The COS FUV detector is a windowless cross delay line (XDL) device that is similar to the detectors used on the Far Ultraviolet Spectroscopic Explorer (FUSE)
. The XDL is a photon-counting micro-channel plate (MCP) detector with two independently-operable segments (FUVA and FUVB). Each segment has an active area of 85 ×
10 mm; they are placed end to end and separated by a 9-mm gap. When the locations of detected photons are digitized they are placed into a pair of arrays (one per detector), each 16,384 ×
1024 pixels, though the active area of the detector is considerably smaller. Individual pixels span 6 ×
m. The long dimension of the array is in the direction of dispersion; increasing pixel number (the detector’s x
axis in user coordinates) corresponds to increasing wavelength. The XDL format is shown schematically in Figure 4.1
. Detector parameters are listed in Table 1.2
The XDL’s quantum efficiency is improved by the presence of a series of wires, called the quantum-efficiency (QE) grid, placed above the detector (i.e.,
in the light path). These wires create shadows in the spectrum that are flagged and corrected by calcos
during data reduction. The XDL also includes an ion-repeller grid that reduces the background rate by preventing low-energy thermal ions from entering the open-faced detector. It acts as a 95% transmission neutral-density filter.
COS is considerably more sensitive than STIS and earlier-generation HST
instruments at comparable spectral resolutions in the FUV. Effective areas for the COS FUV gratings are shown in Figure 5.1
. The maximum count rates for the FUV detector are listed in Table 10.1
. The time-dependent changes in the sensitivity of the COS FUV channel are discussed in Section 5.1.5
The XDL detector segments have extremely low intrinsic dark rates, on the order of 10−6
count/s/pixel; see Section 7.4.1
. Background counts can also be caused be external events, such as proximity to the South Atlantic Anomaly (SAA). Calcos
estimates the dark rate by measuring the counts in an unilluminated region on the detector and subtracting this from the spectrum during processing. Each segment has a distinct dark current that varies with time and may be correlated with the Solar Cycle (see Figure 4.2). The dark rate, in particular for FUVA, has been decreasing recently, which, in concert with improved algorithms in calcos to filter out background counts as a function of pulse height, leads to substantially smaller dark rates. The dark rates are evolving with time, so observers, particular those with faint, background-limited targets, should consider how the changing dark rates may affect their orbital estimates. When planning observations with the G130M/1055 and G130M/1096 modes users should consult http://www.stsci.edu/hst/cos/software/planning/etc/.
The electronics that control the COS detectors have a finite response time t
, called the dead time, that limits the rate at which photon events can be processed. If two photons arrive within time t
the second photon will not be processed. For the FUV channel three factors limit the detected count rate. The first is the Fast Event Counter (FEC) for each segment, which has a dead time of 300 ns. The FEC dead time matters only at count rates well above what is usable, introducing a 1% error at a count rate of 33,500 per segment per second.
is the dead-time constant. For the COS FUV detector t
= 7.4 μ
s, so the apparent count rate deviates from the true count rate by 1% when C
= 1350 counts/s and by 10% when C
= 15,000 counts/s. Note that, when the effect is near the 10% level, then the FUV detector is near its global count-rate limit (see Table 10.1
), so non-linear effects are relatively small.
Finally, the Detector Interface Board (DIB) combines the count streams from the two FUV segments and writes them to a single data buffer. The DIB is limited to processing about 250,000 count/s in ACCUM
mode and only 30,000 count/s in TIME-TAG
mode (the highest rate allowed for TIME-TAG
mode). The DIB interrogates the A and B segments alternately; because of this a count rate that is high in one segment, but not the other, could cause a loss of data from both segments. Tests have shown that the DIB is lossless up to a combined count rate for both segments of 20,000 count/s; the loss is 100 count/s at a rate of 40,000 count/s. Thus, this effect is less than 0.3% at the highest allowable rates. Furthermore, information in the engineering data characterizes this effect.
Corrections for dead-time effects are made in the calcos
pipeline, but they are not included in the ETC
, which will over-predict count rates for bright targets.
Four stim-pulse rates are used: 0 (i.e., off), 2, 30, and 2000 Hz per segment. These rates, which are only approximate, are not
user selectable. Exposures longer than 100 s will use the 2 Hz rate, those between 10 and 100 s use 30 Hz, and those shorter than 10 s use 2000 Hz.
An ultraviolet photon incident on the front MCP of the XDL detector creates a shower of electrons, from which the detector electronics calculate the x
coordinates and the total charge, or pulse height. The number of electrons created by each input photon, or “gain” of the MCPs, depends on the high voltage across the MCPs, the local properties of the MCPs at that location, and the high voltage across the plates. It is not a measure of the energy of the incident photon
. A histogram of pulse heights for multiple events is called a pulse-height distribution (PHD).
Prolonged exposure to light causes the number of electrons per incident photon to decrease, a phenomenon known as “gain sag.” As a result, the peak of the PHD in each region of the detector shifts to lower values as the total (time-integrated) illumination of that region increases. As long as all pulse heights are above the minimum threshold imposed by calcos
there is no loss in sensitivity. However, if the pulse height drops low enough that the pulse heights of the photon events from the target fall below the threshold these events are discarded and the throughput decreases. The amount of gain sag increases with the total amount of previous illumination at that position on the detector, so gain sag appears first in regions of the detector that are illuminated by bright airglow lines, but eventually affects the entire spectrum. Figure 4.4
shows the effect of gain sag on Segment B of the COS FUV detector. These data were obtained using the grating setting G160M/1623/FP-POS=4
. A portion of the extracted spectrum from the same object taken at five different times is shown. The blue curve was constructed using only photon events with pulse heights in the range 4–30 (the limits used by calcos
until December 2010), while the red curve includes pulse heights of 2–30 (the present limits). Two regions that suffer the most serious gain sag are marked: the region near pixel 7200 is illuminated by Lyman-α
when grating setting G130M/1309/FP-POS=3
is used, and the region near pixel 9100 is illuminated by Lyman-α
when the setting is G130M/1291/FP-POS=3
. Initially, the pulse heights were well above either threshold, so the blue and red curves are indistinguishable. As time progressed all of the pulse heights decreased. However, the two Lyman-α
regions decreased faster, causing the blue spectra to exhibit spurious absorption features. This trend continued until the Segment B high voltage was raised in March 2011. The bottom plot shows that increasing the voltage has recovered some of the lost gain, but not all of it.
As the pulse height of a photon event decreases the detector electronics begin to systematically miscalculate its position. On the COS FUV detector this effect, called detector walk, occurs in both x
, but is much larger in the y
(cross-dispersion) direction. The shift is approximately 0.5 pixel per pulse-height bin, which means that the entire spectrum may be shifted by several pixels, and the regions with the lowest gain may be noticeably shifted relative to the rest of the spectrum (Figure 4.5
applies a y
-walk correction to TIME-TAG
data, but the effect remains uncorrected in ACCUM
mode, where no pulse-height information is available. The spectral extraction should remain unaffected, because the extraction regions are large enough to include the misplaced counts.
A range of strategies have been used to minimize the effects of gain sag and detector walk on the science data. Several of these are modifications to calcos
, which means that previously collected data can be improved by reprocessing. Others involve changes to the on-orbit settings, and thus only affect exposures taken after the changes are made.
Pulse Height Thresholdin
g: At present, the lower pulse height threshold used by calcos
when processing TIME-TAG
data has been decreased to near its minimum value. This ensures that as few events as possible are lost due to low gain, but it may have the effect of slightly increasing the detector background. We hope to eventually implement time- and position-dependent pulse height thresholds in calcos
applies a walk correction to TIME-TAG
events. The pulse height and measured position of an event are used to apply a correction factor to its position. Although the walk correction is not time-dependent, it may be modified as we learn more about the walk properties of the detector.
Gain Sag Table
: Low-gain pixels are flagged by calcos
and excluded when combining spectra taken at multiple FP-POS
: In an effort to keep the MCP gain in the spectral regions within the range that gives acceptable position determination, while simultaneously minimizing gain sag, the high voltage on each segment has been adjusted numerous times since launch. The voltages used for a particular exposure can be found in the file headers, but the effects should be transparent to the user, since any effects on the data will be handled by calcos
. More details on the high voltage changes are given in the COS Data Handbook
Change in Lifetime Position
: After years of collecting exposures—and thousands of counts per pixel in the most exposed areas—the gain at certain areas on the MCPs drops so much that none of the techniques described above can return the detector to an acceptable level of performance. Once that occurs the only way to obtain satisfactory data is to move the spectra to a new location on the detector, which can be accomplished by adjusting the position of the aperture and the pointing of HST
. This moves the spectra to a pristine region on the detector, called the “second lifetime position”. Because the optical path is slightly different for each lifetime position the properties of the spectrograph are also slightly different. Thus, the resolving power, throughput, flat field, etc. may differ at different lifetime positions. A keyword in the header of the data files tells calcos
which lifetime position was used, and reference files appropriate to that position are applied when processing the data.
The dark rate varies spatially over the FUV detector. Figure 4.6
shows the sum of approximately 80,000 s of dark exposures taken over a five-month period in 2011. With the standard lower pulse-height threshold of 2, Segment A is relatively featureless away from the edges of the active area, except for a few small spots with a higher rate. Segment B shows several large regions with a slightly elevated rate; they are enhanced by less than a factor of two over the quieter regions.
For most TIME-TAG
observations these features will have a negligible effect on the extracted spectra, because the variation is small and the overall rate is low (see Table 7.1
). In ACCUM
mode, where no lower pulse-height threshold is used, additional features appear. ACCUM
mode is used only for bright targets, so these features should constitute a negligible fraction of the total counts.