Corrected RXTE PCA Background Models

C. Markwardt (U. Maryland/GSFC)

05 Apr 2004


An error was recently found in published PCA "CM" L7 background models. The error prevents the temporal drift term from being subtracted from the model correctly. The maximum error is an oversubtraction of approx. 0.2 ct/s/PCU/layer (approx. 0.1 mCrab, 2-10 keV). This oversubtraction has a smooth linear trend within gain epochs, and resets at gain epoch boundaries. The error will affect observers who need long-term high precision background stability. New background models are supplied which correct the error.


Because the RXTE PCA is a non-imaging instrument, the background must be estimated from a model. The highest precision background model for faint sources is known as the "L7" model, referring to the seven combinations of housekeeping coincidence rates used to generate it.

A 2002 revision of the background model, dubbed the "CM" L7 model, introduced a time variable drift in the background rate. This drift is independent of changes in the L7 rate, which was originally the primary independent variable of the model. Implicit in the introduction of time as a variable is the assumption that there must be some other long-term factors which affects the PCA background rate, perhaps the energy of the particles. For example, the particle energies may change as the spacecraft altitude decreases.

An error was discovered in the construction of the published model (pca_bkgd_cmfaintl7_eMv20030330.mdl). This error prevents the time-variable component from being subtracted properly. Instead, only the constant estimated background from the beginning of the gain epoch interval is subtracted (i.e. no time variations are included). This error was missed in testing because of the sampling of the test datasets.

This error affects all "CM" L7 background gain epochs (3a, 3b, 4, 5). It does not affect the bright source (VLE) model because that model does not have a time variable term.

New models are available from the PCA calibration data FTP site, and will soon appear on the PCA Digest web page. The new models are:

EpochNew Model Name

(See PCA Digest page for epoch descriptions.) The new models have been checked with a range of data and are now correct.


The impact of this error is largest for observers who wish to perform long-term high precision background subtraction. Long-term means datasets that span many months. High precision means scientific programs that demand background subtraction which is precise to a small percentages of the mean background rate.

Plot of the model oversubtraction versus time
Figure 1. Plot of the 2-10 keV model oversubtraction versus time. The epoch boundaries are shown with vertical dashed lines. The horizontal dotted lines represent the 1 sigma uncertainties in the background model.

Figure 1 shows the trend of the oversubtraction versus time. Users can see that the oversubtraction grows linearly from the start of each gain epoch. Epoch 5, being the longest uninterrupted background interval, has the largest accumulated error. The overall maximum oversubtraction is about 0.2 ct/s/PCU/layer, which is approximately 0.1 mCrab ( = 2 x 10^{-12} erg/s/cm^2 ).

While the errors appear large, it should be stressed that except for Epoch 5, the oversubtraction errors are within a factor of two of the uncertainties in the models themselves (see horizontal dotted lines). In general, these are in turn a small percentage of the mean background rate. However, in Epoch 5, the modeling error grows significantly larger than the model uncertainty.

Plot of the model oversubtraction
versus energy
Figure 2. Plot of the model oversubtraction versus pulse height (not energy; see energy-to-channel conversion). The black line represents the oversubtracted amount late in Epoch 5, expressed as a percentage of the mean background level. The red points and error bars represent the unmodeled fractional variances and their 1 sigma uncertainties as a function of pulse height.

Figure 2 shows the energy dependence of the error. The red points give an indication of the ultimate model uncertainties, that is, the performance of the model when the error has been corrected. Generally speaking the oversubtraction is confined to energies below 6 keV, although there may be some small effects near 26 keV. (Also, there are are no corresponding regions of compensating undersubtraction.)

Once the new models are applied, the trends noted in Figures 1 and 2 are eliminated.


Users who demand background subtraction precision better than 0.2 ct/s/PCU/layer over time spans longer than several months should estimate new backgrounds using the new PCA background models, and re-analyze their data. Fresh analysis should of course use the new models whenever possible.


05 Apr 2004 - Initial revision

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.