About ROSAT	ROSAT Home Page	ROSAT Images

Notes on the Spectral Response Matrices

   

PROS stores the photon redistribution files and effective area files (and filter information) separately, while XSPEC uses a file which is a convolution of these. PROS uses a 34 channel matrix.   XSPEC uses a 256 channel matrix but it should be noted that 256 PI channels were not originally intended for use during scientific analysis, rather it was always intended that these raw channels be binned so that known remaining digitization effects could be reduced. There appear to be two subtly different philosophies regarding the correct procedure for this rebinning:

After excluding ``bad'' channels, the 256 channel data should be either:

1. binned such that each new bin contains the same oversampling of the spectral resolution.     This is to prevent dramatically different oversampling factors at various energies skewing the spectral fit. The SASS channels were selected to provide this, but a number of people suggest 19 channels are (statistically) the most appropriate number. Thus rebinning the SASS channels by a factor of two is close.  

   The advantage of using the 256 channel spectra and not the 34 SASS channels is the ability to ignore only the raw bad channels without loosing any good data, the 34 channel spectrum is pre-binned such that some good data can be within a channel also containing bad data.

2. binned such that there are sufficient counts in each new bin so that analysis can be used (if desired), i.e., group to a minimum of 20 counts per channel.  

It should be noted that the skewing of the spectral analysis is only likely to be significant for high S/N data from very steep (or absorbed, flat) sources.   The best recipe may therefore be to use a combination of both methods.

While the PROS xspectraldata area contains other matrices, these are not recommended for use. Dtmat.ieee was an early (pre-launch) matrix and dtmat_5.ieee was the first post-launch matrix, both are non-detector specific and use offar.ieee as the effective area file.

As of 1993 July, it is recommended that a 2% systematic error per channel be used for spectral analysis (until the calibration can be improved). Alternatively, use both matrices to see what the effect on the spectral fitting is.