Spectral Analysis


Projecting on the Energy axis produces of course a spectrum. If produced from the cleaned and linearized event files (SAXDAS), or enabling the correction switch (XAS), LECS, MECS, PDS and HPGSPC spectra will be produced in pulse invariant PI channels (gain corrected PHA channels for each detector). The PDS instrument is composed by four independent crystals, which have slightly different Energy to PI relationship, so to add together the spectra of the crystal, to improve the signal to noise ratio, one has also to equalize the four spectra, i.e. to bring the spectra to the same Energy-PI relationship. This is automatically done for the PDS in the XAS pipeline. Also the MECS instrument is composed by three independent units. So also in this case to add together the spectra of the three units to improve the signal to noise ratio, one has to equalize the spectra. This is not done automatically by neither SAXDAS and XAS. It can be done in SAXDAS setting to yes the appropriate switch of meevelin.

Extraction Region

For the LECS and the MECS the observer should first decide a source extraction region. The size of this region depends on the instrument psf, on the source spectrum and intensity and on the background intensity. The table lists for the LECS and MECS the radius in arcmin of a region containing a given fraction of the source counts, at different energies (assuming that the source is on axis).

50 % 80 % 95 %
MECS 1.5 keV 1.7 2.7 3.5
MECS 6.4 keV 1.2 2.5 4.7
MECS 8.1 keV 1.2 2.7 4.7
LECS 0.28 keV3.7 6.1 8.5
LECS 1.5 keV 1.7 3.0 5.5
LECS 8.1 keV 1.2 3.0 9.5

The figure shows the 50 %, 80 % and 90 % power radius as a function of the energy for the MECS detectors (courtesy of Francesco D'acri).

In SAXDAS linearized event files the Pixel size of both LECS and MECS DETX DETY, and X Y coordinates is 8 arcsec.
The pixel size of the LECS RAWX RAWY coordinates is 14 arcsec, the pixel size of the MECS RAWX RAWY coordinates is 18.4 arcsec.

In both instrument the background is dominated by the Cosmic X-ray background (at least in the central 10 arcmin radius region of the detectors). Therefore even for rather faint sources (F>10-10 erg cm-2 s-1 or F>a few mcrab) a reasonably large region is suggested (4 arcmin for the MECS and 8.5 arcmin for the LECS) For fainter sources smaller regions are maybe a better choice. When using a small extraction region we suggest in any case to check the results obtained this way with those obtained using the above larger "default" regions.

On the SDC anonymous ftp we provide a number of .arf (effective area) files corresponding to various extraction region radii:
2, 4 and 10 arcmin for the MECS and 4, 8.5 and 16 arcmin for the LECS

Background Subtraction

MECS and LECS internal background is low, but it is distributed not very uniformely in the detectors.
At radii larger than about 10 arcmin the energy dependent vignetting of the mirrors is not negligible (a factor 25 % at 5 keV already at 7 arcmin) and therefore it affects significantly the most important component of the background, the Cosmic X-ray background.
Furthermore, the broad scattering wings of the MECS and LECS concentrators will spread a significant amount of source photons (preferably hard photons) on the whole detector, "hardening" any spectrum extracted at off axis positions, if a bright source is present at the center of the detector.
On the other hand the internal background is rather stable on timescales from thousands of seconds to months).

This suggests that if possible, and for spectral analysis issues at least, background should not be evaluated in annuli around the source region but rather from blank fields, using an extraction region similar in size and position to the source extraction region. . Blank fields event files, accumulated on several different pointings of "empty fields" are available on the BeppoSAX anonymous ftp.

There could be cases in which one has the necessity to use a background accumulated during the observation under study (if there were during the observations larges variations of the background). In these cases we suggest to extract the background in an annulus within the central 10 arcmin (if the target is a faint source, (F<a few mcrab or F<10-10 erg cm-2 s-1), or alternatively from regions as far as possible from the position of the calibration sources, e.g. in the two quadrants opposite to the calibration sources.

The observer should then:

  • extract a background lightcurve and check for any background variability;
  • if no variations are seen the observer is strongly encouraged to use as background a spectrum extracted from blank fields using the same extraction region used for the source;
  • if the background is generally stable with the exception of a few segments, and if the source is faint, the observer may decide to exclude those segments from the analysis using XSELEC or generating a new timewindow in XAS, and use again a background spectrum extracted from black fields;
  • if the background is variable over a large fraction of the observation the observer may want to subtract a background spectrum extracted from the same field of the source, as described above.

    For the PDS and the HPGSPC used in the default "rocking" mode, the background is monitored continuously rocking the collimators on and off source every 96 sec (standard mode, the rocking time can actually be reduced or increased, see the SAX Observer Handbook).
    Therefore the standard procedure for background subtraction foresees the accumulation of on source and off source spectra and their subtraction. This is automatically done for each PDS unit independently in the XAS PDS pipeline .

    For the HPGSPC the situation is more complicated as described in the background section. To obtain a proped background subtracted spectrum the observer has to correct the on source minus off source spectrum for the background difference spectrum. This is automatically done by the XAS pipeline. In such a procedure the off- collimator position (corresponding to -210' from the nominal pointing) provided to yield the most stable results and is therefore suggested to be used (default choice in the XAS pipeline).

    Energy/channels boundary

    In the table below the ignore instruction are listed to be used within XSPEC. We strongly recommend the GOs to use them in performing spectral fits.

    DetectorXSPECignoreEnergy range (keV)
    LECSignore **-10 950-**0.1-9.5
    MECS1ignore **-36 238-**1.8-11.5
    MECS2ignore **-36 238-**1.8-11.5
    MECS3ignore **-36 220-**1.8-10.5
    HPGSPCignore **-7. 60.-**7-60
    PDSignore **-20. 200.-**20-200

    Binning

    Binning reduces spectral resolution. Usually we bin data because we want to get gaussian statistics in our newbin, so to be able to apply the chi2 test. However, data follows intrinsically the Poisson distribution, so we could use a method of fitting based on Poisson likelihood. This is actually possible in XSPEC (using the Cash C estimator) but not very popular. The reason is that we know the chi2 distribution and so we are able to associate a probability to a chi2. On the other hand it is not possible to easily asses the goodness of a fit or discriminate between two models using the C statistics.

    If we do not want to fight with complicate statistic problems, but at the same timewe want to employ the full capability of our instruments, we should make sure that we are not highly undersample the redistribution matrix when binning.

    Furthermore, BeppoSAX instruments spectra have a linear canalization (the width of the electronic channels is constant with energy). Since the instrument resolution generally scales with the square root of the energy this means that the lower energies are sampled much less frequently than the higher energies. Therefore the higher energies give more contribution to the chi2 (in case of any deviation, real or systematic). This means that spectral feature at low energies have an higher probability to be missed than their equivalent at high energy.

    To avoid these problems one could rebin the spectra with a binning factor which is not constant with energy.

    For the PDS this is automatically done in the XAS PDS pipeline , which rebins the spectra in a logarithmic way (the observer provides the minimum and maximum energies for the spectrum and the total number of newbin).

    For the LECS and the MECS we generated a number of rebinnnig files to be used in grppha. These files contain a rebinning such to sample the instrument resolution with the same number of channels at all energies 3, or 4, or 5). These files can be found in our anonymous ftp


    Please send questions/problem reports to: helpdesk@sax.sdc.asi.it