|
Документ взят из кэша поисковой машины. Адрес
оригинального документа
: http://xmm.vilspa.esa.es/sas/5.4.1/doc/lccorr/node7.html
Дата изменения: Fri Jan 10 22:37:47 2003 Дата индексирования: Sat Dec 22 07:37:40 2007 Кодировка: Поисковые слова: quasar |
A timeseries is created by evselect from an event list which includes events from the exposure/instrument/ccds of interest. It is customary to use the calibrated, whole-camera event lists which are made as one of the standard set of SAS products: lccorr has been designed under this assumption. The event list parent of the TS must be supplied to lccorr as one of its inputs, and an event list that does not adhere to the product standard may not contain all the information required by lccorr.
The events that end up in the time series represent only a fraction of the photons from that source which impinge on the aperture of the mirror. There are two ways in which events may be lost:
The real world is of course never that simple, and some losses are a mixture of both types - eg events that are discarded by the on-board electronics, or by the processing chain that constructs the event list. Some things that belong to the `involuntary inefficiencies' list also behave like selections (eg the lack of events on a bad pixel), because of their all-or-nothing character.
The primary task of lccorr is to attempt to reconstruct the pre-loss flux values. Methods of doing this are discussed in [2]; the algorithms used in lccorr are derived from the equations within this reference. Essentially, the algorithm consists of modelling the original distribution of events over all the columns in the event list, subjecting this distribution (as far as is possible, and subject also to some user specification) to the same inefficiencies and selections as the real events; the original and the attenuated distributions are then both integrated over all degrees of freedom except time; the ratio between the resulting functions of time is the desired result, the exposure function fracexp.
There are 12 columns in the MOS event list: this suggests that an integration over 11 degrees of freedom (12 - TIME) are needed. Actually it is not that bad:
That brings the number to a much more manageable 5 dimensions: time, x, y, energy and pattern. The pristine event distribution is however only a function of the first four. lccorr makes the simplifying assumption that the shape (as opposed to the amplitude) of neither the spatial distribution nor the spectrum of the events is time-variable. The first is a very reasonable assumption but the latter is more or less forced upon us due to the cumbersome nature - a set of spectra sampled at various times - of the input information necessary to correct the time series in such circumstances. If the object of interest has such behaviour (which may be by no means a rare occurrence), a better alternative is to chop the spectrum into pieces for which the assumption of spectral stability is more reasonable.
The attenuation in the camera is a function of all 5 dimensions. It is assumed that, within a CCD, the time-varying part may be separated from the space-, pattern- and energy-varying part. The latter is assembled in practice from a vignetting function (depends on space and energy), a filter transmission (energy only) and a quantum efficiency (energy and pattern). Note that QE is in principle a function also of spatial coordinates; lccorr assumes that it is spatially uniform.
The selection, which in effect cuts up the 5-dimensional space into chunks, within which the event distribution multiplied by the attenuation function is to be integrated, permits considerable freedom. Any combination of filters on energy, on any of the spatial coordinates or flags, and on pattern will in principle be teased out correctly by the task, although a too-finely divided 5-dimensional space will not be conducive to numerical accuracy. Time filters however are only permitted through good time intervals (GTIs), and may not depend on selection on any of the other dimensions or columns except CCDNR. It is not anticipated that this will in practice greatly restrict the user.
lccorr offers command-line parameters to enable or disable calculation of many of the factors causing loss of events. These are tabulated immediately below and described in greater detail in the next subsection.
|
Factor: |
Corrected? | Parameter: |
|
Vignetting |
`hard-wired' yes at present. | - |
| Filter losses | optional, default=yes | treatfiltertrans |
| Quantum efficiency | optional, default=yes | treatquantumeff |
| Cosmic ray masking | optional, default=yes | treatcosmicrays |
| Dead time between frames | optional, default=yes | treatdeadtime |
| GTI selection | optional, default=yes | treatgti |
|
|
The exposure function, after it is calculated, is stored in a column called FRACEXP in the output TS. Although this column is always present in the output, the values may or may not have been used to correct the rate values in that file. This choice is at the discretion of the user via the parameter `applycorrections', and is recorded in the keywords DEADAPP and VIGNAPP. Note that although the OGIP standard for TS requires the presence of both these two keywords, representing respectively the time- and space-dependent parts of the exposure correction, it was found to be undesirable to separate the exposure contributions in this fashion within lccorr: the keywords are therefore given the same value of T or F by lccorr.
If `applycorrections' is set, the exposure correction is carried out by simply dividing the rates and errors by the relevant value of FRACEXP. For those time bins in which the value of FRACEXP is zero, the rates and errors are set to null (IEEE NaN).