This chapter describes the aspects of the SIS performance which users should be aware of when reducing and analyzing data. In particular, the various SIS data modes are discussed alongside the special analysis techniques they require. Also described is the issue of hot and flickering pixels. For the latest information about the SIS, please consult the Web pages `SIS News' at the URL
and `Calibration Uncertainties' (§1.7).
Various modes are available for observing a source with the SIS, the choice depending on source brightness, the extent of the source and the time resolution desired. For an observation, two concurrent modes are set: the clocking mode (or CCD mode), which determines how the CCDs are read out, and the data mode, which determines how the charge cloud created by a photon is described. Mode-related keywords which appear in SIS data files are described in §4.12.
The SIS CCDs are operated in what is known as `frame-transfer' configuration. Each CCD is divided into two areas, the imaging area, which is exposed to the X-ray optics, and the readout area, which is shielded. Each imaging-area pixel has a corresponding pixel in the readout area to which the charge is transferred (in 16 ms). Once transferred, the charge resides in the readout pixel until the pixel takes its turn to be read out. Readout is effected by applying pulses of voltage (`clock' pulses) across the CCD which move the charge packets - row by row, pixel by pixel - to the amplifier. Reading out an entire CCD in this way takes 4 seconds (the slower the readout, the lower the readout noise). Each SIS camera has four CCDs, and the time resolution is determined by the ordering of the imaging/readout cycles of the four CCDs. The various CCD clocking modes are:
In addition, a special clocking mode known as the `parallel sum mode' is used in conjunction with FAST mode (see below).
Typically, the charge cloud produced by an X-ray photon is not localized to one pixel but spread out over several pixels. There are three basic data modes, FAINT, BRIGHT and FAST, corresponding to three different schemes of analyzing and telemetering this spread to the ground.
The current generation of analysis software cannot handle FAINT mode data directly. For this reason, they are converted on the ground into BRIGHT or BRIGHT2 mode. Converted BRIGHT mode is an exact emulation of the on-board BRIGHT mode (note, therefore, that `BRIGHT mode data' may have been processed on-board or converted from FAINT mode on the ground). BRIGHT2 mode is not a telemetry mode, but a slight variation of the BRIGHT mode which takes advantage of the ability in FAINT mode of applying two corrections, event-by-event `echo' and `DFE' (more about these effects later), which cannot be made in BRIGHT mode. BRIGHT2 mode spectra have 4096 channels, and cannot be combined with BRIGHT mode spectra because of the different way echo and DFE corrections are treated (see §4.6).
Before giving the properties and analysis techniques associated with each mode, here is a list of the most important issues related to SIS data:
Each CCD chip in the two SIS cameras has a small number of catalogable pixels (approximately 10 out of 175,000) which have a higher than normal dark current. One of the eight chips, SIS0-chip0, has about 50 such hot pixels. Even though the number of these pixels is small as a percentage of the array total, each such pixel produces a false `event' which is not rejected by the on-board digital processor and is telemetered.
In addition to the hot pixels, there is also a class of `flickering pixels'. As implied by the name, they are not persistent like the hot pixels, but repeat with a variety of duty cycles on time scales typically ranging from minutes to days. For example, one may stay active for several days, during which it is `detected' in 10% of all exposures. There are several hundred such isolated flickering pixels per CCD.
The flickering pixels are thought to be a symptom of radiation damage, although we do not know the detailed physics of how they are created. As a result, the rates of flickering pixels show a secular increase, and are causing severe problems of telemetry saturation in many modes. Recent updates can be found in the ASCA Observation Planning Guide at the URL
The rate of flickering pixels depends on the clocking mode: Since they are caused by an accumulation of dark currents, the longer the integrate time, the higher the probabilities are that a charge in a given pixel will exceed the event threshold. Thus the flickering pixel problem is most severe in 4-CCD mode, whereas in 1-CCD mode, their rate remains relatively unchanged since launch.
The spectrum of hot and flickering pixels is very soft: most of the counts being below 0.5 keV. However non-negligible effects are possible at higher energies, in some chips.
True X-ray photons are spread over many pixels of the CCDs, because of the vast oversampling of the XRT PSF. Therefore, for a typical observation, the total number of X-ray photons detected in each pixel remains small. In contrast, hot pixels and the more active flickering pixels are `detected' far more often than the surrounding pixels that received a similar X-ray flux. The FTOOL cleansis takes advantage of this fact to remove hot and flickering pixel events from BRIGHT and BRIGHT2 mode data files.
In one version of the algorithm (method 1), a straight cut-off is used (i.e., any pixels with more than N events are removed). Although this method may be advantageous under certain circumstances, and also serves as a backup, the default is a more sophisticated method (method 2). This method uses a comparison with a Poissonian distribution to reject hot and flickering pixels iteratively. For more details, type fhelp cleansis at your system prompt. As with method 1, a filtered events list is extracted.
The processing script runs cleansis in producing the `screened' events files; ascascreen also runs cleansis by default. It is also possible to run it from XSELECT (use command sisclean, followed by save clean).
The use of cleansis may, in principle, remove a small number of real events (either by misclassifying, or by photons landing on hot/flickering pixels). However, the fraction of hot/flickering pixels is small enough that this is not a major concern.
On the other hand, it is impossible to exclude very low duty cycle flickering pixels. Combined with the fact that the flickering pixel rate is increasing, this may lead to an imperfect subtraction of background. For example, if the PV phase blank sky field is used as background for a more recent observation (particularly of low surface brightness extended objects), it may result in a spurious low-energy `excess'. For more details on this subject, please consult the Ueda et al. article in ASCAnews #4 at the URL
Despite this, for most observations, the practical effects of hot and flickering pixels is limited to the danger of telemetry saturation.
The flickering pixels have a predominantly soft spectrum: it is therefore possible to reduce their number in the telemetry by changing the event threshold or by applying a level discrimination.
The event threshold is a fundamental parameter of the on-board processing, and can be changed only by the consensus of the SIS team. After initial testing during the early part of the PV phase, the event threshold was set at 100 ADU for all chips until 1995 Dec 23; after this date, the event thresholds in ADU have been set to different values for different chips such that they all correspond to an energy of 0.4 keV.
Level discrimination, when activated, plays a similar role (all events whose central pixel is below the threshold are discarded during on-board processing). This can be employed in two ways.
Note that data with and without level discrimination, or with different thresholds for level discrimination during an observation, may be mixed in a single sequence in the latter case. The processing script creates separate `unscreened' and `screened' files in such cases.
The CCDs that make up the SIS detector system are sensitive to optical light. Optical photons can be numerous, each creating an electron-hole pair; they can therefore have the effect of raising the dark level. We have discovered, moreover, that parts of SIS-0 Chip 2 and Chip 3 suffer from `light leakage': when the line of sight is close to the bright Earth, these regions produce many spurious `detections', thought to be due to light entering SIS-0 near these regions. They have effects that are similar to the flickering pixels, although the physical origins are distinct.
Since 1993 July 8, the affected areas have been masked off in observations that use these chips, whenever the line of sight is within 25 degrees of the bright Earth.
Most SIS data are reduced and analyzed in BRIGHT mode, the reason being that for observations done in a mixture of FAINT and BRIGHT modes, the largest homogeneous data set is provided by combining on-board and converted BRIGHT mode data. Moreover, many targets do not yield sufficient signal-to-noise to take advantage of the enhancements that FAINT and BRIGHT2 modes afford.
BRIGHT mode is relatively straightforward, but two key issues deserve elaboration: grade, which characterizes the spatial pixel-distribution of an event, and PI (Pulse Invariance), a linearized version of PHA, which allows events from different chips to be analyzed together and secular gain drifts to be corrected. PI data are only available for BRIGHT mode and BRIGHT2 modes.
To eliminate events due to charged particles and to attain the expected energy resolution, X-ray events in the depletion layer from each readout are identified and classified. The first step in finding an X-ray event is to apply the criterion that the central pixel of a 3 x 3 sq-pixel block must have the highest pulse height. In FAINT Mode, if the pulse height of a pixel which satisfies this criterion is higher than the event threshold, the pulse heights of the 3 x 3 sq-pixel block will be sent to the ground.
In BRIGHT Mode, the eight pixels around the central pixel will be compared with the split event threshold which is lower than the event threshold. Based on the distribution of the pixels which have a pulse height higher than the split event threshold, the event is classified, i.e., it is assigned a grade. With this scheme, all the basic charge distributions found by experiment can be included.
The eight grades (0-7) are extensions of four basic groups: S (single), P (single-sided split; subdivided into Vertical, Left and Rights), L (L-shaped) and Q (square-shaped). The extensions arise because of the existence, in some events, of corner pixels which do not belong to the block since they touch the distributions only at the corners (so-called `detached' corners). Usually, charge from an X-ray interaction (less than, say, 5 keV) is split into 1 or 2 pixels, i.e., S events (grades 0-1) and P events (grades 2-5). However, grade-1 events, which consist of single events with one or more detached corners, should be ignored because most are particle events (it is hard for an X-ray event to split charge diagonally). A sizeable fraction of higher energy X-rays (5-10 keV), which tend to be absorbed deeper in the CCDs, are detected as L or Q (grade-6) events, while grade 7 events are dominated by particle events. Since each pixel included in the final PHA calculation contributes its own read-out-noise, grade 0 events have the best resolution and grade 6 the worst. The SIS team has concentrated on a combination of grades 0, 2, 3 and 4; this combination is therefore the best-calibrated. In some instances, however, it may make sense to add grade 6 events (to increase signal-to-noise at high energies) or to concentrate on grade 0.
Between launch and 1993 November 30 07:30 UT, on-board BRIGHT mode telemetry included events with grades 0-4; since then, events with grades 0-6 are telemetered in BRIGHT mode. FAINT mode telemetry contain all events including grade 7; it is customary to throw out grade 7 events during the conversion to BRIGHT or BRIGHT2 data.
Note that in FAST mode, a simpler version of grade is used: FAST mode grade 0 corresponds to BRIGHT mode grades 0 and 2, i.e., to good data. FAST mode grade 1 corresponds to all other BRIGHT mode grades, i.e., to the remaining good plus the bad grades.
SIS BRIGHT and BRIGHT2 data files have a PI (Pulse Invariant) column in addition to the PHA column. The latter is used to record the amount of charge read out from the 3 x 3 sq-pixel area of the chip that contained the event. The PHA values are therefore closely related to the photon energy with the following caveats.
The PI column is incorporated into the events files and is populated during processing (REV1 and later). In addition, the FTOOL sispi can be used to update the column if necessary. The new values will be based on PHA and other columns, as well as a calibration file, which can be obtained by FTP from
The name of this file begins with sisph2pi_ and the remaining six digits indicate the release date in the form DDMMYY. Obviously you should obtain the latest file. As of 1997 March, the latest file is called sisph2pi_110397.fits. The current setting for the parameter calfile should be overridden with the new file name. In populating the PI column, sispi removes the following features of the PHA column:
Tracking of the secular changes in CTE is a continuing calibration task of the SIS team, based on the annual observations of the Cas A supernova remnant and long-term accumulation of Ni line data in the instrumental background. For the latest data, the calculation of PI values inevitably involves some extrapolation of the past trend (by up to 18 months), which can lead to some inaccuracies. Whenever a new calibration file is announced (via the ascanews exploder, the ASCA GOF homepage, and/or the ASCA Newsletter), the release note will include indications of which observations would benefit from re-running of sispi. For the most up-to-date status of the CTE calibration as of 1997 March, please consult the Dotani et al. article in the ASCA newsletter no 5.
It is also necessary to run sispi whenever FAINT mode data are converted to BRIGHT or BRIGHT2 mode data. This is done automatically, if the conversion is performed using ascascreen or XSELECT.
PI is currently not available for SIS FAST mode data because the PHA-to-PI conversion is position-dependent (FAST mode data do not contain positional information). See §4.10 for more information on how to analyze FAST mode data.
PI has been the default column for spectral extractions in XSELECT since the February 1995 release of FTOOLS for BRIGHT and BRIGHT2 mode data. To change the column used for spectral extractions, use the set phaname command. However, it is recommended that you ALWAYS use PI channels, even for early data.
Each event in FAINT mode has nine PHA values, i.e., a 3 x 3 sq-pixel block, containing PHA values which exceed the event threshold and which are centered on the brightest pixel. FAINT mode data, therefore, give the user the greatest freedom to select events based on the shape of the charge cloud and hence the greatest potential to realize the maximum energy resolution of the SIS. However, since conventional X-ray analysis requires that each event has one PHA value, FAINT mode data are converted using the FTOOL faint which is run as part of standard processing. FAINT mode data are converted to two different data modes: (1) BRIGHT mode; and (2) BRIGHT2 mode.
By converting FAINT mode data to BRIGHT mode, users can combine `converted' BRIGHT and `on-board' BRIGHT mode data, thereby creating the largest homogeneous dataset for their observations (see above). However, in the conversion to BRIGHT, the opportunity is lost to make two corrections which can be important with high-signal-to-noise SIS data, namely, the event-by-event dark frame error (DFE) and echo corrections. These two corrections are applied when converting FAINT mode data to BRIGHT2 mode (this makes, however, the BRIGHT2 mode data incompatible with the on-board BRIGHT mode data; these two cannot be combined).
For REV1 and later, BRIGHT2 data files have been provided alongside the `parent' FAINT files and the `sister' BRIGHT files which are converted from the same FAINT files. This means that for observations done in a mixture of FAINT and BRIGHT modes, users with REV1 data will find four types of SIS data file:
One of the ways BRIGHT2 mode data differ from BRIGHT mode data is in how the echo effect is dealt with. This effect arises in the analog electronics on-board which misassign a small fraction (known as the echo ratio) of the charge in a pixel to the next pixel read out. If uncorrected, it results in an apparent PHA value lower than one would predict for a given photon energy, as well as a misassigned grade. The echo ratio
Given the value of the echo ratio, the grade and PHA (related to energy) of each event can be easily corrected provided that the event data are in FAINT mode. With FAINT mode data, the real PHA of each pixel is estimated in the FTOOL faint.
For BRIGHT mode data, it is impossible to correct the echo effect event-by-event. Instead, the echo effect is accounted for by using response matrices which have the aggregate effect built in.
The echo effect is small, and usually can be neglected. To test whether it does affect your scientific results, the results of using BRIGHT (including converted FAINT data) and BRIGHT2 mode data can be compared. In many cases, the advantage of increased statistics from combining converted and on-board BRIGHT data will outweigh the more accurate echo correction that BRIGHT2 data provide.
Even when no events -- photons or particles -- are detected, a small amount of charge flows through the CCDs. This residual level makes up the dark frame level which must be subtracted from PHA values to derive their true values. The dark frame level varies across each chip and depends on the radiation environment of the instrument. Because of these dependencies, the on-board computer calculates a coarse map of the dark frame for each chip and updates the map every read-out.
During the course of the PV phase, it was discovered that the on-board calculation of this level is not always accurate. The uncertainty, known as the dark frame error (DFE), arises because the dark frame is calculated on board by averaging pixels with PHA values between -40 and +40 ADU (1 Analog-to-Digital Unit is about 3.5 eV). When fluctuations in the dark frame occur on time scales less than the averaging period, the dark frame is not correct.
DFE can be divided into three components:
Similar jumps in DFE occur when the satellite leaves the SAA. This happens because the analog electronics of the SIS are turned off in the SAA by the Radiation Belt Monitor, keeping the dark frame at its pre-SAA level. As the particle background raises the dark frame, the dark frame does not attain its true level until several readouts have elapsed.
DFE is corrected in FAINT mode by examining the PHA distribution of the corner pixels of each 3 x 3 sq-pixel event. Only grades 0, 2, 3 and 4 are used because the corner pixels of these grades are not thought to contain charge due to X-rays or particles, i.e., they represent the true dark level. Since DFE varies with time, a `DFE light curve' is calculated - 100 events are needed to determine the peak value of the corner pixel distribution with an accuracy of 1 ADU. The calculation of DFE is performed by the FTOOL faintdfe and the correction itself made by the FTOOL faint.
Although DFE is small, its maximum value of about 0.07 keV means that it should not be neglected in studies of spectral line features. The extent to which your scientific result can be tested, again by comparing BRIGHT and BRIGHT2 mode data.
The algorithm implemented in old versions of faintdfe, distributed with FTOOLS v3.4 and older, worked poorly for data which suffer from the effects of Residual Dark Distribution (see below). Newer versions (FTOOLS v3.5 and later) of faintdfe recognizes 3 different definitions of the zero level.
Please refer to the Dotani et al. article in ASCAnews #4
for full details.
Starting with FTOOLS v3.5 version, faint allows the use of a DFE file (the output of faintdfe) to convert to BRIGHT mode, but only if the DFE file is created with zerodef=2.
The rationale is that the day/night DFE effect can be corrected better in ground processing than on-board. Anticipating this, the SIS team has been using FAINT mode preferentially during orbital day time, and BRIGHT mode during orbital night (when the DFE is small and stable), when a mixture of the two is requested by the Principal Investigator. To take advantage of this, you must first convert the FAINT mode data to BRIGHT mode using faintdfe with zerodef=2, then:
For the first option, no special technique is necessary: we recommend this as the default analysis method that maximizes the statistical quality of the data, while for best calibration we recommend the use of BRIGHT2 mode data only. For the on-board BRIGHT mode data, an additional mkf selection of NSAS should be applied (see §5 and §6).
Radiation damage has not altered the relationship between the energy of incoming X-ray photons and the number of electron-hole pairs generated by them. Rather, it has created a number of charge traps; as the charge is transferred for read-out, each trap removes a constant fraction: this is called the Charge Transfer Inefficiency, or CTI.
The effect of CTI depends on the number of charge traps that between the pixel where the X-ray event was detected and the read-out node. There exist a sufficient number of traps that a statistical prediction can be made of the average reduction in charge, given the RAWX and RAWY positions of the event. The use of PI, rather than PHA, column restores the gain as described above.
However, the distribution of charge traps is not completely uniform, and thus the gain change may differ from one row to the next. This has the effect of spreading the PHA values measured for a mono-energetic input X-ray, i.e., a degradation of the spectral resolution.
The SIS team has derived an empirical formula for this:
where RON is the read-out-noise (5 e RMS at launch), is the Fano Factor, and is the input photon energy, and is the FWHM resolution. The Fano factor was 0.12 at launch; the effect of non-uniform CTI has been modeled by the SIS team as a linear increase in the effective Fano factor:
Where is time since launch in seconds, and is 3.3 for SIS-0 and 2.9 for SIS-1. The effective RON has become significantly (somewhat) worse than 5 e RMS in 4 (2) CCD mode due to RDD effects.
For the latest report on SIS spectral resolution, please read the Dotani et al. article in ASCAnews #5.
Dark frame is calculated on-board by averaging events with PHA in the range 40 to +40 ADU, which would give an accurate indication of the true dark level if the distribution of dark current is symmetric around 0. However, radiation damage has produced a population of pixels with higher dark currents (the extremes of which are detected as flickering pixels), thus skewing the dark frame distribution: This is known as residual dark distribution (RDD; residual after the dark frame correction, on-board or on the ground using faintdfe). Although this is primarily the spectral distribution of dark current values integrated over a chip, the term is often (loosely) applied to the spatial distribution of dark currents. To distinguish, we will use the term RDD map for the latter. RDD has become a major problem for 4-CCD mode, and to a lesser degree, 2-CCD mode.
RDD can lead to mis-classification of event. For example, a grade 0 event may be recognized as a higher grade event because a neighboring pixel is very active. For higher grades, this effect is worse and can even result in a true photon event being classified as grade 7, i.e., as a rejected particle event. Thus RDD causes the effective quantum efficiency of the SIS to become lower. In addition, RDD degrades the spectral resolution: this can be understood because, even in the simplest case of a grade 0 photon correctly recognized as a grade 0 event, the PHA value may be incorrect depending on the level of dark current in the central pixel. When mis-classification is involved, the error in the resultant PHA value can be even larger. Since RDD depends on readout time, it is worse in 4-CCD mode than in 2-CCD and 1-CCD modes.
The SIS response generator, sisrmg, includes a model of the degrading spectral resolution due to RDD. However, it does not yet have a model of the degrading quantum efficiency. Both faintdfe and sisrmg rely on an analytical model of the RDD (the spectral distribution); as of 1998 July, it appears that the RDD spectral function needs a significant revision, particularly for 4-CCD mode data.
It appears that each pixel of each chip of SIS has a characteristic dark current level, which is stable over several months, resulting in a similar dark charge level for a given clocking mode. After some processing to remove particle events and average out fluctuations, we produce the so-called ``RDD map'' files 4.1 In the early days of ASCA, Frame mode data were not taken regularly. The RDD map files exists for 4-CCD mode data only in 1993 Apr, 1994 July (except chip 0 on both SIS-0 and SIS-1), 1994 Nov, and in 1995 Mar; In 1995 Apr, Frame mode data were taken in 1, 2, and 4 CCD mode in all chips that are frequently used in the respective clocking modes. Starting in late 1995, we have started taking Frame mode data regularly. The resultant RDD maps are available at
The file names have the form rdd9304_s0c0m4_t61_62.fits encoding the year and the month, instrument and chip numbers, the clocking mode, and the CCD temperature range.
RDD correction can be applied to Faint mode data file, by using the FTOOLS correctrdd, supplying the input FAINT mode file name (from the unscreened directory) and the appropriate RDD map file names. However, it is necessary to apply both the RDD correction and the DFE correction (which corrects for the dark frame changes within an observation due to, for example, the light leak), and the interplay between the two can lead to errors. It appears that the order faintdfe, correctrdd, then faint minimizes such errors.
Files corrected using the FTOOL correctrdd will have a keyword RDDCOR_V that tells sisrmg not to apply the RDD-related spectral resolution degradation model.
Please note that this procedure is still experimental, and questions remain regarding the current and the ultimate effectiveness of RDD correction. Issues are:
Information in this section (and possible updates) can also be found at:
We have recently discovered another effect that may lead to spurious measurement of N in SIS data: the current version of sisrmg uses a simplified model of how 2-pixel events interact with event threshold. When the event threshold is significantly different from the default, the efficiency predicted by sisrmg for energies event threshold is too high for SIS events of grades 2, 3, or 4. The same observations for which RDD is significant are the ones that tend to suffer most from this effect (since radiation damage also increases the danger of telemetry saturation in a clocking-mode dependent manner). Also, the RDD effects tend to increase the 2-pixel events and make this effect more important (in early data, the fraction of 2-pixel events is 10% below 1 keV).
The situation is much more benign for grade 0 (single pixel) events; we recommend fitting a grade 0 spectrum and a grade 234 spectrum simultaneously, whenever the event threshold is significantly greater than 0.4 keV. More details can be found at:
FAST mode requires special treatment in XSELECT, and analysis techniques for light curves and spectra are different from BRIGHT and BRIGHT2 modes. Although FAST mode provides no imaging information, a crude indication of position is provided by address discrimination.
The high time resolution of FAST mode is achieved at the expense of positional information. Instead of giving the pixel on which an event was centered, FAST mode simply indicates when the event occurred (with a precision of 15.626 ms) and whether it occurred inside or outside the prescribed address discrimination area. In the case of a strong point source, address discrimination is used to allow the exclusion, during reduction and analysis, of the central peak of the PSF which is likely to suffer from photon pile-up. The area itself is specified by the mission operation scientists and takes the form of a horizontal strip. In the FITS science file, the keyword Sn_STAHm gives the starting horizontal pixel (n is 0 or 1, depending on whether the file is from SIS0 or SIS1; m is the chip number), while Sn_ENDHm gives the end horizontal pixel. (To get the value of a FITS keyword, you can use the FTOOL fkeyprint, or, if you're uncertain of the keyword name, you can use the FTOOL fdump which can dump the entire header). Typically, a width of 2 arcmin is used. Setting such a narrow discrimination area before an observation takes place is difficult since it is not possible to predict, with the required precision, where the source will appear on the chip in question. Instead, the source is observed first in BRIGHT mode and the area is then set on the basis of the quick-look data taken after the first pass over the ground station.
In the FAST mode file itself, the arrival time of each event occupies the vertical (V) address of the event and is assigned to the RAWY column. Whether the event is inside or outside the selected area is given in the RAWX column (0 for inside, 1 for outside).
On-board address discrimination is used not only to exclude piled-up photons but to avoid telemetry saturation. This is done by enabling the address discrimination, that is, to telemeter only those events which occur inside or outside the selected area. The keyword Sn_ARENA takes the value 1 if address discrimination is enabled, 0 if it is not. Note that the decision whether to enable address discrimination is made before the observation by the ISAS and GOF duty scientists in consultation with the Principal Investigator. In addition, for FAST mode data without on-board address discrimination, the XSELECT command select FAST provides a direct and transparent way of identifying FAST mode events based on whether they occurred inside or outside the selected area.
The FAST mode events are classified into two grades; FAST mode grade 1 is usually dominated by the particle background, and should not be used in timing or spectral analysis. FAST mode grade 0 roughly corresponds to BRIGHT mode grades 0 and 2.
Before working on the FAST mode data themselves, obtain an image from GIS PH mode data or SIS BRIGHT mode data, and determine the source position in (unbinned) DETX. Note that, for the GIS PH mode data, XSELECT produces unbinned images by default, but for the SIS imaging mode data, it rebins the images by a factor of 4 by default. Therefore you should revert to the original SIS binning by specifying set xybinsize 1 in XSELECT. The source position is necessary for time assignment; an extreme error of the length of a CCD chip (about 11 arcmin) results in an error of about 80 msec in typical cases.
Next use ascascreen; the DETX position, and the detector you used to determine it, are necessary inputs for using ascascreen for FAST mode data.
In addition to the customary questions, you will be given the opportunity of selecting events based on whether they occurred inside or outside the address discriminator. You need not do the selection at this point: such a selection can also be applied later in XSELECT on the output files from ascascreen.
WARNING: In FAST mode you're likely to be dealing with a huge number of events and a similarly large number of time bins. XSELECT might suffer from problems related to running out of memory/diskspace.
The rate of hot and flickering pixels is much lower in FAST mode because the chips are read out faster: there is simply less time for a flickering pixel to accumulate enough charge to cross the event threshold. However, they do occur ( there is one particular hot pixel that plagues many (but not all) sets of SIS-1 Chip-3 FAST mode data), and present a real problem since the image-based methods to remove them cannot be used. This problem is manifested as a DC offset in light curves and as a huge peak at PHA channel around 200.
One aspect of working with FAST mode light curves requires special attention. If the address discrimination area is narrow, then, as the satellite attitude fluctuates, so will the number of events inside the discrimination area, thereby introducing spurious variability. Users can avoid this problem by including events outside the discrimination area in their analysis.
Each SIS camera is made up of four 420 x 422 sq-pixel chips separated by gaps along the spacecraft X and Y axes of 18.5 and 4.8 pixels, respectively. Figure 4.2 makes this clear and also shows the following.
In XSELECT, once the data mode has been set, the various mode-dependent quantities, such as the number of energy channels, are set automatically. It is, however, useful to know some of the corresponding keywords in the SIS data file.