ACIS background

Maxim Markevitch (maxim % head.cfa.harvard.edu)
Last factual update 28 Sep 2001 (updated links 4 Feb 02)


Table of Contents: Related documents:

The ACIS background consists of a relatively soft Cosmic X-ray Background (CXB) contribution and cosmic ray-induced events with a hard spectrum. Most cosmic ray events can be filtered out by applying a grade filter (e.g., rejecting ASCA grades 1,5,7). After such filtering, the CXB component and the cosmic ray component are comparable below ~2 keV (during the quiescent background intervals, see below) and the cosmic ray component dominates above that energy, consistently with the pre-launch estimates.

1. Background flares

A phenomenon not anticipated prior to launch that can seriously affect the scientific value of an observation is background flares, when the count rate can increase by a factor of up to 100. Such flares have been observed anywhere in the orbit, including near the apogee. They are most prominent in the BI chips but also affect the FI chips. The nature of these flares is under investigation (ask S. Virani and P. Plucinsky for an update), but they generally correlate with the increased EPHIN rate for soft electrons. The flares are easily seen in the ACIS light curves; several examples are shown below.

The figure below shows an observation moderately affected by the flares (OBSID 1190). The horizontal axis gives Chandra time; the rates shown are for the BI chip S3 (upper curve) and the FI chip I2 (lower curve), for ASCA grades 02346 and energies below 10 keV. Although celestial sources were not excluded from these light curves, the flux is dominated by the background:

(PS)
Fig. 1

The rate is constant most of the time but there are several strong flares. Note how chip S3 is affected more strongly by these flares. The rates as high as 100 cts/s and 20 cts/s per chip have been observed in the BI and FI chips, respectively, compared to the quiescent rates of 1 cts/s and 0.3 cts/s (0.3-10 keV band, good grades).

The figure below shows the background spectra during a moderate flare and during a quiescent interval (OBSID 1226) in the area of chip S3 free of sources. Their exposures are similar (9-10 ks):

(PS)
Fig. 2

The next figure shows a difference of the above spectra normalized by their exposures, to give an idea of the spectrum of the additional background component arising during the high rate intervals:

(PS)
Fig. 3

The telescope effective area drops off much more steeply than this spectrum does. The nature of this component is beyond the scope of this memo.

1.1. Frequency of flares

The figure below shows cumulative probability for the ACIS background rate to be a certain factor above the respective quiescent rate (see below), for an FI chip S2 and a BI chip S3. Only the good grades 0,2,3,4,6 and energies 0.3-10 keV are used. Celestial sources were masked out whenever possible, otherwise the target flux was roughly estimated and subtracted, but most included observations were dominated by the background anyway. All observations are made without gratings.

The data were divided into 3 time intervals, August 1999 (about 200 ks in total), September - November 1999 (350 ks for chip S3 and 480 ks for chip S2) and December 1999 - January 2000 (about 400 ks). No effort was made to create a "representative" set of observations, but the observations sample all points of the observatory orbit and are selected more or less randomly. The histograms were made by binning the light curves into 260 s bins (80 times the ACIS readout time) so the Gaussian scatter is rather small.

(PS)
Fig. 4

Other FI and BI chips have similar flare probabilities. From the above histograms, 20-30% of the time the background rate in the BI chips is more than twice the quiescent rate, and 5-15% of the time, it is 10 times the quiescent rate. For the FI chips, 5-15% of the time the background rate is twice the quiescent rate. It appears that the flares have become somewhat less frequent with time since August 1999, although the current tendency is not clear. ACIS background flares occur more often during the solar flares, but they are not limited to those periods. Later data as well as various correlations are under analysis.

1.2. What to do with the flares

Unless the target is a bright point source, it will usually be best to discard the time intervals with flares. Even though this can significantly shorten the exposure, the flares are so bright that it may still increase the signal to noise ratio.

The Calibration team has produced experimental quiescent background event files (see Section 2.4 below), created by excluding time intervals with rates outside a factor of 1.2 of the quiescent rate, to keep the background uncertainty below 20%. To use those files, the observation will have to be cleaned similarly. Vertical dotted lines in panels above show this factor of 1.2. For S3, the fraction above this cutoff is between 30-40%, and for S2, it is 10-20%. This is the average fraction of the data that has to be discarded in order to use those background datasets. A work is underway to relax the cleaning criterion for S3 and possibly model some of the weak flare component, so that a bigger fraction of the data could be used.

1.3. Flares and telemetry saturation in VF mode

Two figures below compare the behavior of the filtered S2 and S3 chip rates (0.3-10 keV band and good grades), their total rates (everything that's telemetered in the most common mode with the onboard PHA cut at 3750 ADU or 15 keV) and the total rate from chips I23, S1234 (one of the common 6-chip imaging modes). Figure 5a shows an example of a typical flare that affects only the BI chips but not the FI chips, while Figure 5b shows a less frequent flare species that affects both chip types. Both observations are done in June 2000:

(PS)
Fig. 5a

(PS)
Fig. 5b

During the quiescent time, the total rate is ~10 times the filtered rate for chip S3 and 20-30 times for the FI chips. The total quiescent rates for FI and BI chips are about the same (see Table 4 below). Because flare events look like real photons, the total rate does not increase as steeply as the filtered rate during flares. For example, when the latter increases by a factor of 2 over the quiescent value (probably the maximum tolerable increase for most faint extended source studies), the total rate from the chip increases only by a factor 1.2 or so. Since FI chips exhibit flares less frequently than BI chips, the total rate from all chips most probably increases even less.

The total quiescent rate (Tables 4-5 and Fig. 9 below) is such that, for example, Very Faint mode with 6 chips may be close to the telemetry limit (~68 cts/s), and flares may exceed the limit. However, the above comparison shows that during a flare, the data would probably become useless before the telemetry saturates, provided there is a reasonable telemetry margin, such as a factor of 1.2-1.3 between the telemetry limit and the total quiescent background plus the source flux. If the observer wishes to use VF mode (see this memo for its background benefits) but there is a risk of saturation in the 6-chip mode, the options are (a) to turn one or more chips off, and/or (b) reduce the upper PHA cutoff slightly (see Table 5 below). Note, however, that a cutoff below 13 keV will discard information useful for diagnostics and pileup reconstruction.

When estimating the risk of saturation, note that the figures above correspond to Summer 2000 when the quiescent background rate was relatively low; those with bold imagination may try to guess from Fig. 9 below what it will be during their planned observation.

2. Quiescent background

During the quiescent periods and after the standard event screening, the background appears to be relatively constant with time (see Section 2.2 below). The following tables give observed quiescent background rates in several energy bands, using September 1999 - January 2000 data. The rates are given in cts/s/chip, using only ASCA grades 02346, excluding background flares, bad pixels/columns and celestial sources identifiable by eye. These rates include the CXB component. The energy bands are defined using the 1/20/2000 gain table; this was a tentative table not used in standard data processing, so the rates are only approximate (for more accurate rates at different focal plane temperatures, see this link.)

Table 1: ACIS-I in aimpoint:

--------------------------------------------
E,keV    I0    I1    I2    I3    S2    S3   
--------------------------------------------
0.3-10   0.31  0.31  0.31  0.29  0.32  0.89 
0.5-2    0.07  0.07  0.07  0.07  0.07  0.17 
0.5-7    0.19  0.19  0.19  0.18  0.20  0.40 
  5-10   0.15  0.15  0.15  0.14  0.15  0.51 
--------------------------------------------

Table 2: ACIS-S in aimpoint, no gratings:
--------------------------------------------
E,keV    S0    S1    S2    S3    S4    S5
--------------------------------------------
0.3-10   ...   1.41  0.33  0.86  1.04  0.35
0.5-2    ...   0.17  0.07  0.16  0.52  0.10
0.5-7    ...   0.45  0.19  0.38  0.68  0.22
  5-10   ...   0.97  0.15  0.50  0.16  0.15
--------------------------------------------

Table 3: ACIS-S in aimpoint, HETG inserted:
--------------------------------------------
E,keV    S0    S1    S2    S3    S4    S5
--------------------------------------------
0.3-10   0.27  1.55  0.38  1.03  1.13  0.33
0.5-2    0.07  0.18  0.10  0.18  0.57  0.08
0.5-7    0.19  0.51  0.26  0.53  0.77  0.21
  5-10   0.10  1.08  0.16  0.56  0.16  0.15
--------------------------------------------

As the tables show, the rates for the same chip with either ACIS-I or ACIS-S in aimpoint are close (at least for chips S2 and S3). The rates for different FI chips are very similar, except for S4 that has a defect (a very high background below 2 kev in the form of streaks along the chip x axis; look for J. Houck's destreak for a fix.). The BI chip S3 has a significantly lower high-energy background than the other BI chip, S1 (Mark Bautz says because of the different thickness of the two chips). Finally, the ACIS-S rates with and without HETG are close (but not identical).

There has been a slow decrease of the background rate since August 1999, see Section 2.2 below. In addition, during August 1999, the background in the FI chips has changed because of the large CTI increase that caused a change in the ratio of good and bad grades.

For completeness, Table 4 below gives approximate total (prior to grade and energy screening) quiescent rates, in cts/s/chip, for the most common ACIS mode with the onboard PHA cut at 3750 ADU (about 15 kev) and standard onboard grade filtering. These rates exclude flare periods but include everything else (the targets had negligible flux in the observations used for this estimate). These rates correspond to observations made in late 1999. The secular decline of this unscreened background is more pronounced than for the cleaned background, see Fig. 9 below. These rates may be useful for checking against the telemetry limits.

Table 4: Total rates (cts/s/chip, late 1999):

------------------------------------------------------------------
chip       I0   I1   I2   I3     S0    S1    S2    S3    S4    S5
------------------------------------------------------------------
rate       8    8    9    8      11    11    9     10    10    8 
------------------------------------------------------------------
The total rates with or without HETG are close to within +-1 cts/s.

If one lowers the upper PHA cutoff from 15 keV (now standard), the total quiescent rates for S2 and S3 are reduced as follows:

Table 5: Total rates (cts/s/chip) with different upper PHA cutoffs:

----------------------------------------------------------
Period 		 Aug 1999      Fall 2000 - Summer 2001
                           -------------------------------
Upper E cutoff   15 keV    15 keV  13 keV  12 keV  10 keV 
----------------------------------------------------------
chip S2 (FI)      10         6.3     5.8     5.6     5.0
chip S3 (BI)      11         7.7     5.0     4.2     2.5
----------------------------------------------------------

2.1. Spectrum

The plots below illustrate contributions of the CXB and cosmic ray components to the quiescent background for an FI chip S2 (top) and a BI chip S3 (bottom), with ACIS-S in aimpoint, after the standard grade and hot pixel cleaning and exclusion of obvious celestial sources. Black denotes the data with the ACIS door open but the mirror doors closed (OBSID 62706) and thus only includes the non-CXB component. Red shows the data immediately after the mirror doors opening (OBSID 62568) so it includes all background components:

(PS)
Fig. 6a
(PS)
Fig. 6b

The cosmic ray component dominates above ~5 keV, and it is lower in the FI chips than in the BI chips. A memo by F. Baganoff provides more details on the cosmic ray component from the August 1999 calibration. Both the above observations are performed at the ACIS temperature of -90C and low CTI (for FI chips) so they give only a qualitative picture. At present, the calibration group discusses an observation of the dark side of Moon or Earth to determine the non-CXB component with better accuracy.

Finally, the figure below compares the full quiescent background spectra for the BI chip S3 (red) and a representative FI chip I3 (black). They are extracted from regions covering the whole chip, from the Sep 99 - Jan 00 data with ACIS-S and ACIS-I in the aimpoint, respectively, and exclude the standard bad grades, bad pixels and bright celestial sources:

(PS)
Fig. 7

The above spectra are given here as PHA files that can be used for simulations: I3, S3

Alexey Vikhlinin has recently pointed out that in the Very Faint mode, the particle background component can be rejected more efficiently (compared to a simple rejection of the ASCA grades 1,5,7) by a simple additional cleaning. The background rate can be reduced by a significant factor at low and high energies for both the FI and S3 chips, see this memo.

2.2. Time dependence of the spectrum

2.2.1. Rates in wide energy bands

The plots below show the quiescent rates in different observations vs. time, for the ACIS-I and ACIS-S3 (soon to be updated with recent data) chips in soft and hard energy bands. Obvious celestial sources are excluded. Filled symbols show "good" observations included in the composite background datasets (see below), while open symbols show several observations not included in those datasets either because of their anomalously high soft background, or too high Galactic absorption, or it was an observation of XRB shadows which would obviously affect our rates, or because the target occupied a significant area of the field (in which case the rates shown were corrected for the excluded geometric area). Vertical dotted lines separate periods with different ACIS focal plane temperature. Horizontal lines show average rates for filled dots during the t=-110C and -120C periods. The energy bands for the -100C and -110C data are defined using a nonstandard 1/20/2000 gain table, while for the -120C data, the appropriate standard gain table is used. Therefore, the comparison of the average values for different periods is not entirely meaningful.

(PS)
Fig. 8a

(PS)
Fig. 8b

The plots show that the high-energy rate -- where the background is dominated by its cosmic ray component -- has been remarkably constant (to better than +-10%) since September 1999, after having dropped by up to 30% between August-September (when the appropriate gain tables are applied to the FI chip data, a similar drop is seen). One can notice a slow decline of both the FI and BI chip background during the -120C period until Fall 2000 (with some imagination, it is also visible in the -110C period), after which it leveled off and maybe started increasing, following the full rate (see the figure below). A blue line in each plot shows the approximate decline during Feb-Sep 2000; this will be further discussed below. The low-energy rate shows more random variability, obviously due to the celestial diffuse component, and for S3, due to the possible residual uncleaned flares (because of the different quiescent and flare spectral shapes shown in Figures 2 and 3, such residuals would show up in the 3-6 keV interval).

For comparison, the plot below shows how the total quiescent background for S2 and S3 chips (all that ACIS transmits, before any grade or energy filtering on the ground, but after flare screening; only observations with the standard 15 keV upper PHA cutoff were used). It has been steadily declining since launch until Fall 2000 when the decline have leveled out and apparently started to reverse:

(PS)
Fig. 9

More details on this plot are given in this memo.

2.2.2. Spectral variability

Two panels in Fig. 10 below show the ratio of the spectra from chips S2 and S3 of several ACIS-S observations (cleaned of flares, sources, etc., and normalized by the clean exposure) spanning the Feb-Sep 2000 period, to an average background spectrum for the same period. Black crosses show observations included in that average (and in the quiescent background dataset discussed below), and red crosses show several other obsrevations not included in the average for various reasons, such as the presense of a big object in the field (the object was masked and the rate corrected for the geometric area), too high Galactic NH, or because it was an observation of an XRB shadow. Errors are 1 sigma. Inside each energy interval, the measurements are arranged according to the observation date:

(PS) (PS)
Fig. 10

The scatter is significant. However, at energies above 4-5 keV, the scatter is not random -- there is an apparent decrease with the observation date. Indeed, applying a simple energy-independent correction factor taken from the 5-10 keV panels of Fig. 8 (blue lines; the S2 correction is taken from a similar plot not shown here), we can greatly reduce the scatter at high energies where the background is most important:

(PS) (PS)
Fig. 11

The resulting scatter is well within -10% +20% anywhere above 1 keV for S2 and above 5 keV for S3, for observations not expected to have any unusual background components (black crosses). The behavior of ACIS-I background should be similar to S2. A factor of 1.5 deviation in the lowest-energy bin in one observation (obsid 930), present in both S2 and S3 is under investigation.

For reference, the renormalization factor applied to the spectra is shown below. The users may apply it to the spectra extracted from the -120C Feb-Sep 2000 background datasets (see Sec. 2.4) to reduce the uncertainty of the background modeling. Note that it should not be extrapolated beyond the shown range of dates since the decline has leveled off (see Fig. 8a above); there will be a separate dataset for the constant rate period (hopefully by Nov 2001).

(PS)
Fig. 12

Several S3 observations exhibit a characteristic 20-40% excess in the 3-6 keV interval in the figures above. This must be due to incompletely removed flares -- as mentioned above, due to the different spectral shapes of flares and quiescent background, they should be most visible in this energy interval. All observations have been cleaned of flares by excluding times when the 0.3-10 keV rate was above a factor of 1.2 of the quiescent rate. To improve the accuracy of the background modeling in the 3-6 keV interval, one may need to apply (to both the target data and background data) a more strict cleaning criterion, for example, using the 3-6 keV flux, which would discard more data.

2.3. Spatial structure

The cosmic ray component of the background is spatially nonuniform, as illustrated by the plots below. They show chip x and chip y projections of the 5-10 keV background images for chips S3 and I0 (for the September 1999 -- January 2000 period):

(PS)
(PS)
Fig. 13

There are variations by more than 30%. At lower energies, the background is more uniform.

2.3.1. Time dependence of the spatial structure

The spatial structure appears to stay relatively constant within each time period (except for the FI chips during the t=-100C period when the CTI was increasing rapidly). This is illustrated in the plots for the -120C period below. The plots show chip x and y projections of the 5-10 keV images from 4 ACIS-S observations spanning the Feb-Sep 2000 period (red is the earliest and black is the latest), for chips S2 and S3. Each histogram is rescaled by the time-dependent coefficients used for the spectra in the section above, to correct for the overall rate decline:

(PS)


(PS)
Fig. 14

There is some statistical scatter (each of the shown observations is only 15-40 ks), but there are no apparent systematic changes of the 5-10 keV spatial distribution and it stays within +- 5-10% from the average for both S2 and S3. However, if one compares the structure during different temperature periods, it differs for the FI chips. This is expected, since the CTI and quantum efficiency are different.

2.4. How to subtract the background

Spatial nonuniformity of the background precludes its accurate subtraction using source-free sky regions of the same observation. On the other hand, the observed weak dependence of the quiescent background on time, especially at high energies where it is most important for the analysis of celestial sources, lets us use other observations (cleaned of the flares using exactly the same criteria) to model the background. They can be normalized by the ratio of clean exposures, with a possible small correction due to the slow rate decrease as seen in Fig. 12 above.

A number of source-free observations have been combined to create experimental quiescent background event files (using rather strict flare cleaning criteria) for the three distinct periods in the ACIS life -- the August 1999 (focal plane temperature -100C), September 1999 to January 2000 (t=-110C), and February to September 2000 (t=-120C). There will be a separate dataset Fall 2000 to Summer 2001. These datasets, their detailed description and experimental tools for their use can be found at this link.