/ ciao / caveats / acis_caveats_20160527.html

Caveats about ACIS Pipeline-Processed Event Data

Last Updated: 20 Nov 2017

This document describes potential problems or concerns about ACIS event data processed using the standard pipeline. The issues are categorized according to the nature of the caveat. Users may find the ACIS Data Preparation analysis guide to be a useful companion to this information.

TIME

As of DS 7.6.0 (15 June 2005), the values of TIME associated with events for continuous-clocking mode observations are the times of arrival at the detector, instead of the read-out times. Associated with this change is a change to the good-time intervals (GTIs). The START and STOP times in continuous-clocking mode GTIs are now appropriate for the times of arrival instead of the read-out times. The values of TIME for observations processed using earlier versions of acis_process_events are still the read-out times. The older data have been reprocessed. Note that the times of arrival are at the detector, not the barycenter of the solar system. The tool axbary can be used to compute the times of arrival at the solar system barycenter from the times of arrival at the detector before performing timing analyses.
The value of the keyword TIMEDEL in some ACIS continuous-clocking mode event data files is incorrectly set to 1.4592 s instead of 0.00285 s. While this does not represent a problem if the pipeline-produced files are used for timing analyses, it is a problem if an affected event data file is reprocessed using acis_process_events because the values of the TIMEs in the output file will be corrupted. The problem affects data processed with software versions DS7.6.0 (June 2005) through DS 7.6.7 (May 2006); the correction was made in DS 7.6.8 (June 2006).

To see if your data file is affected, check the ASCDSVER (software version) and/or TIMEDEL header keywords in the evt1.fits file:
```
unix% dmkeypar acis_evt1.fits ASCDSVER echo+
7.6.2

unix% dmkeypar acis_evt1.fits TIMEDEL echo+
1.4592
```
This file was processed between DS 7.6.0 and DS 7.6.7 and has the incorrect TIMEDEL value, so we use dmhedit to update it:
```
unix% dmhedit infile=acis_evt1.fits filelist="" op=add key=TIMEDEL value=0.00285 unit=s

unix% dmkeypar acis_evt1.fits TIMEDEL echo+
0.00285
```
If you do not make this change before reprocessing the data with acis_process_events, the TIME values in the output file will be incorrect.
As of DS7.6.0 (15 June 2005), the START and STOP times of continuous-clocking mode good-time intervals (GTIs) are aligned with the beginning and ending times of frames. The START and STOP times for GTIs processed using earlier versions of acis_build_chip_gti could occur at arbitrary times during a frame. As a result, the keyword EXPOSURE may be inaccurate for older data. This problem affects the computation of the flux of a source, but should be negligibly small for most users (i.e. the uncertainty in the affective area of the detector produces a larger error in the flux than the inaccuracy of the EXPOSURE time). Concerned users should contact the CXC Helpdesk for assistance. The older data have been reprocessed.
The values in the column TIME of the continuous-clocking Level 0 event files were computed incorrectly before 5 October 1999. The contribution to the time associated with the value of TROW was double counted. This problem was propagated to the Level 1 and Level 2 event files. The problem was corrected as of 5 October 1999. The earlier data sets have been reprocessed.
Before version DS 6.9 of the processing software (10 August 2002), the values of the keywords EXPOSURE, LIVETIME, DTCOR, ONTIME and a few others were computed incorrectly in some observations, especially those where the frame time was chosen to be less than the optimal frame time; see also the Time Keywords in Interleaved Data caveat. (For descriptions of the meanings of these keywords, see the ACIS exposure time keywords page.) As a result, analyses of the spectral distribution or total count rate from a source may have produced inaccurate values. This problem has been fixed. Data processed with DS 6.9 or higher should have accurate EXPOSURE values. The older data have been reprocessed. Users concerned about the accuracy of the EXPOSURE for their data should contact the CXC Helpdesk.
The values of TIME associated with data collected using very-long exposures (e.g. the RAW, HISTO, and GRADED_HISTO DATAMODEs) are not reliable because the counter associated with the front-end processor rolls over every 328 s and this behavior affects the time assigned to the events. These data sets are generally analyzed by only the instrument scientists.

PULSE HEIGHT

The charge-transfer inefficiency (CTI) of the front-illuminated (FI) and back-illuminated (BI) CCDs results in a degradation of the energy resolution and an apparent gain shift. These effects are worse for events that occur far from the readout of a CCD (e.g. near the ACIS-I and ACIS-S aim points). The tool acis_process_events can be used to compensate for some of the effects of CTI. Applying the temperature-dependent CTI adjustment significantly improves the energy resolution of the CCDs and removes most of the effects of the apparent gain shift from the ENERGY and PI data (but not the PHA data). The improvements are most noticeable near the tops of the FI CCDs and near the one-quarter and three-quarter chip node boundaries of the BI CCDs.
Since it is practical to calibrate the effects of CTI only every few years, more or less quarterly updates to the calibration of the apparent gain shift are used to apply a time-dependent gain adjustment to the event pulse heights. This adjustment is applied using the tool acis_process_events, which interpolates between epochs. A bug in the time-dependent gain algorithm resulted in a mean -0.5 adu (about -2 eV) shift to the adjusted pulse heights. This problem applies to all ACIS event data processed using the time-dependent gain adjustment in software versions DS 7.3 to DS 7.6.9. Users can reprocess their data using the CIAO 3.4 (or later) version of acis_process_events by running the chandra_repro script or by following the Create a New Level=2 Event File thread to fix the problem.
The values in the columns PHAS_ADJ, PHA, ENERGY and PI of an ACIS event data file may not be accurate for continuous-clocking mode observations. The computation of the CTI and time-dependent gain adjustments (i.e. of PHAS_ADJ and, hence, PHA) and the computation of ENERGY from PHA (and, hence, PI) require knowledge of the CHIPY location of an event. In continuous-clocking mode observations, the CHIPY coordinate is used for timing information instead of spatial information. However, if the right ascension and declination coordinates of the source (i.e. RA_TARG and DEC_TARG) are accurate, then these coordinates and the aspect-solution data can be used to determine the CHIPY location of the source. As of DS 6.5 (12 December 2001), the tool acis_process_events uses the inferred CHIPY location of the source when the values in the columns PHAS_ADJ, PHA, ENERGY and PI are computed. The older data have been reprocessed. Users can reprocess their data using the CIAO 3.3 (or later) version of acis_process_events by running the chandra_repro script or by following the Create a New Level=2 Event File thread.
The energy scale of the ACIS CCDs is calibrated only over the range 0.277-9.886 keV. Users should be cautious about analyzing data near or beyond the endpoints of this range.
Prior to 1 July 2005, the algorithm used to compute the one-dimensional bias maps for continuous-clocking mode observations was susceptible to the deposition of a lot of charge by a cosmic ray. If the charge is deposited in a long vertical streak along the length of one or more adjacent columns, then the bias values for those columns can be anomalously high by up to several tens of adu. As a result, events with small pulse heights are not telemetered because they do not satisfy the minimum pulse height threshold and events with large pulse heights have their pulse height amplitudes (PHAs) systematically underestimated. This problem occurred often for front-illuminated CCDs, but did not affect back-illuminated CCDs. Since a revised version of the algorithm was implemented on 1 July 2005, the rate at which these problems occur has been greatly reduced, far fewer columns are affected, and the bias values in the affected columns are only a few adu too high instead of tens of adu. Therefore, spectral data from continuous-clocking mode observations on or after 1 July 2005 are relatively from such bias problems.

The problem may be diagnosed by plotting a histogram of the bias for a CCD. If the bias exhibits "spikes" in some columns, the bias in the affected columns is inaccurate. The problem of a systematic offset in the reported PHA values for events in the affected columns can be remedied by reprocessing the data using an appropriately adjusted bias. Users should contact the CXC Helpdesk if they think that their data is affected.
The files used to determine the gain and spectral response of the ACIS CCDs have changed several times over the life of the Chandra mission. These changes reflect reductions in the temperature of the focal plane, increases in the charge-transfer inefficiency, and improvements in the understanding of the performance of the detectors. Users interested in the PIs or ENERGYs of events should verify that their data were processed using the latest gain file appropriate for the focal plane temperature of their observation. If necessary, an ACIS event data file can be reprocessed using acis_process_events to update these columns (as discussed in the ACIS data preparation guide). Users analyzing PI spectra should ensure that the P2_RESP (or FEF) file used to produce a PI response-matrix file (RMF) for their data matches the gain file used to process the data. This condition should be satisfied if the latest appropriate versions of the P2_RESP (or FEF) and gain files are used. If users are concerned about a possible mismatch, they should contact the CXC Helpdesk.
If an ACIS event data file is processed with the acis_process_events parameter rand_pha=0 (the default value is rand_pha=0.5), then the values in the column ENERGY are quantized. The difference between one energy and the next is about 4 eV. (The exact value is energy dependent and varies from region to region on a CCD). Since the value of PI is a quantized representation of the value of ENERGY (as described in the CIAO dictionary) and since the width of a PI bin is not an integer multiple of the spacing between adjacent ENERGY values, the quantization of ENERGY can cause periodic spikes in the PI spectrum of a bright source. That is, the width of a PI bin is 14.6 eV by default, which is not an integer multiple of 4 eV. Therefore some PI bins contain three quantized values of ENERGY and other bins contain four.

This problem is avoided by running acis_process_events with the parameter rand_pha=0.5 (the default value). In this case, a random component uniformly distributed between -0.5 and +0.5 adu (i.e. between about -2 eV and +2 eV) is added to the value of PHA when the value of ENERGY is computed. (The value of PHA in the output file is not changed.) Observations performed before 3 July 2001 do not have this randomization enabled. These data were reprocessed to add the randomization. Users can also rerun acis_process_events to perform the randomization themselves or to remove the randomization if it has been applied. Note that this problem only affects PI spectra. PHA spectra are unaffected (even if rand_pha=0).
The meaningful values of the column PI in an ACIS event data file are
```
  n <= PI <= 1023,
```
where n is approximately 6 for the two back-illuminated CCDs (ACIS-S1 and ACIS-S3) and approximately 10 for the eight front-illuminated CCDs. (The exact value of the lower limit varies from region to region on a CCD.) PI = 1024 is an overflow bin. PI = 0 corresponds to the special case where PHA <= 0. This value should not occur normally.
The values in the columns PHA, ENERGY, PI, GRADE, and FLTGRADE of an ACIS event data file may not be meaningful if the observed source is a bright optical source (e.g. Jupiter). Users should contact the CXC Helpdesk for assistance with the analysis of the ACIS data for such observations.

COORDINATES

Prior to DS 7.6.4 (09 November 2005), the data for the SIM rotation angle (DTHETA) in the aspect solution files were used incorrectly when computing the values of DETX and DETY (and, hence, X and Y) from the values of CCD_ID, CHIPX and CHIPY. Since the values of DTHETA are typically quite small (0.0001 rad), the change in the coordinates should generally be less than one ACIS pixel. The problem has been fixed as of DS 7.6.4. Older data in the archive have been reprocessed. Users can reprocess their data using the CIAO 3.3 (or later) version of acis_process_events by running the chandra_repro script or by following the Create a New Level=2 Event File thread.
For sources near the optical axis of the telescope, the size of the point spread function is smaller than the size of the ACIS pixels. The sub-pixel event-repositioning algorithm "EDSER" of Li et al. (2004, ApJ, 610, 1204) has been incorporated into the tool acis_process_events. As of the implementation of version DS 8.4 of the Standard Data Processing (SDP) code in the pipeline (28 June 2011), the EDSER algorithm will be used by default. Older data in the archive will be reprocessed. Users can reprocess their data using the CIAO 4.3 (or later) version of acis_process_events by running the chandra_repro script or by following the Create a New Level=2 Event File thread. The sub-pixel adjustments can be removed by running acis_process_events with pix_adj=NONE or RANDOMIZE.

EVENT FILTERING

In addition to the onboard event filtering, users analyzing ACIS event data should ensure that the data have been filtered to exclude events that may not be "good" X-ray events. The data filtering may include the STATUS or GRADE of an event, events occurring during time intervals where the cosmic-ray background rate is substantially larger than the nominal rate, events associated with a cosmic-ray "afterglow," events occurring on pixels that are hot or that have bad bias values, events found in horizontal "streaks," and so forth.

The column STATUS in ACIS event data files contains a bit-encoded description of possible problems that may be associated with an event. One or more of the bits in the STATUS column is set to one if an event is suspicious. While all telemetered events are included in ACIS "evt1" data files, only those with no STATUS bits set to one are included in ACIS "evt2" data files.
The column GRADE (and FLTGRADE) in ACIS event data files contains a numeric description of the distribution of charge in a 3 x 3 pixel event detection region. (Interested users can refer to the Proposers' Observatory Guide for additional information about GRADE and FLTGRADE.) Some GRADEs are dominated by events associated with cosmic rays instead of X rays. These GRADEs should be excluded from event data analyses. Since the ACIS detectors are calibrated using data that have GRADE = 0, 2, 3, 4, or 6, events with one of these five "good" GRADEs should (in general) be included when performing analyses. Events with GRADE = 1, 5, or 7 (the "bad" GRADEs) should be excluded.
At times, the Chandra satellite passes through regions of relatively high particle fluxes. These occasions may be associated with "flares" in the ACIS background rate. Users may want to exclude the data obtained during the time intervals associated with these flares. The flares can be identified by examining light curves of source-free regions of the back-illuminated CCDs ACIS-S1 and ACIS-S3.
An "afterglow" occurs when a cosmic ray interacts with a front-illuminated CCD to produce a large amount of charge. For timed-exposure mode observations, most of the charge is clocked off of the CCD in a single frame. However, a small amount can be captured in charge traps, which release the charge relatively slowly. As a result, a sequence of events can appear in a single pixel over a few to a few dozen frames. The events need not occur in consecutive frames. There can be gaps of a few frames with no events. In general, the amount of charge released per frame declines with time. However, the trend need not be monotonic, especially near the end of an afterglow. Afterglows can appear to be false sources.

The tool acis_find_afterglow can be used to identify events that may be associated with afterglows. The tool is part of the standard pipeline processing as of DS 8.4.2 (December 2011). It replaced the tool acis_run_hotpix, which is insensitive to afterglows with fewer than 8 events. (See the Cosmic-Ray Afterglow why topic thread for more information.) The tool acis_find_afterglow also finds pixels that are hot.

Afterglows have been observed in continuous-clocking mode observations. However, the existing afterglow identification tools cannot be used to search for afterglows in continuous-clocking mode datasets. If you think your data is adversely affected by afterglows, then contact the CXC Helpdesk.
Horizontal streaks (i.e. streaks at a constant value of CHIPY) are observed on the ACIS-S4 CCD. A streak is the occurrence of several events that have the same value of CHIPY on the same CCD in the same frame of data. Streak events are spatially correlated with the read-out nodes and have small pulse-height amplitudes. These streaks are produced by read-out noise. The tool destreak can be used to identify and remove potential streak events.

The horizontal streaks should not be confused with the vertical "frame-transfer" streaks that are evident for bright sources (see the Remove the ACIS Readout Streak thread). Events in the frame transfer streak are also called "out-of-time" events because the TIMEs recorded in the event file for these events are inaccurate by as much as the time between one frame and the next.
For observations that are performed using TIMED VFAINT mode, the outer 16 pixels of a 5 x 5 pixel event "island" may be used to help reject potential cosmic-ray events. An algorithm to perform this kind of event filtering has been incorporated into the tool acis_process_events. (Interested users should use ahelp to obtain more information about parameters associated with this algorithm: check_vf_pha and trail.) As described in the ACIS VFAINT Background Cleaning why topic, the algorithm can reject up to several tens of percent of good X-ray events for bright sources.
A relatively large number of events are reported for the node boundaries at CHIPX = 256, 257, 512, 513, 768, and 769. Most of the events in these columns are associated with cosmic-ray events that would not normally have been telemetered if the event islands had not been split across a node boundary. Events in these columns and on other "bad" pixels are excluded from ACIS "evt2" data files by default. If you would like to include the events on the node boundaries in your data or modify your bad-pixel file in any other way, see the Customizing an ACIS Bad Pixel File thread.
An existing partial bad-column on ACIS-S1 (aka ACIS-5) is starting to show temperature dependent behavior. As the CCD operates at warmer temperatures, the number of hot pixels on this CCD has started to increase. These hot pixels are associated with a bright "L" shaped feature in the bias images located at (CHIPX,CHIPY) = (233,313) with the new hot pixels extending along the column in the +Y direction. Due to the short, unbroken nature of the feature, these pixels are not automatically identified as bad by the acis_find_afterglow tool.

The hot pixels may be obvious when looking at the Level 2 event file (a bright square shape the size of the dither pattern). Users can also inspect the Level 2 event file in chip coordinates to look for the presence of any hot pixels.
```
unix% dmcopy acis_evt2.fits[ccd_id=5][bin chip=1]" c5.img
unix% ds9 c5.img -pan to 233 313 physical -zoom 4
```
These hot pixels have a very soft spectrum (most events are have energies below 0.4keV) so using an energy filter which excludes low energies may be sufficient to exclude the bad events from the users' analysis.

For observations performed with the gratings, the energies typically covered on ACIS-5 are : HEG 1.0-1.6keV (7.5-12.5Å), MEG 0.5-0.8 keV (15-25Å), and LEG 0.2-0.35keV (35-65Å). The exact energies covered by these bad pixels will depended on the zero-order location.

Users who want to explicitly remove the events can add the location of the hot pixels into the bad pixel file using the acis_build_badpix tool. This procedure is described in the Customizing an ACIS Bad Pixel File thread. The usrfile should look like below
```
#ccd_id chipx_lo chipx_hi chipy_lo chipy_hi time time_stop bit action
#------ -------- -------- -------- -------- ---- --------- --- ------
5       232      234      322      339      0    0         7   include
```

MISCELLANEOUS

Since launch, the ACIS CCDs have experienced fluxes of ionizing radiation. During the time interval from 14 August 1999 to 17 September 1999 and on 18 October 2003, the fluxes were relatively large. As a result, the efficiency with which charge is transferred from pixel to pixel during the read out process is diminished, especially for the front-illuminated CCDs. The increased charge-transfer inefficiency (CTI) means that:
- the total amount of charge (i.e. the PHA) associated with an X-ray of a given energy is smaller far from the read out (i.e. at high values of CHIPY) than near it,
- the energy resolution and detection efficiency are better near the read out, and
- the distribution of event GRADEs for a source depends on the location of the source on a detector.
Since the energy scale and energy resolution are functions of the position of the source on a CCD and the focal-plane temperature, analyses of spectra from different regions of a CCD should use data that were processed using the temperature-dependent CTI adjustment and RMFs that are appropriate for each region.
If more than one X-ray photon interacts in the same event-detection region of a CCD (a 3 x 3 pixel "island") during the same frame, the charge clouds produced by these photon events are said to be "piled." The event-processing software can not distinguish between an event produced by one photon and an event produced by more than one photon. Therefore, the X-ray spectrum of a source that is significantly piled may not resemble the spectrum of the photons that are incident on the detector. A technique (PDF file, 8 pp) has been developed to analyze piled spectra. Pileup models are available in the spectral-fitting packages Sherpa, ISIS, and XSPEC (see the comparison, for low pileup fractions, of the pileup models document).

If pileup is severe, piled events may be lost because the events do not satisfy the telemetry criteria. Furthermore, telemetered events are excluded from Level 2 event files if the events have GRADE = 1, 5, or 7. For example, Chandra images of the Crab pulsar and nebula exhibit a "hole" where the pulsar is located because the X-ray flux is large enough that many X-ray events are detected in a single frame. The resulting, single, piled event does not satisfy the telemetry criteria because it saturates the analog to digital converters or the FLTGRADE of the event is one of the few FLTGRADEs that is dominated by cosmic-ray events and not telemetered to the ground.
Cosmic rays that interact in the detectors can produce particle cascades that deposit charge in many adjacent pixels. Typically, it is not possible to detect a good X-ray event if the event interacts in one of the affected pixels. Therefore, the "dead area produced by these charge cascades (or "blooms") reduces the detection efficiency of the detectors. The average reduction in the detection efficiency depends on the flux of cosmic rays, the position-dependent read-out time, and the CCD being used. For sources near the aim point of the ACIS-I array, the detection efficiency is typically reduced by about 4%. The change in the detection efficiency of the back-illuminated CCDs is negligible.
For X-rays passing through the mirrors, the very bottom of each CCD is obscured by the frame store. As a result, some of the events in rows with CHIPY <= 8 are not detected. (The set of rows affected varies from CCD to CCD.) Since the CIAO tools do not compensate for this effect, the ARFs and exposure maps for sources in these regions may be inaccurate.
The count rates of certain CCD nodes are somewhat higher or lower than the count rates of the other CCD nodes. As a result these nodes appear brighter or fainter, respectively, in full band images of the data. These nodes are working properly. The reason for the differences is that the overclock values vary from node to node, which means that more or less of the events that have very large pulse heights are telemetered. Since these events are typically produced by cosmic rays, not X rays, these differences do not impair the scientific utility of the data.

One way to produce an image that appears more uniform is to filter the data to exclude the highest energy events. For example, if the events that have energies greater than about 12 keV are excluded, then all of nodes on the front-illuminated CCDs should appear to have similar background event densities. Some experimentation with this maximum energy max be necessary to yield optimum results.

Figure 1: ACIS-I image

[Version: full-size]

Figure 1: ACIS-I image

An ACIS-I image with CCD_IDs 0, 1, 2, and 3 labeled. The last quadrant (NODE_ID=3) of CCD_ID 3 has more events than the other nodes on this CCD because it has a higher overclock value, which means that more very-high pulse height events (i.e. cosmic-ray events) are telemetered (see Fig. 2). The same situation applies to CCD_ID 1.

Figure 2: ACIS-I spectra

[Version: full-size]

Figure 2: ACIS-I spectra

Spectra of NODE_ID=3 (black) and of the other three nodes (red) for CCD_ID=3 of the data shown in Figure 1. The red histogram has been divided by a factor of three to compensate for the larger number of nodes. As shown, the only significant differences between these two spectra are at large pulse heights (i.e. above 12 keV).

Figure 3: ACIS-S image

[Version: full-size]

Figure 3: ACIS-S image

An ACIS-S image with CCD_IDs 5, 6, 7, 8, and 9 labeled. The first quadrant (NODE_ID=0) of CCD_ID 5 has fewer events than the other nodes on this CCD because it has a lower overclock value, which means that fewer very-high pulse height events (i.e. cosmic-ray events) are telemetered (see Fig. 4). The last quadrant of CCD_ID 9 has the same features illustrated in Figures 1 and 2.

Figure 4: ACIS-S spectra

[Version: full-size]

Figure 4: ACIS-S spectra

Spectra of NODE_ID=0 (black) and of the other three nodes (red) for CCD_ID=5 of the data shown in Figure 3. The red histogram has been divided by a factor of three to compensate for the larger number of nodes. Again, the most notable differences between these two spectra are at large pulse heights (i.e. above 12 keV).
The front-end processor designated FEP0 had a problem. Half of the bias-map memory for the processor was corrupted. This problem resulted in the loss of some useful X-ray data and the telemetry of many invalid events. To the best of our knowledge, this problem has only affected twelve observations. The effects of the problem should be visually obvious in images of the data: half of the CCD associated with FEP0 has many events and appears bright in the image. A patch to the flight software to address this problem became active in February 2000. Now, after ten bias parity errors have occurred during a science run (an extremely rare event under normal conditions), the patch causes the affected half of the CCD to be disabled until a new bias-map is placed in memory. Since the patch was implemented, the problem has not recurred.
In November 2009, the set of good ACIS flight grades in continous clocking (CC) mode was modified to include flight grade 66 events (all three pixels in the central column of the event island above threshold) that otherwise would have been filtered out in the level 2 event file from the SDP. In December 2014, a new grade file was released in CALDB 4.6.5 that includes a modified grade mapping scheme that reassigns flight grade 66 to a good grade value (2) to recover these events; however, further experience has now shown that the flight grade 66 events add mostly background noise to these observations and should be filtered out rather than remapped to a passable ASCA grade. The new CC-mode grade map introduced in CALDB 4.6.5 is removed in CALDB 4.7.2, returning to the prior behavior of mapping flight grade 66 events to the ASCA grade 7.

Users with DATAMODE='CC33_GRADED' that was taken after DATE-OBS='2009-11-01T00:00:00' should verify that their data were processed with CALDB 4.7.2 or later. If not, the data should be reprocessed with chandra_repro .