The particle models have been parameterised as a function of Master Veto Rate (MVR). They are valid for MVR values between 0 and about 170. Therefore you should produce a set of "clean" times, when your data has an MVR between 0 and about 170, using XRTHK. This will reduce the exposure time by about 10%, but gets rid of periods of high particle activity. If you find that too much data is being lost by excluding the high MVR values, then the high limit may be increased somewhat. NB: The particle model has only been validated for observations upto the end of April 1991; for observations after Sept 1991 it will certainly be inaccurate.
Then produce a source file, using the output file from XRTHK to define the sort times. The effect of particles is well determined between pulse height channels 8 and 249 (essentially no photons are recorded outside this range), so when sorting, you should restrict the corrected PH range to be between these values. It should be noted, however, that observations taken after the PSPC HT voltage was reduced, have useful data between channels 11 and 249. There is a certain amount of particle leakage at the edge of the detector where the vetoing is inefficient. Unless absolutely necessary, source boxes should be wholly within 57 arcminutes of the optical axis. When subtracting full images, you should be aware that regions close to the edge are liable to be poorly subtracted. This is the official line, in practise it seems that there are problems at greater than fifty arcminutes from the centre. The source file may be a spectrum, an image or a spectral image (a time axis may be added to any of these files, however a source time series may only be subtracted using a background time series, in which case spectral information is lost and so the vignetting correction will be less accurate.
Next produce a background file, sorted over the XRTHK times. The background area needs to be as large as possible, to give good statistics, but should cover only a narrow range of off-axis angles, to give a well defined position correction . These contrasting requirements can best be met by using an annular background box of large radius (an inner radius of 0.5 degrees and outer radius of 0.6 degrees has been shown to give good results). The problem then, is that such a box will be disected by the PSPC support structure and will doubtless include several bright sources. One solution is to sort all the data into a spectral image or temporal_image and than set the quality of bright sources and the ribs bad using a combination of WELLARD (now XRTARD) and ARDQUAL (and possibly IGNORE, CIGNORE and CRESTORE). The program will then correctly calculate the area of the background region. NB: This technique minimises the effect of gradients in the background, however, you should check your subtraction using a local background if possible.
Now subtract the two files using XRTSUB. The effect of particles is taken into account by default, although this may be turned off by setting PARTICLES=NO on the command line. A file containing the calculated particle spectrum can be produced by setting PART_FILE=filename on the command line.
In summary the following may be of use in subtracting a spectrum, image or spectral image.