The fit for each energy channel was done INDEPENDENTLY, otherwise we would not be able to represent the spectral evolution with time. For this reason, the signal/noise ratio depends on the energy channel - the code tries to fit Poisson noise even if the signal is expected to be small considering the shape of the spectrum.
The spectral fit was done once for the whole event assuming a constant spectrum and was used only to estimate the detector response matrix including atmospheric scattering (with G.Pendleton's code using precomputed DRM on a grid).
In principle, a better signal/noise representation of the time profiles is possible if one uses a time evolving spectrum and makes a simultaneous four channel fit. Nevertheless, the representation of the time profiles used here is presently the best one for 1024 s time resolution.
File names are not distinguished between triggered and non-triggered events. There are two types of files:
*.sg - signals in the form of the expected total counts with the background subtracted,
*.err - errors squared for signal values. For a typical count rate profile, the squared error just equals the number of counts for the signal + the background. Here, the corresponding value is considerably smaller due to the directional fit.
FORMAT: ascii, 5 columns,
*.sg files:
column 1 - time of the day in seconds,
column 2 - counts in first channel ~30 - 50 keV,
column 3 - counts
in second channel, etc.
*.err files: the same structure but with the errors squared instead of the corresponding counts.