Practical Advice for Users of New RXTE PCA Background Models

C. Markwardt (U. Maryland/GSFC)
K. Jahoda, D. Smith (U. Maryland)
(assisted by L. Brennaman (USRA), S. Duckett (Purdue U.), D. Parker (U. Maryland))

26 Feb 2002

09 Jul 2002 - Revised


This report describes new background models for RXTE PCA observations, and the impact to guest observers. Newer statistical techniques have been employed to refine the existing "L7/240" model.

Results are a new set of "CM" background models with:

  • sensitivity to X-ray variations improved by factor of ~2 over previous models;
  • applies to PCU0 (needs new FTOOLS to fully exploit);
  • new model packaging means fewer files to deal with;
  • gain Epoch 3 is now divided into Epochs 3a and 3b;
  • "CM" stands for "Combined Model," the technique used in their construction.

The unmodeled variance in the background rate, which fundamentally limits the sensitivity to source variations, is 0.04 ct/s/PCU or better, in the top layer 2-10 keV band, and 0.02 ct/s/PCU in the top layer 10-20 keV band. However, for purposes of determining absolute fluxes of sources at arbitrary points on the sky, the cosmic X-ray background is the limiting factor (see below).

The new "CM" models also apply to PCU0, which since 13 May 2000, has been operating without a propane layer. To take full advantage of some of the improvements, use of the now-public FTOOLS release 5.2 is required, but the core functionality of the models can be used older releases of FTOOLS.

Recommendations to Guest Observers

Faint Sources (Relative Flux Measurements) - For the greatest sensitivity to variations in faint sources, guest observers should:

  • Use the new "CM" faint source models (available from PCA Digest page), and a recent PCA SAA History file. "CM" stands for Combined Model. The per-epoch background files should work with all recent release of the FTOOLS.
  • A combined PCA background model for the entire XTE mission is also available. The publicly available version of PCABACKEST (FTOOLS release 5.2) is able to use this mission-long model.

  • The systematic errors are minimized by integrating for at least 1600 seconds per time bin. For shorter time bins, the variance in the L7 rate, the nominal independent variable of the model, begins to become important.

Faint Sources (Absolute Flux Measurements) - For absolute X-ray flux measurements the variation of the cosmic X-ray background is the dominant source of error. (see below)

Bright Sources - For bright sources, users should be aware that the new "CM" VLE background models depend on the PCA SAA History. This is a change from previous models.

PCU 0 - it is possible to generate an appropriate background model for PCU 0. The loss of the propane layer has increased the background level and hence reduced the sensitivity. There are certain background flares which are not modelled and must be filtered out (see below).


The RXTE PCA is a non-imaging instrument, which means that for most work, both in spectroscopy and light curve analysis, the background must be subtracted based on an a priori model. Here, "background" is defined broadly to include anything that contributes non-source counts to the PCA instrument in orbit, including but not limited to:

  • local particle environment;
  • induced radioactivity of the spacecraft; and
  • the cosmic X-ray background.

In general, these components vary as a function of time, so they must be somehow parameterized. The parameterized model is adjusted to fit a set of dedicated observations by the PCA of blank sky regions. Once a good fit is achieved, the same parameterization can be applied to other observations.

The most successful background model to date for faint sources is the so-called "L7/240" model. Here, "L7" is the name of a particular housekeeping rate which the PCA team believes characterizes the X-ray background rate quite well. [ It is the sum of all pairwise and adjacent coincidence rates in each PCU. ] The "240" component refers to a radiactive decay timescale of approximately 240 minutes. The 240 minute timescale may describe the combined effect of several radioactive elements. The L7/240 model is not appropriate for bright sources because the L7 rate can be modified by the source itself.

For bright sources, the "VLE" model has been employed. The Very Large Event rate (VLE) is another housekeeping rate, which characterizes the particle background more crudely than the L7 model, but is more independent of the source rate.

Description of Results

The new faint model is still formally an L7/240 model. It is fundamentally characterized by an explicit linear dependence on the L7 count rate and the effective particle dosages in SAA, and a linear temporal drift term. All of these linear components can be embodied with the existing PCABACKEST framework as it is distributed today.

The model was fitted to the existing PCA background. Observations for gain epochs 3, 4 and 5 were fitted separately, including data through mid-2001. In general a good fit was found, although a few outlier points were found. A day (01 Jan 1997) where extreme background flares were seen, was excluded (see this note by David Smith). After fitting, the unmodelled residuals were computed in several energy bands.

Average PCU 2 Spectrum (Epoch 5)
(click for enlargement | PS file)

Figure: Average PCU 2 spectrum, Epoch 5. The average rate from each layer is shown separately: layer 1 (solid); layer 2 (dotted); layer 3 (dashed).

The largest contributions to the overall spectrum are: (a) X-rays and particles interacting in the top layer; (b) untagged X-ray events from the Americium calibration source; (c) particles from the local space environment; and (d) radioactivity-induced emission from the detector structure.

We find that even after accounting for the statistical error in the X-ray rate, there is still a integration-time-dependent term in the systematic error. We attribute this to fluctuations in the L7 rate, which is one of the independent variables in the model. These fluctuations are minimized by taking observations longer than ~1600 seconds, and therefore guest observers are advised to take integrations at least that long for the greatest sensitivity.

We also see that there is a nonlinear time-dependence of the residuals during Epoch 3. Because Epoch 3 is the longest (~3 years) compared to Epochs 4 & 5, we might have expected this problem to arise. We have solved the problem by dividing Epoch 3 into two separate segments, Epochs 3a and 3b (dates below). There is a separate model for each epoch which guest observers must select correctly, or when the proper tools are available, the mission-wide background model file will work fine.

How to Use the Models

There are only a few caveats to use the new models. First, the new models should be downloaded from the links on the PCA Digest web page. Updates and notices will be available there. Otherwise, you should be able to use the model files for background estimation as you always have (see recipe for instructions).

Second, users should be aware that for the new "CM" Faint model in the time range 15 Apr 1996 through 22 Mar 1999, that the model has been broken into two separate model files. Users must choose the appropriate model file based on the time of observation.

EpochStart TimeStop Time
Epoch 1Launch 21 Mar 1996 18:33
Epoch 221 Mar 1996 18:34 15 Apr 1996 23:05
Epoch 3a15 Apr 1996 23:06 09 Feb 1998 00:00
Epoch 3b09 Feb 1998 00:00 22 Mar 1999 17:38
Epoch 422 Mar 1999 17:39 13 May 2000 00:00
Epoch 513 May 2000 00:00 Present
Table: New Epoch Boundaries for "CM" Faint Model

For bright models, the distinction between epochs 3a and 3b are not made; there is only a single bright source model file for all of Epoch 3.

The new release of the FTOOLS, version 5.2, contains an enhanced version of PCABACKEST. This version of the tool allow you to use the single mission-wide background model file which is available from the PCA digest page. All of the models for the entire mission are stored in a single file, and PCABACKEST automatically chooses the correct ones to apply.

Systematic Errors

Users can use the systematic errors quoted below to estimate their sensitivity to fluctuations of faint sources in a given observation. The "systematic" error is assumed to be any unmodelled variance in the residuals after subtracting the known model components. It also assumes a particular observation length. As already noted above, the recommended observation length (time bin size) is at least 1600 seconds.

The first table represents the systematic error in the 2-10 keV band.

Epoch 3a0.0220.0260.0250.0320.032
Epoch 3b0.0290.0350.0280.0420.042
Epoch 40.0310.0420.0300.0330.032
Epoch 50.1070.0210.0260.0310.019
Table:"CM" Faint Model (ct/s/PCU; top layer; 2-10 keV; 1 sigma)

A similar table is shown for the approximate 10-20 keV band.

Epoch 3a0.0140.0160.0160.0100.011
Epoch 3b0.0090.0090.0140.0170.010
Epoch 40.0130.0150.0100.0170.014
Epoch 50.0480.0190.0160.0160.016
Table: "CM" Faint Model (ct/s/PCU; top layer; 10-20 keV; 1 sigma)

Here are a few interpretations. First, this "CM" release of the faint models has a systematic error which is about a factor of 2 smaller than the previously released models. Second, the systematic errors seem to be consistent and stable across multiple calibration epochs. Third, the loss of the propane layer in PCU0 at the onset of Epoch 5 is apparent as an increase in the systematic error for that PCU. However, usable data is still obtainable from this PCU.

Former PCA background expert David A. Smith has also contributed some comparisons of the new "CM" models versus the old models. His analysis primarily focusses on the 30-70 keV band in PCA observations of sources. This energy band should be source-free in most cases, and so is a good test of the background model. Generally speaking he finds good results.

We will point out again that these systematic errors only apply to relative flux measurements, i.e. to measure the fluctuations of a faint source. For absolute flux measurements, the uncertainty in the cosmic X-ray background is the dominant term. The cosmic X-ray background is approximately 2 x 10^{-11} erg/s/cm^2 (2-10 keV), and 1.2 x 10^{-11} erg/s/cm^2 (10-20 keV). Fluctuations in the background are about 10% of these values. Thus, the expected fluctuations in count rate are:

Energy RangeRate
2 - 10 keV0.16
10 - 20 keV0.04
Table: Cosmic X-ray Fluctuations (ct/s/PCU; top layer; 1 sigma)

PCU 0 Models

It is possible to use PCU 0 for the study of intermediate brightness sources, probably in the 5-100 mCrab range. As already mentioned, PCU 0 lost its propane layer in May 2000. This causes the background rate to be higher and less predictable in PCU 0 than any of the other PCUs, hence PCU0 will be less useful in studying flux levels of faint sources. The released "CM" background models will functionally estimate the background for PCU 0 as for any of the four other PCUs with no extra special processing or tools.

However the background is less stable in this PCU. Background flares are obvious in many observations which are not modelled by the "CM" models (or any other models to date). We suspect that these flares are caused by soft electrons which used to be stopped in the propane layer but are now stopped in the top Xenon layer and no longer vetoed.

Guest observers are urged to use the utmost care during data analysis. Visual inspection of the source+background and model background rates are advised.

A more-or-less mechanical method can be employed for intermediate-brightness sources using the following approach. The basic technique is to filter out background flares by looking for excess counts in the top layer of PCU 0. This analysis requires three steps:

  • Create new filter file with additional columns
  • Compute new L6 column in filter file
  • Apply new MAKETIME expression to filter out flares

WARNING: For bright sources (perhaps >100 mCrab), the source flux begins to contaminate the X1LX1RCNTPCU0 rate which is the basis of this filtering technique. Thus, for bright sources the PCU 0 screening may have to be done by hand.

Step 1. Create new filter file. This step involves re-running the filter file generation process. You need to add the following entries to your apid list before re-running xtefilt:

       74 X1LX1RCntPcu0
       74 X2LX2RCntPcu0
       74 X3LX3RCntPcu0
       74 X1LX2LCntPcu0
       74 X1RX2RCntPcu0
       74 X2LX3LCntPcu0
       74 X2RX3RCntPcu0

NOTE: even with the release of FTOOLS 5.2, you will need to add these apid entries to your appid list file.

Step 2. Compute new column in filter file (prior to FTOOLS release 5.2). The new column is named L6CNTPCU0, and is a housekeeping rate similar to L7. The version of xtefilt included in FTOOLS release 5.2 will automatically calculate L6CNTPCU0 for you, and you can skip this step. Otherwise, you must run FCALC according to the following recipe.

fcalc filtfile filtfile_L6 L6CNTPCU0 \

(note that this expression may be too large to fit the screen.) At this stage you can use the filtfile_L6 filter file as your default filter file.

Step 3. Make good time interval Finally, you need to make a good time interval selection based on these quantities. This is the good time interval file which you should then apply to any of your extractions of PCU 0, spectral or light curve, in addition to any other selections you would normally apply. If you wish, you can also apply the good time interval to other PCUs, at the risk of excising good-quality data.

maketime filtfile_L6 pcu0_noflares.gti \
  expr="X1LX1RCNTPCU0 .lt. 56 + 0.88 * L6CNTPCU0" \
  compact=no time=Time

Bright Source Models: What to Expect

Here are a few brief words on the new "CM" bright source background models. First of all, these new models are derived using knowledge of the history of particle dosages as measured by the PCA SAA History file. Users of the faint source model will know this routine well, but guest observers who focus exclusively on bright sources may not know about this extra file.

The SAA history file measures the integral dose of particles that the HEXTE particle monitor encounters in each pass through the South Atlantic Anomaly. We presume that this dose is directly related to induced radioactivity in the detectors, which persists for several hours after SAA passes. Thus, to make an accurate model, radioactivity must be accounted for. Previous models used orbital position as a proxy for SAA dose.

The new bright source models use the SAA history file instead. Users must be sure to download a recent history file from the XTE calibration area (file pca_saa_history.gz). This file is now generated daily, so a history should always be available within ~24 hours of an observation.

Users should expect improved background subtraction from the Epoch 3, 4 and 5 "CM" Bright VLE models.

Technical Note: Improvements to Model

The following fundamental changes to the analysis were used:

  • The variance in L7 itself was accounted for;
  • An relationship explicitly linear in both L7 and 240 was fitted;
  • Each Standard2 pulse height channel was fitted independently;
  • The modified chi-square approach for low-statistics Poissonian data were used (due to Mighell, 1999, ApJ, 518, 880);
  • The L7 and 240 components were fitted simultaneously rather than separately.
  • Data from both times near and far from SAA were used

The fitted background model for each channel,i, is:

  BKG_i = A_i + B_i * L7 + C_i * DOSE + D_i * (t - t0)
where A_i, B_i, C_i and D_i are the fit coefficients, L7 is the L7 rate in a PCU, DOSE is the summed SAA particle dosages as measured by the HEXTE particle monitor, decayed by a 240 minute folding timescale, and t is the epoch time. This model is linear in all its terms, and allows for a secular drift over time. The background is decreasing as the spacecraft altitude decreases, and as solar activity decreases as well.

As with the previous models, the fitting was applied to multiple, dedicated, PCA background pointings. Each of the points on the sky has a slightly different sky background. We have taken the approach to assign a different set of coefficients A_i for each background pointing. The production background model is determined by taking the weighted average of A_i's for different pointings. Thus, the background model represents an "average" patch on the sky.

Because the same independent variable is being used, the model can still be considered an L7/240 model, and is usable with existing software.


09 Jul 2002 - changed future-looking FTOOLS descriptions to the present tense, since release 5.2 of FTOOLS is now publicly available.

17 Nov 2003 - corrected Mighell citation.

If you have a question about RXTE, please send email to one of our help desks.

This page is maintained by the RXTE GOF and was last modified on Friday, 31-Aug-2012 13:08:42 EDT.