Background Fits for 64-ms Burst Data
The background fits described below are for the concatenated 64-ms burst data. To fit background intervals to bursts, we performed two passes that involve visual inspections. First, we made an attempt at determining background intervals using a simple automated algorithm: the data stream was binned to 1024 ms and the four channels displayed, vertically stacked. The procedure searches for two intervals of contiguous time bins. The first interval has an original starting point at the beginning of the concatenated data stream, and ends 16 s prior to trigger; the second interval begins 32 s after trigger and continues to end of data. That portion of an interval is retained where all count rates are within 3 sigma of their mean; the algorithm succeeds if at least three contiguous points are found in common for channels 1 and 3. The two intervals are then fitted with a quadratic form for all four channels and the background-fitted time profiles are displayed. In any event, and necessarily when the automated interval search fails, the investigator can modify the interval designations. Many considerations would dictate such manual intervention. For instance, besides the apprehensions described above, the algorithm may have determined two intervals which do not entirely bracket the burst. The main objective of this first pass is to attempt a uniform-first approach for all bursts. Unfortunately, close-to-final background intervals are determined for only about half of the events at this stage. An interval that apparently contains the whole burst is also specified manually in this pass.In a second pass, background estimation was refined by careful examination of the concatenated burst record, using a two-screen approach: two investigators simultaneously viewed the data for all eight BATSE detectors, and the four channels of the summed, triggered detectors on two separate displays. On either screen, the data can be plotted either in counts, or in residuals (sigmas) about the background fit. The text from the 4B Catalog Comments Table for each burst is also displayed and reviewed. These comments concern presence in the data stream of occultation steps, varying hard X-ray sources, solar flares, electron precipitation events, etc..., along with their times of occurrence relative to trigger time. (With some experience, we learned to differentiate fairly accurately between these different phenomena by appearance, that is, by their temporal, geometrical and spectral behaviors. The geometrical aspect enters from the directional dependence of emission revealed by the eight-detector display.) Overlaid on the data stream are initial background fits from the first pass (for the option where the chosen ordinate is counts s^-1), along with two pairs of symbols marking the times when 5% and 95% of the burst counts had accumulated. One set of these delimiters represents the 4B Catalog values, and the other set is derived from our own measurements.
This dual viewing approach with attendant information proved to enhance our ability to distinguish between burst and non-burst emissions, particularly near background level, thereby allowing background intervals to be chosen with some confidence. In this second and final pass, the possible number of background intervals is permitted to be larger than two: occasionally a sufficiently long hiatus (apparently at background level) between burst pulse structures should be designated as a third, interjacent background interval in order to capture the background curvature over an extended region. Of the final sample, three (four) background intervals were specified for 78 (7) bursts; five intervals were specified for one burst. Also, the order of polynomial fit could be higher than two; 54 (6) bursts required cubic (quartic) fits.
The specified interval that apparently contains the whole burst, chosen in the first pass, is also displayed and modified when appropriate. The burst interval is always interior to the first and last background intervals, the first point being typically ~10 s after the end of the first background interval, and ~10 s before the beginning of the last background interval. We noticed that since the low-level structure in dim bursts is in practice indistinguishable from non-Poisson background variations, there is a tendency to fit backgrounds in dim bursts closer to, or even including part of, the burst interval.
Obtain a tabulation of the background fits. Obtain a short IDL reader.
If you have a question about CGRO, please contact us via the Feedback form.