Overview

    Figure 7 is a block diagram of the data analysis process for HESSI spectra. The HESSI data analysis software will, like the data, be made public immediately, and should be able to bring users who don't have a detailed knowledge of the instrument to the "dotted line" in the figure: a spectrum in photons/cm²/s/keV with all instrumental effects removed as completely as possible.

    The first stage is to correct for gain drift, deadtime and pileup of nearly simultaneous photons. Gain will be very well known, since there will be many narrow, easily identified background lines. With the two attenuators in operation, we should very seldom exceed 50% dead time, so the deadtime and pileup corrections should remain simple.

    The next task is to identify and subtract background. For most flares, the user will be prompted to select the time just before and after the flare to use as background time. For very long flares, the background might be selected one orbit or even one day before and after the event.

    To convert a background-subtracted count spectrum to a photon spectrum, the response of the instrument must be removed. There are many effects that modify the input spectrum: absorption in the mylar blankets, cryostat windows, and grids; Compton scattering into and out of the detectors; Compton scattering from the Earth's atmosphere (which will dominate the count rate in the rear segments below 100 keV), noise in the electronics, resolution degradation due to radiation damage, the low energy cutoff imposed by the electronics, etc. All of these effects will be accounted for in a database which will be supplied with the software. The database is condensed from laboratory measurements and computer simulations of detector performance and will be updated as the mission progresses. Depending on the user's preferences (and with intelligent defaults), the software will use the various parts of this database to create a response matrix appropriate for each observation.

    When the user is only interested in isolated gamma-ray lines, the response is just the efficiency for photopeak detection, and the conversion from counts to photons is done immediately by dividing by the efficiency. This will also be adequate for hard x-ray flares with no significant component above 100 keV, since the response of the front segments below this point is dominated by complete absorption, not scattering.

    Often, however, the desire will be to correct the entire spectrum, lines and continuum, over a broad energy range. In this case, the usual procedure will be to specify a model form of the spectrum, which can be a combination of either simple functions (power laws, Gaussians, etc.) or physics-based spectral forms (e.g. a set of known nuclear lines from a particular element bombarded by energetic protons or a thin-target bremsstrahlung spectrum from a mono- energetic electron beam). The software will then "fold" this spectrum through the response matrix, check the goodness of fit to the observed count spectrum, and repeat the process, varying the parameters of the input model until the best fit is found. The output of this process is either the best-fit parameters themselves, or else a spectrum created by multiplying the observed count spectrum by the ratio of the model photon spectrum to the model count spectrum.

    The HESSI software will include code that performs the iterative fitting, but we will also provide tools to export the count spectrum and response matrix to the XSPEC package (Arnaud 1996). XSPEC has a wider variety of built-in spectral models than will be available with HESSI. In addition, later versions of the HESSI software will have an algorithm for model-independent inversion of the spectra (Johns & Lin 1992, Smith et al. 1995). This algorithm will probably be most useful for spectra which are dominated by the continuum, not lines, but which extend to high enough energies that simply dividing by the efficiency is insufficiently accurate.