As reported on October 1, 1999 by the HRC instrument team in the Chandra Inflight Status Report, the background rate in both HRC detectors is significantly higher than expected. In the HRC-S, used with the LETG, the current rate is about 7e-5 Hz/arcsecond^2, or 0.12 counts per pixel in 100,000 seconds. This can be reduced to 0.09 counts per pixel/100ksec simply by excluding pha channel 255, with no loss of x-ray events. A dispersed line in the LETGS is between about 20 and 65 pixels tall (cross-dispersion direction) and roughly 7 pixels wide (FWHM), so (using the lower background figure) that works out to roughly 12-40 background counts underneath a line in a 100,000-second exposure. It is expected that the rate will be lower once we get through Solar Maximum.
Fortunately, this background can be reduced even further by appropriate data filtering. We here describe efforts to determine the best filtering criteria, with the aim of reducing the HRC-S background as much as possible with minimal and well known loss of x-ray signal. Ultimately, we hope to provide tables of recommended filter cuts listing the estimated x-ray loss for all diffraction orders, along with filter tables for off-axis sources. At first, we will provide a filter table that preserves ~98% of x-ray events, and will later provide tables for more aggressive filtering (for low S/N situations).
As will be explained later, because of spatial variations in HRC gain and the slight energy dependence of the mean pulse height (we are looking at dispersed photons, remember!), the recommended filter cuts will be position dependent. Laboratory data exist for several energies, although not below 183 eV (longer than 68 Angstroms). There is also a difference between the flight and laboratory gains. Between the lab and flight data, however, can characterize pulse height distributions at all positions and wavelengths pretty well.
The columns in HRC FITS files (level 1) that can be used for background versus x-ray signal discrimination are:
Complementing the work described below, the HRC team is developing a test that screens on the basis of (au3-au1)/(au1+au2+au3) vs. au2/(au1+au2+au3), and similarly for av#. No detailed online documentation is currently available from the HRC team, but very roughly 40% of background events are removed from HRC-I data with less than (probably much less than) 10% x-ray loss.
For the HRC-S, which behaves almost the same on-orbit as during its lab calibration (the HRC-I voltage had to be reduced on-orbit), and which also has extensive flight calibration data with fairly complete coverage across the detector (because of dispersed photons from the LETG), we are developing a PHA filtering method which is potentially more effective and better calibrated than the position filtering. Unfortunately, preliminary indications are that the position filtering method offers no additional benefit. In the future, however, it may be possible to use the position analysis to slightly improve the position resolution of the HRC, and thus the spectral resolution of the LETGS.
As can be seen in 1D binned plots of each of the relevant fits-file columns (pha, au1, sumamps, etc.) which compare the signal distributions from background and x rays (from LMC X-1 and Capella), it is possible to remove background events without losing any (or many) x-ray events. The associated table on that page also shows that filtering the pha column is by far the most effective way to reduce background with the least x-ray loss, and that cuts on the high side are more effective than on the low side.
Minimal filtering of the seven other pulse height columns removes an additional very small fraction of background events with essentially no extra x-ray loss.
Laboratory flat-field data were collected for HRC-I and HRC-S at several energies. Once in orbit, the HRC-I gain was roughly double what it was in the lab, and its voltage was reduced. The HRC-S voltage is the same on-orbit as in the lab, although its gain is roughly 10% higher than in the lab (as described later).
Lab data were collected at 8 energies between 183 eV (B-K) and 6404 eV (Fe-K), and maps of mean pha value were created on a quarter-tap grid. As can be seen in the image below, gain can vary by a factor of two across the HRC-S. There is a similar, though less extreme, variation across the short axis of the detector, too. (Each curve shows the average value for u=15,16,17, which corresponds to half of the crsu=7 tap and all of the crsu=8 tap. In all the u/v plots that follow, data are quoted on a quarter-tap basis, so that crsu=0:15 corresponds to lab u=1:32, and crsv=0:191 corresponds to lab v=1:384.)
Note that the wiggles in the curves are NOT noise--the typical uncertainty in the values plotted is ~1%. The systematic nature of the wiggles is clearly seen below, where each curve is normalized to the average of the curves for B, C, O, and Al-K.
From the existing lab data, one can construct a model of what the expected mean PHA value would be at any point on the detector for photons at any energy, because:
Based on the lab data, we have created a table of u, v, and the predicted mean PHA (in the lab) for a 1st-order diffracted photon at the appropriate position assuming an on-axis source. Interpolation/extrapolation as a function of energy used values at the two nearest lab-measured energies. The results, for u/v values corresponding to the nominal readout region on HRC-S when used with the LETG (crsu=5:10, lab_u=11:22) are shown below. For on-axis sources, the dispersed spectrum generally falls on u=16:18. Note that different curves would be needed for sources that are displaced in v from the nominal aimpoint, or for higher order diffraction.
| u=11:14 | u=15:18 | u=19:22 |
|---|---|---|
 |
 |
 |
These model curves, based on inter(extra)polations of lab data, are being compared with results from flight data. For this we need high S/N, and are using data from Capella (which has many bright lines, particularly at relatively short wavelengths), HZ43 (a bright continuum source at wavelengths longer than about 60 Angstroms), and 3C273 (a continuum source with higher-energy emission).
First, diffracted x-ray data is extracted using tight regions to reduce background contamination. Background regions of the same shape (sometimes wider) are also extracted on either side of the x-ray region. To obtain the most direct comparisons between flight and lab data, we then analyze only data from the flat part of the dither pattern (see below), namely along u=17 or 18, depending on the particular data set. (The corresponding regions, with appropriate offsets, are taken from the background data.) In this way, we will measure the mean PHA over one quarter-tap with fairly uniform illumination, as was done in the lab.
Next, PHA spectra are created for x-rays and background for each quarter-tap, and the background spectra subtracted to yield "pure" x-ray PHA spectra. The background fraction is typically only a few percent. Lastly, the mean PHA value, the sigma of the distribution, and other derived quantities are computed. Some results (Capella lines not yet ready) are shown below. Points with fewer than 30 events were excluded.
| HZ43 obsid 59 |
|---|
 |
As can be seen in the example above, PHA distributions at a specific location on the HRC are quite narrow, with a sigma of <0.25*(mean PHA value) for a given quarter-tap region. The mean value, as expected, also varies greatly from one location to another. In order to look for differences between our mean PHA model and the flight data, we took the ratio of the measured and model mean PHA values and plotted them versus various quantities (FWHM, lab PHA, wavelength, etc.) to look for trends. The most informative are plotted below.
| All data | labu=16 | labu=17 | labu=18 | labu=19 |
|---|---|---|---|---|
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
As can be seen above, the flight data track the lab/model quite well; a few discrepant points near the ends,from "inactive" regions lying outside the CsI photocathode are excluded-- we may try to calibrate these later. One can see a weak correlation between the gain ratio and the mean PHA, and a slight wavelength dependence --we are studying this to try to reduce the amount of scatter in the gain ratio. The bumps around 40 and 65 Angstroms correspond to regions where the background contribution is relatively high, and are probably spurious, although we do not at present understand this.
In addition to knowing how the mean PHA varies with position and energy, we must also know how wide to make the pha cuts. The last panel, with the "width" and "retained x-rays" plots, shows that relative widths may vary with wavelength (although this may be a background-subtraction artifact). From other work we also know that background is reduced more effectively when high PHA channels are removed. This must all be somewhat subjectively balanced to chose the "best" filter cuts.
We hope to assemble a "final" table of filter cuts in Februrary, with one entry for each of the 12x384=4608 quarter-tap regions. Existing CIAO commands to use such a table do not exist, so our crack programming staff (Pete Ratzlaff) has developed a tool. It will probably also filter out bad events such as those with av3>0.7*av2, etc. We are coordinating our efforts with the CXC Science Data Systems group to determine how to best incorporate (or not) this filter tool into standard pipeline processing of LETG/HRC-S data.
The background reduction using this first "98%" filtering table is roughly 50%, depending somewhat on wavelength. Less than 2% of x-rays will be lost. Filtering should be somewhat more effective at longer wavelengths than at shorter, because the peak of the x-ray PHA distribution is further below the peak of the background distribution. Then again, the PHA distributions seem to be wider at long wavelengths, so maybe the filtering will be less effective. We hope the former is true, since background is more of a problem at longer wavelengths, as the dispersed spectrum is wider in the cross-dispersion direction because of astigmatism. This will be our primary subject of study in February.
We will also develop more aggressive filter cuts, that remove up to 10% or 15% of x-rays, but the uncertainty in the lost x-ray fraction will be relatively larger--probably a few or several percent.