-1, Nova Persei 1992, and solar flares along the ecliptic are clearly seen. The remaining background triggers are partly magnetospheric events, and partly GRBs

Artifacts or defects in the data will lead to false triggers. We have systematically searched for gaps in the data, and for constant output numbers (usually zeroes), and, from quality data provided by the BATSE project, for the occurrence of checksum errors. In each case, we set up a time window of exclusion around the defect, so that the automatic application of our search algorithm will not lead to a trigger. Considerable time is lost: the total number of our exclusion windows is over $151\,000$, of which some overlap.

We adopted a limiting S/N ratio for detection in two detectors of 5.0. Our search of the time period TJD 8365-10528 with the criteria described above yielded 7536 triggers. The geographic coordinates of the observatory at the time of trigger are plotted in Fig. 2. Besides some areas near the South Atlantic Anomaly, there are clear concentrations in the south over W. Australia and in the north over Mexico and Texas. The southern concentration has been discussed by Horack et al. (1992). Since these concentrations have nothing to do with cosmic GRBs, we outlined geographic exclusion regions to avoid most of these triggers. With these exclusions in place, we are left with 4485 triggers, which form the basis for classification and discussion.