There has been a lot of recent progress in modelling the
PCA background. The purpose of this note is to compare
the new model ("L7_240") with two models in current use
(the "q6" model in the caldb and the "skyvle" model available
from the PCA team web page).

The "L7_240" model will be documented further shortly.
Very briefly it consists of a model in terms of the
rate of the sum of 7 rates from Standard 2 (L1R1, L2R2, L3R3,
L1L2, R1R2, L2L3, R2R3 - i.e. pairs of adjacent signal anodes)
and a term proportional to the recent integrated SAA doses
which is measured by the HEXTE particle monitor and is modelled
with a 240 minute half life.

For each model, I present 3 plots.

The first contains the
data with errors, the model prediction, and the residuals (data - model).
The data is from 3 PCU, first layer, 2-10 keV (raw channels 0-27).
All of the data is from observations of a particular piece of blank sky
(obsid 20801-05), and each bin represents 256 seconds. The data are obtained
over the course of ~1 year, with 3-4 orbits at a stretch. Examination of
this psuedo lightcurve can separate out the contiguous "SAA" and "non SAA"
orbits.

Rate and residuals for Q6 model ;

Rate and residuals for SkyVle model ;

Rate and residuals for L7_240 model .

The second plot gives the distribution of the residuals (in counts/sec) and
a gaussian fit. For a perfect model, the width of the gaussian should be
about the same as the error on a typical point. The widths of the distributions
are 0.44, 0.34, 0.24 ct/sec for the Q6, skyvle, and L7_240 models. In addition
to having a smaller width, the L7_240 residuals are also more symmetric.

residuals in ct/sec for Q6 model ;

residuals in ct/sec for SkyVle model ;

residuals in ct/sec for L7_240 model ;

Finally, the same data is presented in units of standard deviation. In this
case, the perfect model would have a gaussian width of 1 (typical residual
is just equal to the statistical error). The widths are 1.89, 1.51, and 1.08
for the Q6, skyvle, and L7_240 models, demonstrating that **the L7_240 model is
very close to being a statistically good model of the background.**

residuals in std dev for Q6 model ;

residuals in std dev for SkyVle model ;

residuals in std dev for L7_240 model ;

Coming next: (running this evening) estimates of systematic error per channel

available now - Help running the L7 model on your data.
Questions about this summary? Send mail to keith.jahoda@gsfc.nasa.gov

May 29, 1998