BBXRT Analysis Issue

artist concept of BBXRT in the Shuttle bay in flight

* The Point Spread Function and the Mask

The distribution of photons in the focal plane can be empirically described by a pair of error functions, one narrow to fit the central core of the beam, and one broad to represent a halo, perhaps from geometric scattering. If P(<q) is the total power contained within a given angular radius q (in arcminutes) and N1 and N2 are the relative normalizations of the core and halo components, then

	P(<q)= N1 erf (q/s1) + N2 erf (q/s2)

For our best fitting model we find s1 = 1.8 arcmin, s2 = 5.8 arcmin, N1 = 0.65, N2 = 0.35. The resulting half power radius is approximately 1.3 arcmin.
The silicon detectors consist of five separate cylindrically symmetric detecting elements defined via grooves onto a single silicon block. In order to prevent the detection in multiple elements of X-rays incident near the grooves, an X-ray opaque mask was placed in front of the detector. The mask is shaped to cover the grooves, but occults a width of 0.5mm on each side of the grooves. Projected onto the sky, this corresponds to a width of 1.5 arcminutes. Because of the spreading of the X-rays in the focal plane due to the mirror, the mask reduces the effective area by 20% for an on-axis source, and as much as 50% for a source centered on the intersection between the central circular mask and one of the radial spurs of the mask.
Combining the effects of the PSF and the masking, for an on-axis pointing 62% of all focussed events fall in the central detector, 20% fall on the mask and 17% are distributed among the four outer detectors.

* The Detector Background

From the point of view of background subtraction, the BBXRT observations fall into two categories; those performed during orbit day, and those performed during orbit night. For the orbit night observations, the background files given in the archive are adequate and correct. To first order, the orbit night backgrounds come from the internal background of the detectors described above, scaled directly with the guard rate.
During orbit day, there is an additional background component caused by the resonant scattering of solar X-rays off the Earth's atmosphere. This leads to a substantially increased background flux at energies below 1 keV, particularly a strong 524 eV oxygen line. Channels 20-60 suffer the most from this geocoronal emission. The exact amount of bright Earth contamination of the data at low energies is a steep function of Earth angle as well as guard rate, and this requires a careful selection of background data. We are still working on an accurate way of generalizing background subtraction under these circumstances. Thus, the background spectra in the archive do not include a bright Earth component, and oxygen and nitrogen lines are visible in the data below 1 keV. Upgrades to the archive spectra can be expected when a dependable background subtraction method is finalized.
Above 2 keV the bright Earth contribution becomes negligible and the orbit night backgrounds work acceptably for the orbit day data.
The Mission Quality files referred to above contain six qualities derived from the Shuttle telemetry and standard astronomy programs which can help interpret BBXRT data:
  • the Sun angle, or angle between the pointing direction of BBXRT and the Sun;
  • the Moon angle (we never detected the Moon and it never interfered with observations, but this angle is included for completeness);
  • the Earth angle, or angle between the pointing direction and the center of the Earth (note that the Earth subtends a semi-angle of 72 degrees at the altitude of BBXRT);
  • a 1/0 flag indicating whether the nearest Earth limb to the pointing direction was bright or dark (1=bright);
  • a 1/0 flag indicating whether the Shuttle was in orbit day or orbit night (1=day);
  • and the ram angle, or velocity vector angle, between the velocity vector of the shuttle and the pointing direction.

* Background subtraction

The non-X-ray background for the BBXRT detectors was very low (see Table 2).

Table 2 with background values at different energies

(In units of count s-1 keV-1 field-1, i.e. the sum of data from two detector elements viewing the same part of the sky.)
While the on-orbit background was a factor of two to three higher than we anticipated, it should be kept in mind that the Sun was active during the mission, with two solar flares occurring in the first three days. The background was found to vary with time, and these variations are strongly correlated with the guard rate (see below). For energies less than approximately 4 keV, the diffuse X-ray background dominates the total non-X-ray background.
At this point it may be worth mentioning some details of the anticoincidence methods used. The photoelectrically converted signals in the five detector elements were output to gate leads of jFETs clustered behind the detector in the cryostat. Also in the cryostat, close to the FETs, were LEDs which were occasionally flashed on to dissipate the accumulated charge within the gates. Five separate anticoincidence techniques were applied: events detected in a given element were compared against simultaneous events in that or another element (pulse-pulse and pixel-pixel anticoincidence); against events in any element with deposited charge larger than that expected from a 14 keV X-ray (Very Large Event anticoincidence); against firings of the opto-feedback circuits (LEDs); and against triggering of the guard detector. This guard consists of a pair of photomultiplier tubes coupled to plastic scintillators, which surround 85 percent of the solid angle around each detector. Events failing these tests were flagged accordingly and included in the event lists but not in the accumulations of spectra and light curves.
These anticoincidence techniques were extremely successful in eliminating background; the best example of how effective they were was our ability to observe the bright X-ray pulsar Cen X-3 through the South Atlantic Anomaly with a loss of only 20% of the photons due to anticoincidence flagging. No X-ray detector previously flown has ever been able to observe sources during SAA passage.

* Detector Alignment

Using the beam shape described above, we compared data from observations of Cygnus X-1, Cyg X-2, Cyg X-3 and X0614+09. A number of observations were used in order to ensure that a consistent result was found. Ratios of counts in the five pixels of each detector were compared with those predicted using a series of Monte Carlo simulations of the beam in the focal plane, for various off-axis angles and azimuths, to determine the appropriate beam spread and off-axis position. The bottom line is that there is a mean displacement of 0.822 arcminutes (49 arcsec) between A and B, and an azimuth of -63.24 degrees. The azimuth of the displacement is measured from the A3/A4 boundary, and corresponds to a rotation of 63 degrees through B4 towards B1. So, for a source on-axis in A, the distribution of counts in the outer pixels of B should be maximized in B4, with a slight excess in B1. Similarly, for a source on-axis in B, the counts in the outer pixels of A should be maximized in A2, with a slight excess in A3.

* The Off-axis angle; Matrices and Aspect Solution

Observations performed off-axis suffer from a measure of vignetting by the mirror. Because of this effect, it is important to use the response matrix constructed for the off-axis angle relevant to the observation being studied. Response matrices covering the full range of off-axis angles are provided in the BBXRT archive, but it should be noted that although the energy vignetting is handled correctly in these matrices, no absolute flux correction for losses due to the mask have been applied.
We have two types of pointing information for the BBXRT observations. The attitude files (part of the 1993 February release) contain pointing information for the entire mission; a right ascension, declination and roll angle every second, in 1950 coordinates, including slewing intervals. This pointing information was constructed by adding the TAPS (Two-Axis Pointing System) gimbal angles to the position of the Orbiter -Z-axis. This is only accurate to maybe 10 or 20 arcminutes, which makes it useful as a guide to the start and stop times of observations and as a sanity check for what area of the sky we were looking at, but not directly useful as an absolute position reference for the calculation of the off-axis angle. A more accurate position can be determined from the aspect camera solutions.
We have aspect solutions for more than 60% of our observations, and the mean aspect solution for each pointing is contained in the observation catalogues accessible via BROWSE or anonymous FTP. For many observations, no aspect solution exists. The principal reason for this is that the aspect camera was turned off during orbit day, because the illumination was too great for it, and during the SAA.
In addition, there were several problems in the first three days with the processing electronics associated with the aspect camera. These were eventually eliminated by modifying the settings of a few onboard processing parameters. There were also intervals when the aspect camera was flooded with reflected light or Earthglow; sometimes the star field observed was too sparse, and too few stars were available for an aspect solution, and for some LMC pointings the field was too rich. Finally, there are of course no solutions for times when we were Earthblocked or slewing.
The aspect camera boresight was determined from observations of the Crab, Cygnus X-1, Cyg X-2, Cyg X-3 and X0614+09. A couple of iterations were performed before settling on the final boresight on which the aspect solutions contained in the database are calculated. For some bright sources, the off-axis angle of the observation can be determined using ray-tracing; these angles are also included in the released database. Where no formal off-axis angle can be determined, the relative count rates in each pixel can be used to make an inspired guess. Off-axis vignetting corrections (i.e., a given off-axis response) are not sensitive to a 1 to 3 arcminute error in the off-axis angle, depending on source count rate.

* Microphonics

From examination of data taken when the covers were closed on the BBXRT instrument, we found evidence for bursts of spurious events that are probably connected with microphonics. Typically such a burst lasts less than two milliseconds, and contains up to five events from within a given detector (usually A) but from several pixels (more often than not A4, A1 and A0). The microphonic events form a major contribution to the low-energy background for very-low-countrate sources. For brighter sources, the contribution from microphonics is negligible.
A routine was written to remove such events during the accumulation of background or low-countrate sources. For observations where the total countrate (including events rejected by anticoincidence flags) is below 5 counts per second, the accumulated spectra in the database were constructed using this tool, to ensure no pollution by microphonics. The background files were also created using this routine.

* Timing information

Table 3 shows the launch and landing times for BBXRT in a range of different systems:

Table 3

There are several clocks on board the shuttle that are relevant to this discussion. BBXRT has its own internal clock, and so do TAPS and the Orbiter. Of the three, the Orbiter clock is the most reliable; every 60.000000 seconds, this clock emits a so- called MIN pulse. The MET time of the last-received MIN pulse is embedded in the BBXRT raw telemetry. Examining the MET and MIN times throughout the mission, it was found that
1 BBXRT minute = 59.99651 +- 0.00001 seconds.
There is no evidence of a day-to-day variation in this value. Thus, the BBXRT clock loses 5.026 seconds per day, or about 45 seconds over the whole mission. This estimate agrees well with a pre-launch estimate of 5.04 seconds/day derived using a 0.73-day calibration run.
We also have determined that the TAPS clock runs 9 seconds behind the Orbiter clock. All derived pointing information includes this correction.
As the the BBXRT observations were so short and BBXRT's strength is spectroscopy rather than timing analysis, no barycentric corrections have been applied to the data.

* Discriminator Thresholds

For the 1992 September archival release all the channels have been written to the spectral files, however it should be noticed that the first few channels are below the discriminator cutoff and always have zero counts. (The number of affected channels varies with detector and with the discriminator setting at the time of the observation.) Also, channel 512 is the overflow channel and should be ignored in spectral fits. Future versions of the PHA files may include a QUALITY column to indicate the 'bad' channels.
As a rule of thumb, the central elements had an effective low energy threshold of around 300 eV. For the outer elements, this threshold was closer to 500 eV. The central pixels can go lower because of their intrinsically lower noise and better shielding, surrounded as they are on all sides by active detector elements.

* Deadtime

Deadtime is negligible for all but the brightest sources we observed, namely the Crab, Cygnus X-2 and (marginally) Cygnus X-3. It is a steep function of countrate; for the on-axis observation of the Crab (total count rate including flagged events approx 2000 cts/sec) the deadtime is about 35% in total, while it is less than 10% for the on-axis Cyg X-2 observation (1150 cts/sec) and less than 5% for Cyg X-3 (278 cts/sec). All other BBXRT observations have fewer than 200 cts/sec.
The following is a summary of our current knowledge of possible BBXRT deadtime effects.
There are four reasons why X-rays incident on the BBXRT detectors might not make it into the telemetry stream:
  • The telemetry can only hold 252 events per frame for a total rate of 2016 events per second. To first order, once the telemetry buffer is full additional events that occur before the start of the next frame will be lost. (Actually, the data from each pixel is buffered, so that one event from each of the ten pixels can be saved after the telemetry buffer is filled. These saved events are then put into the telemetry buffer at the start of the next frame.) Note that if the event rates for the pixels are different, the dead time will be different for each pixel; pixels with large rates have larger deadtimes.) For the Crab (start MET: 3.055), a plot of event time versus event number shows glitches every 0.125 seconds, corresponding to the filling of the telemetry buffer. The glitches have an average dead time of 14.5 msec or about 12% of the frame duration.
  • Events may be flagged with one or more of the five quality flags (LED, guard, large event, pixel-to-pixel, or pulse pile-up). BBXRT could be commanded not to put these events into the telemetry, although we never did this. Spectral data are normally selected to exclude events with any of these flags set (the archive spectra and lightcurves were accumulated using only unflagged events).
  • Pulse pile-up can also prevent events from getting into the telemetry. If an event occurs within an interval of about 100 microseconds of a previous event in the same pixel, the earlier event is flagged and the second event is not analyzed. The interval was chosen so that unflagged events are completely unaffected by subsequent events. Many of the flagged events will only be minimally affected. The exact length of the interval varies by about 10% from pixel to pixel. If a third event occurs more than 100 microsec after the first event but less than 100 microsec after the second, the third event will be analyzed and have the pile-up flag set. For the Crab data referred to above, 13% of events are flagged due to pulse pile-up, contributing to a flagging ratio of about 20% overall.
  • Events in the same pixel that may occur within 10-20 microsec may not be sensed as different events in the BBXRT electronics. The pulse heights will then pile up in a nearly linear fashion. For large rates this can affect the resulting spectrum. To be strict, the convolution of the spectrum with itself times the fraction of such events should be subtracted from the spectrum. In practice, considering the count rates we encountered, we consider this to be unnecessary.

* Co-adding pixels

When analyzing data for which a significant signal was detected in several pixels (i.e. almost every observation) it may be tempting to combine data from different pixels into a single spectrum. However, this is strongly discouraged; the resolution of the detectors is so good and the gains are sufficiently different that smearing of spectral features and distortion of continuum information can result from such an addition. The best thing to do (and the technique which is almost universally adopted in BBXRT papers originating from the LHEA) is to fit the data from the brightest pixels simultaneously, with chained values for continuum parameters such as photon index, temperature etc., and separate normalizations.
It should go without saying that for an extended source, one should be cautious about even this approach to spectral fitting and carefully treat the pixels separately before attempting to combine data in a joint fit.

* The Detector Response

The BBXRT detector spectral response matrices have been painstakingly constructed based on data from pre-launch, in-flight and post-flight calibrations. Before launch, measurements of the effective area of the mirrors and the detector efficiencies were made at GSFC. In flight, spectral calibrations were performed by moving an internal Fe 55 source into the field of view and observing the Mn Ka and Kb lines. The Crab Nebula was observed on several occasions.
More detailed modeling of partial charge in the detectors, and adjustments for the gold M edge in the mirrors and Al and Si edges in the detectors, have been performed since the flight of BBXRT, using further ground calibration data obtained at GSFC and the National Institute for Standards and Technology in 1991 and 1992. Using the most recent response matrix, the systematic residuals from a power law fit to data from the Crab Nebula show no features larger than 5% between 0.5-8 keV and 10% between 8-12 keV, and no narrow features in the spectrum with an equivalent width >12 eV. A 10% residual uncertainty still persists at ~0.4-0.5 keV. A full description of the BBXRT calibration process can be found in Weaver et. al, 1994. ApJSupp (in press).
[About BBXRT] [Archive] [Software] [Gallery] [Publications]