The IRAC detector pixels are limited in the number of photons (actually, electrons) they can accurately accumulate and detect. Once this maximum number is reached, the detector pixel is “saturated,” and additional photons will not result in an increase in read-out data numbers. Prior to this, the detector becomes effectively less sensitive as more photons are received, an effect referred to as “non-linearity.”
The saturation value varies slightly pixel-to-pixel, and substantially from detector to detector. The IRAC InSb (3.6 and 4.5 µm) detectors typically have saturation values of approximately 44000 DN in the raw data (30000 DN for warm mission data). The Si:As detectors (5.8 and 8.0 µm) have saturation values closer to 55000 DN. The IRAC pipeline automatically detects pixels that exceed a pre-defined threshold, and marks them in the data mask. IRAC uses a Fowler-sampling scheme where the returned DN are the difference between a set of readouts at the end of the integration (signal reads) and a set at the beginning (pedestal reads). Thus, once a pixel has saturated the signal reads, the DN for that saturated pixel will actually start to decrease, and as a result of this double-valued nature the DN value alone is not a reliable saturation indicator. Examining the images containing very bright sources is necessary in order to evaluate saturation based on the observed spatial structure of the source. Very bright sources, for example, will appear to plateau or even develop a dark hole in the center. For point sources, a rough estimate of the flux in the saturated pixels can be made by fitting the wings of the PSF to the linearized pixels in the BCD image. If the data were taken in the high dynamic range mode, the IRAC pipeline will automatically identify pixels in the long frame times that are saturated based on the observed flux in the short frame times. The short frame time data can then be used to recover saturation in the long frame time data (this is not done automatically). This replacement is accurate to about 10% at the peak of bright sources as the ≈ 0.1 arcsecond jitter of the telescope coupled with pixel phasing in channels 1 and 2 and charge diffusion in all channels will cause the measured flux densities between short and long frames to vary.
The IRAC arrays are slightly nonlinear at all signal levels. At levels above 30000 DN (in the Level 0 raw data) the response is low by several percent. As part of pipeline processing, the data are linearized based on ground calibrations (which have been verified in flight) of this effect. The BCD data are linear to better than 1% up to about 90% of full well, which is defined to be the level where we no longer can fully linearize the data, and at which saturation, by definition, begins. Below 20% of full well the nonlinearity in the raw data is negligible.
In detail, there are four places in the electronics where a pixel may saturate: the detector diode, the unit cell source-follower in the Read-Out Integrated Circuit (ROIC), the output source-follower in the ROIC, and the analog-to-digital converter (ADC) in the warm electronics. In most cases, it is the ADC that saturates first, at 0 or 65535 units. ADC saturation produces a discontinuity in the second derivative of the measured Fowler DN versus the flux. The other saturations are smooth, with no discontinuity. In the other cases, depending on the channel, the detector diode may saturate before or after the source-followers.
In principle, for any source for which we already know the spatial variation of its intrinsic surface brightness, we can determine whether the pixel is above or below saturation, and therefore, its flux. In practice, we do not know the gains of the source-followers very well near saturation, nor do we know enough about the detector diode saturation, to make a good estimate of the flux. Therefore, we flag pixels which are above the range of our linearization correction (see Sections 5.1.13 and 5.1.21 for more information on how the pipeline handles linearity and saturation).
The users should note that the pipeline does a saturation correction for bright point sources (see Section 5.2.2). In some cases of a bright and variable background, such as the center of a globular cluster, the pipeline may be fooled to insert an artificial “star” into the CBCD image. Users are encouraged to carefully examine the CBCD images in such cases and use the BCD images if any new artificial-looking features are found in the CBCD images. An example of such an erroneous saturation correction star substitution is shown in Figure 7.2. For point sources, the saturation corrected photometry should be good to roughly 10% accuracy.
Figure 7.2: Erroneously inserted saturation correction stars in a CBCD image (left), compared to the original BCD image (right). These data are from PID=46, AORKEY 12484352. The arrows point to the sources erroneously inserted by the pipeline. There are column pull-down and banding correction artifacts in the CBCD image as well, inserted by overactive pipeline corrections.
7.2.2 Muxbleed (InSb)
Multiplexer bleed, or “muxbleed,” (see also Section 5.1.11) appears in IRAC channels 1 and 2 (3.6 and 4.5 μm). It looks like a decaying trail of pixels, repeating every fourth column, with an enhanced output level trailing a bright spot on the same array row. The effect can wrap around to subsequent rows, but it does not wrap from the last row to the first. Since columns are read simultaneously in groups of four, one for each mux output, the next pixel read out on any single output is four pixels to the right, in array coordinates. As the BCDs for channels 1 and 2 are flipped in the y-direction when compared to the raw images, the read direction is top to bottom for these BCDs, and muxbleed-triggering pixels will affect rows beneath the source. Muxbleed is usually accompanied by a pinstripe pattern (“muxstripe;” every fourth column) that may extend over part of the image preceding or following the pixel. It is caused by a slow relaxation of the mux following the momentary disequilibrium induced when a bright pixel's voltage is placed on an output FET during pedestal and signal reads. Although the pixel rise and fall times are fast (2.6 and 1.0 microseconds, respectively) compared to the 10 microsecond time to clock the next pixel onto an output, longer relaxation times are involved for an output FET to fully recover after the voltage from a bright pixel is briefly impressed on its gate. The decaying trail has a time constant of tens of microseconds, and the pinstripe, tens of seconds.
Stars, hot pixels, and particle hits can generate muxbleed, and the characteristics of the pinstripe depend on the frame time and the Fowler number. Hot pixels may show muxbleed in a raw image, but in the BCD the muxbleed induced by hot pixels may not be present because it was canceled in either the labdark subtraction or in the skydark subtraction. The pinstripe pattern is nearly constant in areas of a single image that do not contain a saturating star, particle hit, or hot pixel. The characteristics of muxbleed from particle hits depend on when the hit occurs within the frame.
Muxbleed was characterized long before the launch of Spitzer, and it is reasonably well understood, and it is fully corrected in the final IRAC pipeline. The pinstripe is strongest in channel 2, particularly in 12 second frames. In channel 2 mosaics, even with overlap correction, there may appear to be bright and dark patches everywhere, about the size of one frame or part of a frame. Upon close inspection, though, individual patches are revealed as areas of a nearly constant pinstripe pattern that runs between the edges of the array, bright stars, hot pixels, and particle hits. A systematic and automated pinstripe correction scheme has been implemented in the pipeline.
Figure 7.3: Images showing the muxbleed effect (the horizontal line on both sides of a bright stellar image). The pixels on the left side of the bright source are pixels on rows following the row in which the bright source was located (and have wrapped around in the readout order of the array). The vertical (white) lines are due to the so-called “column pull-down” effect. These are 12-second BCD frames in IRAC channel 1, taken from IRAC PID=618, AORKEY 6880000. Pinstriping is also seen in these images.
The amplitude of the effect decays as one moves away from the bright spot, and this decrease can be nicely described by a simple function. In general, muxbleed decays rapidly within 5 – 10 reads and plateaus at a roughly constant value. The functional form of muxbleed is frame time independent. However, the amplitude does not scale linearly with the flux at the brightest pixel or the integrated flux of the triggering source, and this often leaves over/undercorrected muxbleed in BCD frames. For this reason, an additional muxbleed correction by fitting the functional form of the muxbleed pattern to the actual muxbleed incidence is performed after the BCD frame creation (i.e., CBCD frames), and this corrected muxbleed below the rms noise level of the image.
Figure 7.4: Demonstration of the S18 pipeline muxbleed removal. The image on the left is before and the one on the right is after the correction. These are First Look Survey channel 1 data, taken from AORKEY 4958976. Note that the brightest star in the upper-left corner is heavily saturated and the current muxbleed scheme can correct muxbleed from a saturated source also.
Figure 7.5: A typical bandwidth effect trail in channel 4, in a 30 second frame. These data were taken from PID=1154, AORKEY 13078016.
An example of muxbleed correction is shown in Figure 7.4 (see also Figure 5.8, and see Section 5.1.11 for information on muxbleed correction in the pipeline). It can be seen that at least cosmetically the effect can be greatly reduced without introducing new artifacts. With an additional correction to residual muxbleed during the CBCD pipeline, resultant images should be nearly muxbleed-free.